Genetic Analysis of Collective Motility of Paenibacillus sp. NAIST15-1

Bacteria have developed various motility mechanisms to adapt to a variety of solid surfaces. A rhizosphere isolate, Paenibacillus sp. NAIST15-1, exhibited unusual motility behavior. When spotted onto 1.5% agar media, Paenibacillus sp. formed many colonies, each of which moved around actively at a speed of 3.6 μm/sec. As their density increased, each moving colony began to spiral, finally forming a static round colony. Despite its unusual motility behavior, draft genome sequencing revealed that both the composition and organization of flagellar genes in Paenibacillus sp. were very similar to those in Bacillus subtilis. Disruption of flagellar genes and flagellar stator operons resulted in loss of motility. Paenibacillus sp. showed increased transcription of flagellar genes and hyperflagellation on hard agar media. Thus, increased flagella and their rotation drive Paenibacillus sp. motility. We also identified a large extracellular protein, CmoA, which is conserved only in several Paenibacillus and related species. A cmoA mutant could neither form moving colonies nor move on hard agar media; however, motility was restored by exogenous CmoA. CmoA was located around cells and enveloped cell clusters. Comparison of cellular behavior between the wild type and cmoA mutant indicated that extracellular CmoA is involved in drawing water out of agar media and/or smoothing the cell surface interface. This function of CmoA probably enables Paenibacillus sp. to move on hard agar media.


Introduction
Colonies formed by robust motile Paenibacillus bacteria form intricate patterns on agar media. P. vortex forms a highly branched colony pattern on agar media [24]. When grown on hard agar media, hundreds to millions of P. vortex cells assemble and generate rotating colonies (also referred to as wandering colonies). These rotating colonies move forward, leaving behind trails of cells; the latter form the 'branches' of the colony pattern. P. alvei forms a nebula colony pattern on agar plates, which comprises randomly scattered clusters of bacteria [23]. Although its colony pattern is quite different from that of P. vortex, P. alvei also forms wandering colonies [3], indicating that P. vortex and P alvei share common mechanisms for motility. P. dendritiformis forms a chiral branching colony pattern, where cells are aligned parallel to their neighbors [25]. At the tips of the growing branches, each cell moves back and forth along their neighbor [28]. Since these cells do not form wandering colonies, the mechanism underlying the motility of P. dendritiformis appears to be different from those of P. vortex and P. alvei. The molecular mechanisms underlying motility and colony pattern formation are very interesting but are unclear because these Paenibacillus bacteria are not amenable to genetic manipulation. For instance, although flagella are thought to produce the driving force, this remains unproven.
Here, we describe the motility behavior of a genetically tractable isolate of Paenibacillus sp. NAIST15-1 and its genetic analysis. Paenibacillus sp. showed peritrichous hyperflagellation in response to growth on hard agar media. Hyperflagellated cells formed moving colonies that were able to migrate across the surface of solid media containing >1.5% agar. We found that the cmoA gene (colony movement A) was necessary for motility on hard agar media. CmoA is a large extracellular protein that envelops moving cells clusters. CmoA is conserved in Paenibacillus bacteria that form moving colonies. The cmoA mutant were unable to move on 1.5% agar media but able to spread on 0.6% agar media. However, its cellular behavior on 0.6% agar media was quite different from that of the wild-type strain. We discuss possible roles of CmoA in motility on hard agar media.

Motile behavior of Paenibacillus sp.
Paenibacillus sp. NAIST15-1 (hereafter referred to as Paenibacillus sp.) was isolated from spore-forming bacteria that showed antagonistic activity against a plant pathogen Fusarium oxyosporum from weed roots and associated soil in the course of the previous study [29]. The bacterium produced swollen sporangia (S1 Fig), which is typical of the genus Paenibacillus [13]. Consistent with this, the 16S rDNA sequence of this bacterium was quite similar to those of Paenibacillus bacteria (S2 Fig); it showed 99% identity with that of Paenibacillus alvei. These observations indicated that the bacterium belonged to the genus Paenibacillus. Paenibacillus sp. exhibited unusual colony formation. When inoculated onto the center of plates containing 2×YT/1.5% agar media, Paenibacillus sp. spread over the surface and formed many discrete colonies ( Fig 1A). This colony scattering phenotype appears similar as that observed for Paenibacillus alvei, and is described as either "wandering colonies" or "nebula pattern formation" [3,23]. We were interested in this phenotype and a molecular mechanism behind it. Our isolate Paenibacillus sp. was suitable for the analysis because it was amenable to genetic manipulation (Methods, see below). The colony spreading pattern was greatly affected by the agar concentration ( Fig 1A). Paenibacillus sp. formed a featureless mat all over plates containing 0.3% or 0.5% agar, but formed a mat within the central portion of 1% agar plates, with multiple colonies around it. Colony spreading was partially inhibited on 2.0% agar, and completely inhibited when the agar concentration exceeded 2.5%. We then examined the behavior of motile cells at the leading edge zones of colonies grown on 0.3%, 0.5%, and 1.5% agar media by light microscopy ( Fig 1B). Because cells grown on 0.3% agar media moved independently in three dimensions within a thick fluid layer, we were not able to obtain well-focused images. The behavior looks like swimming motility. Cells on 0.5% agar media were observable in a single plane, and often formed small groups that moved together. These groups were unstable and frequently changed their members and shape during movement. On 1.5% agar media, most cells formed groups in which the cells were closely aligned along their long axis; these cells often formed bundles (Fig 1B). The number of cells in these clusters varied from a few to over thousands, and the larger clusters contained stacks of cells (Fig 1B; lower-right panel). Neither these densely packed clusters nor large clusters were observed at the leading edge zones of colonies grown on 0.3% and 0.5% agar media. Among these clusters, larger clusters moved forward smoothly, while single cells and small clusters moved very slowly or not at all (S1 Movie and S2 Movie). Thus, cluster formation appears to facilitate cell movement on the surface of 1.5% agar media. While the clusters were moving forward, some cells often fell away from the clusters, suggesting that these detached cells comprise static single cells and small clusters at the leading edge zone. Swarm bacteria often exhibit elongated cellular morphology during swarming motility [5,6, and references therein]. However, Paenibacillus sp. did not show elongated cell morphology on 1.5% agar media.
We next investigated how the formation of moving cell clusters led to the colony scattering phenotype on 1.5% agar. For this purpose, a suspension of Paenibacillus sp. was spotted onto 1.5% agar and the process of colony formation analyzed under a stereo microscope (S3 Movie). During the first few hours post-inoculation, multiple tiny colonies appeared at the inoculation site, which then, while growing, moved around actively over the surface of the media. The speed of the moving colonies was 3.6 μm sec -1 (an average of 12 colonies was monitored for 10 min). Each moving colony appeared to have polarity; a fixed forward region led colony movement. The shape of moving colonies was not constant and sometime became very long. Moving colonies sometimes coalesced and divided. Some colonies then began to rotate and form vortices. Vortex formation began at the inoculation site, which contained a high cell density, and all colonies formed vortices as the colony density increased. Both clockwise and counterclockwise vortices were observed. The rotation speed of the vortices gradually decreased until static colonies were formed. These observations show that moving microscopic clusters of cells grow up to form visible colonies that move actively on hard agar (Fig 2). We called these moving  The wild-type strain was  inoculated onto the center of 2×YT plates containing a variety of agar concentrations and grown at 37˚C for 18 h. Plate diameter, 9 cm. (B) Morphology of motile cells. The wild-type strain was inoculated onto the center of plates containing 0.3%, 0.5%, or 1.5% agar and incubated at 37˚C for 6 h. Coverslips were placed directly on the surface of the leading edge zones of the colonies and cell morphology observed under a light microscope. Stacks of cells can be seen in the lower-right panel, and are indicated by an arrow. Scale bar, 5 μm.
doi:10.1371/journal.pgen.1006387.g001 colonies "wandering colonies", as first described by Henrichsen [3]. Each wandering colony then rotated and finally became a static round colony (Fig 2). These unusual behaviors are the reason why Paenibacillus sp. forms many discrete colonies on hard agar media. Moving colonies have also been described for Paenibacillus vortex, which forms a highly branched colony pattern on hard agar media [24]. At the tips of growing branches, tens to thousands of cells form a spiraling colony that moves forward on hard agar media [24]. However, unlike for Paenibacillus vortex, Paenibacillus sp. colonies moved forward without spiraling as wandering colonies; the formation of spiraling colonies by Paenibacillus sp. was observed in transition from the active moving phase to the sessile phase (S3 Movie). Thus, colony movement and spiraling are separate entities in Paenibacillus sp. at least under the growth conditions tested herein.
Draft genome sequencing of Paenibacillus sp. NAIST15-1 The entire genome of Paenibacillus sp. was sequenced using a massively parallel sequencing platform, MiSeq (Illumina). A total of 1.7 Gb of 300 base paired-end reads were assembled into 42 contigs (>1 kb) using CLC genomic workbench ver. 6.5 (CLC bio, Qiagen). The N 50 value, which is a statistical measure of average length of contigs, was 264,241 bp and the longest contig was 824,207 bp. The draft genome of Paenibacillus sp. contains 6,768,284 bp, with a G+C content of 46.3%. The genome comprises 6,034 coding sequences, 5 ribosomal RNA partial sequences, and 78 transfer RNAs.
Flagellar genes were identified at five different loci within the Paenibacillus sp. genome. The largest cluster was the 30 kb fla/che operon, which contains genes for the flagellar hook-basal body complex, chemotaxis proteins, and an alternative sigma factor, σ D (Fig 3). Another large flagellar gene cluster contains genes that encode anti-σ D FlgM, the CsrA translational regulator, flagellin, and the filament cap. The other loci were the flhOP operon, which is probably required for hook assembly, and two stator operons (motAB and motCD). The composition and organization of these flagellar gene clusters were very similar to that of a Gram-positive model bacterium, Bacillus subtilis (Fig 3). These bacteria are expected to have a similar system for flagellar formation. However, B. subtilis has SwrA, which is required for transcriptional activation of the fla/che operon [30], but no swrA homolog gene was found in the Paenibacillus sp. genome. swrA is located between ftsEX-ctpB and uvrBA in the B. subtilis genome. Interestingly, these homologs are found upstream and downstream of motAB in the Paenibacillus sp. genome (S3 Fig).
Type IV pili are involved in motility in some bacteria [31]. Here, pili-related genes were also identified at two loci (S4A Fig). These gene products were similar to Flp pilus assembly proteins, Tad pilus assembly protein, and pre-pilin peptidase. However, our homology search did not identify all of the genes involved in Type IV pili biosynthesis, such as genes for pilin, and inner membrane core protein.
Flagella are essential for motility on hard agar media Genome sequencing suggests that Paenibacillus sp. possesses at least two potential organelles for motility: flagella and pili. To determine which might be responsible for motility, we carried out gene disruption. Flagellar genes fliF and hag encode the M ring of the flagellar basal body and flagellin, respectively. We constructed bacteria harboring in-frame deletions of fliF or hag using the pMAD plasmid [32]. The growth rates of the fliF and hag mutants in 2×YT liquid media were indistinguishable from those of the wild-type strain (S5 Fig). The motility of these disruption mutants was then tested on media containing agar concentrations varying from 0.3% to 1.5%. Disrupting fliF and hag completely abolished motility on all agar media (Fig 4). The deleted fliF and hag alleles were then restored using a pMAD plasmid carrying the wildtype sequence of fliF or hag. The resulting complemented strains exhibited normal motility To test whether flagellar rotation is required for motility, we constructed disruption mutants of these operons. Whereas disrupting motAB did not affect motility on all agar media tested, disrupting motCD prevented motility on 1.0% and 1.5% agar media (Fig 4). Disrupting both motAB and motCD completely abolished motility on all agar media tested (Fig 4). We then cloned motAB or motCD in the multicopy plasmid pHY300PLK [34] and carried out complementation tests. Introduction of multicopy motAB (pHYmotAB) only partially restored motility to the motCD mutant, whereas introduction of pHYmotCD fully restored motility to the motCD mutant ( Fig 5). We also carried out complementation tests using the motABmotCD quadruple mutant. Introduction of pHYmotAB restored motility to the motABmotCD mutant only on 0.3% and 0.5% agar media whereas introduction of pHYmotCD restored motility to the motABmotCD mutant on all agar media tested ( Fig 5). These results indicate that either motAB or motCD is sufficient for motility on 0.3% and 0.5% agar media, but that only motCD can facilitate motility on 1.0% and 1.5% agar media. The requirement for both flagella and stator proteins indicates that flagellar rotation generates the impetus for motility on both soft and hard agar media.
The number of flagella is a key factor for robust motility [35][36][37][38]. Increasing the number of flagella allows bacteria to swarm on harder agar media or to swim through more viscous   media. We hypothesized that increasing the number of flagella might be required for motility on hard agar media. To test this hypothesis, we first carried out fractionation experiments to identify the flagellin protein in Paenibacillus sp. protein extracts. Approximately 8 × 10 6 cells were spread over the surface of a 9 cm diameter 1.5% agar plate. After a 5 h incubation at 37°C, Paenibacillus sp. formed many moving colonies on 1.5% agar plates. These bacterial cells were suspended in buffer, collected, and the bacterial proteins separated into three fractions: secreted proteins, cell-surface associated proteins, and cellular proteins, as described in Methods. These fractions were then analyzed by SDS-PAGE (S8 Fig). We expected that flagellin would mainly be restricted to the cell-surface associated protein fraction, in which two strong protein bands (150 kDa and 28 kDa) were detected on SDS-PAGE gels (S8 Fig). LC-MS/MS analysis identified these two proteins as the S-layer protein (SpaA; Mw 111.5 kDa) and flagellin (Hag; Mw 29.6 kDa) (S9 Fig). The size of the S-layer protein on SDS-PAGE was much greater than its deduced molecular weight. The S-layer protein of Paenibacillus alvei is glycosylated [39,40]. Also, Paenibacillus sp. possesses a putative S-layer glycosylation gene cluster (PBN151_2313 to PBN151_2300] in its genome. These observations suggest that the S-layer protein in Paenibacillus sp. is also glycosylated. The size of the flagellin protein on SDS-PAGE was very similar to its deduced molecular weight. These results confirm that flagellin is a major cell-surface associated protein. We then compared flagellin levels in the cell-surface associated protein fraction under three different growth conditions: liquid media and solid media containing 0.5% or 1.5% agar. Since 0.3% agar media are fragile, we used liquid media rather than 0.3% agar media for this analysis. Flagellin levels in 0.5% agar and 1.5% agar were the same ( Fig 6A); however, those in liquid media were much lower at both the exponential-and early stationary growth phases ( Fig  6A). To confirm this result, we tried to visualize flagellar filaments on cells grown in liquid or 1.5% agar media. To achieve this, a TCA (Ser) to TGC (Cys) substitution was introduced into the 161st codon of hag on the genome, which would allow the flagellin filaments to be labeled with a sulfhydryl-reactive fluorescent dye. The S161C substitution did not interfere with flagellin function because the hag S161C strain showed normal motility (S6 Fig). After fluorescent labeling, filaments were visible in the hag S161C strain but not in the wild-type strain (Fig 6B), indicating that the visualized filaments were indeed flagella. Consistent with the results of SDS-PAGE analysis, cells grown in liquid media had a few flagella, whereas those grown on 1.5% agar media exhibited peritrichous hyperflagellation ( Fig 6B). We further compared the transcription of flagellar genes (hag, motAB, and motCD) in liquid, 0.5% agar, and 1.5% agar cultures. Northern blot analysis revealed that transcription of hag, motAB, and motCD was low in liquid, and strongly induced in 0.5% and 1.5% agar media ( Fig 6C). Thus, hyperflagellation was supported by the transcriptional induction of flagellar genes on solid media. These results demonstrate that flagellar formation is strongly induced in response to growth on the surface of media containing agar at concentrations of 0.5% or above. Wandering colony formation was observable on media containing >1.0% agar, as described above. Flagellin levels and flagellar gene expression between 0.5% and 1.5% agar media were the same, indicating that flagellin levels per se are not a discriminator of wandering colony formation. Additional factors may be required for wandering colony formation on hard agar media.

Identification of cmoA
When cellular proteins were fractionated, flagellin was observed not only in the cell-surface associated protein fraction but also in the secreted protein fraction in the wild-type strain ( Fig  7, lanes 1 and 2). We found that one large, unknown protein (120 kDa) exhibited the same distribution as flagellin (Fig 7, lanes 1 and 2). Moreover, the 120 kDa protein was induced on solid media, as observed for flagellin ( Fig 6A). Interestingly, the level of the 120 kDa protein was greatly reduced in the hag mutant (Fig 7, lanes 5 and 6). Therefore, we hypothesized that the 120 kDa protein might be involved in motility on hard agar media.
LC-MS/MS analysis revealed that the 120 kDa protein was the product of PBN151_2348 (S9 Fig), which is located downstream of the flagella gene cluster (Fig 3). PBN151_2348 is hereafter called cmoA (see below). To confirm that the 120 kDa protein was the product of cmoA, we constructed two new strains: an in-frame deletion mutant of cmoA and a cmoA-mCherry fusion strain. The fractionation analysis of cellular proteins revealed that deleting cmoA caused the 120 kDa protein to disappear from both the secreted and cell-surface associated protein fractions (Fig 7, lanes 7 and 8). The introduction of the mCherry tag into the 3-terminus of cmoA resulted in the disappearance of the 120 kDa protein and the appearance of a new 150 kDa protein (Fig 7, lanes 3 and 4). Unfortunately, the 150 kDa protein was hidden behind the abundant S-layer protein in the cell-surface associated protein fraction, and was only observable in the secreted protein fraction. The size of this protein (150 kDa) was consistent with the deduced molecular weight of the CmoA-mCherry fusion protein. These results confirm that the 120 kDa protein is the product of cmoA.
The motility assay revealed that the cmoA deletion mutant spread on 0.3% and 0.5% agar media but not on 1.0% and 1.5% agar media (Fig 4). The complemented strain exhibited normal motility, comparable with that of the wild-type strain (S6 Fig). Thus, CmoA is specifically required for motility on hard agar media. SDS-PAGE analysis showed that the cmoA mutant grown on 1.5% agar media expressed high levels of flagellin (Fig 7, lanes 7 and 8). Indeed, when water was poured on cmoA mutant cells grown on 1.5% agar, the cells immediately moved actively in water (S4 Movie). These results indicate that the cmoA mutation causes defects in motility on hard agar media without preventing flagellar formation. Based on these observations, we designated this gene cmoA (colony movement gene A).
CmoA is a large protein comprising 1,064 amino acids and contains an N-terminal signal sequence required for secretion, a single vWFA (von Willebrand factor type A) domain, and eight tandem IPT/TIG (immunoglobulin, plexins, transcription factors-like/transcription factor immunoglobulin) domains ( Fig 8A). The vWFA domain is observed in cell adhesion and extracellular matrix proteins in eukaryotes [41], while the IPT/TIG domain is observed in cell surface receptors and transcription factors in eukaryotes [42]. However, their function in bacteria is unknown. By protein blast search on the National Center for Biotechnology Information (http://blast.ncbi.nlm.nih.gov), we identified 29 CmoA homologs; 24 were found in Paenibacillus spp., and the others were found in Brevibacillus thermoruber, Aeribacillus pallidus, Domibacillus indicus, Cohnella thermotolerans, and Clostridium sp. BL8. Except for Clostridium sp. BL8, these are Paenibacillus and its closest relatives. Among them, four species are reported to exhibit wandering colonies or the colony scattering phenotype in literatures: P. vortex [24], P. alvei CCM2051T [43], P. assamensis [44], and Paenibacillus sp. Y412MC10 [45], The wildtype strain was grown to OD 600 = 0.6 or 2.1 at 37˚C in 2×YT liquid media or at 37˚C for 5 h on 2×YT/0.5% or 2×YT/1.5% agar media. Cells were collected from each culture and resuspended in 10 mM Tris buffer. The number of cells in each sample was adjusted according to the OD 600 value. Cell-surface associated proteins were extracted by boiling in 0.1% SDS. The extracted proteins were then separated by SDS-PAGE, and strained with CBB. S-layer, CmoA, and flagellin proteins are indicated by arrows. The positions of molecular weight markers are shown. (B) Fluorescent staining of flagella. Wild-type and hag S161C (P148) strains were grown to OD 600 = 0.6 in 2×YT liquid media or grown for 5 h on 2×YT/1.5% agar plates at 37˚C. Cells were collected and incubated with Alexa Fluor 594 C5 maleimide in the dark. After a brief wash, cells were observed under a fluorescence microscope. Scale bar, 5 μm. (C) Comparison of flagellar gene expression. Total RNA samples were isolated from wild-type cells grown in liquid media, on 0.5% agar, or on 1.5% agar media. Transcription of the indicated genes was analyzed by Northern blotting. The positions of 23S and 16S ribosomal RNAs (rRNAs) are indicated by arrow heads. rRNAs stained with methylene blue are shown as a loading control.   Table 1) were grown at 37˚C for 5 h on 1.5% agar media. The secreted protein (lanes 1, 3, 5, and 7) and cell-surface associated protein (lanes 2, 4, 6, and 8) fractions were prepared and separated by SDS-PAGE. The positions of CmoA-mCherry, CmoA, and flagellin are indicated by arrows.   Table 1. Plate diameter, 9 cm. (D) Localization of CmoA. Wild-type and cmoA-mCherry (P205) strains were spotted onto 1.5% agar media and grew until small colonies moved out from the inoculation site. Coverslips were placed directly onto the moving colonies on the plates and examined under a fluorescence microscope. Merged phase-contrast (false-colored red) and mCherry (false-colored green) images are shown. Scale bar, 5 μm. (E) The leading edge zones of expanding wild-type and cmoA mutant colonies on 0.6% agar media. The swarm colonies were spreading from the upper left to the lower right in these photographs. The leading edge of the cmoA mutant swarm colony is indicated by arrows. Scale bar, 20 μm. F. Northern blot analysis of cmoA. Total RNA samples were isolated from wild-type cells grown in liquid media, on 0.5% agar media, or on 1.5% agar media. The positions of 23S and 16S rRNAs are indicated by arrow heads. rRNAs stained with methylene blue are shown as a loading control.
doi:10.1371/journal.pgen.1006387.g008 though while the colony phenotype of the other species is unknown. On the other hand, P. dendritiformis, P. polymyxa, P. larvae, all of which do not form wandering colonies, have no CmoA homolog. In Paenibacillus sp. cmoA is expected to constitute an operon with upstream genes yvyC, fliD, fliS, and PBN151_2349 in the flagellar gene cluster because there is no terminator sequence between these genes (Fig 8B). Among them, FliD and FliS are similar to the filament cap protein and the flagellar type III export chaperon specific for flagellin, respectively. YvyC and PBN151_2349 are unknown small proteins (111 and 99 amino acids, respectively). Except for PBN151_2349, the yvyC-fliD-fliS-(PBN151_2349)-cmoA operon is conserved in Paenibacillus vortex, Paenibacillus alvei CCM2051T, Paenibacillus sp. Y412MC10, and Paenibacillus assamensis that form wandering colonies. Each bacterium has a different small gene at the position corresponding to PBN151_2349. yvyC, fliD, and fliS are also conserved as members of the flagellar gene operon, yvyC-fliD-fliS-fliT,in Bacillus subtilis [46]. These observations indicate that cmoA may be a member of flagellar regulon.

The role of CmoA in motility
CmoA has the N-terminal signal sequence required for secretion, and was detected in extracellular protein fractions, indicating that CmoA might have an extracellular function. To address this, we tested whether the cmoA mutant exhibited extracellular complementation. The secreted protein fraction was prepared from the wild-type strain grown on 1.5% agar, sterilized by passing through a membrane filter, and spread over the surface of a 1.5% agar plate. When the cmoA mutant was inoculated onto the center of the plate, it spread over almost all the plate within 18 h of incubation (Fig 8C). In contrast, the comA mutant did not spread on 1.5% agar plates supplemented with the secreted protein fraction prepared from the cmoA mutant ( Fig  8C). As described above, the hag mutant exhibited decreased CmoA (Fig 7). The secreted protein fractions derived from the hag mutant restored motility to the cmoA mutant only partially (Fig 8C). Those secreted protein fractions did not affect the motility of the wild-type strain, and did not restore motility to the hag mutant (Fig 8C). These results indicate that extraneous CmoA can restore motility to the cmoA mutant. Thus, CmoA plays an extracellular function in terms of motility on hard agar media.
We next examined the localization of CmoA using the cmoA-mCherry strain. The CmoA-mCherry fusion is functional because the cmoA-mCherry strain exhibits normal motility (S6 Fig). To examine its localization during motility on hard agar media, the cmoA-mCherry strain was spotted onto 1.5% agar plates and grown until small colonies moved out from the inoculation site. Coverslips were placed directly onto the moving colonies and examined under a fluorescence microscope (Fig 8D). The bacterial cells showed strong fluorescence emitted by mCherry, which appeared between cells and enveloped clusters of cells. Weak fluorescence was also observed across the surface of the media, indicating that a small part of CmoA diffuses into the media during colony movement. No fluorescence signal was observed in wild-type strain under the same conditions. We tried to examine the effect of the cmoA mutation on cellular behavior. However, we were unable to examine cellular behavior on 1.5% agar media. This is because motility mutants of cmoA and hag did not move when spread over the surface of 1.5% agar media; rather they formed cell aggregates by dividing within the inoculated spots. The shape of the aggregates was quite similar to that of moving cell clusters formed by the wild-type strain (S10 Fig). The cmoA mutant was able to spread on media containing up to 0.6% agar. The clusters formed by the wild-type strain on 0.6% agar media tended to be larger than those formed on 0.5% agar (S11 Fig). Therefore, we compared cellular behavior of both the wild-type strain and the cmoA mutant on 0.6% agar media using video light microscopy. The density of the wild-type cells at the leading edge zone of expanding colonies was very low (Fig 8E), and the cells frequently formed small clusters of several cells (S5 Movie). The clusters of wild-type cells moved more rapidly and smoothly than single cells, and moved forward into the unexplored region of agar media without stalling (S5 Movie). By contrast, the cmoA mutant formed a clear leading edge with a high cell density (Fig 8E). cmoA mutant cells in the area immediately inside the edge moved but did not form cell clusters (S6 Movie). When cmoA mutant cells reached the leading edge, some stalled and others turned back. The behavior of the cmoA mutant is very similar to that of other swarm bacteria, in that the cells are active within the swarming colonies but stall at the swarming edge on hard agar media. Swarming bacteria attract water from agar media and swarmer cells within the swarm colony move in the resulting fluid layer [7][8][9][10]. However, the fluid layer at the edge is very thin and cannot support cell movement until sufficient fluid is supplied from inside the swarm colony [10,47,48]. Thus, our observations indicate that expansion of the cmoA mutant probably depends on the fluid layer, as observed for other swarm bacteria. The cmoA mutant is probably unable to move into the unexplored region until sufficient fluid is supplied from inside the swarm colony. The leading edge region of the wild-type strain had very low cell density. Since a high cell density is thought to be important for extracting water from agar media and maintaining the resulting fluid [10], wild-type cells do not appear to acquire water in the same manner, as the cmoA mutant and other swarm bacteria. In fact, single cells of the wild-type strain were able to move slowly by themselves at the leading edge zone on 0.6% agar media (S5 Movie). These observations indicate that the wild-type strain has a water acquisition mechanism that does not depend on cell population, and that the cmoA mutant loses this mechanism. One possible explanation is that cell associated CmoA may facilitate extraction of water from the agar media and/or smooth the cell surface interface, which enables the wild-type strain to move on harder agar media.
We hypothesized that exogenous CmoA may stimulate motility of other bacteria. To test this possibility, we examined motility of B. subtilis, Paenibacillus alvei, and E. coli on plates supplemented with or without CmoA. The result showed that exogenous CmoA slightly enhanced B. subtilis motility: these cells formed slightly larger swarm colonies on 1% agar plates supplemented with the Paenibacillus sp. wild-type lysate than on plates supplemented with the cmoA mutant lysate (S12 Fig). A similar tendency was observed for the B. subtilis srfAC mutant, which does not produce the surface active compound surfactin involved in swarming motility, on 0.8% agar [49]. However, these effects were not very marked: therefore, we cannot conclude that CmoA stimulates B. subtilis motility. Exogenous CmoA did not affect motility of Paenibacillus alvei and E. coli (S12 Fig).

Low wetness is required for wandering colonies
We compared cmoA expression in liquid and agar cultures. Northern blot analysis revealed that cmoA transcription was low in liquid culture and strongly induced in 0.5% and 1.5% agar cultures (Fig 8F), which was the same with flagellar genes, hag, motAB and motCD (Fig 6C). The size of cmoA transcript was much bigger than 23S rRNA (Fig 8E), suggesting that cmoA is probably cotranscribed with upstream genes yvyC, fliD, and fliS as expected from DNA sequence (Fig 8B). Wandering colonies were observable on media containing >1.0% agar. However, expression of cmoA and flagellar genes was the same between 0.5% and 1.5% agar media, indicating that the induction of cmoA and flagellar genes is insufficient to support wandering colony formation. As shown in S5 and S6 Movies, wild-type cells formed moving cell cluster in a CmoA-dependent manner on 0.6% agar. However, these moving clusters did not grow to visible wandering colonies on 0.6% agar media, suggesting that moving clusters are less stable on 0.6% agar than on 1.5% agar. We hypothesized that wetness might affect the stability of wandering colonies. To test this hypothesis, a small amount of water was gently poured on wandering colonies grown on 1.5% agar and cellular behavior was immediately observed. We found that wandering colonies disassembled easily and quickly in water (Fig 9), and individual cells moved randomly in water (S7 Movie). These observations indicate that wandering colonies are very sensitive to wetness and cannot maintain their shape in high wet conditions. These observations indicate that low wet conditions are required not only to induce cmoA and flagellar gene expression but also to maintain wandering colonies. Thus, Paenibacillus sp. forms wandering colonies specially to move on low wet surfaces.

Discussion
Here, we describe cooperative motility behavior of Paenibacillus sp. NAIST 15-1 and its genetic analysis. Paenibacillus bacteria have commonly been isolated from diverse environmental samples, such as soil, hot springs, blood samples, honeybees, and plants [14-18, 22, 44, 45]. However, their genetic studies have been limited due to the difficulty in their genetic manipulation. We demonstrated that Paenibacillus sp. was amenable to genetic manipulation. Thus, this strain will be useful for genetic analysis of Paenibacillus bacteria. Paenibacillus sp. exhibited distinct types of flagella-driven motility, which was dependent on the agar concentration. Individual cells swam in liquid media or in a thick fluid layer on 0.3% agar media, but exhibited hyperflagellation and formed small moving groups, when grown on the surface of 0.5% agar media. Bacterial motility is often inhibited in solid media containing agar at concentrations above 1%. However, Paenibacillus sp. formed wandering colonies, which moved across the surface of media containing >1.5% agar.
We showed that flagella are essential for the motility of Paenibacillus sp. Despite its special motility behavior, the organization and composition of flagellar genes in Paenibacillus sp. are quite similar to those in B. subtilis. Flagella formation by Paenibacillus sp. was induced on hard agar media, which led to peritrichous hyperflagellation. Hyperflagellation was supported by the transcriptional induction of flagellar gene expression. Transcriptional induction of flagellar gene expression in response to growth on hard agar media is also observed in other swarm bacteria including B. subtilis, Vibrio spp., Proteus mirabilis, and Serratia marcescens [36,[50][51][52][53]]. An artificial increase in the number of flagella allows bacteria to swarm on harder agars or to swim through more viscous media in B. subtilis, Salmonera enterica, Proteus mirabilis, Pseudomonas aeruginosa [35][36][37][38]. Thus, an increase in the number of flagellar is a key factor for motility on hard agar media. Flagellar stators generate the torque required for flagellar rotation using an electrochemical gradient of H + or Na + across the cytoplasmic membrane [33]. Paenibacillus sp. possesses two sets of flagellar stators, MotAB and MotCD. The gene disruption experiment showed that both MotAB and MotCD facilitate motility on 0.3% and 0.5% agar, but only MotCD facilitates motility on 1.0% and 1.5% agar. Multicopy MotAB was unable to restore motility to the motCD mutant on hard agar media. These observations indicate that MotCD plays a specific role in motility on hard agar media at least under the conditions tested herein. Some bacteria harbor multiple stators that play distinct roles in motility. Vibrio spp. has two types of flagella: polar and lateral. Polar flagella are driven by the Na + -dependent stator and provide the impetus for swimming motility [54]. Lateral flagella are driven by the H +dependent stator and provide the impetus for swarming motility [54]. The formation of lateral flagella and their stators is transcriptionally induced in response to growth on hard agar media [53]. However, unlike Vibrio spp., transcription of both motAB and motCD operons of Paenibacillus sp. was induced on hard agar. A recent study shows that Paenibacillus sp. TCA20 also has two sets of stator operons, motA1motB1 and motA2motB2 [55]. Phylogenetic analysis of MotB1 and MotB2 along with other MotB stator subunits showed that MotB2 is highly similar to H + -dependent MotB stators, whereas MotB1 belongs to a different stator cluster from known H + -dependent or Na + -dependent MotB stators [55]. The MotB1-type stators are unique to Paenibacillus bacteria, and the MotA1MotB1 stator is suggested to use Ca 2+ or Mg 2+ as coupling ions [55]. Phylogenetic analysis showed that MotB of Paenibacillus sp. was classified into the group of MotB1-type stators, whereas MotD was classified into the group of H +dependent MotB stators (S7B Fig). These observations indicate that MotAB and MotCD may use different coupling ions for flagellar rotation. Although at present the reason for the functional difference between MotAB and MotCD is unclear, the finding that flagella and stator proteins are required for motility strongly suggests that flagella rotation provides the impetus for Paenibacillus sp. motility on both soft and hard agar media.
Swarming motility is defined as flagella-driven group motility on a solid surface [3][4][5][6]. According to this definition, the motility by Paenibacilus sp. on hard agar media is classified as swarming motility, although it has unique characteristics. Since flagella rotation pushes a cell forward against the surrounding water, surface water is a critical element for swarming motility. Swarming bacteria on hard agar media can attract water to the surface from the agar matrix [7][8][9][10]. Cell growth (a high cell density), cellular secretions, and flagella rotation are thought to attract water to the surface, from where it diffuses throughout the swarming colony [7-11, 47, 48]. As a result, the swarming colony maintains a fluid layer within itself, in which each cell moves actively. However, the edge of the colony contains only a very thin layer of fluid [10]; thus swarmer cells stall there until fluid is supplied from the inside of the swarm colony [11,47]. The flagella of stalled cells at the edge of the swarm colony frequently orient toward the unexplored region of the media and rotate; this likely pumps fluid outwards, thereby aiding expansion of the swarming colony [11,47]. Thus, swarmer cells cannot move into unexplored regions of hard agar media without first expanding the fluid layer, and swarm bacteria usually formed single swarm colony when inoculated onto a single site on hard agar media. Paenibacillus sp. formed many clusters on hard agar media, which grew up to form visible wandering colonies. Cells in wandering colonies were associated with neighbor cells and did not move within wandering colonies. Cell clusters and wandering colonies were able to leave the fluid layer of the mother swarm colony and move around the unexplored region on hard agar media by themselves. As a result, Paenibacillus sp. formed multiple colonies even when inoculated into single site on hard agar media.
We identified CmoA as a protein specifically required for motility on hard agar media. CmoA is a large protein of 1,064 amino acids and comprises a single vWFA domain and eight tandem IPT/TIG domains. The vWFA and IPT/TIG domains are mainly conserved in eukaryotes and many of these domains are found in extracellular proteins such as cell adhesion proteins, extracellular matrix proteins, and cell surface receptors [41,42]. Proteins containing these domains are also found in bacteria, but are not well characterized. The N-terminus of CmoA is a typical signal sequence for secretion. Indeed, CmoA was detected in extracellular fractions. Moreover, the motility of the cmoA mutant was restored in media supplemented with the secreted protein fraction derived from the wild-type strain, but not that from a cmoA mutant. Live imaging of cells expressing the CmoA-mCherry fusion protein plated on hard agar media showed that CmoA was mainly localized around the cells. CmoA-mCherry was present between the cells within clusters, but also covered entire clusters. These observations show that CmoA has an extracellular function, as observed for vWFA and IPT/TIG domain proteins in eukaryotes. The cmoA mutant was able to induce flagella in response to surface growth condition and moved on solid media containing up to 0.6% agar. However, its cellular behavior was quite different from that of wild-type cells. On 0.6% agar media, the wild-type strain formed small cell clusters that moved smoothly into the unexplored region of the media, whereas the cmoA mutant neither formed clusters nor moved smoothly into the unexplored region. Cells of the cmoA mutant moved actively inside the swarming colony, but stalled at the swarm edge and formed a clear leading edge. Interestingly, this behavior is similar to that observed for swarm cells of other bacteria such as B. subtilis and E. coli [11,[47][48][49], whose spread on hard agar media is dependent on the acquisition of fluid. Thus, the cmoA mutant seems to spread on hard agar media in a similar manner to other swarm bacteria. We also observed that the leading edge region of the wild-type strain had very low cell density. Under the condition, cell population-dependent water extraction from agar media is unexpected. In other word, Paenibacillus sp. has a water acquisition mechanism that does not depend on cell population. We propose that cell-surface associated CmoA may be involved in drawing water out of agar and/or smoothing cell surface interactions. Lacking this function, cmoA mutant cells may be unable to leave the mother swarming colony and move into the unexplored region on hard agar media. Proteus mirabilis secretes high molecular weight polysaccharide, which is responsible for extracting water from agar media [56]. E. coli secretes some osmolytes, probably lipopolysaccharides, to extract water [57]. Several swarm bacteria secrete surface active molecules, e.g. surfactin secreted by B. subtilis [49] and rhamnolipids by Pseudomonas spp. [58]. CmoA may be the first identified protein responsible for extracting water from agar media or for reducing surface tension to facilitate swarming motility.
We also observed that clusters of the wild-type cells moved much faster than single cells. Many swarm bacteria form cell clusters, called rafts, which also move faster than single cells in the fluid layer of swarm colonies. For instance, B. subtilis cells recruited to a raft moved with the group, whereas cells lost from a raft stop moving [49]. The formation of rafts is thought to facilitate movement on hard agar media partly by reducing viscosity/drag on individuals [12]. This though probably applies to cell clusters of Paenibacillus sp. As rafts of swarm bacteria are unstable and dynamically labile in terms of both members and shape, no specific substances or matrices appear to maintain rafts stability [5, 6, and references therein]. However, cell clusters of Paenibacillus sp. may have different features. Cell clusters of Paenibacillus sp. were able to leave the fluid layer of the mother colony and move alone into unexplored region. Moreover, on hard agar media containing >1% agar, those clusters grew up to form visible wandering colonies, in which cells are associated with neighbor cells. Therefore, the cell clusters of Paenibacilus sp. on hard agar media appear to be more stable than rafts of other swarm bacteria. We speculate that Paenibacillus sp. may have a mechanism for stabilizing cell clusters. Many swarm bacteria such as B. subtilis and E. coli, cannot move on media containing 1% agar. However, some bacteria, called robust swarmers, can move on media containing >1.5% agar [6, and references therein]. One such robust swarmer, Proteus mirabilis, forms highly organized multicellular rafts during swarming motility. Electron microscopy shows that adjacent cells in those rafts are connected via intercellular bundling of flagella [59]. Likewise, cells in wandering colonies of Paenibacillus vortex are also connected by filament networks, which are probably flagella. [24]. Since these electron microscope images were taken using fixation procedures, further confirmation using live cells will be required. On the other hand, these observations suggest that there may be a mechanism that stabilizes cell clusters in robust swarmers in which flagella may play an important role. The cmoA mutant did not form cell clusters and did not move on hard agar media containing >1% agar. CmoA was located around cells and enveloped cell clusters. Based on these observations, we speculate that CmoA may also play a role in stabilizing cell clusters. For example, cell-surface associated CmoA may interact with flagella, which form cell-cell connections. The social Amoeba, Dictyostelium discoideum forms multicellular fruiting bodies in response to nutrient exhaustion. During this process, cell-cell adhesion is mediated by binding between extracellular proteins TgrB1 and TgrC1 via mutual IPT/TIG domains [60]. CmoA itself may form cell-cell connection via its multiple IPT/TIG domains. However, cell-cell connections that support wandering colony formation are completely different from stable cell-cell connections observed in bacterial biofilms [61]. Wandering colonies were easily disassembled when water was added. The cell-cell connection in wandering colonies may be fragile or sensitive to wetting. Further work will be required to determine how Paenibacillus sp. forms wandering colonies and to examine role of CmoA in wandering colony formation.
Professor Ben-Jacob's group isolated and extensively analyzed pattern forming Paenibacillus bacteria [23-25, and references therein]. However, these strains were isolated as contaminants from a B. subtilis stock and their natural habitat is unclear. Here, we isolated Paenibacillus sp. from the rhizosphere of natural weeds. We found that this type of Paenibacillus bacteria was easily isolated from natural weed roots and associated soils (S13 Fig). The natural habitat of pattern-forming Paenibacillus bacteria seems to be soil and its related environment including the rhizosphere. This seems to be the reason why Paenibacillus sp. exhibits complex motility behavior. Soil includes various substances, and its surface properties vary. The water content in soil is often variable and is sometimes very low. Thus, soil bacteria appear to employ multiple motility systems to overcome these environmental challenges. For instance, a soil bacterium, B. subtilis, exhibits flagellar-dependent swimming and swarming motility [49]. Recently, two groups reported that B. subtilis also exhibits sliding motility under conditions in which flagella do not work [62,63]. Myxococus xanthus, which does not have flagella, also exhibits two types of motility, individual gliding motility and social twitching motility, which have different advantages on different surfaces [64]. Pseudomonas aeruginosa exhibits flagella-dependent motility, Type IV pili-dependent twitching motility, and sliding motility [65]. The results presented herein reveal that Paenibacillus sp. has evolved to use flagella under a wider range of conditions. To overcome conditions of low wetness, Paenibacillus sp. forms multicellular wandering colonies. When wandering colonies encounter wet conditions, they quickly disassembled and individual cells can swim or swarm in the water layer. These properties will be advantageous for Paenibacillus sp., enabling it to disperse its cells rapidly and widely in soil.

Isolation of Paenibacillus sp. NAIST15-1
Plant weeds were uprooted from the ground within Nara Institute of Science & Technology (NAIST) and most of the soil around the roots removed. The plant roots were then cut, placed in 15 ml tubes, and soaked in 2 ml LB medium. After a brief vortex, the extract was incubated at 85°C for 10 min and then plated onto LB at appropriate dilutions. Twelve strains, all of which showed a colony scattering phenotype and antagonistic activity against Fusarium oxysporum, were isolated. Among these, Paenibacillus sp. NAIST15-1 was selected for analysis based on its strong growth and transformability. The 16S rDNA region was amplified from its genome DNA with primers 16S-F1 (5'-TTAGCGGCGGACGGGTGAGT) and 16S-R1 (5'-TGACGGGCGGTGTGTACAAG). Nucleotide data for partial 16S rDNA sequence has been submitted to the DDBJ/EMBL/GenBank databases under the accession number, LC185074.

Strains and growth conditions
The Paenibacillus sp. strains used in this study are listed in Table 1. All strains were derivatives of Paenibacillus sp. NAIST 15-1. Paenibacillus sp. was grown in LB (LB Lennox, Difco) or 2×YT (16g l -1 tryptone (Difco), 10g l -1 yeast extract (Difco), 5g l -1 NaCl) medium. For plates, media were solidified with the indicated concentrations of agar (for Microorganism Culture, Nakalai Tesque). Single colonies were isolated on 2.5% agar plates. Antibiotics were used at the following concentrations: erythromycin, 2.5 μg ml -1 ; tetracycline, 10 μg ml -1 ; and chloramphenicol, 2.5 μg ml -1 . For the motility assay, strains were grown overnight in 2×YT liquid media at 28°C with shaking. One microliter of culture was then spotted onto the center of 2×YT plates. The overnight cultures generally contained an average of 8 × 10 7 cells ml -1 . For the motility assay, freshly prepared plates were allowed to dry overnight on the laboratory bench and used the next day. For the protein and RNA analyses, overnight cultures were diluted 10-fold, and 100 μl were spread on each 9 cm diameter plate. Cells were then cultivated at 37°C for 5 h prior to analysis. Escherichia coli HB101 was used for plasmid construction and was cultivated in LB medium. Ampicillin was used at 50 μg ml -1 when necessary. Plasmids were prepared using the FastGene Xpress Plasmid PLUS kit (Nippon Genetics) prior to electroporation of Paenibacillus sp. Paenibacillus alvei NBRC3343 was obtained from National Institute of Technology and Evaluation. E. coli W3100, B. subtilis NCIB 3610, and its srfAC mutant were described previously [29].

Microscopy
Colony formation was analyzed by time-lapse light microscopy. Briefly, 1 μl of Paenibacillus sp. suspension was spotted onto 2×YT/1.5% agar medium in a 35 mm diameter plate. The plate was then incubated at 37°C in a microscope incubation system (Tokai hit). The process of

Draft genome sequencing
Genomic DNA from Paenibacillus sp. was fragmented (500 bp fragments) using a sonicator (Covaris S2, Covaris). A library for sequencing was prepared using a NEBNext DNA Library Prep Master Mix Set for Illumina (New England Biolabs) and paired-end sequenced (2 × 300 bp) on a MiSeq sequencer using MiSeq reagent kit v. 3 (Illumina K.K.). Raw reads were trimmed and de novo assembled using CLC Genomics Workbench 6.5 (CLC bio, Qiagen).

Nucleotide sequence accession numbers
The raw read data were deposited in the DRA database at DDBJ with the following accession numbers: DRA ID: DRA003610; BioProject ID: PRJDB3476; and BioSample ID: SAMD00030729. The draft genome sequence and annotation for Paenibacillus sp. NAIST15-1 were registered in DDBJ/EMBL/GenBank with the following accession numbers: BBYF01000001-BBYF01000042 (42 entries).

Strain construction
Gene disruption was performed using the plasmid pMAD, which contains a thermosensitive origin of replication for Gram-positive bacteria, erm, and bgaB, which encodes a thermostable β-galactosidase [32]. pMAD was a kind gift from Dr. Kazuya Morikawa (Tsukuba University) and Dr. Michel Débarbouillé (Institut Pasteur). DNA fragments involved in gene disruption via double crossover recombination were prepared by PCR and cloned into pMAD. The plasmid pEpGAP-mCherry [67] was used to construct the cmoA-mCherry strain. pEpGAP-mCherry was a kind gift from Dr. Neta Dean (Stony Brook University). The primers used for PCR are listed in S1 Table. A detailed explanation of the methods used for plasmid construction and gene disruption are provided in S1 File.

Fractionation of cellular proteins
Overnight cultures grown at 28°C in 2×YT were diluted 10-fold and 100 μl spread over 2×YT/ 1.5% agar plates. After 5 h of incubation at 37°C, cells from two plate cultures were suspended in 600 μl of 10 mM Tris-HCl (pH 7.6) and collected in 1.

LC-MS/MS analysis
After the cell-surface associated protein fraction was separated by SDS-PAGE, protein bands were stained with Coomassie Brilliant Blue and excised. Peptides were prepared for LC-MS/ MS analysis by in-gel digestion with trypsin. LC-MS/MS analysis was performed using a LTQorbitrap XL system (Thermo Scientific), as described by Tanaka et al. [68]. The MS/MS spectra of the identified peptides were searched against our own database, which contains SpaA, flagellin, and CmoA sequences, and against non-redundant protein sequences in the National Center for Biotechnology Information (NCBI) database, using a MASCOT server (Matrix Science).

Flagellar staining
Cells were collected from 2×YT liquid or plate cultures and pelleted by centrifugation at 5,000 rpm for 2 min. Cells were then suspended in T-base buffer [69]. Samples were mixed with Alexa Fluor 594 C5 maleimide (Life Technologies; final concentration, 15 μg ml -1 ) and incubated at room temperature for 1 h in the dark. The samples were then washed twice with Tbase buffer and observed under a fluorescent microscope.

Northern blot analysis
Cells were collected from 2×YT liquid or plate cultures. Total RNA was extracted and Northern blot analysis performed as described previously [70]. The primers used for RNA probe preparation are listed in S1 Table.

Extracellular complementation
Overnight cultures grown at 28°C in 2×YT were diluted 10-fold and 100 μl spread over 2×YT/ 1.5% agar plates. After 5 h of incubation at 37°C, cells from two plate cultures were suspended in 800 μl of 2×YT liquid media and collected in 1.5 ml tubes. After brief vortexing, the sample was separated by centrifugation at 15,000 rpm for 5 min. The supernatant was sterilized by passing through a 0.45 μm PVDF membrane filter and 100 μl of the resultant sample was spread over a 1.5% agar plate. The plate was left for 10 min on the bench and used for the swarm assay. ΔmotABΔmotCD, P162; and ΔcmoA, P198 shown in S1 Table. (TIF)