We used an automated gene annotation pipeline provided by PATRIC 19 to predict genes in S. meliloti. This pipeline uses a combination of gene prediction programs, Glimmer 20, GeneMark 2122, TICO 23 and RBSfinder 24 to predict genes and compares with genes in RefSeq. A total of 512 new protein-coding genes (with length >90 nt) in the intergenic regions of the genome were predicted through this automated pipeline (Additional file 1). The number of predicted genes in different length ranges is shown in Figure 1. Most of the predicted new genes are relatively small (length <400 nt). The average length is about 200 nt. These genes were BLASTed against the NCBI non-redundant (NR) protein database 2526. The result showed that 159 candidates had BLAST hits in the NR database with E-values less than 0.01, whereas the remaining 353 of the candidates had no significant hits. Small gene size and lack of BLAST hits may be the reasons that the predicted new gene candidates were missed in the original genome annotation process.

Total RNAs were extracted from S. meliloti 1021 cells grown to mid-exponential phase in the TY medium and treated with DNase I to remove genomic DNA (Methods). The 16s and 23s rRNAs were depleted and the RNA samples were then amplified to produce cDNA fragments of average length about 150 nt (Methods). With two test cDNA samples loaded on to 4 lanes per sample of a 16 lane sequencing plate, the titration run generated a total of 19,005 high quality reads with average length of 136 nt (Table 1). Although our rRNA removal step indicated that more than 90% rRNAs were depleted as judged by Agilent 2100 Bioanalyzer, approximately 90% of the reads still aligned to the rRNA operons (Figure 2). This may be due to relative low mRNA population in the S. meliloti cells. Out of 17092 reads aligned to the rRNA operons, 3 reads matched 5s rRNA, 2860 reads matched 16s rRNA, 13983 reads matched 23s rRNA, and the remaining 246 reads aligned to the integenic regions between the rRNA genes in the rRNA operons. For the 1854 non-rRNA sequences, 1774 matched to 737 of the 6271 RefSeq genes (proteins and RNAs) and 59 matched to 32 of 512 new protein-coding genes predicted through our gene prediction pipeline. The remaining 21 sequences mainly matched sequences either immediately before or after a coding region, presumably 5' UTR or 3' UTR.

Twenty new gene candidates with multiple GS FLX sequence hits or long sequence hits (>80 nt) were chosen for further verification using RT-PCR analysis of the original total RNA samples. These new gene candidates were predicted by PATRIC pipeline as protein-coding genes. Figure 3 shows an example of a new gene candidate (VBISMc1000) in the Sinorhizobium Genome Browser, which was built using the GBrowse software 27. All 20 genes were detected in the RT-PCR experiment and the PCR products were sequenced and confirmed (Figure 4). The negative controls indicated that there was no genomic DNA contamination in the RNA samples tested.

To test whether the new genes are co-transcribed with their upstream or downstream flanking genes (if any), a set of primer pairs were designed (Figure 5, Additional file 2) to detect RT-PCR product of the transcripts with the flanking genes. The results are summarized in Table 2. Columns "co-transcribed with upstream gene" and "co-transcribed with downstream gene" indicated whether the RT-PCR products of the transcripts, which span from the upstream flanking gene to the new gene or from the new gene to downstream flanking gene, were detected. Ten of the predicted genes were not detected to be co-transcribed with either upstream or downstream flanking genes (Table 2).

The sequences of putative new genes were searched against the NR database from NCBI and SwissProt from EBI 28 using BLASTX 25 and Smith-Waterman programs (29; Table 2). Both programs produced very similar results: 10 of the 20 new genes had no significant hits, and the other 10 had either a full length or partial match with proteins in the NR database (Table 2). The genes with no significant hits were relatively short with lengths ranging from 120 to 366 nt.

Four predicted genes showed significant matches with genes with known or putative functions (Table 2). VBISMc2940 had 89% similarity to a conserved hypothetical signal peptide protein from S. medicae WSM419. VBISMb0839 had 69% similary to a two component transcriptional regulator, LuxR family from S. medicae WSM419. VBISMa1337 partially matched to a putative dioxygenase with 25% similarity. VBISMc3282 matched to nodulation protein NolR with 59% similarity. The NolR protein is a transcriptional regulator for common nodulation genes as well as the three nodD copies present in S. meliloti. Previous studies have shown that nolR gene in S. meliloti strain 1021 has a single insertion in the C-terminal coding sequence which abolishes the DNA-binding ability of the NolR protein 3031. Thus, Rm1021 has no NolR activity. NolR^- strains nodulate host plants less efficiently than NolR⁺ strains. In the RefSeq database, VBISMc3282 was not previously annotated as a gene although it is a mutant form of the nolR gene. Here, we demonstrate and confirm that this mutant gene is expressed.

No neighboring genes were detected to be co-transcribed with VBISMc2940, VBISMb0839 or VBISMa1337, while VBISMc3282 was co-transcribed with the downstream gene SMc01535, a hypothetical protein. The other six genes with BLAST hits matched only hypothetical proteins (Table 2).

Because the RNA amplification step was linear (Methods), we expect that the cDNA samples we prepared represent relative mRNA levels in the cell. Thus, for a full sequencing run, with high coverage, the number of sequences would be a good indication of gene expression levels. Due to the low coverage of our pilot experiment, we cannot yet estimate gene expression levels based on number of sequences for each gene. However, we expect that most of the genes that showed five or more matches to our transcriptome sequences should be highly expressed in the cell population (Additional file 3). The known genes with high sequence copy number are consistent with our knowledge about the high expression level of those genes under the same growth condition (our unpublished microarray data).

Sinorhizobium meliloti strain1021 was grown at 30°C in TY medium 34 to mid-exponential phase (OD₆₀₀ = 0.6). Cell growth was stopped by adding 1/9^th volume of stop solution (5% buffer equilibrated phenol pH 7.4 in ethanol) and placed on ice. Cells were collected by centrifugation in a microcentrifuge at maximum speed for 3 minutes. The cell pellets were stored in -80°C. Total RNA was isolated by using Qiagen RNeasy bacterial RNA purification kit (Qiagen, Valencia, CA). The total RNA was treated with DNase I on mini-RNeasy column before eluted with RNase free water. For RT-PCR experiments, an additional DNase I treatment was done after RNA was eluted from the RNeasy mini column to ensure that there was no genomic DNA contamination. 20 μl of total RNAs eluted from the RNeasy mini-column were treated with 5 μl DNase I (Qiagen) in 10 μl RDD buffer, 1 μl RNase inhibitor (Invitrogen, Carlsbad, CA) and 64 μl RNase free water (Qiagen) at 25°C for 30 minutes. The RNAs were then extracted with phenol/chloroform and precipitated with ethanol using standard protocols. 16s and 23s rRNAs were then depleted using the MICROBExpress™ Bacterial mRNA Enrichment Kit (Ambion, Austin, TX). Total RNAs and rRNA depleted RNAs were quantified and analyzed on the Agilent 2100 Bioanalyzer. 7 μg of total RNA per reaction was used. After 16s and 23s rRNAs were depleted, about 0.5 μg (7%) RNAs was recovered. The total RNA samples had RNA integrity number (RIN value) of 8.0 or better. As shown in Figure 2, more than 90% 16s and 23s rRNAs were depleted. We analyzed more than 10 independent rRNA-depleted preparations on the Agilent 2100 Bioanalyzer. The results were consistent and showed that the 16s and 23s rRNA peaks were greatly reduced in these preparations but not completely removed (Figure 2). In addition, two small peaks immediately before 16s and 23s rRNAs could not be removed by the Ambion MicrobExpress kit. The two peaks were consistently present in all of our RNA preparations.

Two RNA samples were prepared for RNA amplification. Sample 1 was 16s and 23s rRNA depleted RNA sample as described above. Sample 2 was the 16s and 23s rRNA depleted RNA sample ligated to a 3' RNA adptor (5'-PO₄-UUCGCUGUUC UUAGCGGCCG CAUGCUC-idT-3'; idT: 3' inverted deoxythymidine) (Dharmacon Research, Lafayette, CO) and a 5' RNA adptor (5'-OH-AUGUGCGCGA CUUCCUGUAG ACGGAACGCU AGAAGAAA-OH-3') (Dharmacon Research). 3' and 5' adaptor ligations were done as described in Argaman et al. 2001 35.

To obtain enough cDNA for sequencing, the 16s and 23s rRNA depleted RNAs (sample 1 and 2) were amplified using Nugen WT-Ovation Pico RNA amplification system 36. 5 ng of starting RNA was used. The SPIA™ amplified single strand cDNA (2.5 ug) was then taken through a second strand cDNA synthesis, using the following conditions: 5X 2nd strand reaction mix 30 μl (Invitrogen), dNTP, 10 mM 3 μl (Invitrogen), E. coli DNA ligase 1 μl (Invitrogen), E. coli DNA polymerase I 4 μl (Invitrogen), RNase H 1 μl (Invitrogen), RNase-free water 91 μl (Ambion). The reaction mix was incubated at 16°C for 2 hours. The cDNA was then purified using the Qiagen PCR clean up kit resulting 4 ug of cDNA quantified by using the Nanodrop spectrophotometer. 1 ug of cDNA of each sample was size selected (>100 bp) using Roche's GS FLX library Preparation Guide recommendations (no nebulization was necessary due to the size range of the cDNA GS FLX library Preparation Guide), and a single stranded library was created.

The DNA sequencing libraries for the two samples were combined with the sequencing beads in 4 different concentrations to determine the optimal conditions for emPCR amplification. All 8 preparations were sequenced in 8 lanes of a GS FLX sequencing plate using the standard Roche/454 protocols. Sequencing data was obtained after a 7 hour run on the GS FLX. The 54,162 raw reads from GS FLX sequencing run that passed the sample key code filter (initial bases TCAG) were further filtered by the 454 software to eliminate 10,521 mixed reads (with two or more different DNA strands/bead), 15,604 excessively short reads (less than about 50 bp), and 9,032 interrupted reads ("dots"). 35% of the raw reads passed all filters in this titration run to provide the 19,005 high quality reads used in this study.

GS FLX sequences passed filtering criteria were BLASTN-aligned to S. meliloti genome. Sequences with matching to rRNA operons were filtered. The remaining sequences were BLASTed against RefSeq genes and our predicted new genes from our gene annotation pipeline.

5 μg DNase I treated total RNA was reverse transcribed using superscript II with 4 pmoles of equally mixed gene-specific primers for each candidate gene selected (cm012b-cm0031b, Table S1). Primers were designed using Primer3 37. For PCR, each 40 μl reaction includes 0.5 μl of 40 μl reverse transcription reaction, 20 μl of 2X GoTaq Green Master Mix (Promaga, Madison, WI) and 0.5 μM of primer pair of each gene (IDT, Coralville, IA). PCR conditions were 95°C 2 min, 30 cycles of 95°C 45 s, 52°C 45 s, 72°C 60 s, and a final cycle of 72°C for 10 min. PCR products were examined by electrophoresis in a 2.5% agarose/TAE/EtBr gel. Sequencing was performed using the BigDye Terminator Cycle Sequencing Kit (Applied Biosystems) and analyzed on an Applied Biosystems model 3730 automated capillary DNA sequencer.