Correction: Unravelling the Transcriptome Profile of the Swine Respiratory Tract Mycoplasmas

Table S2. Transcriptome mapping from Mycoplasma flocculare. Copyright: ß 2014 The PLOS ONE Staff. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.


Introduction
Mycoplasma hyopneumoniae, Mycoplasma flocculare and Mycoplasma hyorhinis are the most important species that have been identified in the swine respiratory system [1,2]. M. hyopneumoniae is the etiological agent of porcine enzootic pneumonia, whereas M. hyorhinis, in addition to being present in swine respiratory tracts, may cause swine serofibrinous to fibrinopurulent polyserositis and arthritis [3]. M. flocculare is also widespread in swine herds, but no disease has been associated with this species [4]. Based on a 16S rRNA sequence and genome comparison, it has been shown that M. hyopneumoniae and M. flocculare are phylogenetically closely related species [5,6]. M. hyopneumoniae can adhere to the cilia of tracheal epithelial cells and cause damage. Although M. flocculare can also adhere to cilia, no important resulting damage has been observed, suggesting that M. hyopneumoniae and M. flocculare may possess different mechanisms of action involving the recognition of different receptor sites on the cilia [7]. In addition to the ability of M. hyorhinis to colonize other host sites [4], its potential role in human cancer development and acceleration to malignant phenotypes has been proposed [8][9][10].
Similar to other mycoplasmas, M. hyopneumoniae, M. flocculare and M. hyorhinis possess small genomes with limited biosynthetic potentials. Six M. hyopneumoniae genome strains have been sequenced [6,[11][12][13][14] in addition to four M. hyorhinis strains [10,[15][16][17] and one M. flocculare strain [6]. Although a significant amount of data has been produced by genome sequencing and comparative analyses in these species, very limited information related to transcriptional mechanisms and regulation is available for these organisms.
Analyses related to transcriptional unit (TU) organization, transcriptional regulation and promoter sequence composition have been performed on the M. hyopneumoniae 7448 genome [18][19][20]. Global assessments of the TU organization of the M. hyopneumoniae genome by both in silico and in vitro approaches have suggested that the ORFs are continuously transcribed (cotranscribed) in large clusters (TUs). The authors have predicted that each TU is transcribed in the same direction with no intervening gene transcribed in the opposite orientation [18]. Similar predicted TU organization to that of the M. hyopneumoniae genome have also been reported in the M. flocculare and M.
hyorhinis genomes [6] in addition to that of Mycoplasma pneumoniae [21]. The advancement of knowledge of the whole transcriptome organization should contribute to the understanding of the unexplored mechanisms of mycoplasma transcriptional regulation.
Considering the importance of the three Mycoplasma species that inhabit the swine respiratory tract and the limited knowledge available regarding mycoplasma transcriptional regulation, we aimed to analyze the whole transcriptomes of M. hyopneumoniae, M. flocculare and M. hyorhinis. In this work, we have sequenced, analyzed and constructed single nucleotide resolution transcriptome maps for the three species. A comparative analysis of the whole gene expression profiles between these swine respiratory mycoplasmas was also performed. Furthermore, the organizations of the genomes into large TUs (polycistronic mRNA), as previously predicted, were estimated and compared between the three species.

Materials and Methods
Bacterial strains, culture conditions and RNA preparation M. hyopneumoniae strain 7448, M. flocculare ATCC 27716 and M. hyorhinis ATCC 17981 were used in this study. The bacteria were grown in 25 ml Friis broth [22] at 37uC for 24 hours with gentle agitation in a roller drum.
Total RNA was isolated with the RNeasy Mini Kit (Qiagen, USA). For cell lysis, 0.7 ml of RNeasy Lysis Buffer (RLT buffer) in 0.134 M b-mercaptoethanol was used per cultivation flask. The purification was performed according to the manufacturer's instructions, with on-column DNase I digestion using the RNase-Free DNase Set (Qiagen, USA) and a second round of treatment with DNase I (Fermentas, USA). The absence of DNA in the RNA preparations was monitored by PCR. The extracted RNA was analyzed by gel electrophoresis and quantified using the Qubit system (Invitrogen, USA). RNA quality and integrity were determined by the evaluation of the RNA Integrity Number (RIN) using the Agilent 2100 Bioanalyzer (Agilent, USA). Values that were greater than or equal to 9.5 indicated sufficient quality.
The ribosomal RNA (rRNA) depletion was performed using Terminator 59-phosphate-dependent exonuclease (TEX -Epicentre, USA), according to the manufacturer's instructions. Briefly, total RNA (10 mg) was combined with TEX 10X Reaction Buffer A, 20 U RiboGuard RNase Inhibitor, 10 U TEX (1 U/mg mRNA) and nuclease-free water in a final volume of 40 ml and incubated at 30uC for 60 minutes. The reaction was stopped with EDTA (0.5 M), and the mRNA was precipitated with 3 M sodium acetate and 2.5 volumes of ethanol. The mRNA was quantified using the Qubit system (Invitrogen, USA), and the effectiveness of the reaction was assessed by the Agilent 2100 Bioanalyzer (Agilent, USA). The absence of 16S and 23S rRNA in the posttreatment sample indicated a successful reaction.

cDNA preparation and sequencing
The cDNA library preparation and pyrosequencing were performed using GS-FLX Titanium series reagents following the manufacturer's instructions (Roche Diagnostics, Germany). An equal amount of mRNA (200 ng) from each strain was used to synthesize the first and second strands of the cDNA, according to the cDNA Rapid Library Preparation Method Manual GS FLX Titanium Series (Roche Diagnostics, Germany). After the library construction, the samples were quantified using the Qubit system (Invitrogen, USA), and average fragment sizes were determined using the Agilent 2100 Bioanalyzer (Agilent, USA). Three cDNA libraries for each strain were generated, totaling nine libraries, which were sequenced using the Roche/454 GS-FLX system. Following the analyses of the reads, each of the triplicate sets of libraries was pooled.

Assembly and mapping
For the transcriptome assembly, Newbler v2.8 was used with the default parameters. The FastQC software (http://www. bioinformatics.babraham.ac.uk/projects/fastqc/) was used for the read quality assessments, and the FASTX-Toolkit v0.0.13 (http://hannonlab.cshl.edu/fastx_toolkit/) was employed for trimming. Reads were aligned with Bowtie2 v2.1.0 [23] using the ''sensitive'' default option. The resulting alignment files were filtered with SAMtools v0.1.19 [24] to assess mapping quality (MQ; . = 10). BEDtools v2.17.0 [25] ''multicov'' was used to count reads mapped to specific genomic regions. Finally, matches were counted and the reads per kilobase per million (RPKM) computed [26]. Raw and normalized counts are available for each gene. To map the 454 reads, the M. hyopneumoniae 7448, M. flocculare ATCC 27716 and M. hyorhinis HUB-1 genomic sequences were used as references, and the rRNA sequences were filtered. The results were parsed, and GenBank (gb) format files were generated for analysis with the Artemis Release 10.5.2 software package [27], which included information about the coverage of each region and its respective gene identities. Circular genome plots were generated with DNAPlotter [28].
The Standard Flowgram Format (SFF) files of the sequence data generated in this study have been deposited in the Short Read Archive (SRA) database at NCBI under the accession number PRJNA255516.
The operon assessment was based on published experimental data from M. hyopneumoniae 7448 [18]. Based on this study, the M. flocculare and M. hyorhinis TUs were predicted [6]. Thus, using the RNA-Seq results and in-house Perl scripts, we analyzed the predicted TUs for the three mycoplasma species, accounting for the following criteria: i) the previous prediction of the structure by Siqueira et al. [18] and Siqueira et al. [6] ii) the expression of all genes; iii) the transcription of the genes in the same direction; and iv) the expression of the intergenic regions between the genes. Overlapping pairs of such genes were joined together to identify large operon structures.

Results and Discussion
cDNA sequencing, assembly and mapping of M. hyopneumoniae, M. flocculare and M. hyorhinis transcriptomes A total of 1,204,156 raw sequencing reads with an average length of 274 bp were generated using the Roche GS FLX Sequencing Platform. After cleaning and removing the adapters from each sequence in addition to low-quality sequences (quality scores ,20) and rRNA sequences, a total of 683,385 high-quality reads were obtained and used for the de novo assembly. The cDNA libraries were constructed with total mRNA from M. hyopneumoniae 7448, M. flocculare ATCC 27716 and M. hyorhinis ATCC 17981. The RNA preparations from the three species were obtained from bacteria grown under the same culture conditions. Table 1 shows a detailed comparative analysis of each library assembly. To determine the transcribed regions in the genomes, we estimated the average coverage depths of the reads mapped per nucleotide/base. We used a pileup format, which produces a signal map file for the whole genome in which the alignment results (coverage depth) are represented reported per base. The resulting reads were mapped to the M. hyopneumoniae 7448, M. flocculare ATCC 27716 and M. hyorhinis ATCC 17981 genomes. The sequence coverage per base was subsequently plotted and visualized using the genome browsers Artemis and DNAplotter [27][28] (Figure S1).
The M. hyopneumoniae library allowed for the generation of 415,265 reads, of which approximately 94% mapped to the reference genome sequences of M. hyopneumoniae strain 7448. Most of the 695,313 reads (93%) from the M. flocculare library also mapped to the reference genome sequences of M. flocculare strain ATCC 27716. Similarly, 86% of the 93,578 reads from the M. hyorhinis library mapped to the reference genome sequences of M. hyorhinis strain HUB-1. The mapped M. hyopneumoniae, M. flocculare and M. hyorhinis genes represented 92%, 98% and 96% of the total predicted ORFs in these genomes, respectively ( Table 1). The few unmapped genes were mainly identified as hypothetical genes with unknown functions (over 60%, considering the three transcriptome mappings). Full lists of the transcribed genes in M. hyopneumoniae, M. flocculare and M. hyorhinis under the culture conditions used in this study are presented as supplementary data (Table S1; Table S2; Table S3, respectively). Therefore, our results showed that the majority of genes were transcribed, demonstrating whole basal-level expression profiles for the mycoplasmas of the swine respiratory tract ( Figure S1). The maps shown in Figure S1 represents the genomes (gray circles) and transcriptome maps (green circles) of M. hyopneumoniae, M. flocculare and M. hyorhinis, respectively, enabling the visualization of the high levels of gene coverage achieved by the cDNA assembly and mapping of the three Mycoplasma genomes.
As shown in Table 1, the total contig numbers were similar for the three libraries, and the coverages ranged from 1 to over 2,000 reads per contig, with most (62%) of the contigs covered by at least 15 reads. The contig length distributions were similar for the M. hyopneumoniae, M. flocculare and M. hyorhinis contigs ( Figure  S2). The majority of the contigs (approximately 25%) ranged in size from 200-500 bp, but many possessed lengths ranging from 1,000-3,000 bp (approximately 15%) ( Figure S2).
Genome annotation refinement is an important aspect of a transcriptome analysis. The transcriptome map of M. flocculare allowed for the identification of five novel protein-coding genes that were not present in the initial annotation. The BLASTX search showed that they were homologous (similarities of 89% or more, sequence coverages of 100%) to genes from M. hyopneumoniae and other mycoplasma species. Three of these genes encoded hypothetical proteins, with the gene IDs MHP7448_0560, MHP7448_0709 and MHP7448_0138 in the M. hyopneumoniae 7448 genome, to which 127, 10 and 146 reads, respectively, were mapped in the M. flocculare transcriptome. Another novel identified gene encoded a lipoprotein with the gene ID MHP7448_0366 in M. hyopneumoniae, to which 31 reads were mapped in the M. flocculare transcriptome. A gene encoding a type III restriction-modification system: DNA methylase with the gene ID MHP7448_0386 in M. hyopneumoniae was also identified, to which 4 reads were mapped in the M. flocculare transcriptome.
According to the current annotation of the M. flocculare genome, the ftsZ gene is located in contig 04 and is positioned from nucleotides 8844-9491, which is located 647 bp downstream of a hypothetical gene (MF01093) in the opposite strand. However, the transcriptome map constructed in this study allowed for the identification of the incorrect annotation of the start codon of the ftsZ gene. The gene MF01093, with a highly expressed level (RPKM = 139,138.8) (see Table S2), is in fact a part of the ftsZ gene. Thus, the ftsZ gene is positioned from nucleotides 8510-9505 (995 bp in length) in contig 04. This observation was confirmed by BLASTX comparisons against homologous genes in other Mycoplasma sp.
The transcriptome map of M. hyorhinis allowed for the identification of two novel genes. The first encoded a hypothetical protein, HUB-1, which was positioned from nucleotides 721079-721572 (with 12 associated reads). BLAST comparisons indicated that this gene shared 98% identity (with an e-value of 2e-85) with MYM_0631, which is a hypothetical protein that is positioned from nucleotides 719192-719647 in the M. hyorhinis GDL-1 genome and totals 453 bp in size. The second non-annotated gene was rnpB, to which 7,876 reads were mapped. The BLASTX analysis showed that it shared homologies (similarities of 98% or more, sequence coverages of 100%) with the genomes of other M. hyorhinis strains and other Mycoplasma species. Additionally, in this work, we detected the endoribonuclease P (RNase P) ncRNA in the M. hyorhinis HUB-1 genome. RNase P, which is encoded by the rnpB gene, is a ribonucleoprotein complex that removes 59 leader sequences from tRNA precursors during tRNA biosynthesis. RNase P is present in all cells and subcellular compartments that synthesize tRNA, but catalytic activity by the RNA alone has been demonstrated only for bacterial RNase P RNA. Bacterial RNase P RNAs have been separated into two main structural classes. Type A (rnpA) is the most common structural class, and type B (rnpB) is found in low G+C gram-positive bacteria [29]. Mycoplasma sp. genomes typically contain both classes of RNase P.
The availabilities of transcriptome maps for M. hyopneumoniae, M. flocculare and M. hyorhinis now allow for the detection of novel transcripts that have not been previously annotated in these genomes. Such transcripts include noncoding RNAs (ncRNAs), which were shown in other microorganisms to regulate important processes, such as pathogenesis, iron metabolism, and quorum sensing [30][31][32]. However, intergenic regions within TUs could represent portions of long expressed transcripts and be misclassified as ncRNAs. Therefore, these regions were excluded from this analysis. We manually analyzed the expressed intergenic regions located between the TUs to find evidence of novel ncRNAs. We identified 78, 130 and 72 putative novel ncRNAs in the M. hyopneumoniae, M. flocculare and M. hyorhinis transcriptomes, respectively (Table S4). Indeed, the BLASTX search provided evidence that these transcripts are ncRNAs. However, we believe that some of the putative ncRNAs that was detected in M. flocculare was possibly an artifact because this genome was not completely assembled.
Although it is known that bacteria express sRNAs that are 50-500 nucleotides in length [33], we examined all reads with lengths ranging from 30-600 nucleotides (Table S4). The distributions of ncRNA lengths were very similar in the three organisms, and the majority of ncRNAs presented with lengths of between 51 and 200 nucleotides ( Figure 1). This length variation may indicate the diversities of these elements, which may be related to RNA-based regulatory strategies. The putative ncRNAs detected here can represent some antisense RNA (asRNAs), but our approach was not designed to detect asRNAs. Further studies involving the determination of the strand specificity of expressed novel transcripts are necessary to fully characterize these regions.
In our M. hyorhinis transcriptome, we further identified a highly conserved ncRNA that is present in many bacteria, including Mycoplasma species, such as M. hyopneumoniae and M. flocculare, but not in any of the previously mapped M. hyorhinis genomes. This ncRNA is the signal recognition particle (SRP) ribonucleoprotein complex. The SRP RNA belongs to a group of small 4.5S RNAs that are found in bacteria, which direct the traffic of proteins within the cell and allow for their secretion. The SRP RNA, together with one or more SRP proteins, contribute to the binding and release of signal peptides [34].
Comparative gene mapping of M. hyopneumoniae, M. flocculare and M. hyorhinis transcriptomes The read distribution of the M. hyopneumoniae transcriptome showed that 92% of the predicted genes possessed significant alignments with one or more sequences (Table S1). A detailed analysis demonstrated that approximately 80% of the mapped genes have RPKM $5. Some of genes with the best RPKM values are summarized in Table 2. Notably, the products of the genes with the highest expression level were mainly associated with basal metabolism and also included chaperone proteins, adhesins, surface proteins and transporters; the highest RPKM values was 159,921.4 (corresponding to 30,423 reads) that match to the transcript of the MHP_0104 gene (rnpB), which encodes the RNA component of RNase P. Other genes that were also associated with high RPKM values included those encoding the cell division protein MraZ, S-adenosyl-methyltransferase, cell division protein FtsZ, p216 surface protein and hexosephosphate transport protein.
The transcript mapping of the M. flocculare genome revealed that 98% of the predicted genes significantly aligned to one or more sequences of the M. flocculare transcriptome (Table S2). A detailed analysis demonstrated that approximately 94% of the mapped genes have RPKM $5. Some of genes with the best RPKM values are summarized in Table 3, and similar to the findings observed for M. hyopneumoniae, those genes with high numbers of reads were related to basal metabolism, chaperone proteins, adhesins and unknown products. Similar to M. hyopneumoniae, in M. flocculare, the mraZ and ftsZ gene, which encode cell division proteins, present a higher transcriptional level compared with the other genes. A gene that is exclusive of M. flocculare, encoding hypothetical protein (MF01463), and another four hypothetical genes (MF00750, MF00857, MF01460 and MF01461) with orthologs in the M. hyopneumoniae genome also presented with high RPKM values (Table 3).
Surface proteins are potentially related to bacterial pathogenesis. The M. hyopneumoniae colonization of respiratory ciliated epithelium relies specifically on the expression of two functionally redundant adhesin families that are paralogs of P97 and P102. The M. hyopneumoniae gene encoding P216 adhesin (MHP7448_0496) presented with a significant amounts of transcripts (RPKM = 10,796.4) ( Table 2), similarly to previous studies that predicted this protein to be a highly expressed cilium adhesin in M. hyopneumoniae [35,36]. P216 is a proteolytically processed cilium and heparin binding protein [35], and recent data have highlighted the importance of endoproteolysis as a means of generating surface diversity in M. hyopneumoniae [36]. Other genes encoding surface proteins, such as MHP7448_0656 (lipoprotein p65), MHP7448_0198 (protein p97-copy 1) and MHP7448_0497 (p76 membrane protein), which are possibly involved in host cell adhesion, were also among the genes with high transcriptional level ( Table 2).
Previous analyses have shown that the orthologs of M. hyopneumoniae genes encoding MHP7448_0198 (protein p97copy 1) and MHP7448_0199 (protein p102-copy 1) are the only adhesin genes absent in the M. flocculare genome [6]. In the current study, different RPKM amounts, that suggest different expression level, was found in other genes encoding adhesin-like proteins in the M. flocculare genome compared with those of the M. hyopneumoniae transcriptome (Table 4), and even higher numbers of reads mapped to MF00472 (protein p97-copy 2, with RPKM = 1,089.7) (the M. hyopneumoniae ortholog was MHP7448_0108 with RPKM = 234.7) and MF00475 (protein p102-copy 2, with RPKM = 151.8) (the M. hyopneumoniae ortholog was MHP7448_0107 with RPKM = 51.1). The finding of higher numbers of reads mapping in the M. flocculare adhesins compared with those of the M. hyopneumoniae orthologs was unexpected (see Table 4), but according to a previous gene organization and location analysis, the regions containing these orthologs in M. flocculare display inversions and rearrangements in contrast with those of M. hyopneumoniae [6], which may be related to changes in the products' functionalities [46]. Further-more, the p97-copy 2 protein of M. flocculare lacks the R1 domain [6]. Specific cleavages in repeat regions, which are designated as R1 and R2, play key roles in M. hyopneumoniae adherence [39,44]. These R1 and R2 repeats are absent from the M. flocculare p97-like protein and its orthologs from M. hyopneumoniae (p97-like adhesin). Thus, even with the high read numbers of some of the adhesins that were observed in the current study, M. flocculare possesses a weak adhesion ability [7], which may be due to the variants of these proteins that lack the functional domains that are associated with adhesion capacity or antigenicity [38,39,44].
Interestingly, the expression level of MHP7448_0198 (p97-copy 1), which has no ortholog in M. flocculare, was elevated, as shown in Tables 2 and 4. Some studies have shown that p97 and p102copy 1 are important to the pathogenic capacity of M. hyopneumoniae [41,42]. Moreover, a significant number of reads (25,208 reads; RPKM = 1,0796.4) mapped to cilium adhesin P216, which was highly expressed in M. hyopneumoniae, compared to the M. flocculare ortholog with 278 reads (RPKM = 72.5) ( Table 4). The P216 protein is cleaved, generating the heparin-binding proteins p120 and p85, which bind to porcine cilia and are recognized by serum antibodies. These two proteins play significant roles in the interactions of M. hyopneumoniae with host cells by playing immunoreactive roles in the humoral immune response of swine during the natural course of infection, and they are important components of the surface architecture of Mycoplasma cells [35,36]. In summary, the p97-copy 1 and P216 adhesins may be related to the pathogenic capacity of M. hyopneumoniae based on their presences and expression patterns.
The transcriptome analysis of M. hyorhinis was performed with the strain M. hyorhinis ATCC 17981. Because the genome of this strain was not available, the transcriptome assembly and mapping was based on the M. hyorhinis HUB-1 genome. The percentage of reads mapped (86%) was the lowest among the three mycoplasma cDNA libraries (Table 1). This result indicates that an assembly with a different genomic strain can produce inaccurate mapping, especially in intergenic regions. Ninety-six percent of the predicted M. hyorhinis HUB-1 genes were identified in our transcriptome (Table 1), and approximately 95% of the mapped genes have RPKM $5. Some of genes with the best RPKM values are summarized in Table 5. The gene encoding the MraZ protein was the gene with more represented expression level in M. hyorhinis transcriptome (RPKM = 8,637.4) ( Table 5), followed by other genes that were mainly related to basal metabolism, including hexosephosphate transport protein, gap, pdhA, rdhB, and tuf. Somewhat similar to the other two transcriptomes, the mraW gene, which encodes S-adenosyl-methyltransferase, was associated with a significant RPKM measurement (RPKM = 1,456.1)   ( Table 5). The genes vlpF, vlpE, vlpB and vlpG which encode variant surface proteins that are exclusive to M. hyorhinis, in addition to MHR_0162 (surface antigen) also were aligned by expressive numbers of reads. M. hyorhinis contains an exclusive variable lipoprotein (Vlp) system that constitutes its major coat proteins [47] and provides a strategy for evading the host immune system. Different M. hyorhinis strains carry variable numbers of vlp genes [47]. M. hyorhinis HUB-1 is characterized by the presence of seven vlp genes displayed in the order 5-vlpD-vlpE-vlpF-insertion sequence (IS)-vlpG-vlpA-IS-vlpB-vlpC-39 [15]. Although all seven vlp genes were identified in the M. hyorhinis transcriptome map, the transcript numbers mapping to the different vlp genes were variable. vlpF, vlpE and vlpB were mapped by more reads compared with vlpG, vlpD and vlpC (see Table S3). No reads were mapped to vlpA in this analysis. Previous reports have shown that structural variations affect the abundances and functionalities of Vlps on the mycoplasma surface [48], but the full consequences of these variations are yet to be understood. An organism carrying multiple vlp genes has the capacity to generate a large number of variants expressing antigenically distinct Vlp products, either alone or in combinatorial mosaics on the cell surface. Citti et al. [49] argued strongly for the associations of specific functions with each product, which are presumably selected for in the natural host niche. Most likely, the ability of M. hyorhinis to evade the host immune system is related to the variable expressions of the vlp genes, and therefore, future research may improve knowledge regarding the variation of Vlps in M. hyorhinis.
Cell wall biosynthesis by gram-positive and gram-negative organisms is a complex process that involves numerous proteins and has both cytoplasmic and membrane components [50]. The processes of cell wall biosynthesis and cell division are therefore tightly coupled. The protein products of the mraW, ftsH, ftsY, ftsZ and mraZ genes are involved in bacterial cell wall biosynthesis and/or cell division [51] and are the currently representative proteins in swine respiratory mycoplasmas. All of these genes were transcribed in the three mycoplasma maps with abundant transcript counts ( Table 2; Table 3; Table 5).
Abundant numbers of transcripts were identified in the three transcriptome maps that aligned to S-adenosyl-methyltransferaseassociated genes, that participate in cell wall biosynthesis (mraW) and cell division (mraZ). Notably, the gene ftsZ, which encodes a cell division-related protein, is among the genes mapped with the largest RPKM amounts in the M. hyopneumoniae and M. flocculare transcriptomes ( Table 2; Table 3). However, the transcriptional levels of this gene were lower in M. hyorhinis (RPKM = 839.9). Most bacteria produce the tubulin homolog FtsZ [52], which forms a cytoskeletal filament that is essential for membrane constriction and the coordination of septal peptidoglycan synthesis [53]. Although mycoplasmas possess a gene encoding FtsZ and some evidences suggested this gene as essential to these organisms [54]. However, more recently, it was experimentally demonstrate that ftsZ is non-essential for cell growth and in the absence of the FtsZ protein, M. genitalium can manage feasible cell divisions and cytokinesis using the force generated by its motile machinery [55]. Furthermore, in M. pneumoniae, very low levels of both ftsZ mRNA and FtsZ protein are present [56], supporting the hypothesis that mycoplasmas do not actually use FtsZ solely for cell division and revealing a potentially unusual function of this protein.
Another functionally related group of transcripts with significant RPKM amounts encoded heat-shock proteins (HSPs) ( Table S1;  Table S2; Table S3). HSPs are induced by environmental stress and play important roles in stimulating both the host innate and adaptive immune responses [57,58]. These proteins can be classified into six families as follows: the large-molecular-weight HSP family, HSP90 family, HSP70 family, HSP60 family, smallmolecular-weight HSP family, and ubiquitin [59]. HSP70 has been extensively studied and is known to function as a molecular chaperone, anti-cell apoptosis agent, and antioxidant and to induce immune responses, improve stress tolerance, and promote cell proliferation, cytoskeleton formation, and repair [60]. Both HSP70 and HSP60 induce immune responses that protect hosts against bacterial and mycoplasmal infections [61][62][63]. Furthermore, it has been shown that monoclonal antibodies generated against a portion of M. hyopneumoniae HSP70 are capable of blocking the growth of M. hyopneumoniae [64]. In the three transcriptomes we observed high numbers of reads that were mapped to the DnaK gene (HSP70), which forms a chaperone protein complex with the DnaJ and GrpE genes [63] Additionally, DnaJ and GrpE mapping also resulted in a significant number of reads. HSP60 (GroEL) was absent in the three mycoplasmas of the swine respiratory tract [6,12,15]. A comparative genome analysis between M. hyopneumoniae, M. flocculare and M. hyorhinis has been recently described [6], showing highly levels of similarity in the genome contents and organizations of M. hyopneumoniae and M. flocculare. These two closely related mycoplasma species [5,6] present with pathogenic differences; M. hyopneumoniae is considered to be pathogenic, and M. flocculare is considered to be a commensal species. This genome comparison showed many differences that may help to explain their differing behaviors. However, they were not able to identify specific virulence determinants that could explain the differences in their pathogenicities [6]. This same comparative analysis showed that the M. hyorhinis genome also shares significant numbers of genes with both M. hyopneumoniae and M. flocculare. The differences in M. hyorhinis gene composition seem to be related mainly to the ability of M. hyorhinis to colonize different hosts and sites.

Transcriptional Units structures
We have constructed transcriptome maps of M. hyopneumoniae, M. flocculare and M. hyorhinis. The RNA-Seq data enabled us to determine the structures of the TUs on a genome-wide scale, which is important for the identification of co-expressed genes and for the understanding of the coordinated regulation of the mycoplasma transcriptome. Our results support previous predictions that the genes are preferably transcribed in polycistronic mRNA, forming TUs, in the mycoplasma genomes [6,18]. We have identified that co-expression occurs for approximately 90% of the predicted genes for all three species (Table 1; Table S5;  Table S6; Table S7). By joining consecutive overlapping reads or reads that mapped uniquely (Figure 2), we have constructed a common scheme of gene transcription in mycoplasmas. Recently, a comparative genome analysis of M. hyopneumoniae, M. flocculare and M. hyorhinis was described [6]. The arrangements of the TUs as determined by in silico prediction revealed that they are present in similar numbers in these three mycoplasmas genomes [6], indicating that genes are preferably organized into TUs in M. flocculare and M. hyorhinis similar to M. hyopneumoniae [18]. Moreover, as previously described for M. hyopneumoniae, the overall gene distributions within the TUs in M. flocculare and M. hyorhinis are highly variable with respect to gene number and the functional categories of the encoded products [6]. As expected, these TUs predicted by in silico and in vitro approaches in M. hyopneumoniae in addition to the computationally predicted TUs in the M. flocculare and M. hyorhinis genomes were confirmed in the three transcriptomes analyzed here.
Over 80% of the TUs predicted for the three species were either confirmed or partially confirmed by our transcriptome analysis (Table 1; Table S5; Table S6; Table S7). Table 1 shows the confirmed TUs, for which all ORFs exhibited cotranscription, besides showing the partially confirmed TUs, for which the majority of the ORFs exhibited cotranscription. Among the predicted TUs for M. hyopneumoniae and M. hyorhinis [6,18], only two were not associated with any transcriptional products in the present transcriptome maps. All genes belonging to these TUs, in both genomes, encoded hypothetical products. Remarkably, in the M. flocculare transcriptome, all of the TUs predicted by the previous in silico analysis [6] possessed at least one gene that was transcribed ( Table 1).
The overlapping reads mapping to each reference genome generated a surprisingly large number of long contigs ( Figure S2). Some examples are the contigs that mapped to nucleotides 268881-277986 with sizes of over 9,000 bp, and the mapped contig at position 242981-245879 with a total length of 2,898 bp, which were both observed in M. hyopneumoniae 7448. The M. hyorhinis map has the largest mRNA contig mapped, with a size reaching of 13,000 bp (locations: 485595-498603 bp and 779997-788652 bp). Figure 2 illustrates a representative TU for each of the analyzed transcriptomes. Most of the observed TUs possessed overlapping reads (as shown in Figures 2a, 2b and 2c, green arrows). Interestingly, the transcript levels of genes belonging to the same TU were variable (see Figures. 2a, 2b and 2c), indicating that such staircase-like expression is a widespread phenomenon in bacteria. The presence of consecutive genes within operons possessing different expression levels has been observed in other bacterial species [21,65]. The M. pneumoniae transcriptome data showed a natural polarity with respect to the TU, in which the first gene of the TU exhibits the most prominent level of transcription, and levels progressively decreased for each successive gene in the TU [21]. Thus, although genome reduction leads to longer operons accommodating genes with different functions [18,21,66], the last gene can still retain internal initiation and termination sites under certain conditions. Mycoplasmas are characterized by the presence of long TUs containing genes that encode highly functionally variable products. Moreover, all the M. hyopneumoniae TUs are preceded by putative promoter sequences [20], providing evidence of gene organization as an important factor in the regulation of transcription. A subset of genes within TUs that are able to be transcribed by alternative internal promoters have been demonstrated and associated with possible complex transcriptional organization in M. hyopneumoniae genomes [18][19][20]. According to previous predictions [20], 70% of M. hyopneumoniae TUs possess internal promoter sites, which is in agreement with the high variations in expression levels that were observed within the TUs. These findings are also similar to others that have been described in many bacterial species, including those in the M. pneumoniae transcriptome, which has been characterized by the frequent occurrence of polycistronic operons with alternative transcripts [21], and the Bacillus subtilis transcriptome, in which 20% of the genes in polycistronic operons are transcribed from more than one promoter [67]; additionally, approximately 6% of the polycistronic operons contain internal read-through terminators, at which partial continuation of transcription occurs [68]. Similarly, in the Halobacterium salinarum, 40% of the operons are expressed in condition-dependent manners [69]. These findings indicate that various activators and/or repressors can regulate transcription. Thus, a TU could respond to different inputs, enabling higher regulatory responsiveness [33]. Taken together, these data suggest a complex relationship between genomic organization and gene expression in prokaryotes.
Overall, we provided comprehensive assessments of the transcriptomes of three important Mycoplasma species that inhabit the respiratory tract of swine. We have analyzed the RNA populations present in M. hyopneumoniae, M. flocculare and M. hyorhinis in detail and used the data to map transcript boundaries and operon structures on a genome-wide scale. Moreover, we showed that almost all of the genes in these mycoplasma genomes are expressed at some basal level and that the majority of the genes are co-expressed, confirming the previously predicted TUs. Our results expand upon current knowledge regarding the coordination of transcription in swine respiratory mycoplasmas, in which genes are preferably transcribed in TUs. The determinations of all of the functional elements in the M. hyopneumoniae, M. flocculare and M. hyorhinis is a prerequisite for formulating holistic approaches to delineating the complex transcriptional regulation that occurs in these mycoplasma species.        hyopneumoniae Myo-inositol transcription unit (TU) composed by ten genes. This TU is on the forward strand in the genome and has a staircase behavior, meaning that the consecutive genes have lower and steady expression levels. (B) M. flocculare TU structure composed by five genes. This TU is located on the reverse strand in the genome. (C) M. hyorhinis TU structure composed by three genes. This TU is located on the forward strand in the genome. The position on the chromosomal sequence is indicated in base pairs (bp) below both termini of the bars. Visualization by the software Artemis. doi:10.1371/journal.pone.0110327.g002