The genome of P. polymyxa ATCC 842 was sequenced by using SMRT sequencing. This resulted in 82,421 continuous long reads (CLR) with an average (total) length of 11,724 base pairs (bp) (Supplementary Figure 1). From the CLRs, 129,188 subreads (i.e., individual fragments) of an average length of 7,454 bp were extracted. Using the Pacific Bioscience de novo genome assembly algorithm (HGAP v. 2.0), we obtained a single 5.97-Mb contig with an average 128-fold coverage and a confidence score > 99.99% (Supplementary Figure 2). The genome sequence of P. polymyxa ATCC 842 obtained with SMRT sequencing is highly similar to what was originally documented in the NCBI database and less similar to that of E168 (Supplementary Figure 3). Whereas, the G + C content of the contig was 45.1%, slightly higher than that of the contig (44.9%) of P. polymyxa ATCC 842 originally deposited in the NCBI database (Jeong et al., 2011). The distribution of GC contents and GC skew in the genome is depicted in Figure 1. Annotation was performed with the NCBI Prokaryotic Genome Annotation Pipeline, showing a total of 5,414 coding sequences (CDSs), in which over 1,300 CDSs were not annotated in the previous study (GenBank ID: CP024795; Supplementary Table 1). A number of CDSs are classified into groups of substance transport and metabolism, according to the COG analysis¹ (Supplementary Figure 4) and the KEGG analysis² (Supplementary Figure 5). Besides, by using the antibiotic resistance genes database (ARDB) (Liu and Pop, 2009), 148 ARGs were predicted (Supplementary Table 2). Minimum inhibitory concentration (MIC) test showed that P. polymyxa ATCC 842 was hypersensitive to erythromycin and vancomycin (MIC < 1) and moderately sensitive to tetracyclin, kanamycin, and chloramphenicol (Supplementary Table 3).

Our SMRT sequencing analysis also revealed the presence of a 45.5-kb plasmid in P. polymyxa ATCC 842 (named pATCC842 hereafter). The distribution of GC contents and GC skew of the plasmid is depicted in Supplementary Figure 6. In pATCC842, 38 CDSs were predicted (Supplementary Figure 7 and Supplementary Table 4). CDS2 is supposed to encode a protein homologous to ParM, a segregation protein which provides the force for driving copies of the plasmid to the end of a bacterium. CDS4 is supposed to encode a replication-relaxation protein. CDS6 is homologous to TcpC, which is structurally homologous to the periplasmic region of VirB8, a component of the type IV secretion system from Agrobacterium tumefaciens (Porter et al., 2012), and therefore considered to be required for conjugative transfer. Sequences of three plasmids (i.e., pPPM1a, pSb31l, and pSC2) of P. polymyxa were previously deposited in the NCBI database. By using a nucleotide comparison software (i.e., progressiveMauve) (Darling et al., 2010), the sequence of pATCC842 was compared with those of pPPM1a, pSb31l, and pSC2. Interestingly, the sequence of pATCC842 showed little homology to sequences of pPPM1a and pSC2 and moderate homology to pSb31l (Supplementary Figure 8). We then performed an iterative tBLASTx of pATCC842 against the NCBI nucleotide database by using BLAST pairwise alignments. The result showed that pATCC842 was most similar to the sequence of pHD05 from Paenibacillus sp. IHBB 3084 (44% sequence coverage, 90% identity) (Supplementary Figure 8), implicating that pATCC 842 and pHD05 may share the same ancestor.

Eight DNA methylation motifs were predicted in the genome of P. polymyxa (Figure 2 and Table 5). The cognate RM system was analyzed with the REBASE database (personal communication with Dr. Richard J. Roberts). A pair of reverse-complement motifs a1 and a2 (CNAGNNNNNTTGK, MCAANNNNNCTNG), which can be recognized by N-6 adenine-specific methyltransferases in the two reverse complementary strands, are predicted to be recognized by a type I RM system PpoAI (Table 5). Motifs a1 and a2 are evenly distributed in the genome of P. polymyxa ATCC 842, without significant GC% or GC skew or ORF bias (Figure 2). CDSs encoding M, S and R subunits of PpoAI were identified in loci away from MGEs, indicating that this RM system is inherent to this strain (Table 6). By contrast, no corresponding RM systems were predicted for other putative methylation motifs that can be recognized by N-6 adenine-specific methyltransferases (b, c, d, f) or by cytosine-4-specific methyltransferases (e and g). Bacterial methylation is often complete or very close to 100%. Considering that the percentages of methylation of motifs a1 and a2 are more than 98%, they are likely to be a real methylation motif. Because only 7.92% to 25.29% of the other six putative motifs are methylated, it is questionable whether these motifs can be considered “real” methylation motifs.

To functionally characterize PpoAI, genes encoding R, M, and S subunits were replaced by a selective marker (i.e., kanamycin resistance gene) in P. polymyxa (Figure 3A and Supplementary Figure 11). The ppoAI null-deletion mutant was confirmed through PCR with a pair of primers targeting on the up- and downstream of the ppoAI locus (Supplementary Figure 12), as well as the flanking sequence of the ppoAI locus and the kanamycin resistance gene (data not shown). A plasmid named pWBUC02 containing a second selective marker (i.e., erythromycin resistance gene) was constructed to examine transformation efficiency in the ppoAI null-deletion mutant (Supplementary Figure 13). To characterize the function of PpoAI, pWBUC02 was isolated from the wildtype strain and the ppoAI null-deletion mutant, respectively, and used as the donor DNA for transformation assays. We observed that, with pWBUC02 isolated from the ppoAI null-deletion mutant as the donor DNA, although transformation efficiency of the ppoAI null-deletion mutant was ∼3.7 × 10³ CFU/μg, no transformants of the wildtype strain were detected (Figure 3B). The result reflected that pWBUC02 from the ppoAI null-deletion mutant was not methylated and therefore cleaved by the REase of PpoAI in the wildtype strain. Although pWBUC02 isolated from the ppoAI null-deletion mutant was unable to transform the wildtype strain, it well transformed the ppoAI null-deletion mutant, with a transformation efficiency of ∼3.7 × 10³ CFU/μg (Figure 3B). The data revealed that the ppoAI null-deletion mutant was unable to restrict unmethylated pWBUC02. Taken together, the plasmid transformation assay clearly shows that the ppoAI null-deletion mutant is defective in DNA modification and restriction.

To further explore the phylogenetic distribution of the PpoAI RM system, we performed BlastP alignments of the endonuclease subunit (R), the specificity subunit (S), and the methyltransferase subunit (M). Homologs of both R and M subunits were exclusively found in Firmicute (Figure 4 and Supplementary Figures 14A,B). In contrast, no homolog of S subunit was found in the database (Supplementary Figure 14C). The region (CCR) between two target recognition domains (TRDs) was reported to be conserved (Gubler and Bickle, 1991; Kim et al., 2005). Comparative analysis of the CCR of the S subunit showed that the corresponding homologous sequences were found not only in Firmicute but also in species of other phyla (Figure 4C and Supplementary Figure 15). Of note, although Clostridium and Paenibacillus belong to different classes of Firmicute, homologs of R and M subunits, as well as the CCR of the S subunit, were found in a number of species in both of the two genera (Figure 4 and Supplementary Figures 14, 15).

Some MGEs carry genes encoding RM systems which help MGEs parasitize in the host genome in a manner comparable with toxin–antitoxin systems (Mruk and Kobayashi, 2014). Providing that genes encoding PpoAI do not lie in MGEs (Table 6), the RM system should be evolved independent of HGT. The presence of PpoAI would restrict the transfer of MGEs carrying cognate DNA motifs, leading to MGEs containing few such motifs. To evaluate whether the PpoAI played a role in the evolution of the genome of P. polymyxa driven by HGT, we calculated the frequency of its cognate motif in the prephages and genome islands, as well as in the genome. The frequency of the motif of PpoAI is 0.188 per kb in the genome, which is significantly higher than that (0.122 per kb) in MGEs (i.e., pre-phages and genome islands) (Figure 4C and Table 7). Components (M, R subunits and CR1 of the S subunit) of CcaP7III in Clostridium carboxidivorans are highly similar to those of PpoAI (Figures 4A,B, 5). SMRT sequencing data of Clostridium carboxidivorans P7 have been deposited in the REBase database. In this strain, the frequency of the methylation motif of CcaP7III in the genome was estimated to be 0.0792 per kb, remarkably higher than that (0.0144 per kb) in MGEs (i.e., prophage and genome islands) (Figure 4C and Table 7). These findings indicate the RM systems belonging to PpoAI family could impose a selective pressure favoring MGEs containing fewer targeting DNA motifs.

Horizontal gene transfer drives the evolution of genomes by introducing extracellular genetic elements (Gogarten et al., 2002; Frost et al., 2005; Treangen and Rocha, 2011; Syvanen, 2012). RM systems often limit MGEs with cognate DNA motifs (Vasu and Nagaraja, 2013; Koonin et al., 2017). In this study, we showed distribution constraints of recognition motifs in MGEs by PpoAI with regard to that in genomes (Figure 4), adding new evidence supporting direct impact of RM systems on genome evolution driven by HGT. Of note, prophage P4 contained remarkably more DNA motifs of PpoAI than the average level in the genome (Table 7). Two explanations were provided. First, the ancestor of P4 may belong to a group of bacteriophages that entered cells as the form of single-stranded DNA or RNA or expressed anti-R-M genes to avoid the constraint of RM systems (Rusinov et al., 2018). Second, the ancestor of P4, which initially contained few DNA motifs of PpoAI, entered the cell earlier than other bacteriophages, and evolved more such DNA motifs through random mutation during evolution.

A type I RM system is consisted of R (restrictase for cleaving DNA), M (methyltransferase for methylating DNA), and S (sequence-specific recognition) subunits (Murray, 2000; Loenen et al., 2014). Homologs of both R and M subunits of PpoAI were found in ∼10 species belonging to Paenibacillus and seldomly found in other closely related genera (Figure 4 and Supplementary Figure 14). The S subunit of a type I RM system determines the specificity of the type I RM system (Calisto et al., 2005; Kim et al., 2005; Loenen et al., 2014). The homolog of S subunit of PpoAI was not found through BlastP in NCBI database (Supplementary Figure 14C). Accordingly, the cognate DNA methylation motif was found to be unique in the REBASE database. In Enterococcus faecium, a type I RM system was found to be associated with subspecies separation (Huo et al., 2019). We propose that PpoAI could serve as a barrier to HGT and contribute to species separation in Paenibacillus. Interestingly, homologs of both R and M subunits of PpoAI were found in ∼20 species of Clostridium (Figures 4A,B and Supplementary Figure 14), which are phylogenetically distant to Paenibacillus and belongs to a different class in Firmicutes. Although S subunits are diverse, sequences of their CCRs are conserved among different species (Gubler and Bickle, 1991; Kim et al., 2005). Homologous sequences of CCR of the S subunit of PpoAI were found in several Clostridium species, in addition to some bacteria even beyond Firmicute (Figure 4C and Supplementary Figure 15). The frequency of its cognate DNA methylation motif also shows a distribution constraint: it was more than sixfold lower in MGEs than that in the genome (Figure 5). These observations implicate that the ancestor of PpoAI should have conferred immunity against extracellular DNA beyond Paenibacillus. A number of bacteria belonging to both Paenibacillus and Clostridium are valuable to agriculture, industry, and human health but intractable to genetic manipulation. Characterization of their shared RM system would provide a foundation for constructing genetic tools in strains of both genera.

Single-molecule real-time sequencing revealed the presence of methylation at both adenine and cytosine sites in P. polymyxa (Table 5). In all, 164,207 DNA methylation motifs were detected in the genome. There are 125,208 and 34,424 copies of motif e (CNNNNRNH) and motif g (CNVNDYBH), respectively, and they both have 4-methylated cytosines (Table 5). The two motifs comprise ∼98.4% of the total copies of DNA methylation motifs. In contrast, only less than 1.6% detected methylation nucleosides locates in the rest of the six motifs with N-6 methylated adenines (Table 5). Therefore, methylation on cytosines could be more prevalent than that on adenines in P. polymyxa ATCC 842. Their cognate RM systems and impact to genomic evolution remain to await further investigation.

Belonging to the generally recognized as safe species, strains of Paenibacillus can produce a number of metabolites of agricultural, medical, and industrial values (Grady et al., 2016; Wang et al., 2018). Establishing a model strain for understanding genetics of Paenibacillus would provide guidance for constructing cell factories in this species. Although genetic engineering tools were available in P. polymyxa SC2 which seems to not have strong RM systems (Zhang et al., 2013), phenotypic variability of this strain could limit its use as a model strain (Hou et al., 2016). In contrast, P. polymyxa ATCC 842 was phenotypically and genetically stable from one generation to another, making it ideal as a model strain of Paenibacillus. Our previous work established a method for transfer and expression of exogenous genes in P. polymyxa ATCC 842 (Shen et al., 2018). In this study, we have successfully inactivated its chromosomal genes (Figure 3). Together, we have established a complete genetic manipulation system for expressing exogenous genes and inactivating endogenous genes in P. polymyxa ATCC 842. Moreover, deleting the type I RM system further increased transformability of this strain.

Nevertheless, there are still obstacles for establishing ATCC 842 as a model strain. Our previous work revealed the presence of an additional type IV RM system which efficiently degrade Dam methylated DNA in P. polymyxa ATCC 842 (Shen et al., 2018). Deleting the gene encoding the type IV RM system would further simplify the genetic manipulation process in P. polymyxa ATCC 842. In addition, we have only identified two types of plasmids (i.e., pWB980 derivative and pRN5101) which were able to replicate and two selective markers which conferred kanamycin and erythromycin resistance genes in this strain. Further expanding the genetic toolbox for P. polymyxa ATCC 842 would expedite the study and use of Paenibacillus.

In this study, we analyzed the genome of P. polymyxa ATCC 842 with SMRT sequencing data. IS and GEI are hallmarks of HGT in bacteria (Syvanen, 2012; Lang et al., 2017). Comparative genome analysis revealed 78 ISs and six GEIs in the genome of ATCC 842 (Table 7 and Supplementary Table 5), indicating that HGT events in this species were frequent. Considering that most ISs and GEIs do not contain phage-like or conjugative elements, these MGEs mostly could be acquired, which is normally mediated by a conserved DNA uptake machinery (Sun, 2018; Dubnau and Blokesch, 2019). In the genome of P. polymyxa, a set of DNA uptake and processing gene homologs was identified (Supplementary Figure 15 and Supplementary Table 8). These genes include the comEA which potentially encodes the DNA-binding protein, comEC which potentially encodes the inner membrane channel protein, comGA and comC which could be required for the assembly of competence pseudopili. The competence pseudopilin was encoded by comGC in B. subtilis, and the ortholog of this gene was not identified in P. polymyxa. Assembly of the pseudopilus is mediated by the integral membrane protein ComGB in B. subtilis (Dubnau and Blokesch, 2019). Although neither comGC ortholog nor comGB ortholog was identified in P. polymyxa, two genes (tadB and tadE), which potentially encode the Flp pilus, were found in its genome. In natural transformation of Micrococcus luteus, Flp pilus was proposed to function as the competence pili (Angelov et al., 2015). The helicase ComFA may pull ssDNA into the cytoplasm on the expense of ATP in B. subtilis (Dubnau and Blokesch, 2019). However, no comFA ortholog was identified in P. polymyxa. In the cytoplasm, the incoming ssDNA is protected by the widely conserved ssDNA-binding protein DprA/Smf (Mortier-Barriere et al., 2007). Interestingly, a dprA/smf homolog was found in P. polymyxa ATCC 842, but absent in P. polymyxa SC2.

Although DNA uptake machinery is well conserved in bacteria, the mechanisms of competence induction for natural transformation are extremely diverse (Aas et al., 2002; Ogura et al., 2002; Claverys et al., 2006). In gram-positive bacteria, the pheromone stimulates the two component system which activates the transcription regulator for expressing late competence genes (including DNA uptake and processing genes) (Ogura et al., 2002; Claverys et al., 2006). Because pheromones and proteins involved in the competence signal transmission are often species specific, conditions for inducing natural transformation varies among different bacterial species (Ogura et al., 2002; Claverys et al., 2006; Shanker et al., 2016). It would be interesting to explore the condition for inducing natural transformation of P. polymyxa in the future study. Spontaneous plasmid transformation and cell-to-cell plasmid transfer without the assistance of DNA uptake machinery and conjugative machinery have been repeatedly documented in bacteria (e.g., E. coli) (Sun et al., 2006, 2009; Wang et al., 2007; Sun, 2016, 2018; Hasegawa et al., 2018). It is unclear whether these unconventional DNA transfer mechanisms are involved in the evolution of the genome and the plasmid in P. polymyxa.

In conclusion, we first analyzed the genome of P. polymyxa with SMRT sequencing, uncovering frequent HGT events in the genome and the presence of an originally unappreciated plasmid. Next, we identified a pair of reverse complementary DNA methylation motif and genes encoding their cognate RM system PpoAI in the genome. Genetic analysis confirmed a role of PpoAI in protecting P. polymyxa against extracellular DNA. Furthermore, we found that the frequency of recognition motifs in MGEs was lower than the average level, implicating that PpoAI influenced incorporation of extracellular DNA into the genome of P. polymyxa. Wide distribution of homologs of component of PpoAI in different classes of the phylum Firmicute implicates important role of its ancestor in the evolution of genomes of other species. Taken together, our work would not only reveal the impact of the RM system on bacterial genome evolution but also establish a model strain of Paenibacillus for gene function analysis and cell factory construction.

Genomic DNA of P. polymyxa was isolated with the bacterial genomic DNA isolation kit (Takara Biotech Co., Ltd.) and sequenced and assembled according to PacBio guidelines (refer to Supplementary Materials for the detail). DNA methylation was detected by using the RS_Modification_and_Motif_Analysis protocol within SMRT Portal v2.30, with a standardized in silico false-positive error < 1%. The motifs with a mean modification quality value (QV) higher than 50 and a mean coverage of > 100 × were validated as being modified³. The consensus sequences of all motifs were depicted as sequence logos that were obtained by the WebLogo 3 server⁴ (Crooks et al., 2004). Antibiotic resistance genes were predicted with ARDB⁵ (Liu and Pop, 2009). RM systems and the DNA motifs with methylation were analyzed with Rebase⁶ (Roberts et al., 2015). The insertion sequence was predicted with the ISsaga online server⁷ (Varani et al., 2011). Prophages were predicted with an online phage search tool (PHAST)⁸ (Zhou et al., 2011; Arndt et al., 2019). CRISPR finder⁹ was used for identifying CRISPRs (Zhu et al., 2016). Genomic islands (GIs) were predicted with Island Viewer¹⁰ (Bertelli et al., 2017). Structural regions of the S subunit of PpoAI were predicted with the online tool Uniprot¹¹.

All strains, plasmids, and primers used in this study were listed in Table 8. E. coli and P. polymyxa were cultured in Luria-Bertani medium containing 0.5% (w v^–1) yeast extract, 1% (w v^–1) typtone, and 1% (w v^–1) NaCl or on LB agar plates supplemented with or without appropriate antibiotics. The 10 × Spizizen minimal medium was prepared as follows: 4% (w v^–1) (NH4)₂ SO₄, 18.3% (w v^–1) K₂HPO₄⋅3H₂O, 6% (w v^–1), KH₂PO₄, 1% (w v^–1), Na₃C₆H₅O₇⋅2H₂O, and 0.2% (w v^–1) MgSO₄⋅7H₂O. B. subtilis was cultured in GM1 containing 0.8% (w v^–1) glucose, 0.04% casein hydrolysate and 0.1% yeast extract supplemented with 10% (v v^–1) of the 10 × Spizizen minimal medium, or GM2 medium containing 0.8% (w v^–1) glucose and 0.02% casein hydrolysate supplemented with 10% (v v^–1) of the 10 × Spizizen minimal medium, or on LB agar plates supplemented with appropriate antibiotics. All strains were incubated with shaking at a speed of 180 rpm under 37°C or 30°C. Plasmids were isolated from bacteria with the SanPrep column plasmid mini-preps kit (Axygen Biotech Co., Ltd.). To isolate plasmid from B. subtilis and P. polymyxa, lysozyme (4 mg/ml) was added to the cell resuspension solution and incubated at 37°C for 1 h.

Plasmid transformation of E. coli was performed according to the documented chemical transformation method (Cohen et al., 1972; Sun et al., 2013). When the cell culture was grown in LB medium, to an OD₆₀₀ of 0.4, cell pellets were collected by centrifugation, washed with 100 mM CaCl₂ solution twice, and resuspended in 100 mM CaCl₂ solution on ice. Plasmid was transferred into the competent E. coli cell with heat shock for 90 s, and transformants were screened on LB plates supplemented with ampicillin (100 μg/ml).

Plasmid transformation of B. subtilis was performed by using the Spizizen’s method (Anagnostopoulos and Spizizen, 1961). A single colony of B. subtilis was inoculated into 5 ml GM1 medium. The overnight grown culture was transferred to 5 ml of fresh GM1 at an inoculum of 0.2% (v v^–1). The bacterial culture was further grown to the end of exponential stage (1∼2 h) before being transferred to 5 ml of the GM2 medium. After 1∼2 h of incubation in GM2 supplemented with or without 1% xylose, plasmid DNA was added to the cell culture before screening on plates supplemented with kanamycin (50 μg/ml) or erythromycin (25 μg/ml).

Plasmid transformation of P. polymyxa was performed by using our previously established method (Shen et al., 2018). A fresh single colony of P. polymyxa was inoculated into LB medium and incubated at 37°C overnight. The overnight grown culture was inoculated into 50 ml of fresh LB supplemented with 20 mM sorbitol. Exponentially growing cells were collected by centrifugation and washed with the precooled solution A three times, followed by resuspension in the solution B. Plasmid isolated from either B. subtilis or P. polymyxa was transformed into 50 μl aliquots of electrocompetent cells. Transformation efficiency was calculated as the number of transformants per μg of plasmid DNA. Refer to the Supplementary Materials for more details about plasmid transformation of P. polymyxa.

The temperature-sensitive plasmid pRN5101 was previously used in deleting chromosomal gene in P. polymyxa SC2 (Zhang et al., 2013). A derivative of pRN5101 (pRN5101-1) for deleting genes encoding PpoAI was constructed by inserting the DNA fragment containing a kanamycin resistance gene flanked by upstream and downstream of genes encoding PpoAI into the vector by using the one-step cloning kit (Vazyme Biotech Co., Ltd.). To examine transformability of P. polymyxa, pWBUC02 conferring erythromycin resistance was constructed by ligating the DNA fragment containing an erythromycin resistance gene from pRN5101 with linearized and egfp-deleted plasmid pWBUC-egfp. Refer to Supplementary Materials for details about construction of pRN5101-1 and pWBUC02. To construct the ppoAI deletion mutant, pRN5101-1 was isolated from E. coli DH5α and then transformed into B. subtilis with the Spizzen’s method as described. pRN5101-1 from B. subtilis was then electroporated into P. polymyxa and screened on LB-agar plates supplemented with erythromycin (25 μg/ml). Transformants were cultured with shaking at a speed of 180 rpm in LB supplemented with kanamycin (50 μg/ml) at 37°C. Every 12 h, 1% (v v^–1) of the culture was inoculated into fresh LB medium containing kanamycin (50 μg/ml) and incubated with shaking at 37°C. After three rounds of inoculations, the serial diluted cell culture was spread on LB agar plates supplemented with kanamycin (50 μg/ml), followed by screening for colonies unable to grow on LB plates containing erythromycin (25 μg/ml) at 30°C. Kanamycin-resistant erythromycin-sensitive colonies were selected and examined through colony PCR with the primer pair P11–P12.