Short genome report of cellulose-producing commensal Escherichia coli 1094

Bacterial surface colonization and biofilm formation often rely on the production of an extracellular polymeric matrix that mediates cell-cell and cell-surface contacts. In Escherichia coli and many Betaproteobacteria and Gammaproteobacteria cellulose is often the main component of the extracellular matrix. Here we report the complete genome sequence of the cellulose producing strain E. coli 1094 and compare it with five other closely related genomes within E. coli phylogenetic group A. We present a comparative analysis of the regions encoding genes responsible for cellulose biosynthesis and discuss the changes that could have led to the loss of this important adaptive advantage in several E. coli strains. Data deposition: The annotated genome sequence has been deposited at the European Nucleotide Archive under the accession number PRJEB21000.


Introduction
Biofilms are ubiquitous microbial communities growing in close association with surfaces present in natural and anthropic environments [1]. Biofilm bacteria often selfassemble by producing a cohesive extracellular matrix that protects these multicellular aggregates against environmental changes and maintains the integrity of the biofilm structure [2]. Cellulose is a homo-polysaccharide composed of glucose units linked by β-1 → 4 glycosidic bonds and is the most common organic compound on Earth [3]. Initially studied in Gluconacetobacter xylinus, cellulose production is a widespread phenomenon shared by many commensal and pathogenic Betaproteobacteria and Gammaproteobacteria, including Salmonella enterica serovar Typhimurium, Klebsiella pneumoniae, Burkholderia mallei, Shigella boydii, Yersinia enterocolitica, Vibrio fischeri, Pseudomonas putida and many Escherichia coli strains [4,5]. Here we report the complete genome sequence of E. coli 1094, a biofilm forming and cellulose-producing strain isolated from the feces of a healthy human male. E. coli 1094 lacks virulence factors commonly associated with pathogenic E. coli and is a member of E. coli phylogenetic group A [6,7].

Classification and features
Escherichia coli is a Gram-negative, rod-shaped, non-spore forming and facultative anaerobic species belonging to the Enterobacteriaceae family. They are commonly found in the intestines of endotherms and are taxonomically placed within the Gammaproteobacteria of the Proteobacteria phylum (Table 1).
E. coli 1094 is a commensal strain isolated from feces of a healthy human male. Like many other natural E. coli isolates, 1094 produces cellulose as the main component of its biofilm extracellular polymeric matrix ( Fig. 1) [8], and has been used as a model for studying both transcriptional regulation and function of cellulose biosynthesis genes (bcs genes) [8,9], as well as to analyze the structure of the cellulose secretion machinery [10]. Here, we investigate global genomic differences between representative members of E. coli phylogenetic group A and discuss their functional consequences on cellulose production.
E. coli 1094 was previously classified as a member of the phylogenetic group A based on triplex PCR with a combination of primers amplifying two genes (chuA and yjaA), and an anonymous DNA fragment designated TSPE4.C2 [7]. Using all available complete genomes of phylogenetic group A, we performed a core genome alignment to produce a phylogenetic tree, and found it to be in agreement with previous phylogenetic classification [11], showing that E. coli strain 1094 is closely related to strains ATCC 8739 and HS (Fig. 2).

Genome sequencing information
Genome project history E. coli 1094 has been used as a model to study different aspects of cellulose biosynthesis and biofilm formation [8,9]. Cellulose production requires the expression of bacterial cellulose synthesis genes clustered in two divergent operons, as well as genes involved in general glucose metabolism [8], In order to further elucidate the genetic bases of cellulose synthesis, we chose to sequence E. coli 1094 using two approaches: Illumina and PacBio sequencing. While Illumina sequencing and subsequent downstream analysis generated 204 contigs, PacBio sequencing and assembly produced 4 contigs. A summary of the project information and its association with "Minimum Information about a Genome Sequence" [12] are provided in Table 2.
Growth conditions and genomic DNA preparation E. coli 1094 was cultivated in LB medium overnight at 37°C. High quality genomic DNA for sequencing was extracted using the Wizard Genomic DNA Kit (Promega) (for Illlumina approach), or the QiaAMP DNA Mini Kit (QIAGEN) (for PacBio approach).

PacBio sequencing
Library preparation, sequencing, and assembly were performed by the Earlham Institute. PacBio sequencing libraries were prepared from 10 μg DNA using standard Pacific Biosciences protocols (Pacific Biosciences, Menlo Park, CA). Following construction, libraries were size selected, from 7 to 50 kb, using the Sage Science BluePippin™ system with a 0.75% gel cassette. Libraries were run on the Pacific Biosciences RSII sequencer at 350pM using P6/C4 chemistry. A Single Molecule Real Time (SMRT) cell was used, yielding 150,292 reads, and 1213 megabases of sequence data. Reads were assembled using PacBio's hierarchical genome-assembly process (HGAP.3), with filtering and adaptor trimming performed as previously described [13]. The minimum seed read length used was 6 kb, with a quality score cutoff of 0.8. The Celera Assembler was used to produce 4 large contigs, using pre-assembled, errorcorrected reads. The maximum contig length produced was 4,903,991 bases.

Illumina & PacBio comparison
Illumina single-end reads were mapped against the four large contigs generated by PacBio reads using the singleend mode of Bwa mem v0.7.4 [14] with default parameters (Fig. 3). Output SAM files were converted to BAM files using SAMtools v0.1.19 [15]. Sequencing coverage was computed using BEDtools v2.17.0 0 [16] and values were normalized to 1 (genomes are haploids), from the median coverage over the large contigs (354×). The mapping coverage along the four PacBio contigs validated the sequence assemblies. Some peaks of high coverage are observed in unitig_0, unitig_1 and unitig_2 ( Fig. 3a), which suggests that multiple copies of these regions exist. By contrast, the coverage of the complete unitig_3 indicates that there is likely more than one copy per chromosome within each cell.
SPAdes v 3.9.1 [17,18] was used for assembling Illumina reads, for detecting putative plasmids sequence with the options '-plasmid' , '-k 21,51,71' , 'cut-off auto' and '-careful'. Illumina reads were re-mapped onto the contigs using the single-end mode of Bwa mem v0.7.4 with default parameters. After converting output SAM files to BAM files by SAMtools, putative plasmid reads were extracted using  SAMtools (option 'view -F 4') and recorded in a Fastq file by picardtools 'SamToFastq' [19]. Putative plasmid reads were mapped against the four large PacBio contigs using the single-end mode of Bwa mem v0.7.4 with default parameters. The coverage computed by BEDtools (Fig. 3b) indicates that the complete unitig_3 is classified as a putative plasmid sequence and also appears to exist in high copy number. For unitig_0, unitig_1 and unitig_2, the mapping coverage shows some portions of large contigs classified as putative plasmids. This may correspond to plasmids with similar coverage to the chromosome, due to low copy number, or to misclassification by plasmidSPAdes.

Sequence circularization
PacBio scaffold sequences were compared against themselves with the bl2seq BLASTN algorithm [20], and ACT [21] was used for synteny visualization. The resulting overlapping sequences were easily identified between the beginning and the end of each large contigs, suggesting that all four PacBio large contigs are circular. To determine the size of the chromosome and each plasmid, the size of the overlapping region (Unitig_0: 15,155; Unitig_1: 8,919; Unitig_2: 13,971; Unitig_3: 9,380) was subtracted from the length of each contig (Unitig_0: 4,903,991; Unitig_1: 123,705; Unitig_ 2: 118,720; Unitig_3: 30,364); the final sizes are reported in Table 3.

Genome annotation
The complete genome sequence was annotated with the RAST server [22] which predicted 5028 coding sequences and 110 RNAs.

Genome properties
A summary of the genome of E. coli 1094 is included in Table 3. The genome statistics are provided in Table 4.
Three putative plasmids were identified, and found to be circular. The genome of strain 1094 has a total length of 5,176,780 base pairs and a G + C content of 50.9%. The majority of the protein-coding genes were assigned a putative function (78.8%) while the remaining ones were annotated as hypothetical proteins. Genes in internal clusters were detected using BLASTclust with thresholds of 70% covered length and 30% sequence identity [23]. CRISPR, transmembrane helice, signalP and Pfam protein families predictions were done using CRISPRFinder [24], TMHMM Server v.2.0 [25], SignalP 4.0 [26] and Pfam 29.0 [27], respectively. The distribution of genes into COGs functional categories is presented in Table 5.
Insights from the genome sequence E. coli 1094 sequence was aligned against selected genomes belonging to E. coliphylogenetic group A. Alignment and posterior NJ (neighbor joining) phylogenetic reconstruction was carried out with Parsnp, a fast core-genome multialigner, using default parameters [28]. Alignment and tree visualization was done with Gingr, a dynamic visual platform (Fig. 2) [28].
We selected five E. coli genomes that are representative of the multiple clades within phylogenetic group A, and performed orthologous clustering to examine the genomic differences of E. coli 1094. [E. coli K12 W3110 (accession n°: AP009048), E. coli BL21-Gold(DE3)pLysS AG' (accession CP001665), E. coli HS (accession n°: CP000802), E. coli ATCC 8739 (accession n°: CP000946) and the prototypical enterotoxigenic strain of E. coli ETEC H10407 (accession FN649414)] (Fig. 4). Our analysis reveals a core genome of 3409 genes shared among all strains (Fig. 4). E. coli ETEC H10407 possesses the highest number of specific genes (834), followed by E. coli 1094 (499), E. coli The analysis of the location of genes only present in E. coli 1094 identified eight 1094-specific regions, most of which are prophage or phage-associated proteins; two regions represent a putative complete prophage, containing genes encoding the phage capsid tail and replication proteins (Fig. 4). Other 1094-specific regions of interest contain CRISPR associated proteins. We also identified clusters only present in some, but not all E. coli strains  analyzed, including (i) a putative complete type VI secretion system (present in 1094 and HS), (ii) a type III secretion system cluster (absent in BL21-Gold(DE3) and K12 W3110 strains of E. coli) (iii) a cluster of genes encoding invasins and an iron acquisition system, (present in 1094 and H10407), and (iv) several cellulose biosynthesis genes present in all strains, except for E. coli HS. We performed a basic BLAST (BLASTN 2.6.1+, [29]) of each smaller contig identified, and report the following results. Unitig_1 shows sequence homology to Salmonella enterica subsp. enterica serovar Senftenberg strain 775 W plasmid pSSE-ATCC-43845 (Accession: CP016838.1), with 33% of the contig showing 99% identity. Unitig_2 displays sequence homology to Klebsiella pneumoniae subsp. pneumoniae strain 234-12 plasmid pKpn23412-362, with 39% of the contig showing 99% identity (Accession: CP011314. 1); other regions of the contig show homology to multiple E. coli strains and plasmids. Unitig_3 displays sequence homology to Citrobacter freundii strain B38 plasmid pOZ182; 68% of the contig shows 96% identity (Accession: CP016765.1); in addition, two separate regions of 7.5 and 2. 1 kb are highly homologous to E. coli plasmid pV004-a DNA and pV001-a DNA (Accession: LC056140.1 and

Not in COGs
The total is based on the total number of protein coding genes in the genome LC056078.1). Taken together, this suggests that 1094 expresses 3 distinct circular plasmids.

Comparative analysis of the bcs (bacterial cellulose synthesis) region
We compared the region corresponding to the E. coli 1094 bcs operon with corresponding regions in the strains E. coli W3110, HS, BL21-Gold(DE3)pLysS AG' , ATCC 8739, and H10407, which are representative of phylogenetic group A. This analysis shows that whereas only partial fragments of the external genes of the bcs operon (bcsC and bcsG) exist in E. coli HS, the five other strains analyzed contain all genes within the bcs operon (Fig. 5). K12 derivative strains do not produce cellulose, as they contain a premature stop codon in the gene bcsQ, due to a single nucleotide polymorphism (SNP) in the region TTG/TAG (17 T > A) (Fig. 5) [30]. Serra et al., repaired this SNP, which resulted in cellulose production in E. coli K12 W3110, suggesting that the premature stop codon in bcsQ could either affect the function of BcsQ, or has a polar effect on neighboring genes [30]. The other SNPs (relative to E. coli 1094) observed in the bcs operons of E. coli K-12 strains result either in synonymous codons or in conservative amino acid exchanges. By contrast, comparison of the bcs operon sequence between E. coli 1094 and E. coli ATCC8739 revealed 100% sequence identity between bcsQ, bcsR, bcsA and bcsB genes. Finally, the sequence of the bcs operon in E. coli ATCC8739 contains one SNP in bcsE that leads to a premature stop codon, and may also have a polar effect on bcsF and bcsG, which are essential for cellulose production [8] (Fig. 5). E. coli strains BL21-Gold(DE3)pLysS AG' and H10407 contain multiple SNPs in bcs genes that are essential for cellulose biosynthesis. Some of these SNPs lead to amino acid changes that could negatively affect the function of these proteins. bcsA is particularly interesting, as a pairwise genetic comparison in all six E. coli strains analyzed revealed the existence of multiple bcsA SNPs (between 21 and 29, depending on the strain pair compared). None of the SNPs resulted in amino acid exchange or a premature stop codon, suggesting a strong selection pressure for this specific amino acid composition. Our phylogenetic analysis of the phylogenetic group A allowed us to localize these evolutionary events, by following a parsimony criterion, which allowed us to infer that mutations leading to premature stop codons took place in two separate evolutionary events: the branch of the strain ATCC8739 and in the K12 common ancestor (Fig. 2). These mutations resulted in sequence interruption of bcsE and bcsQ respectively.
Cellulose production of E. coli strains Cellulose production was tested in the six E. coli strains analyzed in this study. We monitored colony fluorescence on LB-calcofluor plates, which is a common assay used to detect cellulose biosynthesis [31] (Fig. 6). This analysis revealed that only E. coli 1094 strain produced detectable levels of cellulose. As was previously shown, laboratory E. coli K12 strains do not produce cellulose The gray color code represents the pairwise Blast matches between one strain, and the strain directly below it; darkness corresponds to higher similarity between the two strains, and strains were ordered based on phylogenetic relatedness. The star in red represents a SNP leading to a premature stop codon. The figure was generated using the Easyfig software [41] Fig. 6 Cellulose production in six E. coli strains. Cellulose secretion was evaluated by the binding of calcofluor (CF), as visualized by colony fluorescence under ultraviolet light (fluorescence appears white in image). Five microliters of an overnight culture was spotted onto LB agar plates, containing 0.02% of CF and incubated for 24 h at 30°C. Images were acquired using a G:BOX imaging system (Syngene, Frederick, MD) [30]. Here we show that, as expected, E. coli HS is not able to produce this polymer, as it was known to lack all bcs genes. Interestingly, cellulose production was not detected in E. coli ATCC 8739, suggesting that the SNP in the region TGT/TGA (303 T > A) of bcsE described above could have a polar effect on the downstream genes. While the bcsEFG operon is essential for cellulose production [8], bcsE is only necessary for maximal cellulose production [32], suggesting that the inability to synthetize detectable levels of cellulose may be due to a polar effect of the bcsEFG operon. Finally, cellulose production was not detected in strains BL21-Gold(DE3) pLysS AG' and H10407, which could be explained by the presence of multiple amino acid changes that potentially affect the biological activity of BcsB, BcsC and BcsQ proteins, which are essential for cellulose production.

Conclusions
The human commensal intestinal isolate Escherichia coli strain 1094 naturally produces cellulose, a polysaccharide known to contribute to adhesion and biofilm development. We compared genomic content and cellulose production of closely related E. coli strains in phylogenetic group A, and found 1094 to be the sole strain to produce measureable levels of cellulose, and conclude that these strains lack the capacity to produce cellulose due to one or several SNPs in cellulose biosynthesis genes: non-synonymous SNPs in bcsB, bcsC, bcsQ, and a nonsense mutation in bcsE. The genome sequencing and annotation here provides valuable information for future study of the regulation of the bcs genes and cellulose production.