First genomic insights into members of a candidate bacterial phylum responsible for wastewater bulking

Samples

Methanogenic sludge samples reported in a previous study (Soo et al., 2014) were used in the present study for shotgun sequencing and fluorescence in situ hybridization (FISH). Briefly, two sludge samples (A1 and A2) were taken from the system at different sampling dates (A1, 25th December, 2012; A2, 16th September, 2010), and sample A1 was further separated into two parts (flocculant sludge [F1] and granular sludges [G1]) by gravimetric settlement (Soo et al., 2014) (Table S1). Each sample had been divided into two parts: one used for obtaining DNA via bead-beating and phenol chloroform extraction and the other fixed in 4% paraformaldehyde for FISH (Soo et al., 2014).

Fluorescence in situ hybridization

KSB3-specific FISH probes were designed in ARB v5.5 (Ludwig et al., 2004) using 16S rRNA genes identified in the KSB3 genomes and KSB3 16S rRNA gene sequences available in the current Greengenes database (May 2013 version) (McDonald et al., 2012). In order to maximize the specificity and fluorescence intensity of the probes, helper probes were also designed (Table S5). FISH was performed as described previously (Sekiguchi et al., 1999) using the probes listed in Table S5 and all probes were hybridized overnight. Probes were labeled with either Alexa488 or Cy3 fluorophores, and probes with different fluorophores were used together for dual-staining FISH. Images were captured using an epifluorescence microscope (Axioplan 2; Carl Zeiss) equipped with a cooled charge-coupled device (CCD) camera (DP72; Olympus) and subsequently processed using imaging software (DP2-BSW, version 2.2; Olympus). Super-imposed images were generated using Adobe Photoshop (CS5.1; Adobe).

Metagenome sequencing

Previously sequenced paired-end and mate-pair metagenomes for samples A1, A2, F1, and G1 (Soo et al., 2014) were supplemented with additional data generated in this study using the same DNA. Paired-end Nextera libraries were prepared for each sample according to the manufacturer’s instructions, quantified using the QuantIT kit (Molecular Probes) and sequenced (2 × 250 bp paired end) on an Illumina MiSeq using the Reagent Kit v2 (Illumina) at the National Institute of Advanced Industrial Science and Technology, Japan (AIST; Table S1). The extra metagenome sequencing generated 7.3 Gb, 13.1 Gb, 1.9 Gbp, and 7.9 Gbp for A1, A2, F1, and G1 samples, respectively (yielding a total of 56.2 Gb for all samples combined). For scaffolding, two additional large-insert mate-pair libraries (∼3 kbp and ∼7.5 kbp) were constructed from sample A1 using the Mate Pair Library Preparation Kit v2 (Illumina) and sequenced on Illumina MiSeq system (MiSeq Reagent Kit v2) yielding 2.3 Gb and 3.0 Gb, respectively (for a total of 7.3 Gb from sample A1 when combined with mate-pair data; Soo et al., 2014).

Community profiling

16S rRNA gene amplicon sequencing of all UASB sludge samples using the Illumina MiSeq system was previously reported (Soo et al., 2014). Community composition was also examined by extracting all 16S rRNA reads from the metagenome datasets using the closed reference picking script in QIIME v1.6.0 (Caporaso et al., 2010b) with the Greengenes database (de-replicated dataset at 97%, March 2013 McDonald et al., 2012) as reference (otu_ picking_method, uclust_ref; similarity cutoff value, 0.95). All reads from paired end sequencing were quality filtered using a QIIME script (split_ library_ fastq.py) with the following stringent parameters to ensure only high quality reads were included in the analysis: read trimming with a Phred quality threshold of 17 (-q 17) and discarding reads shorter than 50% of the input read length (-p 0.5), and then the quality filtered single-end reads were used for the closed reference picking to generate OTU tables.

Assembly and binning

Metagenome assembly and population genome binning followed the approach previously described (Albertsen et al., 2013). A graphical illustration of the workflow is shown in Fig. S1. Briefly, paired-end metagenome reads in fastq format were merged with SeqPrep (https://github.com/jstjohn/SeqPrep) using default settings and Illumina sequencing adapters were removed. Unmerged reads were quality trimmed and filtered using Nesoni v0.112 (https://github.com/ Victorian-Bioinformatics-Consortium/nesoni) with removing low quality bases from reads with a Phred quality threshold of 17, removing homopolymers reads, and eliminating trimmed reads shorter than 30 bases (clip –quality 17 –homopolymers yes –length 30). The merged and trimmed reads from the four metagenomes were co-assembled using SPAdes v2.5.0 with the following parameters suited to a complex metagenomic assembly: –only-assembler -k 67 –sc. Reads from respective samples were separately mapped to scaffolds using BWA v0.7.4 (Li & Durbin, 2010) with the BWA-MEM algorithm using default parameters. Population genome binning using differential coverage (Albertsen et al., 2013) was performed using GroopM v0.1 (Imelfort et al., 2014) with the initial core formation based on contigs/scaffolds longer than 1,500 bp. Manual refinement of population genome bins, and subsequent recruitment of contigs/scaffolds longer than 500 bp was performed using the GroopM tools.

Identification of conserved marker genes

All contigs/scaffolds in each genome bin were translated into six reading frames, and hmmsearch in HMMER3 (Eddy, 2011) was used to identify 111 single copy marker genes conserved in most bacteria (Dupont et al., 2012), 83 phylogenetically-informative marker genes (Soo et al., 2014), and the 38 marker genes proposed by PhyloSift (Darling et al., 2014). To determine the completeness and contamination of each genome bin, the distribution and number of the 111 conserved single copy marker set was determined using CheckM (Parks et al., 2014) with default settings.

Refinement of population genome bins

Scaffolding of metagenome contigs using the mate-pair data was performed with SSPACE v2.0 (Boetzer et al., 2011). SSPACE was run with the following two sets of parameters: lower stringency for minor population genomes with relatively low coverage (e.g., UASB270), -k 2 (minimum number of links to compute scaffold) -a 0.7 (maximum link ratio between two best contig pairs) -x 0 (no extention of the contigs using paired reads) -p 1 (making .dot file for visualization) and higher stringency for major population genomes (e.g. UASB14): -k 4 -a 0.7 -x 0 -p 1. The resulting dot files were used for visualizing contig connections using Cytoscape v.2.8.1 (Shannon et al., 2003). In addition, Cytoscape attribute files were generated with coverage, length, and bin number (bin name) information for each contig/scaffold. Based on the coverage information and number of connections between contig/scaffolds, external contig/scaffolds are manually added to each bin. In addition, contig/scaffolds with a small number of connections to other contig/scaffolds in their respective bins were excluded. Refined sets of contig/scaffolds were then scaffolded with SSPACE with the following low stringency parameters: -k 2 -a 0.7 -x 0 -p 1. For further refinement, shotgun mate-pair reads were mapped to the newly generated scaffolds using CLC genomic workbench v6.0 (CLC Bio) using default parameters with the exception of a similarity fraction of 0.98 and exported in SAM format. The assembly was visualized using Circos (Krzywinski et al., 2009) and used for manual inspection of the assembly as previously described (Albertsen et al., 2013). Manual correction of misassembly and mis-scaffolding was performed using the microbial genome finishing module in CLC genomic workbench v6.0 (CLC Bio).

Genome tree

Finished bacterial and archaeal genomes were downloaded from IMG (release 4.1) (Markowitz et al., 2014), from which the 38 universally (Darling et al., 2014) or 83 single-copy proteins broadly conserved in bacteria were identified using HMM searches (Soo et al., 2014). To evaluate the robustness of the protein trees (genome trees), four different outgroup taxon configurations (two data sets for 38 marker genes, two data sets for 83 marker genes) were made (Table S3). Homologous proteins obtained from the KSB3 and reference genomes in each taxon configuration were aligned using hmmalign in HMMER3, and subsequently concatenated. A mask was generated for the concatenated alignment using Gblocks (Talavera & Castresana, 2007) with only conserved positions found in more than half of the sequences considered. All tree topologies were tested for robustness using the maximum likelihood methods from FastTree v2.1.7 (with default parameters, JTT model, CAT approximation) (Price, Dehal & Arkin, 2009) and RAxML v7.7.8 (JTT and Gamma models with rapid 100 times bootstrapping) (Stamatakis, 2006). The PHYLIP SEQBOOT module (Felsenstein, 1989) was used to generate 100 resampled alignments and FastTree was used to analyze the resampled alignments (-n 100). A script (CompareToBootstrap.pl) included in the FastTree package was used to compare the original tree to the resampled trees and generate bootstrap values. Generated trees were imported into ARB (Ludwig et al., 2004), where they were rooted, left-hand ladderized using the “beautify” tool and grouped into phylum-level clades. A representative tree (tree no. 1; Table S4) was exported from ARB and visualized using iTOL (Letunic & Bork, 2011).

16S rRNA gene phylogeny

KSB3 related 16S rRNA genes were manually curated using the Greengenes database (version May 2013; McDonald et al., 2012) in ARB (Ludwig et al., 2004). 16S rRNA genes from binned population genomes were aligned with PyNAST (Caporaso et al., 2010a), imported into ARB, and the alignments were manually corrected using the ARB EDIT tool. Sets of taxa (>1,300 nt) were selected in ARB and their alignments were exported applying Lane mask filtering. One set of taxa included representatives across all recognized bacterial phyla to determine the relative position of KSB3 in the bacterial domain (Fig. S6). A second set of taxa included all KSB3 sequences to determine the relative position of the two UASB filament genomes within the KSB3 phylum (Fig. 1A). Neighbor joining trees were calculated from the masked alignments with LogDet distance estimation using PAUP* 4.0 (Swofford, 2003) with 100 bootstrap resamplings. Maximum likelihood trees were calculated based on the masked alignments using RAxML v7.7.8 (GTR and Gamma models + I) with rapid 100 time bootstrapping. Maximum parsimony trees were calculated using PAUP* 4.0. A heuristic search was used with a random stepwise addition sequence of 10 replicates and nearest-neighbor-interchange swapping. Bootstrap analyses on the maximum parsimony trees were run with 100 times resampling for each best tree. Generated trees were re-imported into ARB for visualization.

Genome analysis

The two assembled KSB3 population genomes were initially annotated with PROKKA v1.7 using default settings (Seemann, 2014), and manually curated by comparison to UniRef90 (Suzek et al., 2007), IMG (Integrated Microbial Genomes, finished genomes, release 4.0) (Markowitz et al., 2014), COGs (Clusters of Orthologous Groups) (Tatusov et al., 2000), PFAM (Punta et al., 2012), and KEGG (Kyoto Encyclopedia of Genes and Genomes) (Aoki-Kinoshita & Kanehisa, 2007) databases. Bi-directional best-blast matches were performed for proteins with matches to UniRef90 and IMG using a bit score threshold of 300, and one-way BLASTP matches with a bit score of 60 (Castelle et al., 2013). For COGs, RPS-BLAST against COG PSSMs from the CDD database (Marchler-Bauer et al., 2013) was performed using an e-value cutoff of 0.01, with the top hit retained for each protein domain. The amino acid sequences were also searched for conserved motifs with PFAM (Punta et al., 2012) using HMMR3 (Eddy, 2011) and PfamScan with default settings (with family noise cutoff). Protein domain structure of some gene products were additionally evaluated using InterProScan search (Quevillon et al., 2005). For manual annotation of the KSB3 genomes, we ranked the resulting annotations as follows: bi-directional best-blast matches with UniRef90 and IMG data; one-way matches with UniRef90, IMG, and COGs; PFAM matches; hypothetical proteins (Castelle et al., 2013). For comparison of gene sets with other genomes, we downloaded the full IMG database (release 4.1) containing all genomes in IMG and their annotations (e.g. PFAMs and COGs). In addition, a list of all finished bacterial genomes and associated metadata (e.g. taxonomic affiliation and genome size) was obtained though IMG. Ribosomal RNA copy number was estimated by determining the ratio of average genome coverage to 16S rRNA gene coverage for each KSB3 genome calculated using BWA read mapping. CRISPR loci were identified using CRT v1.2 (Bland et al., 2007). Presence/absence of some gene sets related to cell envelope structure (Albertsen et al., 2013), complex bacterial lifestyle, and adaptability to fluctuating environmental conditions (‘social IQ’) (Sirota-Madi et al., 2010) were evaluated based on IMG annotation for finished genomes and annotated KSB3 genomes, and the resulting abundance matrix was visualized using R and ggplot2. Orthologous proteins between the two KSB3 genomes were identified using pairwise bi-directional best hit BLASTP searches. Glycoside hydrolases were identified using the CAZy database (Lombard et al., 2014) (dbCAN HMMs v3.0, Yin et al., 2012) with HMMER3 (default settings). Transmembrane proteins were predicted using TMHMM Server v. 2.0 (Moller, Croning & Apweiler, 2001).

Gram-staining and gliding motility

Gram-staining for KSB3 filaments was performed based on the method of Hucker, (Doetsch, 1981). Gliding motility of KSB3 filaments was evaluated using fresh sludge samples examined under an epifluorescence microscope (Axioplan 2; Carl Zeiss) equipped with an automatic thermo-control system (Thermo Plate, MATS-55SFG-FT; Tokai Hit). Fresh sludge samples were placed on glass slides, and a cover glass carefully positioned over the sample to minimize exposure to air. To maintain anaerobic conditions on the slides, reducing agents (Na₂S and/or L-cysteine) were added to the samples. The temperature of the microscopic stage was maintained at 37 °C, and time-lapse images were recorded with a cooled CCD camera (DP72; Olympus) equipped with the imaging software (DP2-BSW, version 2.2; Olympus). Glucose, maltose, ribose, mannose, galactose, arabinose, raffinose, sucrose, xylose, fructose, lactate, ethanol, propionate, nitrate, nitrite (final concentration of 5–10 mM) and yeast extract (0.1%) were used as candidate stimuli to induce motility. Each potential stimulant was mixed with the cells prior to the observation, or placed at the edge of the glass cover creating a gradient as the stimulant diffused into the sample.

KSB3 populations in the UASB system

Two UASB sludge samples taken two years apart (A1 and A2), and flocular (F1) and granular (G1) fractions derived from sample A1, reported previously (Soo et al., 2014), were used in the present study. The UASB system had a history of periodic bulking caused by KSB3 filaments (Yamada et al., 2007; Yamada et al., 2011). Inspection of 16S rRNA gene amplicon community profiles of these samples (Soo et al., 2014) revealed two KSB3 16S rRNA phylotypes accounting for 4.9 and 3.7% of total sequencing reads from samples A1 and A2, respectively. The dominant phylotype, representing ∼94% of the KSB3 reads, was identical to the previously reported bulking phylotype (clone YM-1, AB218870; Yamada et al., 2007), and the minor phylotype, representing ∼6% of the KSB3 reads, was identical to a low abundance clone detected in the UASB reactor during normal operation (clone SmB78fl, AB266927; Narihiro et al., 2009). The internal transcribed spacer (ITS) region of the bulking phylotype was sequenced to confirm that it was the same strain present in the bulking and normally operating UASB sludge (Fig. S2). Previously reported metagenomes (Soo et al., 2014) and additional shotgun sequencing of samples A1, F1, G1 and A2 were used for recovering high quality draft population genomes of the two KSB3 phylotypes (Table S1). Based on detection of 16S rRNA genes in the shotgun paired-end read datasets, the KSB3 phylotypes comprised up to 10 and 11% of the A1 and A2 metagenomes respectively with the dominant KSB3 phylotype having approximately 10 fold higher abundance than the minor phylotype (Fig. S3), broadly consistent with the amplicon results.

Recovery of KSB3 population genomes

The four metagenomes (59 Gb in total, Table S1) were co-assembled, generating 504,757 contigs/scaffolds (>500 bp) with a combined length of 906 Mb, an N50 of 3 kb and a longest scaffold of 506 kb. Population genomes were recovered from the assembly by exploiting variations in population abundance (coverage) between individual sample metagenomes (differential coverage binning, Albertsen et al., 2013) using the automated binning tool GroopM (Imelfort et al., 2014). The completeness and contamination of the population genomes were estimated by detection of single copy marker gene sets widely conserved in the domain Bacteria (Dupont et al., 2012). Thirty-nine bacterial population genome bins were obtained with >65% completeness (>73/111 markers) and <10% contamination (<11/111, marker genes with >1 copy in a population genome indicate presumptive contamination with another organism). These genomes were refined by tracking mate-pair reads in network graphs to further improve the completeness and reduce contamination of the bins, and to recruit repeat sequences, notably ribosomal RNA operons, which can evade differential coverage binning if present in multiple copies (Albertsen et al., 2013).

We identified 16S rRNA gene sequences in two refined population genome bins (UASB14 and UASB270, Fig. S4) that were identical to the amplicon sequences from the dominant and minor KSB3 phylotypes, respectively (Table 1). Despite careful manual curation of both genomes, their estimated completeness based on 111 conserved single copy marker genes (Dupont et al., 2012) is only ∼93%, and both also have an inferred ∼6% contamination based on these markers. Inspection of the marker genes with no or >1 hit, however, show a high degree of overlap between the genomes, suggesting that these particular genes are either actually absent, duplicated, or laterally transferred based on phylogenetic inference and gene neighborhood (Table S2). This may not be unexpected given the phylogenetic novelty of the lineage. A revised estimate of completeness and contamination based on a prediction that six of the 111 marker genes are absent and five are duplicated is >98% and <2% respectively (Table 1). To estimate the number of rRNA operons in each KSB3 genome, we compared average genomic coverage to 16S rRNA gene coverage, which indicated that UASB14 and UASB270 have three and two rRNA operons respectively (Fig. S5).

KSB3 phylogeny and morphology

The relative position of the two KSB3 genomes within the phylum was assessed by comparative analysis of their 16S rRNA gene sequences with publicly available full-length sequences. UASB14 and UASB270 represent two of several major lines of descent in the KSB3 phylum (Fig. 1A). According to Greengenes classification (McDonald et al., 2012), UASB14 belongs to an unnamed class-level lineage and UASB270 is a member of class MAT-CR-H3-D11 for which we propose the names Moduliflexia and Vecturitrichia, respectively (Table 1; Supplemental Information 1). To confirm the status of KSB3 as a candidate phylum, as inferred by 16S rRNA comparative analyses (Yamada et al., 2007, Fig. S6), we constructed phylogenetic trees based on a larger genomic sampling. Two sets of marker genes broadly conserved in all domains of life (38 markers) (Darling et al., 2014) or in Bacteria (83 markers) (Soo et al., 2014) were obtained from the KSB3 genomes and up to 354 publicly available reference genomes (Markowitz et al., 2014). Each gene family was independently aligned and ambiguous and/or non-informative positions removed, and then the filtered alignments were concatenated for maximum-likelihood inference. Four sets of outgroup configurations were used including representatives of all major genomically sampled bacterial phyla (Table S3). The two KSB3 genomes form a robustly monophyletic group in all analyses, and did not reproducibly affiliate with any other phyla (Fig. 2, Fig. S7, Table S4), consistent with the original proposal that KSB3 is a candidate bacterial phylum (Yamada et al., 2007). The average amino acid identity (AAI) between UASB14 and UASB270 is 60.3% (Fig. S8), and supports their assignment to separate classes as it falls within the range of known class-level AAI values (44–61%, Konstantinidis & Tiedje, 2005).

To confirm the filamentous morphology and relative abundance of the KSB3 phylotypes, we designed 16S rRNA-targeted fluorescence in situ hybridization (FISH) probes specific at the phylum and class level and combined them with previously applied KSB3-specific probes (Yamada et al., 2007) (Table S5). We detected only filamentous KSB3 morphotypes in the UASB sludge and these comprised the majority of observed filaments in sample F1 (Figs. 1C–1D). The relative abundance of the two KSB3 phylotypes inferred from both amplicon and metagenome data was also consistent with FISH analyses; that is, filaments belonging to the class Moduliflexia (presumably mostly UASB14) greatly outnumbered those belonging to the Venturitrichia (presumably mostly UASB270) (Fig. 1B), noting that the two filaments were indistinguishable by light microscopy alone (Figs. 1C–1D).

General features of the KSB3 genomes

Both KSB3 genomes are large by bacterial standards, >7 Mb (Fig. S9) and have median GC content, ∼50% (Table 1). Since UASB14 and UASB270 are not close relatives, large genome size may be a characteristic feature of the KSB3 phylum or at least of the two classes that they represent (Fig. 1A). A total of 5,989 and 7,048 open reading frames (ORFs) were identified in the UASB14 and UASB270 genomes, respectively (Table 1). For both genomes, approximately two thirds of the ORFs had a predicted function and the remaining third were hypotheticals. Reciprocal BLASTP best matches between the predicted gene products of the two KSB3 genomes indicate a shared set of 3,296 orthologs, representing approximately half of the gene inventories in each genome. Included in this common set are conserved genes for translation, nucleotide transport and metabolism, and construction of a diderm (Gram negative) cell envelope including lipopolysaccharide synthesis (Fig. S10). We identified a full complement of rRNA and tRNA genes in UASB14, but not UASB270, which were likely missed in the latter genome (Table 1; Table S6). The UASB14 rRNA genes are estimated to be present as three nearly identical operons (Fig. S5) collapsed into a single large repeat during the assembly process. A number of large clustered regularly interspaced short palindromic repeats (CRISPR) were identified in both genomes (Table 1). CRISPR, together with associated cas genes, constitute a recently described defense mechanism against invading foreign DNAs and have been found in a majority of bacterial genera and most Archaea (Sorek, Kunin & Hugenholtz, 2008). A COG category analysis of the KSB3 genomes indicates that both have high relative proportions of carbohydrate metabolism and transport (G) and signal transduction (T) relative to the bacterial average (Fig. 3). More detailed inferred metabolic properties of the two KSB3 genomes are described below.

Strictly fermentative metabolism

Both KSB3 representatives have an incomplete tri-carboxylic acid (TCA) cycle and lack most electron-transport chain complexes including terminal oxidases, indicating a strictly fermentative metabolism (Fig. 4). They encode, however, both superoxide reductase and thioredoxin reductase, suggesting oxidative stress tolerance. Both genomes have a large complement of transporters and enzymes for importing and degrading complex and simple carbohydrates, which can then be fed into a complete glycolysis (Embden-Meyerhof-Parnas) pathway (Fig. 4; COG category G in Fig. 3). Both filaments also are likely capable of hydrolyzing polymers such as cellulose and starch via a range of glycoside hydrolases (Table S7). They redundantly encode four different enzymes for converting pyruvate to acetyl–coenzyme A (acetyl-CoA), namely pyruvate dehydrogenase, pyruvate-formate lyase, pyruvate ferredoxin oxidoreductase, and pyruvate-flavodoxin oxidoreductase. Both generate adenosine triphosphate (ATP) by converting acetyl-CoA to acetate via two enzymes (acetate kinase and phosphate acetyltransferase) commonly found in general fermentative anaerobes (Mai & Adams, 1996), in addition to glycolysis. They may reoxidize NADH produced during glycolysis by converting pyruvate to D-lactate and acetyl-CoA to ethanol (Fig. 4).

The KSB3 genomes possess a large complement of enzymes for conversion and transport of amino acid and peptides (Fig. 4; COG category E in Fig. 3), including numerous proteases and peptidases (Table 1). Peptide and amino acid degradation in the KSB3 filaments may produce pyruvate, oxaloacetate, succinyl-CoA, and possibly propionyl CoA (Fig. 4). Notably, a complete set of genes for the methylmalonyl CoA pathway was identified in UASB270, suggesting a role in either amino acid degradation, propionate oxidation and/or propionate formation as a fermentative end product in this organism. Some fermentative anaerobes are known to produce hydrogen to scavenge excess electrons generated during metabolism (Sieber, McInerney & Gunsalus, 2012). In both KSB3 genomes, we identified several hydrogenase genes (Fig. 4). By examining domain structure and gene neighbourhoods (Fig. S11), we predict that some of these genes encode catalytic enzymes. This may permit them to engage in syntrophic interactions with hydrogenotrophs, such as methanogens, in the sludge granules (Sieber, McInerney & Gunsalus, 2012). However, based on FISH experiments highlighting KSB3 and archaeal cells, we did not observe a close proximity between the two groups that would facilitate syntrophy (data not shown). Some of the hydrogenase genes are located next to signal transduction genes raising the possibility that they are involved in signal transduction and chemotaxis (Fig. S11).

Sensory capabilities and motility

One of the most striking features of the KSB3 genomes is the presence of extensive regulatory networks, including two-component signal transduction systems (Table 1; Table S8). Signal transduction genes (COG category T) are among the highest represented categories in both genomes (Fig. 3). Two-component systems respond to a broad range of extracellular and intracellular signals, and play a role in many cellular processes including growth, motility, and the cell cycle (Galperin, 2004; Skerker et al., 2005; Kirby, 2009). UASB14 and UASB270 encode 135 and 114 putative transmembrane sensor proteins likely used for environmental signaling (Galperin, 2004), and 131 and 116 putative response regulators containing CheY-like domains, respectively (Table S8). They each contain over 60 methyl-accepting proteins and numerous Che-like chemotaxis proteins (Table S8). Even when compensating for their relatively large genome sizes, both KSB3 genomes possess high proportions of environmental sensory networks compared to other sequenced bacterial and archaeal genomes (Fig. 5; Figs. S12 and S13; Table S9). The high representation of sensory components in the KSB3 genomes is on par with social Myxococcales such as Sorangium cellulosum and Stigmatella aurantiaca,  both of which exhibit complex, self-organizing behavior in response to environmental stimuli (Huntley et al., 2011).

Sensory capabilities are an important component of a bacterium’s overall social “intelligence” or social IQ (Ben-Jacob et al., 2004), a metric recently proposed based on the abundance of two-component systems, transcription factors, defense mechanisms and transport systems (Sirota-Madi et al., 2010). We determined that the KSB3 filaments have among the highest social IQ scores of any sequenced bacterial and archaeal species to date, scoring particularly well in the two-component and transport system categories (Fig. S14). This suggests that the filaments are sensitive to their surroundings and capable of adaptable behavior in response to changes in their local environment. Key to this adaptability is motility. No genes for flagella production were identified in either KSB3 genome, so to determine if KSB3 bacteria are indeed motile, we observed filaments enriched from UASB granules by wet mount microscopy under a range of conditions. KSB3-specific FISH of samples taken in parallel confirmed that the majority of filaments in these samples were members of the KSB3 phylum (Fig. 1). Initially no motility was observed, therefore based on the metabolic reconstruction of the KSB3 genomes, we added a range of compounds (mostly simple sugars, see ‘Methods’ section) to the edge of the microscope slides to create a gradient that could be sensed by the filaments to stimulate a motility response. We observed gliding motility at rates of between 20 to 30 µm/min only when a glucose or maltose gradient was applied under conditions mimicking the UASB reactor operation (Movie S1). Both KSB3 genomes encode a number of the genes necessary for type IV pili formation (pilB, pilC, pilG, pilT, pilV, and flp pilus assembly protein) that may enable gliding via extension and retraction (Jarrell & McBride, 2008). However, the full gene complement for pili formation (Mauriello et al., 2010) was not detected and the mechanism for KSB3 gliding motility remains to be determined.