Staining of total RNA from bacteria resolved on gels reveals abundant sRNAs 4849505152. We used this method to analyse S. meliloti small RNA (<400 nt: Figure 1A). Four intense bands were detected, three of which correspond to the sizes predicted for 5S RNA (band 5), 4.5S RNA (band 2), and the 5' end of tmRNA (band 4, 41). Band 6 (≈70 nt) corresponds to a length and migration profile compatible with tRNAs. No RNAs migrating faster than 70 nt were detected. Two less intense bands (1 and 3) were observed and were compatible with sizes predicted for rnpB RNA and for the 3' end of tmRNA 41. Northern blotting confirmed the identity of each of these bands (Figure 1B). Surprisingly, no additional RNAs were visualized with either RNA extraction method (Trizol™ or Qiagen™, not shown) or stress (see conditions in Material and Methods section, data not shown). This experiment indicated that small RNAS do not constitute an abundant class of RNAs in S. meliloti, in contrast to what has been observed in marine Cyanobacteria and in Staphylococcus aureus 505152. As a consequence, we could not employ direct elution and sequencing to identify new sRNAs 53 in S. meliloti. We thus used computational prediction to identify candidates for further testing.

ncRNAs in AT-rich hyperthermophiles can be located on the basis of local variations in genomic base composition 1011. Although the S. meliloti chromosome is GC-rich (62.7%), we scanned the regions containing tmRNA, rnpB and ffs (Figure 2) for base composition signals. Unlike AT-rich genomes, no difference was observed between sRNA and background genomic sequences (Table 1). BLAST alignments of the three sRNAs against all sequenced bacteria revealed significant similarities with sequences from the alpha-2 subgroup of the class Proteobacteria. However maximum likelihood trees (see Additional file 1) showed that most identical sequences were from different genomes (R. etli and leguminosarum for ffs and tmRNA; Mezorhizobium for rnpB). To conclude, no apparent dominant bias was observed for the three known S. meliloti ncRNAs, except primary sequence conservation in IGRs of related alpha-proteobacteria genomes (especially subgroup-2).

Most bacterial sRNA searches were conducted using the E. coli genome as a reference, so we first compiled the sRNAs available in the Rfam database for that organism (strain K12, riboswitch/cis-regulatory genes excluded) to try to identify sRNA orthologs using BLAST alignments (e-value < 1) against the complete S. meliloti chromosome sequence. No significant similarities were detected except for 4.5S and rnpB (see Additional file 2). ssrA (tmRNA) was not detected, presumably because the S. meliloti tmRNA is in a two-piece "permuted" form 41. Surprisingly, the "ubiquitous" ssrS (6S RNA) was not detected either. Similar analysis with Bacillus and Pseudomonas sRNAs [RFAM] gave no further hits. This first approach confirmed a key point: most sRNAs cannot be detected by inter-phylum primary sequence identity.

RNA-specific tools described in the literature were then assessed. However, most rely on preliminary structural knowledge, knowledge which is poor for this phylum. Thus only a subset of these tools was tested: those reported to be able to detect new sRNA gene candidates in prokaryotic species. ISI (Intergenic Sequence Inspector, 54) relies on inappropriate filters (in particular GC%), and was far too sensitive if the filters were disabled (close to 300 candidates). The use of QRNA 7 requires preliminary alignments. We tested it by aligning the 61 sRNAs annotated in the E. coli (K12) genome against S. meliloti's chromosome, but only 4.5S and rnpB were predicted by this tool. Finally, we assessed sRNApredict2 18, which combines commonly sRNA-associated genetic features; it was used to search the S. meliloti chromosome IGRs for sRNAs, by comparison with A. tumefaciens and R. etli. As shown in the output tables (see Additional file 3), 13 sRNAs were predicted, seven being common to the two genomes compared. However, a detailed analysis of each candidate revealed that only one (pred1) was likely to correspond to a genuine small RNA gene whereas the others are almost certainly 5' UTRs (in view of the small distances and nature of the surrounding genes) or repeated elements (RIME, Sm- 5556).

We ended up selecting for further analysis: the IGRs containing the origin of replication oriC (as a negative control) and the tmRNA, rnpB and 4.5S RNA genes; 22 IGRs longer than 450 nt; 13 IGRs showing short sequence identity in alpha-proteobacteria (see Additional file 10); and 28 IGRs selected using the Artemis Comparison Tool (ACT) with related genomes (see Additional file 11). Indeed, the sequence identities marked by the ACT were mostly confined to actual open reading frames (ORFs), even between two close chromosomes (S. meliloti and A. tumefaciens). Although hits in intergenic regions were extremely rare (less than 40), all three S. meliloti known sRNAs (tmRNA, 4.5S and rnpB) were included in these hits: we thus concluded similar ACT-visualized IGRs represented good sRNA gene candidates. Among these IGRs, those containing large repeats (Sm-1 to Sm-5) or RIMEs were however excluded, as were IGRs smaller than 50 nt (to avoid inclusion of 5' or 3' UTR elements).

This selection accounted for 67 candidate IGRs (see Additional file 4), which throughout this paper shall be referred to as sra, followed by a number corresponding to their order of apparition on the chromosome (starting from oriC).

Microarray hybridization was the first method used to test for transcription of the sRNA candidates; an IGR-dedicated array was used (see Materials and Methods). The initial goal of the time-course microarray experiments was to detect ncRNA gene transcripts in various conditions. We first monitored gene expression over time during growth in minimal (MV) and rich (LB) media. The average log₂ SNR of the three known ncRNAs (tmRNA, rnpB and 4.5S) and the ribosomal RNA 5S (included as a control) was very low (0.49) in MV medium, but was higher (4.13) in LB medium (see Additional file 5). We applied various stresses, as described in Materials and Methods, but detected no specific induction (not shown). We therefore used LB medium for expression analyses and the results are expressed as a heat map (Figure 3). We also checked the expression of all of the candidates by Northern dot blotting. Twenty-five candidates were found to give a strong hybridization signal (including rnpB and 4.5S RNA genes) and nine gave weak but detectable signals. Among the dot blot-"positive" IGRs, ten did not show significant expression on microarrays (Fig. 3, group B1), perhaps due to insufficient fluorescent labelling (e.g. the RNA was too small or labelling was hampered by secondary folding). Twenty-three IGRs gave no transcription signal either in Northern dot blots or in microarrays (Fig. 3, group B2). These included sra01, which corresponds to the S. meliloti origin of replication (oriC or smc04880) and was used as an internal negative control. Indeed, various methods (including qPCR and blotting) have been used to demonstrate that there is no transcription at this site (not shown). Lastly, 11 IGRs giving a transcription signal on microarrays were not confirmed by Northern dot blotting; in these cases the signal may have originated from surrounding coding genes.

Northern blotting was used to detect sRNAs. The same probes as for Northern dots were used but after hybridization of newly purified RNA from fresh cultures. For some candidates, probes were redrawn to target areas common to all considered subgroup-2 alpha-proteobacteria. sra15 was discarded because it appears to correspond to a pseudo phe-tRNA gene (not shown). This locus is possibly a vestige or a target of DNA insertion; indeed, a 500-kb symbiosis island has been shown to integrate into a phe-tRNA in the M. loti chromosome 57.

Four sRNA candidates (sra51, sra37, sra47, sra59) gave large transcripts (~1000 nt, not shown), inferred to be part of adjacent ORFs for three reasons: (i) they were larger than the intergenic region; (ii) they were large enough to encompass one of the flanking coding genes; and (iii) the strand detected corresponded to the orientation of at least one of the flanking genes. We did not pursue investigations to determine whether these 5'/3' UTRs encompass cis-regulatory RNA structures.

The signals detected for sra48, sra58 and sra62 were very strong and composed of multiple bands (data not shown). For sra48 and sra58, the explanation is sequence-specific hybridization of the probe to various transcripts due to small imperfect DNA repeats (sm-2 fragment in the case of sra58 and a previously undescribed repeat in sra48, data not shown). The sra62 candidate corresponds to a repeated region that coincides with the fixT loci, present in three copies in the genome of S. meliloti 40. In the two first copies, fixT genes (fixT1 and fixT2) are co-located with fixK genes (fixK1 and fixK2). The third copy region of fixT gene (fixT3) encompasses sra62, which corresponds to a fixK3 pseudogene (not shown).

However, it cannot be concluded that the 23 candidates (group B2, Figure 3) for which no transcript was detected by microarray or Northern blot experiments correspond to false predictions: some of them may possibly be true sRNAs, but which are poorly or not expressed in the conditions tested.

We further investigated the 17 sRNA candidates that yielded signals with a suitable size in Northern blots (<300 nt, Figure 4A). We excluded (Table 2) sra10, that contains a possible 101 amino acid ORF and sra29, which is a probable leader for rpsF (S6 ribosomal protein gene) 5859; we also excluded sra24 that may encode a short peptide (38 aa long) rather rich in cysteine (10%). The 23 first aa (

MALFFKPHCFLSLYCCLLSQRG...

Of the 14 remaining RNAs, six correspond to S. meliloti-specific sRNAs (called orphan sRNAs, Table 3). All except sra61 lie within a genomic island 60 (sra14 in Smc19T; sra11, sra12a and sra12b in Smc21T; sra66 maps at the 5' end of Sme80s, just downstream from the insertion tRNA^ser gene smc03779). All these sRNAs are absent from Agrobacterium and other Rhizobium species, consistent with the absence of the three genomic islands from these related genomes. The "coding potential" of sra11, sra12a, sra12b and sra14 was assessed and only sra66 could translate a small non-conserved 16 aa peptide. There may be a relationship between their biological functions and their presence in genomic islands as bacterial islands are often related to eukaryotic host cell colonization (virulence or symbiosis). Indeed bacterial sRNAs were recently detected in the pathogenicity island of Staphylococcus 51 and in Salmonella (InvR 36). Multiple 5' and 3' RACE mapping of these candidates was unsuccessful, possibly due to insufficient cDNA or because they are too small for satisfactory cloning. With no available orthologous sequences, determination of extremities through alignment and structure prediction based on covariations are not possible. The functions of these new non-coding genes are thus unknown. However, many small RNAs in bacteria act as post-transcritptional regulators via base-pairing action, so we used TargetRNA 61 to assess their capacity to act as antisense RNAs. No significant putative target was found for sra11, sra12a, sra14 or sra61, but sra12b and sra66 match with mRNA targets (see Additional file 12). One interesting TargetRNA prediction was the interaction between sra66 and a tolR-exbD-like gene (smc03957). TolR-ExbD proteins are membrane-bound transport proteins essential for ferric ion uptake in bacteria 62, and small RNAs modulating the free intracellular iron pool have indeed been identified in other bacteria (e.g RyhB, 63).

In E. coli, sRNA genes were considered to be conserved if the alignment had an E-value lower than 0.001 64. On this basis, the sequences of the eight remaining expressed S. meliloti sra genes are highly conserved in related alpha-proteobacteria (e-value < 10^-10; Table 4). sra56 is even an analog of Escherichia coli's 6S RNA (SsrS) that has recently been corroborated by RFAM. sra56 may thus be considered as a fortuitous positive control, validating our experimental approach. All these conserved sra genes we describe were then subjected to transcriptional element analysis, structure prediction using covariation, and potential mRNA target prediction.

S. meliloti chromosomal sRNA-encoding genes have a slight distribution preference, 70% of them being on the left replicore, but no preference was observed concerning the leading or the lagging strand (Figure 4B). The conservation of gene adjacency (synteny) may be associated with functional relationships (e.g. gcvB and ssrA are functionally associated to their conserved adjacent genes, respectively gcvA and smpB) 3565. Therefore, we assessed the conservation of neighbouring genes for the eleven alpha-proteobacterial sra, including tmRNA, rnpB and 4.5S (see Additional files 10 and 11). We found only three (4.5S, tmRNA and sra41) where both flanking genes were conserved and four with one conserved flanking gene (including rnpB). Three display no synteny (including sra56 = 6S). However, our observations suggest that in alpha-proteobacteria, functional association between sRNAs and conserved adjacent genes is uncertain, as even smpB is not linked to tmRNA (separated by 1,1 Mb).

The computational identification of promoter sequences is difficult because signals are weak 4. Consensus binding sites have been proposed for the S. meliloti σ⁷⁰ and σ⁵⁴-dependent promoters (CTTAGAC-n17-CTATAT 66 and TGGCACG-n4-TTGCW 67, respectively). Using relaxed regular expressions (i.e. pattern matching), we scanned the 5' UTR region of each sra gene to detect any such consensus (max. 4 mismatches). In the same way than for the orphan candidates, the end-mapping experiments remained unsuccessful for all new candidates but sra32 (5' end-AAACAGGCAGGAA and 3' end-CTTGTTTTTTT), thus the exact sites of the extremities of the sra genes are unknown. However, sequence alignment between several alpha-proteobacteria was informative. As for ORFS, no significant nucleotide identities were apparent in the 5' and 3' regions (not shown). Thus, the start and stop of the alignment and the sizes deduced from electrophoretic motility were used to define the ends of each sra gene. We only detected consensus σ⁷⁰-promoters in the ubiquitous ncRNAs (ffs, tmRNA, rnpB and 6S) and the 5'-UTR mapped region of sra32 (Table 5). Even with a larger spacer between the -35 and -10 boxes (17 to 20), no consensus sequence was detected in the other sra genes. Similarly, only the ffs 5'-UTR matches a possible σ⁵⁴ binding site. However, S. meliloti contains a large number of predicted sigma-factors (ca. 16), it is therefore possible that sra genes are dependent on them. Unfortunately, Melina-based 68 motif extraction from the promoter regions of these genes was inconclusive and did not show any consensus.

We also looked for potential classical L-shaped terminators (stable stem-loops followed by U-stretches) at the alignment-deduced 3' end of each sra. As most terminators in GC-rich bacteria have no long consecutive U-stretches 69, we also looked for I-shaped (stable stem-loop with no U-stretch) and V-shaped (two consecutive hairpins 69) terminators (Table 6). Terminator structures could be predicted for all conserved sra genes, except sra30 and sra34, for which the alignments are too short for the ends to be accurately mapped. All three shapes were found, I-shapes being the most common. This supports the proposal that "orphan" terminators in IGRs may indicate the presence of an sra gene, although because of the high GC content of the S. meliloti genome, stable stem-loops are frequent and consequently not very informative.

The structures of non-coding RNAs are important for their function. Consequently, we made a conserved secondary structure prediction for all alpha-proteobacteria sra genes (except sra30 and sra34 because of too short alignments).

We first compared the 4.5S and 6S Alifold 7071 and RNAz 72, outputs with the equivalent RFAM structure (Additional file 13). The two tools give different structures for 4.5S, the Alifold prediction being closer to that of RFAM than the RNAz prediction. However, bulge A, the minimal site required for binding to EF-G 73, is similarly folded in the three models. The three predicted structures of sra56 (6S) are similar, 6S being folded as a largely double-stranded RNA with a single-stranded central bulge. However, the 3' side of the central bulge can form a stable stem-loop in alpha-proteobacteria, as previously described 74.

The predicted structure of sra03 is well-conserved, composed of three long hairpins: the central one, ending with a 15 nt loop, is strictly conserved, and the 3' and 5' stem-loop structures present co-varying nucleotides. Analogous design (three stem-loops) was found for the E. coli sRNAs drsA, micC and qrr 46. No significant target mRNA was found using the full RNA but the 15-nt loop presents complementarity with the 5'-UTR of the smc03977-smc03976 operon (not shown). These two genes have paralogs, mostly in alpha-proteobacteria genomes. The function of Smc03977 is unknown but Smc03976 belongs to the ZapA cell division family of proteins 75. ZapA binds the ftsZ cell division gene, which has been reported to undergo antisense inhibition in E. coli 22, so sra03 is possibly implicated in cell division regulation.

sra25 was only identified in Rhizobia (meliloti, etli and leguminosarum), however with remarkable primary sequence identity. Its predicted secondary structure is also highly conserved, composed of a long stem with one central bulge and a small hairpin. The 3' end structure was predicted to act as a stabilization terminator. A possible pairing with the 5' UTR of a gene encoding a putative membrane protease protein (Smc04020) was proposed by TargetRNA, although with low significance (p-value > 10^-3). This protein resembles the E. coli HflKC protein, known to interact with the cell division protease FtsH 76. As for sra03, we propose that sra25 may have a role in the S. meliloti cell cycle.

Only Alifold proposed a consensus structure for sra33: it resembles other short bacterial sRNAs including RyeE 46. Predictions for sra33 showed strong pairing ability with the 5' ends of smc00899 and of rkpJ. The first gene is organized in an operon structure with smc00900 (encoding a PilT-like toxin), defining a post-segregational killing toxin-antitoxin (TA) system. This type of mechanism, involving sRNAs in the inhibition of TA systems, has been extensively described in E. coli 77. Finding such a mechanism in S. meliloti is interesting as we recently detected approximately 53 TA loci (accounting for 95 genes) in the complete genome sequence 78.

sra32 is the only small RNA predicted both by sRNApredict2 and ACT analysis. For this gene, hybridization was tested by Northern blot and a signal was obtained for two of the most distant members: A. tumefaciens (140 nt) and R. elti (132 nt) (Figure 5A and 5B). The secondary structures of sra32 were predicted by both Alifold and RNAz; 59% of the nucleotides can pair in seven conserved regions, those at the 3' and 5' ends (TB1 and TB7) possibly forming stabilization stem-loops. Only the primary sequence of TB3 is conserved, all other structures being supported by covariations. The search for potential mRNA targets yielded significant predictions for fliM (flagellar motor switch) and smc01800 (cytochrome C oxidase). The same sequence of sra32 (nt 60 to 95, TB4 to TB6) is predicted to pair with the 5' leader region of both mRNA targets. Discovery of a putative flagella antisense RNA in S. meliloti and related bacteria is interesting. Indeed, all these bacteria interact with plant or mammal cells and flagella are cell-surface components required for eukaryotic-prokaryotic cells interactions 37.

Finally, structure and target prediction of sra41 needed preliminary analysis as (i) two signals were detected by Northern hybridization (a major ~106 nt band and a minor ~68 nt species) and (ii) sra41 is present in three imperfect copies, two in tandem in the same chromosomal IGR and the third, with a more divergent primary sequence, maps in pSymA. Additionally, sra41 is well-conserved in alpha-proteobacteria and is present in two to three copies, on both chromosomes and (mega-)plasmids except in Agrobacterium (chromosomal loci only). Translatable small ORFs are predicted within some sra41 candidates (see Additional file 6) whereas others are devoid of complete coding frames. A similar conserved structure was proposed for the three S. meliloti copies, composed of three stem-loops as described for sra03. Target mRNAs were predicted separately for each of the three sra41 copies in S. meliloti. The only predicted target for the first chromosomal copy is smc02392, which encodes a hypothetical protein with a Sel1 repeat-containing domain. The second copy is predicted to interact with two mRNAs encoding Smc00317, a transmembrane protein with homologies to efflux carrier proteins, and Smc01118, a glucoprotease-like protein possibly involved in chaperoning processes. No significant target was predicted for the pSymA copy.

In all experiments, S. meliloti strain 1021 (streptomycin resistant, Str^R) was grown at 30°C in the presence of 25 μg.ml^-1 of streptomycin. Bacteria were initially grown to the stationary phase in LB (Luria-Bertani) medium, collected, washed and then resuspended in MV (Vincent minimal medium, 1970 80) or LB medium. Stresses were applied in MV at mid-exponential phase (OD600 nm = 0.4) as follows: 10 minutes with a sub-lethal dose of hydrogen peroxide (10 mM); salt shock (0,4 M NaCl), pH 5 and pH 9 generated with 10 N HCl or NaOH were also applied during 10 minutes; cold (10°C) and heat shocks (40°C) lasted 15 minutes. The other alpha-proteobacteria used in Northern blot and sequencing experiments are listed in Additional file 7.

Cells were harvested by centrifugation (5 min, 2600 × g, room temperature) and immediately frozen at -80°C. These pellets were subsequently re-suspended in 200 μl of TE buffer (10 mM Tris-HCl pH8 – 1 mM EDTA) containing lysozyme (1 mg.ml^-1) and total RNA was extracted using the RNeasy^® mini kit (Qiagen™) and treated with RQ1 Rnase free DNase (1 U.μg^-1, GE Healthcare) in the presence of 6 mM of MgCl₂ for 20 min at 37°C.

RNAs shorter than 300 nt were eluted from a 2.5% high-resolution agarose gel (Sigma) in 0.5 × TBE buffer 81 and used to build a cDNA library using the MessageAmpII-Bacteria kit (Ambion). The RNAs were 3' end-polyadenylated and first strands were synthesized using oligo-T7-dT (AATACGACTCACTATAGGGCGAA dT25) then treated overnight at 37°C with terminal deoxynucleotidyl transferase in the presence of 1 mM of dTTP. The second strands were obtained with a 5'-primer (GGAATACTAGTGACACCAGACAAGTTG dA15). PCR was carried out at 55°C using 1 μl of the cDNAs and 4 pmoles of the appropriate specific primers (see Additional file 8): one corresponding to the targeted gene and the other specific to either the 5' or 3' tagged ends. The products obtained were inserted into pGEM-T (Promega) and sequenced using the BigDye terminator v3 protocol (Applied Biosystems). RNA genes of interest were amplified by PCR with a 50°C to 40°C touch-down program, using the 5' and 3' end primers designed after mapping, from 12 genomic DNA preparations from several Rhizobia (see Additional file 8). The resulting products were sequenced using the protocol described above and analyzed in silico for co-variation and conserved secondary structure.

DNase-treated, S. meliloti total RNA (10 μg per dot) was denatured (by incubating for 10 min at 70°C then chilling on ice) and spotted under vacuum onto a nylon membrane (Zeta Probe) using Bio-Dot (Bio-rad). These dot membranes were baked for 30 min at 80°C, pre-hybridized, hybridized and washed according to Ambion's instructions (Oligoprobes), and hybridization was visualized on InstantImager (Packard). 20-mer oligonucleotides (see Additional File 3) were 5' P-end labelled with [gamma³²P]ATP. For each candidate, two oligonucleotide probes (one per strand) were selected to identify the strand of transcription. Aliquots of 62 pmoles of each oligonucleotide probe were incubated for 1 hr at 37°C with 1 μl of T4 polynucleotide kinase, 2.5 μl of 10 × kinase buffer (Ambion) and 2 μl of ³²P-ATP [5,550 KBq (222,000 KBq.mmol^-1)]. The radiolabelled probes were then purified through MicroSpin G25 columns (GE Healthcare). The efficiency of probe labelling was monitored in a liquid scintillation analyser (1600 TR Packard).

S. meliloti was grown in LB medium to mid-exponential phase (OD_{600 nm}= 0.8). Total RNA (10 μg) was extracted, denatured, and subjected to electrophoresis on an 8% acrylamide/bis acrylamide (19:1) denaturing gel (8 M urea) in 0.5 × TBE buffer 81. The RNA was then transferred to a nylon membrane (Ambion) in the same buffer and the membrane was baked for 30 min at 80°C. Biotinylated oligoprobes were used for hybridization and bound probes were detected according to Ambion's instructions (BrightStar^® BioDetect™).

A PCR-based microarray was designed (see Additional file 9) with primer3-designated oligonucleotides derived from the 67 IGRs (intergenic empty regions) identified by screening (see Additional File 4). Seventy bp were excluded at the 5' and 3' ends of each IGR to avoid expression contamination by adjacent genes. The IGRs with lengths >550 nt were amplified as more than one PCR fragment. Each PCR involved 200 ng of S. meliloti 1021 genomic DNA, 1.25 units of TaqDNA polymerase (Promega) and 4 pmol of each oligonucleotide in a final volume of 50 ml for 40 cycles (Peltier Thermal Cycler, MJ Research Inc). The PCR buffer contained 5 μl MgCl₂ (25 mM), 1 × Taq buffer (Promega) and 4.3 μl dNTPs. The length and quality of each PCR product was assessed by electrophoresis on 2% agarose TBE (Tris-Borate EDTA 0.5 × final) gels. The IGRs-PCR probes were dried, resuspended in 30 μl betaine, 1.5 M-3 × SSC and spotted in triplicate onto GAPS II™-coated glass slides (Corning) through robotic arraying (Microgrid II, BioRobotics).

The labelled cDNA was resuspended in hybridization buffer (3 × SSC/0.1% SDS/50% formamide) and cohybridized with 10 μg of salmon sperm DNA to the microarray glass slides overnight at 55°C. The slides were washed for 5 min at 55°C in wash buffer (2 × SSC/0.1% SDS) then 2 min in high-stringency buffer (0.2 × SSC/0.1% SDS, twice), 2 min in 0.2 × SSC (twice) and 2 min in 0.1 × SSC, and dried by centrifugation (3 min, 210 g). Hybridized microarray slides were scanned (GenePix4000, Axon Instruments, Inc.) with independent excitation of the fluorophores Cy5 and Cy3 at 10-μm resolution. Our microarray time series experiment was planned with a dye-swap design; Dabney & Storey recently showed that a simple average dye-swap removes dye bias without affecting the biological signal and preserves the ordering of true expression means 82. To determine if ncRNA candidates showed detectable expression as assessed by microarray analysis, we used log₂ SNR (signal to noise ratios), the "noise" value being the average intensity of spots containing spotting buffer. All genes with log₂ SNR lower than 2 (a signal less than 4-fold greater than the technological noise) were considered as untranscribed under the conditions tested and their value changed to 0 (filtering). The resulting expression profiles are illustrated as a heat map.

Intergenic regions (IGRs) were determined as genome areas with no gene annotation on either of the two strands. The S. meliloti database used 83 includes original annotations of all ORFs, as well as tRNA, rRNA and repetitive elements. The Artemis Comparison Tool (ACT 84) from the Sanger Centre 85 was used for whole genome comparisons of IGRs between the S. meliloti chromosome, as the reference, and other alpha-proteobacteria genomes: Agrobacterium tumefaciens strain C58 Cereon, Rhizobium etli CFN42, Rhizobium leguminosarum bv. vicae 3841 and Mesorhizobium loti MAFF303099.

ISI 54 was used taking the initial embl record file corresponding to the S. meliloti chromosome (AL591688) as an input, and applying an IGR length threshold of 7 bp. Parameter 'w' in BLAST was set to 7 in order to refine word detection.

In order to produce the input alignment file for QRNA v.2.0.3.c 7, the 61 ncRNAs annotated in the E. coli K12 Refseq file (NC_000913) were extracted, and compared to the chromosomal IGRs of S. meliloti thanks to WU BLAST 2.0, with various e-value thresholds (10^-5, 10^-2, 0.1 and default: 10). QRNA was then run with scanning window set to 100 nt.

sRNAPredict2 18 was tested using the S. meliloti chromosome ORF list (NC_003047.ptt) from the Refseq repository 86. The ".coords" TIGR file is unavailable for S. meliloti. The t/rRNA database was compiled from data in the TIGR_CMR RNA list 87 and from the S. meliloti sequencing consortium website 83. The terminator database was derived from TransTerm 8889. The S. meliloti sRNA training set was generated from data available at the corresponding RFAM database page 46. Two reference genomes were used: Agrobacterium tumefaciens str. C58 and Rhizobium etli CFN 42. Consequently, two blast output files were generated, where S. meliloti intergenic regions were compared to the cited genomes using WU BLAST 2.0 with same parameters as Livny 18 (E = 10e^-5, V = 10000 et B = 10000). Finally, databases of regions of predicted conserved secondary structure were produced using QRNA, again with parameters set to the same values as in 18. No input files corresponding to RNAMotif search results or to promoter/transcription binding sites were provided to sRNAPredict.

TargetRNA 61 was used (default parameters) to identify sra candidates' mRNA targets. Potential base pair binding interactions were considered significant only if their P-value fell below 0.001. Nucleotide sequences were analyzed by ClustalW 90 and resulting multiple alignments were used in ALIFOLD 71 to model RNA secondary structures. Similar predictions were realized using the RNAz program 72.