The presented prediction was based on the work of Argaman et al. 10, Wassarman et al. 19 and Rivas et al. 20. It was based on sequence conservation in the intergenic regions (IGRs) of fully sequenced genomes, co-localized transcription terminators and genomic arrangement of the predicted sRNA genes.

In Streptomyces, two species have been fully sequenced to date, S. coelicolor and S. avermitilis, and their genomic sequences were used for the prediction. Using TIGR annotations, IGR sequences were identified in both genomes. There were 3753 and 4292 IGR sequences with lengths between 40 and 1000 nt identified in S. coelicolor and S. avermitilis, respectively. In these IGR sequences, the conservation was computed using BLAST 21. BLAST databases were created for the S. avermitilis and S. coelicolor IGR sequences, and single IGR sequences were BLASTed against the database of the other species. The BLAST parameters were -r 1 -q -1 -G 1 -E 2 -W 9 -F "m D" -U -m 8.

For the conserved IGR sequences, co-localized Rho-independent terminators were identified. They had to start not farther than 50 nt downstream of the 3' end of the conserved sequences, oriented appropriately. The terminators were predicted by TransTermHP 22 with confidence > 75%. TransTermHP identifies the terminators by searching for a common mRNA motif: a hairpin structure followed by a short uracil-rich region. For each terminator, a score is assigned reflecting hairpin stability and related to the likelihood that it arose by chance.

For estimation of the BLAST E-value cut-off for significant sequence conservation, three different cut-off values, 1 × 10^-5, 1 × 10^-10 and 1 × 10^-20, were applied. They produced 1666, 1233 and 710 conserved IGR sequences, respectively. Among these, 63, 51 and 37 of the conserved IGR sequences had co-localized terminators. These decreased numbers showed that the screen was not sensitive to the terminator filter, but instead depended solely on the cut-off value, as more conserved sequences could have more co-localized terminators. Therefore, another indication for the cut-off estimate was required. To this end, the E-values between the three known S. coelicolor and S. avermitilis sRNAs, tmRNA, M1 RNA and 4.5S RNA, were computed. Their E-values were 2 × 10^-130, 9 × 10^-94 and 2 × 10^-42, respectively. In our dataset, we also identified two tRNAs that could also be considered as sRNAs with conserved structures. They had E-values of 6 × 10^-26 (tRNA ala) and 2 × 10^-27 (tRNA gly) (Table 1). These E-values were relatively much higher, while still representing strong conservation. If the cut-off was derived from the E-values of the three known sRNAs, sRNAs between 6 × 10^-26 and 2 × 10^-42 could be missed by the prediction. Therefore, the three known S. coelicolor sRNAs were BLASTed against the corresponding E. coli sRNAs. The E-values were 1 × 10^-14, 6 × 10^-6 and 0.068 for M1 RNA, tmRNA and 4.5S RNA, respectively. Assuming that the sRNA conservation between S. coelicolor and S. avermitilis should be stronger than between S. coelicolor and distantly related E. coli, the BLAST E-value cut-off was chosen to be 1 × 10^-20, lower than any of the three E-values.

Using the E-value cut-off of 1 × 10^-20, we obtained 710 conserved sequences in IGRs of the S. coelicolor and S. avermitilis genomes. Out of the 710 conserved IGR sequences, 37, with co-localized terminators, were considered to be potential sRNA genes (Table 1 and Table S1 in Additional file 1). Two sRNAs were predicted within one IGR (denoted as ID # 413.1 and 413.2 in Table 1), since two Rho-independent terminators were localized within a single IGR sequence. The 37 predicted sRNAs included two known Streptomyces sRNAs (tmRNA, 4.5S RNA), three tRNAs (tRNAs for alanine, lysine and glycine) and a 5S rRNA (Table 1) that were not included in the annotations used for the prediction. They therefore could not be excluded from the prediction and served as standards.

Two known Streptomyces sRNAs (tmRNA, 4.5S RNA) were identified in the previous section using Rho-independent terminators. However, the third known Streptomyces sRNA, M1 RNA, was not found. This was due to its lack of a Rho-independent terminator. Instead of the terminator, we found a C-rich stretch with C-content of 73% (Figure 1a). It also had a structure with properties similar to the stem-loop structure of Rho-independent terminators 10 (Figure 1c). The termination of the M1 gene differed in S. coelicolor and S. avermitilis, as in S. avermitilis, a Rho-independent terminator was found at genomic coordinates 7094081 – 7094112 (Figure 1b). The terminator was predicted by TransTermHP with a high confidence (87%).

These data suggested that Streptomyces sRNAs might be terminated by a C-rich stretch stem-loop different from the classical Rho-independent terminator. We thus searched the conserved IGR sequences that lacked the co-localized Rho-independent terminators for the C-rich stem-loop. To this end, C-rich stretches were sought in 50 nt segments flanking the ends of the conserved IGR sequences. The stretches with G/C content > 75% and C content > 60% were folded using the RNAstructure program 23. The structures were required to have at least 10 nt in the loop and 3 – 5 base pairs in the stem for them to be considered possible terminating factors. These properties were derived from the C-rich stem-loop of the M1 gene. Besides the M1 RNA gene, one gene was identified with such a C-rich stem-loop (ID # 60; Figure 1d). Together with the 31 sRNAs predicted using the Rho-independent terminator, the total of new predicted sRNAs was now 32.

Out of the 32 new predicted sRNAs and known Streptomyces sRNAs left in the dataset as standards (three known Streptomyces sRNAs, three tRNAs and a 5S RNA), 10 were localized on the same strand as the genes that flanked their 5' ends (5'-flanking genes) (Table 1). These 10 sRNAs needed to be distinguished from the 3' UTRs of mRNAs. Therefore, their 5'-flanking genes were inspected for transcription termination. Either Rho-independent terminators or C-rich stretches indicating the Rho-dependent termination were considered. The Rho-independent terminators were predicted by TransTermHP 22 and they were required to be localized in between the 5' end of the predicted sRNA and 100 nt upstream of the 3' end of the 5'-flanking gene, oriented appropriately. Three predicted sRNAs were found with such a genomic arrangement: two newly predicted sRNAs (ID # 445 and 472) and the tRNA for lysine (ID # 482) (Table 1).

To detect the Rho-dependent transcription termination, C-rich 30-mers were sought in between the 5' ends of the predicted sRNAs and 100 nt upstream 3' ends of the 5'-flanking genes. The C-rich 30-mers with C content between 35% and 60% have been reported to be required for one type of Rho-dependent transcription termination 2425. Here, 30-mers with C content > 60% were required, as Streptomyces have G/C rich genomes. Six sRNAs were identified with such a genomic arrangement: three predicted sRNAs (ID# 115, 126 and 341; Table 1) and three known Streptomyces sRNAs (tRNA ala, tmRNA and M1 RNA, Table 1 and 2).

Expression of the predicted sRNAs was examined by microarray analysis and RT-PCR. Internal oligonucleotides for the 32 predicted Streptomyces sRNAs were designed and spotted on microarray slides (see Materials and Methods). Specificity of the microarray signal was tested in two ways: 1. The housekeeping RNAs (5S rRNA and tRNAs for Ala, Arg, Gly, Ser and Lys) and two known S. coelicolor sRNAs (tmRNA and M1 RNA) were included in the experiment as standards; 2. The oligonucleotides for tmRNA, M1 RNA, 5S RNA and two predicted sRNAs (ID # 390 and 389) were designed with an increasing number (0, 2, 4, 6, 12, 24) of internal mismatches. As expected, the mismatches decreased the microarray signal by decreasing specificity of hybridization. Out of the 32 predicted sRNAs, 20 were found to be expressed (Table 1 and 2).

The expression of sRNAs was further verified using RT-PCR. The 20 predicted sRNAs whose expression was detected by microarrays were reverse-transcribed using primers (18-mers) designed to match the predicted sRNA sequences immediately upstream of the transcription terminators. RNA samples were acquired in those growth phases that corresponded to the highest microarray signal. The RT-PCR recognized expression of 9 predicted sRNAs (Figure 2, Table 1 and 2).

The predicted Streptomyces sRNAs were analyzed for conserved RNA structures using RNAz 2627. The analysis showed that 29 out of the total 32 predicted new Streptomyces sRNAs had a strongly conserved secondary structure (Table S4 in Additional file 1). For the three remaining (ID # 96, 146 and 234), no conserved structure was found or no prediction occurred because of too many gaps in their alignments. The 29 sRNAs were predicted to be functional sRNAs, for most of them (23) the probability was highly significant (> 0.9). The probabilities were included in the last column of Table 1.

To identify sequence similarities, the predicted sRNAs were computationally compared to sequences of known sRNAs of other species in Rfam database. The sequence similarity was computed using BLAST. The BLAST parameters were set as follows: -r 5 -q -4 -G 10 -E 6 -W 7 -F "m D" -U. The BLAST bit score cut-off for the Rfam database was 23.3 (the log₂ of the size of the target database). The seven known Streptomyces sRNAs among the predicted sRNAs were identified correctly with bit scores > 100. Other BLAST hits (with bit scores between the cut-off and 100) were strongly ambiguous (except for one, as explained later), and they were most likely not biologically significant. Therefore, the biologically significant bit score cut-off was considered to be 100. No other BLAST hits with bit scores higher than this cut-off were found.

The predicted sRNAs were also matched against Rfam families using the Infernal package 28. The Infernal method scores sequence and structure similarity at the same time. Moreover, Infernal uses consensus sequences and structures of the query sRNAs instead of single sequences. To this end, the sRNA sequences of S. coelicolor and S. avermitilis were aligned by ClustalW 29, and consensus structure was predicted using RNAalifold 30 from the alignments. Infernal was applied to the consensus structures in the local mode and without the HMM filter. The local mode was used since the Streptomyces sRNAs were supposed to have sequences and structures very different from most species in Rfam. Nevertheless, they might retain local similarities especially in the conserved functional sites. The local mode increases the probability of detection of such local similarities, as it allows for detection of partial similarities of the query model and target sequence. The HMM filter was not used in order to attain as sensitive a search as possible. The three known Streptomyces sRNAs, three tRNAs and the 5S RNA were among the predicted sRNAs that were identified correctly with bit scores > 50, and therefore the biologically significant bit score cut-off was estimated to be 50. However, no other hits with bit scores higher than this cut-off were found.

We identified function features within the predicted sRNAs. The function features were derived from the known sRNAs of other species that were also expected to function in Streptomyces. They were M1, tm, 4.5S, spot42, 6S, oxyS, csrB, rprA, ryhB, dicF and micF RNAs 31. The function features were identified based on expert knowledge, and were characterized by local sequences and local structures. They were mostly localized in the function sites of the sRNAs. The characteristic sequences and structures were identified in the sequences and structures of the predicted Streptomyces sRNAs. Significant hits were obtained in four cases: the three known Streptomyces sRNAs and the predicted sRNA ID # 389. The ID # 389 sRNA and the 4.5S RNA are shown in Figures 4 and 3, respectively, to demonstrate their analogous similarities to the known sRNAs of other species. In the case of Streptomyces 4.5S RNA, the conserved function features were identified at nt 38 – 59 and nt 26 – 35 (Figure 3a). Analogously for the ID # 389 predicted sRNA, the conserved function features were identified at nt 75 – 90, 117 – 120, 124–128 and 148 – 161 (Figure 4a).

We therefore concentrated on the ID # 389 sRNA. It was the only unambiguous BLAST hit to the Rfam with a bit score (27.3) between the biologically significant and minimal significant bit score cut-offs. Its sequence corresponded to the only unambiguously identified promoter, when the promoters were sought manually for the predicted sRNAs. It was the Streptomyces σ^H transcription factor that corresponded to the σ⁷⁰ factor that is specific in E. coli for 6S RNA. Based on the promoter position, the length of the ID # 389 sRNA was determined to be 181 nt, which was similar to the lengths of known 6S RNAs (188 and 184 nt for B. subtilis and E. coli, respectively).

S. avermitilis and S. coelicolor genomic sequences were imported from the NCBI ftp site 40. ORF annotations were obtained from TIGR Comprehensive Microbial Resource 41, including tRNAs and rRNAs ORFs. The computation and algorithms used in this study was made using MATLAB and Bioinformatics toolbox 37.