Soil samples were collected from seven sites, which were located in the Inner Mongolia Autonomous Region (IM), Hebei Province (HB), Henan Province (HN), Jiangsu Province (JS), Jiangxi Province (JX), Fujian Province (FJ), and Guangxi Zhuang Autonomous Region (GX), respectively. The location information of the sampling sites is shown in Supplementary Table S1. Each site was planted with one dominant vegetation type. In each site, surface soil samples (0–20 cm) from three randomly selected plots (0.5 m × 0.5 m) were collected and mixed to represent a site, after removing the surface litter. After the soil samples were sieved and passed through a 2 mm diameter mesh, they were stored at 4°C for less than 2 weeks before a series of experiments, including Fe(III)-reducing bacteria enrichment, MFCs operation and soil property measurement.

Each of the seven soil samples was divided into three aliquots as replicates for MFCs operation. Before the operation, soil physiochemical properties of each aliquot were analyzed using routine methods (Page et al., 1982). Briefly, soil texture was determined by the sieve and pipette method. Soil maximum water holding capacity (MWHC) was determined by the difference between dry and soaked soil weights. Soil pH was measured at 1:2.5 (soil:water) and soil electrical conductivity (EC) at 1:5 (soil:water). Soil cation exchange capacity (CEC) was analyzed by the compulsive exchange method. Soil organic carbon (OC) was determined by K₂CrO₄ oxidation; total nitrogen (TN) by Kjeldahl digestion; and total phosphorus (TP) by colorimetry following NaOH digestion. Dissolved organic carbon (DOC; extracted by 0.5 M K₂SO₄) and humic carbon (HC; extracted by 0.1 M Na₄P₂O₇ and 0.1 M NaOH) were measured using a TOC analyzer (TOC-L, Shimadzu, Kyoto, Japan). Soil total dissolved iron (DFe_T) was extracted by 0.2 M H₂C₂O₄-(NH₄)₂C₂O₄ (McKeague and Day, 1966) and measured by flame atomic absorption spectroscopy (AA240, Agilent Technologies, Santa Clara, CA, USA). Soil microbial biomass carbon (MBC) was measured by the fumigation-extraction method (Vance et al., 1987).

A picture and schematic diagram of soil MFCs are shown in Supplementary Figure S1. Twenty-one air-cathode MFC reactors were built in beakers with a 6 cm diameter and 13.5 cm height. Square carbon felt (Haoshi, Lanzhou, China) and platinized carbon paper (Hesen, Shanghai, China) were used as the anode and cathode, respectively, with the same area of 9 cm² (side length 3 cm). In each reactor, the anode was embedded with 250 g soil (dry weight) and the cathode was placed on the soil surface. Deionized water was gently poured into the reactor to keep soil flooded. The two electrodes were connected to an external circuit with a resistance of 1000 Ω using titanium wire. The MFCs were operated in triplicate at a constant 30°C in an incubator. Voltage data generated by the MFCs were recorded every 10 min using a data acquisition module. Deionized water was added every 24 h to compensate for water evaporation and maintain the initial state. After 270 days of MFCs operation, flooded soil in each MFC was air dried under open circuit conditions at 30°C and sieved through a 2 mm diameter mesh to measure the soil properties using the same methods as detailed in the previous section. To confirm that the voltage originated from microbial processes rather than chemical reactions, another seven control MFCs were operated under the same conditions but with chloroform fumigation-sterilized soil (Deng et al., 2015; Jiang et al., 2015).

The enrichment of Fe(III)-reducing bacteria in the soil samples was conducted under anaerobic conditions (10% CO₂, 10% H₂, and 80% N₂) in an anaerobic workstation (MiniMacs, Don Whitley Scientific, Shipley, UK). For each soil sample, 2.0 g soil (dry weight) was inoculated into 100 mL basal medium. One liter of basal medium (BM) contained 8 g peptone, 1 g yeast extract, 0.12 g NH₄Cl, 16 g sodium acetate, 5 g NaCl, 1.2 g K₂HPO₄, 1 g cysteine hydrochloride, 1 mg resazurin, 5 mL mineral, and 5 mL vitamin solutions (Sigma-Aldrich, Co., St. Louis, MO, USA) (Lovley and Phillips, 1988). The electron acceptor was 25 mM ferric citrate. The medium was boiled with N₂ for 20 min and then autoclaved in sealed bottles. The pH of the autoclaved medium was 6.7. The inoculated medium was incubated at 30°C for 7 days under dark conditions. The enrichment procedure was repeated three times (Kim et al., 2005).

Five-milliliters of the enrichment product were collected after the anaerobic culture and centrifuged at 14,000 × g for 10 min. The genomic DNA was immediately extracted from the precipitates using a Fast DNA SPIN kit for soil (BIO101, MP Biomedicals, Carlsbad, CA, USA) following the manufacturer’s instructions. The purity and the quantity of the extracted DNA were determined using a nanodrop UV-Vis spectrophotometer (ND-1000, NanoDrop, Wilmington, DE, USA) at 230, 260, and 280 nm.

The bacterial 16S rRNA genes of the seven enrichments were amplified using universal primers 515F (5′-GTG CCA GCM GCC GCG GTA A-3′) and 907R (5′-CCG TCA ATT CMT TTR AGT TT-3′) (Weisburg et al., 1991). The PCR reactions were quantified and the products were then purified. At least 24,000 reads were conduct for each sample using the Illumina MiSeq platform (Illumina, San Diego, CA, USA) using 2 bp × 250 bp paired end flow cells and reagent cartridges. The Illumina sequencing data were analyzed by Mothur (Schloss et al., 2009) using the MiSeq standard operating procedure (Kozich et al., 2013). The raw data were deposited in the NCBI Sequence Read Archive database with the accession number SRP071622.

After the three rounds of 7 days anaerobic incubation, 0.5 ml of the enrichment products were separated by the spread plate method after 10-fold dilution for bacteria isolation using solid BM. The plates were incubated for 5 days under anaerobic condition, after which single black colonies were picked out and inoculated into liquid BM. The procedures were repeated three times to obtain pure cultures. All the experiments were conducted in the MiniMacs anaerobic workstation.

The phylogenetic analysis of the isolates was conducted based on the 16S rRNA gene sequences, which were PCR amplified using the primers 27F (5′-AGA GTT TGA TCM TGG CTC AG-3′) and 1492R (5′-GGT TAC CTT TGT TAC GAC TT-3′) (Suzuki and Giovannoni, 1996). The PCR products were cloned using the Peasy^TM-T3 Cloning Kit (TransGen, Beijing, China) according to the manufacturer’s recommendations. GeneScript (Nanjing, China) sequenced six clones for each isolate. The vector sequences were removed using DNASTAR Lasergene (version 7.1). The gene sequences of the isolates were subjected to taxonomic assignments using BlastX. The sequences of all the isolates were submitted to GenBank with the accession numbers KT889276–KT889290. One strain each of isolate 1 (CGMCC 1.5212) and isolate 2 (CGMCC 1.5211) were deposited in the China General Microbiological Culture Collection Center.

One of the strains phylogenetically related to each species was randomly selected and the exoelectrogenic activity of the strains was characterized by polarization curve measurement using H-type dual chamber MFC reactors (100-I, Fuxiao, Changshu, China). The anode and cathode were both rectangular carbon felt with the same area of 18 cm² (3 cm × 6 cm), connected to an external resistance of 1000 Ω using titanium wire. LB medium, which is favorable for power generation by exoelectrogenic bacteria, was used in the subsequent tests of the isolates (Feng et al., 2014). The anodic chamber and cathodic chamber were filled with 120 mL LB medium and 120 mL potassium ferricyanide [100 mM K₃Fe(CN)₆ in 50 mM, pH 7 PBS], respectively. The two chambers were separated by a cation exchange membrane (32S, Qianqiu, Hangzhou, China). The reactors and LB medium were autoclaved before use. Voltage data were recorded every 20 min using a data acquisition module.

The polarization curves of the MFC reactors with pure isolates were measured in the fed-batch mode (Lovley and Phillips, 1988). Briefly, the liquid medium containing isolated cells (200 μL) was inoculated into the anodic chamber. Eighty-milliliters of medium were replaced after the voltage of MFCs reached its peak. The reactors were then operated under open circuit conditions for 4 h and with different loads (50,000, 10,000, 5,000, 1,000, 800, 500, 300, 200, and 100 Ω) for 20 min for each load. All the MFCs were operated at a constant 30°C in an incubator.

The charge generated from the MFCs, defined as the quantity of the generated electrons, was calculated as previously described (Deng et al., 2015). Current density was calculated from the external load and cell voltage according to Ohm’s law (I = U/R) and normalized to the surface area (m²) of the cathodic electrode. The power density was calculated by using P (mW m^-2) = 10 × U²/(R × A), where U (mV) is the recorded voltage, A (m²) is the surface area of the anode, and R (Ω) is the external load (Cheng et al., 2006).

Significant differences between means were determined by one-way ANOVA at a level of P < 0.05, using the least significance difference (LSD) test. Cluster analysis of the Fe(III)-reducing bacteria community was conducted based on the square Euclidean distance by the between-groups linkage method. Redundancy analysis (RDA) was carried out by Canoco for Windows (version 4.5) between the Fe(III)-reducing bacteria community and environmental or electrical variables, which were selected using the Monte Carlo permutations test (499 permutations). All statistical tests, except for RDA, were performed using SPSS software (version 18.0).

The electricity generation by MFCs comprising the seven soils lasted for 270 days (Figure 1A). The voltage curves of the MFCs were characterized by a single peak. The seven soils varied in their peak voltage and also the time from the beginning to the peak. It took about 40 days for JS, 60 days for both FJ and GX, and 100 days for JX to reach to their peak voltages. IM, HB and HN reached their peaks at round 130 days. The peak voltages of JS, GX and JX were 148.1, 123.3, and 102.6 mV, respectively. They were significantly higher (P < 0.05) than FJ (32.9 mV), HN (8.4 mV), HB (7.4 mV), and IM (6.2 mV) (Figure 1B). After the peak data, the voltage continued to decrease until the end of MFCs operation, when the voltage data of all soils were below 30 mV. During the 270 days of MFCs operation, JX generated the highest charge (1052.4 C), followed by GX (842.2 C), JS (644.7 C), FJ (302.2 C), HN (66.4 C), HB (63.5 C), and IM (41.9 C) (Figure 1C). No voltages were detected in any MFCs containing sterilized soil samples, indicating that the electricity generation originated from microbial catalysis rather than chemical reactions.

The soil physicochemical and microbial properties varied between soils (Table 1). Before MFCs operation, the CEC, OC, TN, and HC values in soils from JS and JX were significantly higher (P < 0.05) than the other soils. JS exhibited the highest MBC, followed by JX and GX. JX also had the highest DFe_T. Soils from IM, HB and HN exhibited higher soil pH (>8) but lower OC, DOC, HC, and MBC compared with the other soils. Compared with soil physiochemical and microbial properties before MFCs operation, after 270 days MFCs operation, the OC and DOC of IM, HB and HN did not change significantly; however, those of JS, JX, FJ, and GX decreased significantly (P < 0.05). The MBC of all the seven soils decreased significantly after MFCs operation.

The sequences of Fe(III)-reducing bacteria in the seven soils were assigned to the known phyla, class and family. Four phyla were observed, with Firmicutes being the overwhelmingly dominant phylum (Supplementary Figure S2A). Six classes were detected, including α-, β- and δ-Proteobacteria, Clostridia, Bacilli, and Actinobacteria (Supplementary Figure S2B). The most abundant class was Clostridia, which accounted for over 90% of the total composition in the seven soils. At the family level, the composition of Fe(III)-reducing bacteria was different between soils (Figure 2). In IM, HB and HN, Clostridiales Family XI. incertae sedis accounted for 30.5, 26.0, and 18.3% of total composition, respectively. However, it was a minor bacterial group in FJ (2.6%), GX (0.3%), JS (<0.1%), and JX (<0.1%). By contrast, the relative abundances of Lachnospiraceae in IM (7.9%), HB (3.7%), and HN (7.4%) were lower than those in JS (14.7%), JX (25.7%), and GX (29.9%). GX was dominated by Clostridiaceae (57.5%) and Lachnospiraceae (31.4%). The relative abundance of Peptostreptococcaceae in GX was only 0.1%, which was much lower than that of the other six soils (16.5∼38.5%). Eubacteriaceae and Ocillospiraceae were detected in FJ, although their relative abundances were less than 0.1%. Cluster analysis of the Fe(III)-reducing bacterial community at the family level revealed that IM, HB, and HN were grouped in one cluster, while JS, JX, and FJ were in another cluster. GX was not grouped with either cluster (Figure 2).

Redundancy analysis showed that the first two components (RDA1 and RDA2) together explained 65.5% of the total variation of the Fe(III)-reducing bacteria (Figure 3). Along RDA1, IM, HB, and HN were separated from FJ, JS and JX, and GX was separated from the two groups. RDA2 mainly separated the group comprising FJ, JS and JX from the other four soils. The relative abundances of the families Clostridiales Family XI. incertae sedis, Ruminococcaceae, Bacillaceae, Dehalobacteriaceae, Symbiobacteriaceae, and Syntrophomonadaceae were positively related to soil pH and were increased in IM, HB, and HN. The relative abundances of the families Peptostreptococcaceae, Hyphomicrobiaceae, Burkholderiaceae, and Veillonellaceae were positively related to soil OC and were increased in JS, JX and FJ. The families Clostridiaceae and Planococcaceae were more abundant in GX, and their relative abundances were positively correlated with soil DOC. Peak voltage and charge, which increased along with RDA1 and RDA2, were positively correlated with the families Hyphomicrobiaceae, Burkholderiaceae, Paenibacillaceae, Lachnospiraceae, Planococcaceae, and Clostridiaceae.

Culturable strains isolated from the seven soils were phylogenetically related to 15 different species, of which 11 were Clostridium spp. (Table 2). Isolate 12, belonging to the Lachnospiraceae, was isolated from JS. Isolates 13–15, belonging to Ruminococcaceae, Eubacteriaceae, and Ocillospiracea, respectively, were isolated from FJ. Polarization and power density curves of the 15 isolates are shown in Figure 4. The MFCs catalyzed by the 15 isolates showed an open circuit voltage ranging from 400 to 630 mV. The cell voltage decreased, while the electrical current increased, with decreasing external load. The voltage drops of the polarization curves showed activation losses, ohmic losses and concentration losses. The power density curves of the 15 isolates peaked at the range of 16.4∼28.6 mW m^-2. Isolate 6, which was related to both Clostridium amylolyticum and Clostridium mesophilum, exhibited the highest P_max (28.6 mW m^-2) of all the isolates.