Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Genome-Wide Discovery of Small RNAs in Mycobacterium tuberculosis

  • Paolo Miotto,

    Affiliation Emerging Bacterial Pathogens Unit, S. Raffaele Scientific Institute, Milan, Italy

  • Francesca Forti,

    Affiliation Dipartimento di BioScienze, University of Milan, Milan, Italy

  • Alessandro Ambrosi,

    Affiliation University Statistical Center for Biomedical Sciences – Università Vita-Salute S. Raffaele, Milan, Italy

  • Danilo Pellin,

    Affiliation University Statistical Center for Biomedical Sciences – Università Vita-Salute S. Raffaele, Milan, Italy

  • Diogo F. Veiga,

    Affiliation Department of Systems Biology, The University of Texas MD Anderson Cancer Center, Houston, Texas, United States of America

  • Gabor Balazsi,

    Affiliation Department of Systems Biology, The University of Texas MD Anderson Cancer Center, Houston, Texas, United States of America

  • Maria L. Gennaro,

    Affiliation Public Health Research Institute, University of Medicine and Dentistry of New Jersey, Newark, New Jersey, United States of America

  • Clelia Di Serio,

    Affiliation University Statistical Center for Biomedical Sciences – Università Vita-Salute S. Raffaele, Milan, Italy

  • Daniela Ghisotti,

    Affiliation Dipartimento di BioScienze, University of Milan, Milan, Italy

  • Daniela M. Cirillo

    Affiliation Emerging Bacterial Pathogens Unit, S. Raffaele Scientific Institute, Milan, Italy

Genome-Wide Discovery of Small RNAs in Mycobacterium tuberculosis

  • Paolo Miotto, 
  • Francesca Forti, 
  • Alessandro Ambrosi, 
  • Danilo Pellin, 
  • Diogo F. Veiga, 
  • Gabor Balazsi, 
  • Maria L. Gennaro, 
  • Clelia Di Serio, 
  • Daniela Ghisotti, 
  • Daniela M. Cirillo


Only few small RNAs (sRNAs) have been characterized in Mycobacterium tuberculosis and their role in regulatory networks is still poorly understood. Here we report a genome-wide characterization of sRNAs in M. tuberculosis integrating experimental and computational analyses. Global RNA-seq analysis of exponentially growing cultures of M. tuberculosis H37Rv had previously identified 1373 sRNA species. In the present report we show that 258 (19%) of these were also identified by microarray expression. This set included 22 intergenic sRNAs, 84 sRNAs mapping within 5′/3′ UTRs, and 152 antisense sRNAs. Analysis of promoter and terminator consensus sequences identified sigma A promoter consensus sequences for 121 sRNAs (47%), terminator consensus motifs for 22 sRNAs (8.5%), and both motifs for 35 sRNAs (14%). Additionally, 20/23 candidates were visualized by Northern blot analysis and 5′ end mapping by primer extension confirmed the RNA-seq data. We also used a computational approach utilizing functional enrichment to identify the pathways targeted by sRNA regulation. We found that antisense sRNAs preferentially regulated transcription of membrane-bound proteins. Genes putatively regulated by novel cis-encoded sRNAs were enriched for two-component systems and for functional pathways involved in hydrogen transport on the membrane.


Regulatory RNA species can modulate transcription, translation, mRNA stability, DNA maintenance, and gene silencing. They function through a variety of mechanisms, including changes in RNA conformation, protein binding, base pairing with other RNAs, and interaction with DNA [1][3]. It has been established that small RNAs (sRNAs) are crucial elements regulating gene expression in Gram-positive and Gram-negative bacteria [4][7]. The prototype of a bacterial sRNA is a non-coding RNA of 50–300 nucleotides (nt) in length that acts by imperfect base pairing with trans encoded RNA target(s). sRNAs may also be cis-encoded and transcribe antisense (AS) to the target RNA. Recent research has shed some light on the relevance of sRNAs in bacterial pathogenesis, including modulation of expression of virulence factors in response to environmental and host signals [2], [8][14].

Intracellular survival of Mycobacterium tuberculosis (MTB) results from the ability of this pathogen to sense the host environment and to switch to a non-replicating, persistent form that is resistant to host insult and to treatment with most antibiotics. Multiple genes associated with persistence have been identified in MTB. However, the complex post-transcriptional regulation underlying the phenotypic switch to persistence remains poorly understood. In particular, our knowledge of the regulatory networks involving sRNAs is still in its infancy. Understanding how sRNAs control bacterial virulence may open a new prospective in the fight against tuberculosis (TB) and contribute to the goal of TB eradication [15][18].

About 40 species of sRNAs have been reported so far in mycobacteria (including MTB) by cloning approaches and computational predictions [19], [20]. Additional sRNAs have been identified by RNA sequencing (RNA-seq) [21], [22]. RNA-seq results for sRNAs however heavily depend on threshold definition for reads coverage and on the criteria used for transcript annotation, thus requiring further validation [3]. Here we report the first genome-wide expression analysis of sRNAs in MTB by microarray followed by size determination, mapping and computational target prediction.


Genome-wide Expression Profiling of sRNAs in MTB

To identify sRNAs in MTB, we performed microarray expression analysis on small-size RNA enriched fraction extracted from exponentially growing cultures of M. tuberculosis H37Rv. We used a set of 1373 sRNA species previously identified by RNA-seq (Type A candidates, [22]), which was updated to reflect changes in annotation in the genome of MTB (type A, Table S1).

For microarray analysis, we considered as expressed only the candidates showing an expression mean significantly higher (p-value <0.05) than the positive control genes (this group included genes expressed during exponential growth such as sigA and rrs and the short ribosomal 5S RNA; for a complete list of control genes, see Materials and Methods). The microarray analysis validated 258 (18.8%) species (Figure 1), about a third of which (n = 97) showed a length below 50 nucleotides. When we calculated the minimum free energy (MFE) associated p-value, which represents a measure of secondary structure stability, we found that the proportion of sRNAs showing an MFE-associated p-value <0.05 was nearly double for “array-validated” candidates (37/258, 14.3%) than for the remaining “non validated” candidates (85/1115, 7.6%).

Figure 1. Heat-map representing the expression profiles of microarrays.

As positive controls we used reference sigA, rrs and the short ribosomal 5S RNA. Normalized control genes showed a mean expression level of 7.16±1.57. Expression level: yellow to blue (highly expressed to lowly expressed). Statistics: red = p-value <0.05; green = p-value >0.05.

Classification of MTB Putative sRNAs

A functional classification of the selected sRNAs identified was performed based on genome location (see Materials and Methods and Figure S1). We found 22 intergenic sRNAs, 84 candidate sRNAs mapping within 5′ or 3′ UTRs, and 152 antisense sRNAs (Table 1).

Table 1. Classification of validated sRNAs according to their genomic position.

The two more represented classes identified by microarrays were AS RNAs (60%) and 5′/3′ UTRs (32.6%); of the latter class, most were 5′UTRs (5′UTRs: 90%). RNA species identified within the 5′ UTRs could represent cis-acting regulatory elements such as riboswitches. Between the 5′UTRs we found genes involved in cell wall processes, intermediary metabolism, information pathways, PE/PPE, virulence, detoxification and adaptation, and lipid metabolism (Table S2).

Statistical Analysis of the Frequencies of Specific Nucleotides at the −10 Region

To show that the different RNAs validated by microarray expression analysis were transcribed, we compared the frequency of the nucleotide’s distributions at specific positions within the −10 region between sRNA candidates, annotated coding sequences (CDSs), and random sequences. We could confirm the effective transcription of sRNAs identified by microarrays by showing a clear non-random distribution of nucleotides within the −10 region. In particular, we showed that our candidates show a −10 region more similar to the one of CDSs rather than to random genomic sequences not annotated for transcripts. To achieve this goal we used a statistical approach previously described by Dornenburg [23]. We compared the frequency of nucleotides at the −10 consensus hexamer between sRNA candidates, annotated CDSs, and random sequences. The hexamer TANNNT recognized by the sigma factor A was used as reference consensus. For the set of annotated CDSs, −10 hexamers match the consensus, on average, 1.7 times out of 3 known consensus nucleotides −10 match score). In contrast, randomly selected sequences match the consensus only 1.5 times out of 3 (control match score) (Figure S2). Due to the high number of terms compared for each category (see Materials and Methods), this difference is highly significant (P<0.0001). The −10 match score for the sRNAs is not significantly different from that of the set of CDSs. The comparison of Type A candidates only to the control group of random sequences and to the consensus sequences of annotated CDSs showed similar results. Considering microarray-validated candidates only (score 2.0), the difference with the dataset from CDSs increased whereas that with the random sequences slightly decreased.

Characterization of sRNA by Northern and Identification of the 5′-end by Primer Extension

Microarray data were further validated by Northern blot analysis. We selected 23 candidate sRNAs (Table 2) from all the functional categories identified (see Table 1). Genomic mapping of these sRNAs is shown in Figures S3 and S4.

The candidates chosen were:

  1. four intergenic (#149, 161, 224 and 1096);
  2. eight 5′/3′ UTRs of genes (#1149, 1414, 1502, 1542, 1565, 106, 189, and 193);
  3. ten AS (#54, 135, 1029, 540, 1080, 1556, 488, 1137, 294, and 498).

We also included candidate 1359, although it was mapped within the Rv2165c gene, since it is highly expressed by microarray analysis. As a negative control we used a 5′UTR sRNA not validated by microarray (#76). Since Northern blot analysis allows relative comparison of transcript abundance, RNA was extracted from exponential and stationary phase cultures in order to evaluate differential expression associated with growth state. Total RNA was extracted from the strain H37Rv. For a smaller pool (candidates #224, 540, 1137, and 1565) we also tested the attenuated strain H37Ra.

The results of the Northern blot showed that in most cases (20 out of 23), one or more specific bands were detected (Table 2, and Figure 2). The Northern blot of candidate #189 showed a single band of predicted length; in other cases (#1565) a band of the predicted size was detected among several others. For most candidates, however, the length observed by Northern analysis differed from the results of the RNA-seq-based computational mapping (for example, Figure 2, #498 and 1137). sRNAs #106, 1029 and 1414 and the negative control #76 were negative by Northern blot analysis (data not shown). No difference was observed between exponential and stationary phase expression for all sRNAs cultures (Figure 2). Only candidate #224was expressed at higher levels in M. tuberculosis H37Rv than in H37Ra.

Figure 2. Northern blot results for selected candidates.

The RNAs were extracted as indicated in M&M: exp: M. tuberculosis H37Rv in exponential growth phase; stat: M. tuberculosis H37Rv in stationary growth phase; Rv: M. tuberculosis H37Rv; Ra: M. tuberculosis H37Ra. Arrows indicate the predicted length.

Since computationally predicted length and that observed by Northern blot analysis differed in almost all species, we set out to identify the 5′ ends of eight selected sRNAs (#488, 498, 1080, 1137, 1149, 1359, 1502 and 1565) by primer extension. In all cases, we could identify one or more bands, one of which corresponded to the predicted 5′ end with +1 or −1 nt difference (data not shown). A −10 consensus sequence for sigma factor A (TANNNT) was identified for 7 out of 8 sRNAs mapped, suggesting that transcription of the sRNAs is under sigA control. Upstream of the candidate 1359 a good consensus sequence for −10 and −35 sigma B is present (NGTGG – N14–18–NNGNNG).

Identification of the −10 sigA Consensus Sequences and Putative Terminators in sRNA Candidates

Since the 5′ ends identified by primer extension well matched those predicted by RNA-seq, the discrepancies between the length observed by Northern blot and computational analysis might mainly be due to inaccurate 3′-end mapping. First, we performed the analysis of all the sRNAs upstream sequences to detect the presence of −10 consensus for sigA. We decided to search for sigA consensus motifs because this is the primary sigma factor that regulates gene expression during the exponential growth phase. We searched for −10 sigma A consensus sequences (TANNNT) within −50 and +15 nucleotides from the 5′-end mapped according to RNA-seq results. At the same time, we searched for putative terminators downstream of the predicted sRNAs. The 3′-end region (−20/+270 nucleotides) was sought for intrinsic terminators using the GeSTer algorithm [24], and for rho-independent terminators, according to Gardner et al. [25]. In 60.5% of cases we detected a sigA promoter consensus sequence upstream of the presumed 5′-end. Terminator consensus motifs were present in 22.1% of the regions downstream of the sRNA. Only a minor percentage (13.6%) of sRNAs presented both promoter and terminator sequences, whereas one third (31.0%) of predicted candidates did not show consensus motifs either for sigma A promoters or terminators (Table 3).

Table 3. Promoter and terminator consensus sequences found among 258 M. tuberculosis sRNAs by bioinformatics approach.

Classification of all 1373 candidates, promoter and terminator consensus sequence analyses, and genomic annotation are summarized in Table S1.

In silico Identification of Pathways Regulated by Antisense sRNAs

An important question in sRNA biology is to identify the genes that these molecules regulate. We performed an in silico analysis to identify pathways regulated by the set of novel sRNAs discovered in MTB. We focused on AS RNAs because the direct targets for this class of sRNAs are known. When we performed microarray analysis we found 152 AS sRNAs that overlap the same number of unique target genes in the opposite strand (Table 1). Using functional enrichment analysis, we identified molecular functions and pathways enriched in AS-regulated genes in MTB. In particular, hydrogen transport on the membrane (GO:0015078, GO:0015077, GO:0006818, GO:0015992, GO:0034220) and two-component systems (mtu02020) emerged as function potentially subjected to antisense regulation. When we analyzed cellular location, we found that membrane-bound proteins were preferentially targeted (GO:0044425, GO:0005886, GO:0031224). These results are summarized in Table 4. The corresponding targeted genes can be found Table S3.


Here we report on the integration of experimental and computational methods to identify 258 novel antisense and intergenic sRNAs in MTB. The present work derives from a previous, genome-wide identification of candidate sRNAs by RNA-seq [22]. Large RNA-seq dataset require accurate validation, especially for genome annotation of novel transcripts. We considered reliable the sRNAs identified with high reads coverage and validated these sRNAs by microarray and, for representative species of the microarray-validated pool, by Northern blot and primer extension analyses. This approach provided more details on the annotation of the candidates; indeed, the particular design strategy we adopted for microarray probes allowed acquiring important data on the size of the sRNAs. The inclusion of probes targeting well-known control genes permitted the collection of expression data for these newly discovered sRNAs.

Our results lead to several conclusions. As supported by −10 region and 5′ end mapping analyses, our computational approach applied to Illumina sequencing results provided reliable detection of sRNA transcripts. All but one of the sRNAs mapped experimentally are preceded by a sigA promoter consensus sequence. In addition, an initial analysis of sRNA target pathways indicated that AS sRNAs preferentially target genes involved in two-component systems and membrane activity. Our results suggest that expression of the sRNAs described may be not as part of the stress response, but are actively involved in the regulation of basic metabolism in MTB.

Together, our study provides a comprehensive top-down demonstration of methods whereby large sets of RNA-seq data on sRNAs can lead to detailed sRNA mapping and identification of target pathways. Our analytical path differs from previous studies on sRNAs of M. tuberculosis, which reported only few sRNAs and did not provide a prediction of putatively regulated pathways. Indeed, despite Arnvig and colleagues performed RNA-seq to identify novel sRNAs in MTB [21], they provided an overview on genome-wide data and focused their attention characterizing only few of the sRNAs detected. In the present report we further characterized several candidates previously identified by us (e.g. #84, #1169, #1137, #488) [22] and by Arnvig and colleagues (MTS0479, MTS0997, MTS0903, MTS2458, respectively) [21]. Our work provides more comprehensive data on recently discovered sRNAs in MTB and statistics analysis support our experimental data: in fact, the genomic regions where we mapped sRNA candidates showed consensus sequences for active transcription indicating the probability to be transcribed.

Our starting dataset included a high number of candidates (>1000) as reported by other studies based on RNA-seq approach [26]. By microarray expression profiling, we validated nearly 20% of the previously identified putative sRNAs. These sRNA species showed to be enriched in higher secondary structure stability, as supported by MFE analysis. We found ∼100 sRNAs shorter than 50 nucleotides. sRNAs are usually between 50 and 200 nt in length; however, smaller RNAs of the size of microRNAs have been described in bacteria [23], [27]. The functional significance of these bacterial microRNAs remains to be elucidated.

Most of the sRNAs identified are antisense. This is in agreement with recent findings that extensive antisense transcription occurs in bacteria, as it does in Eukarya and in Archaea [14], [23], [28][31]. The target gene for AS sRNAs is likely to be the antisense CDS, whose expression is negatively controlled either by interfering with transcription/translation or by favoring mRNA degradation. The analysis of pathways regulated by AS sRNAs identified enrichment for two-component systems and cell-wall components. This preliminary finding suggests that cis-acting RNAs may inhibit stress-responsive modules, such as the two-component systems [32][34], and they may also affect other critical functions such as the metabolism of the cell wall, which is a virulence determinant of the tubercle bacillus [35][37].

We have hypothesized a possible function for some of the 5′/3′ UTR sRNAs validated by Northern blot. Some of our sRNAs have a role in transcription attenuation: candidate 1502, upstream of gene ilvB (Figure S4), coincides with the previously described mcr9-mpr14 [20]. However, while DiChiara et al. detected this sRNA only in Mycobacterium bovis BCG, we detected mcr9-mpr14 in MTB H37Rv. Since the ilvB gene is part of the isoleucine-valine operon, the presence of the short peptide encoded by the candidate 1502 (with its typical structure “3 valine, 1 isoleucine”) suggests that also in mycobacteria, as in most bacteria investigated so far [38], this sRNA is part of an attenuator system for controlling valine-isoleucine synthesis. In order to confirm the existence of attenuation in mycobacteria, we analyzed the sRNA candidate that was mapped upstream of leuA. This candidate (#1938; see Table S1) was classified among class C sRNAs [22] and was not included in our initial selection. By sequence analysis we found a 13 amino acids long coding region, containing a stretch of 3 consecutive leucine codons, and followed by a stem-loop structure suggesting a possible role in the control of leuA expression. The identification by Northern analysis of multiple bands at 85, 65 and 45 nt, respectively (Figure S5) is in good agreement with an attenuator role of this candidate. Interestingly, this sRNAs is the only species tested in this work that showed differential relative signal intensity between virulent and avirulent M. tuberculosis strains (Figure S5), suggesting a different control of this biosynthetic pathway in these two mycobacteria.

Our work suggests the identification of alternative RNA-based regulatory tools such as riboswitches within the 5′-end of mRNAs. Riboswitches are part of the mRNA molecule they regulate, usually the 5′UTR, and hence act in cis. They can adopt different conformations in response to environmental signals, including stalled ribosomes, uncharged tRNAs, elevated temperatures, or small molecule ligands [39]. Thus, riboswitches include the above-mentioned attenuators. In accordance with the findings of Arnvig et al. [21], our data suggest that in MTB this kind of regulation is rather widespread, as it was found in other Gram-positive bacteria like B. subtilis [40], [41].

Arnvig and colleagues used the reads-coverage ratio between the 5′ UTR and the downstream CDS to better identify putative riboswitches [21]. We performed RNA-seq only on the small-size RNA fraction, therefore the comparison of reads-coverage between candidates and CDSs could not be performed. However, by combining our experimental observations and computational analysis we could detect some putative riboswitches.

Indeed, our data suggest that candidate 1149 might be co-transcribed with TB8.4 and could act as a riboswitch, regulating TB8.4 expression, encoding a low molecular weight T-cell antigen [42] The candidate 1096 could also be a riboswitch regulating pstS2, one of the three components of the inorganic phosphate detection component of the membrane-associated phosphate-specific transporter (Pst) [43], [44]. pstS2 knockout strains showed attenuated virulence [44]. Moreover, pstS2 and pknD genes are encoded within the same operon, and pknD knockout strains do not survive in inorganic phosphate poor-growing conditions [43]. Thus, candidate 1096 may have a crucial sensing role in controlling the pstS2-pknD operon. Interestingly, the candidate 135 mapped antisense to the 5′ region of the pstS2 gene (Figure S3). Thus, the operon encoding for this gene seems to be under tight control.

Other putative riboswitches we have identified are candidate 161, controlling expression of a nitrate reductase (narGHJI operon) [45], [46] (Figure S3) and candidate 1565, mapping within the 5′UTR of groES, encoding for a co-chaperonin protein [47]. Arnvig and colleagues [21] identified the expression profile for the 5′ UTR associated with the groES gene. We suggest that this sRNA may be a thermosensor, as frequently found in other bacteria upstream of genes with a function similar to groES.

Several predicted intergenic sRNAs have been confirmed by Northern analysis (149, 161, 224 and 1096 in Table 2). A more careful analysis of other sRNAs indicated that some of them were likely to be also intergenic. This is the case for candidate 1359. Looking at the annotated sequence we observed that this sRNA maps within Rv2165c (mraW, S-adenosyl-methyltransferase). However, by comparing the polypeptide encoded by MTB with the protein encoded by M. bovis BCG we found that homology starts 53 amino acids downstream of the indicated start codon, where a GTG codon is present (nucleotide 2429447) preceded by a Shine-Dalgarno sequence (GGGGAGG) (Figure S6). Thus, candidate 1359 is transcribed upstream of Rv2165c, in the intergenic region.

Candidate 488, previously reported by Arnvig et al. (MTS2458; [21]), and originally classified in this work as antisense to 3′UTR of Rv3191c, may be intergenic since it was not overlapped to the 3′end of the Rv3191c, as supported by the length of the transcript visualized on Northern blot.

Candidate 498 is transcribed in opposite direction to both neighboring genes TB7.3 and Rv3321A. Its predicted length (300 nt) classified this sRNA as a 3′UTR to Rv3321A. However, the length observed by Northern blot, the presence of both a sigma A consensus sequence and a possible stem-loop structure support the hypothesis that the candidate 498 is a trans-acting sRNA, whose targets have to be identified, rather than a sRNA overlapping to the Rv3321A-3′ end.

Candidate 190 mapped upstream of 16S rRNA gene being entirely contained in mcr3 sRNA gene [20]. Interestingly, upstream of the 5′ end of mcr3 no promoter consensus sequences were present. On the contrary, upstream of the 5′end of 190, which is internal to mcr3, we found −10 and −35 sigA consensus sequences, that can be identified with the most active rRNA operon promoter, P3 [48][50]. Thus sRNAs such as 190, 191, and 192 mapping in the ribosomal DNA could be cis-acting regulatory elements controlling unique MTB rRNA operon.

The candidate 193 (predicted length: 60 nucleotides) mapped between the ribosomal RNAs 23S and 5S (rrl and rrf genes, respectively). According to the positive signal obtained by Northern blot, this putative sRNA might be a residual RNA derived from ribosomal RNA processing. Additional bands at higher and lower molecular weight observed by Northern blot support this hypothesis. Very recently, ribosomal noncoding RNAs have been described in eukaryotic cells [51]. Despite similar mechanisms have not been described in bacteria, further experiments are needed to confirm the small RNA nature of this candidate originating by the ribosomal genomic region.

Overall, our data greatly increase the complexity of sRNA complement in MTB and suggest that RNA-mediated regulation in this organism may be as common and multifaceted as in other bacteria [3], [52]. Furthermore, we detected several new sRNAs and riboswitch-like regulators, which provide new insights into the dynamic regulation of gene expression in MTB. Indeed, riboswitches represent a cost-effective means for genetic regulation compared with protein synthesis and they also provide an immediate feedback response.

Materials and Methods

Ethics Statement: N/A

RNA extraction.

For analysis of sRNAs, M. tuberculosis H37Rv was grown in Middlebrook 7H9 medium supplemented with 10% OADC (Oleic Acid, Albumin, Dextrose, Catalase). Exponential growth phase culture was harvested at OD600 between 0.5 and 0.8 and washed twice in PBS. The small-size RNA (approx. <200 nucleotides) enriched fraction was extracted using the mirVana™ miRNA Isolation Kit (Ambion, Austin, TX, USA) according to manufacturer’s instruction. Both small-size RNA enriched and total RNA (depleted sRNAs) fractions were collected and analysed with Agilent 2100 Bioanalyzer (Agilent Biotechnologies, Santa Clara, CA, USA).

Candidate sRNAs, Promoter and Terminator Consensus Sequences

A custom analysis pipeline was developed for analyzing Illumina reads (details are reported in Pellin et al. [22]): after computational processing of reads to remove the poly-A tail, reads have been mapped to the whole genome using SOAPv1 tool, and those regions coding for proteins or for functional RNA molecules (tRNAs and rRNAs) were excluded from the analysis. To identify putative loci encoding for sRNAs a coverage map and a conservation map were superimposed. Coverage value of a specific genome base derives from RNA-seq data and corresponds to the count of reads overlapping that specific position. The calculation of the conservation value is based on the analysis of a BLASTN 2.0 output file, regarding the pairwise alignment of each MTB intergenic sequences and a database containing all intergenic sequences extracted from all the others genomes in the MTB genus. The conservation value then corresponds to the weighted count of genomes within the MTB genus, with at least one pairwise alignment hit that satisfies high quality criteria.

To identify sRNA candidates the following criteria were considered:

1- region coverage by consecutive nucleotides;

2- region length ≥30 and ≤550 nucleotides.

And at least one of the following conditions must be satisfied:

3.1 RNA-seq coverage value ≥50 or,

3.2 Genomic region conservation value ≥1.77 or,

3.3 RNA-seq coverage value ≥22 and genomic region conservation value ≥0.97.

Coverage and conservation thresholds were established on the empirical distribution functions and determined by means of statistical criteria according to data characteristics.

In order to assess candidates’ reliability, many different characteristics of each putative sRNA were analyzed. In particular, Minimum Folding Energy (MFE) and relative p-value (MFE p-value) were taken into account.

The application of such pipeline (fully described in [22]) to our data resulted in the identification of 1948 sRNA candidates, classified as Type A (based on criteria 1, 2 and 3.1; 1378 candidates), Type B (based on criteria 1,2 and 3.3; 114 candidates) and Type C (based on criteria 1,2 and 3.2; 456 candidates). In this work we focused on Type A candidates only. In fact, we would expect that the most promising candidates are included in this category, whereas Type B and Type C could include higher number of false positives due to the criteria used for their identification.

For the present work, an update on the annotation of the candidates has been performed in January 2012 to include in the genome most recently annotated CDSs [53]. In particular, the analysis published in Pellin et al. [22] was performed using as reference the NC_000962.gff database; whereas for the update the NC_000962.2.gff database was used. Candidates obtained by RNA-seq were further analyzed for the presence of promoter and/or terminator consensus sequences. For the analysis of the promoter region we used the consensus sequences described by Sachdeva and coworkers [54]. In particular, we analyzed the region between −50 and +15 nucleotides from the predicted 5′-end of the candidates for the presence of a Sigma A (sigA) −10 consensus motif (mismatch: 0). In particular, we analyzed the region between −50 and +15 nucleotides from the predicted 5′-end of the candidates for the presence of a Sigma A (sigA) −10 consensus motif TANNNT (with allow mismatch equal to 0) by means of a custom script written in R and using BioConductor package Biostrings [55].

In order to identify putative terminators, we applied both GeSter and RNIE algorithm.

Gester analysis was performed by means of the web based version called WebGester [56] with parameters set to default. Results reported in this paper regards to putative terminators satisfying ΔG cut-off equal to −16.98.

The RNIE algorithm was run in the “gene mode” that is optimized to individually annotate the downstream regions of genes or, more generally, putative encoding regions. We followed the same pipeline suggested by RNIE authors in [25]. Initially we took all the annotated coding sequences from the MTB complete genome and extracted subsequences from −20 to +80 nt around all annotated gene termini, according to available annotation file. Each of these was folded using the RNAfold routine from the Vienna package [57] and then subjected to a permutation test where native MFEs were compared to the pooled distribution of MFEs for 1000 permuted sequences with the same dinucleotide content in each termini. The regions that had a P<0.001 where subsequently fed into the alignment and folding algorithm Cmfinder [58]. The three most significant covariance model obtained from MTB CDS and TRIT model proposed in [25] were deployed to annotate the downstream regions of putative candidates.

Microarray for sRNA Candidates Expression

CombiMatrix microarray technology (Irvine, CA, USA) has been adopted to validate putative sRNA candidates by expression evidence. Probe sets for candidates have been designed according to the following criteria: the probes were non-overlapping sequences of 30 nt, spanning all the candidate length. Probes were located at least 5 nt from each other and from the candidate boundaries. As result, probe sets were made of one single probe when candidates were shorter than 75 nt of length; two probes when candidates length were between 75 and 110 nt; three probes when candidates length were between 110 and 180 nt, and four probes for longer candidates. To each probe, a control mismatch probe was provided (mismatch: two bases). As positive controls, probes targeting the following genes have been included: sigA (Rv2703), 16S (rrs, Rvnr01), and 5S (rrf, Rvnr03). As negative controls we considered the same number of probes of the positive controls, made of random sequences not mapping in the M. tuberculosis H37Rv genome. Each array contained two complete probe sets and relative mismatches, each in duplicate. Small-size RNA fraction and total RNA fraction depleted of small-size RNAs were prepared using mirVana miRNA Isolation kit (Applied Biosystems/Ambion, Austin, TX, USA). The ULS labeling kit for CombiMatrix arrays (Kreatech Biotechnology BV, Amsterdam, The Netherlands) was used to label RNA samples with biotin and for hybridization on a 12K array followed by electrochemical detection on the ElectraSense reader performed according to manufacturer’s instructions. Two independent experiments were carried out starting from two biological replicates each (M. tuberculosis H37Rv grown in Middlebrook 7H9 10% OADC, OD600 between 0.5 and 0.8). Microarray hybridization was carried out twice (technical replicate) for each biological replicate; every microarray was acquired at least twice. Results were analyzed applying Robust Multichip Analysis (RMA) procedure on the logarithm of the background (mismatches) corrected perfect match electrical current intensities.

We compared the candidates expression with controls by means of a t-type statistic test. P-values were computed by means of permutations, to avoid any distributional assumption, and adjusted for multiplicity by means of the Holm-Bonferroni method [59], [60]. Candidates showing an expression mean significantly higher than the expression of controls (p-value <0.05) were considered as validated. Then, a combined p-value (namely p-adj) for each candidate was calculated with Fisher rule considering both RNA-seq and microarray results.

Criteria for the Classification of sRNAs

We decided to consider approximately 60 nucleotides at the 5′ end of a gene as putative promoter region. As reported by Gardner and colleagues [25], the majority of termination signals are within 80 nucleotides from the 3′-end of a gene. Thus, we have considered as intergenic those sRNAs mapping more than 80 nt upstream or downstream of a neighboring CDSs. Candidates mapping within ≤80 nucleotides from a CDS on the same strand were considered potentially synthesized as untranslated regions (5′/3′ UTRs). Candidates synthesized by the strand opposite to a CDS or a UTR are antisense (AS) sRNAs, and can be divided in subclasses depending on their relative position.

Northern Blot Analysis

10 µg of total RNA were separated on a 6% denaturing polyacrilamyde gel, blotted on a positively charged membrane (Hybond N+, GE Life Sciences), and hybridized to specific 32P-labeled riboprobes or oligonucleotides, as described previously [61]. Membranes were exposed to a phosphor screen and visualized with a phosphorimager. The oligonucleotides used are reported in Table S4 and Table S5, respectively.

Primer Extension Analysis

The oligonucleotide probes were 5′ end labeled with T4 polynucleotide kinase (Promega). 5′ ends were identified by primer extension analysis, as described in Boorstein and Craig [62]: an annealing mix (10 µl), containing 10 µg of total RNA, 1 U/µl RNasin (Promega), 0.5 pmole radiolabeled oligo and 1X ss-hybridization buffer (300 mM NaCl, 10 mM Tris HCl pH 7.5, 2 mM EDTA), was denatured at 80°C for 4 minutes and incubated for 2 hours at 50°C for the annealing. Then, to the annealing mix were added 40 µl of 1.25 X RT-buffer (1.25 mM of each dNTP, 12.5 mM DTT, 12.5 mM Tris HCl pH 8, 7.5 mM MgCl2), 5 U RNasin (Promega) and 10 U AMV Reverse Transcriptase (Finnzymes) and incubated 30 min at 50°C for the extension. RNA was hydrolyzed with NaOH and all samples were neutralized with HCl and then precipitated and dissolved in 6 µl stop mix (95% formamide, 20 mM EDTA, 0.05% bromophenol blue, 0.05% xylene cyanol FF). Reaction products were separated on a 6% denaturing polyacrylamide gel along with sequencing reactions made with the same labeled oligonucleotides used for the primer extensions.

Statistical Analysis of the −10 Region

We used a statistical approach to further validate sRNA candidates. The approach described by Dornenburg [23] was adapted for this scope.

In summary, we compared the frequency of specific nucleotides at fixed positions within the −10 consensus region. Due to the lack of consistent published data on −10 consensus sequences in MTB, we considered only the TANNNT hexamer recognized by the sigma factor sigA. Therefore, we analyzed the frequency of T in position −13, A in position −12 and T in position −8 for our sRNA candidates, for annotated CDSs in the genome of MTB and for a set random sequences, all of the same sample size. Random sequences were generated and used as control. A score of 3 was attributed when all the 3 nucleotides TAT were found at the expected position (−13, −12, −8, respectively); a score of 2, 1 or 0 was attributed when 2, 1 or 0 nucleotides, respectively were found at expected positions. An average of the scores was then calculated for each of the three categories (candidates, annotated CDSs and control sequences). A Mann-Whitney test statistic was used to compare scores and p-values computed by permutation methods. For comparison we considered a number of random sequences equal to the sample size considered by the permutation. Sample sizes were as follow: 1948 candidates (all), 1373 type A candidates, 260 validated candidates, 4048 CDSs. All statistical analyses were performed in R and Bioconductor environment [55].

Functional Enrichment of Antisense Regulated Genes

We obtained functional annotation for MTB genes from three different sources: (i) GO terms from the DAVID database [63], (ii) metabolic pathways from the KEGG database [64], and (iii) functional categories from the TubercuList database ( We then applied the Fisher’s Exact Test to determine biological terms that are enriched among antisense-regulated genes, based on the contingency table shown in Table S6. The test was slightly modified by removing one gene from the list of interest (i.e. a-1) before computing the Fisher exact probability; this modification removes the effect of terms based on single genes only [65].

Supporting Information

Figure S1.

Schematic representation of sRNA candidates classification according to their mapping position. Thick black arrows indicate coding sequences (CDS); thin arrows indicate sRNAs; dotted arrows the distance from the 5′/3′ ends of CDS to the putative 5′/3′end of the sRNA; AS: antisense.


Figure S2.

Schematic representation of the distance between the −10 consensus hexamer in candidate sRNAs, annotated coding region and random sequences. Picture of the relative mean score differences among sequences groups. To preserve as best as possible these dissimilarity, we obtained the plot coordinates from a non-linear multidimensional scaling of the pairwise absolute differences, taken as inter-groups distance matrix.


Figure S3.

Schematic representation of the mapping position of sRNA candidates validated by Northern blot (from 54 to 1029). Empty arrows represent coding sequences (CDS), whereas sRNA candidates are reported as grey arrows. Dotted arrow (candidate 1359) represents the possibly erroneous annotation of the CDS Rv2165c. Dotted lines represent UTRs.


Figure S4.

Schematic representation of the mapping position of sRNA candidates validated by Northern blot (from 1080 to 1565).


Figure S5.

Northern blot for candidate 1938 in M. tuberculosis H37Rv and H37Ra (exponential growth phase).


Figure S6.

Schematic representation for the mapping position of the sRNA 1359 validated by Northern blot.


Table S1.

Dataset including all the candidates (A, B, C types) representing the starting point of the present work.


Table S2.

Functional categories of genes showing RNA species within the 5′ UTR. Classification was performed according to TubercuList.


Table S3.

Functional enrichment of cis- regulated genes.


Table S4.

List of oligonucleotides used in the Northern blot. * Oligo: 32P-labelled oligonucleotides used as probes in the Northern blot assays; FWD or REV: primers used to amplify the genomic regions to be used as templates in the in vitro transcription of the riboprobes.


Table S5.

List of other oligonucleotides used for the primer extension. DAVID database and TubercuList have been used for functional annotation of MTB genes. GO: gene onthology; mtu: KEGG pathway. TBL: TubercuList.


Table S6.

Contingency table for evaluating functional enrichment of antisense-regulated genes. a = number of AS-regulated genes annotated with term j; b = number of AS-regulated genes not annotated with term j. The quantities c, and d are similarly defined for the remaining genes not overlapped by AS sRNAs. N is the total number of M. tuberculosis genes for which annotation is available.


Author Contributions

Conceived and designed the experiments: PM DMC DG MLG. Performed the experiments: PM FF. Analyzed the data: PM FF AA DP DFV. Contributed reagents/materials/analysis tools: CDS GB. Wrote the paper: PM DMC MLG DG.


  1. 1. Storz G, Altuvia S, Wassarman KM (2005) An abundance of RNA regulators. Annu Rev Biochem 74: 199–217.
  2. 2. Waters LS, Storz G (2009) Regulatory RNAs in bacteria. Cell 136: 615–628.
  3. 3. Storz G, Vogel J, Wassarman KM (2011) Regulation by small RNAs in bacteria: expanding frontiers. Mol Cell 43: 880–891.
  4. 4. Viegas SC, Arraiano CM (2008) Regulating the regulators: how ribonucleases dictate the rules in the control of small non-coding RNAs. RNA Biol 5: 230–243.
  5. 5. Repoila F, Darfeuille F (2009) Small regulatory non-coding RNAs in bacteria: physiology and mechanistic aspects. Biol Cell 101: 117–131.
  6. 6. Thomason MK, Storz G (2010) Bacterial antisense RNAs: how many are there, and what are they doing? Annu Rev Genet 44: 167–188.
  7. 7. Romby P, Charpentier E (2010) An overview of RNAs with regulatory functions in gram-positive bacteria. Cell Mol Life Sci 67: 217–237.
  8. 8. Waddell SJ, Butcher PD, Stoker NG (2007) RNA profiling in host-pathogen interactions. Curr Opin Microbiol 10: 297–302.
  9. 9. Toledo-Arana A, Repoila F, Cossart P (2007) Small noncoding RNAs controlling pathogenesis. Curr Opin Microbiol 10: 182–188.
  10. 10. Massé E, Salvail H, Desnoyers G, Arguin M (2007) Small RNAs controlling iron metabolism. Curr Opin Microbiol 10: 140–145.
  11. 11. Levine E, Hwa T (2008) Small RNAs establish gene expression thresholds. Curr Opin Microbiol 11: 574–579.
  12. 12. Papenfort K, Vogel J (2010) Regulatory RNA in bacterial pathogens. Cell Host Microbe 8: 116–127.
  13. 13. Zhou Y, Xie J (2011) The roles of pathogen small RNAs. J Cell Physiol 226: 968–973.
  14. 14. Georg J, Hess WR (2011) cis-antisense RNA, another level of gene regulation in bacteria. Microbiol Mol Biol Rev 75: 286–300.
  15. 15. Pieters J (2008) Mycobacterium tuberculosis and the macrophage: maintaining a balance. Cell Host Microbe 3: 399–407.
  16. 16. Russell DG, Vanderven BC, Glennie S, Mwandumba H, Heyderman RS (2009) The macrophage marches on its phagosome: dynamic assays of phagosome function. Nat Rev Immunol 9: 594–600.
  17. 17. Atun R, Raviglione M, Marais B, Zumla A (2010) Tuberculosis control is crucial to achieve the MDGs. Lancet 376: 940–941.
  18. 18. Rohde KH, Veiga DFT, Caldwell S, Balazsi G, Russell DG (2012) Linking the transcriptional profiles and the physiological states of Mycobacterium tuberculosis during an extended intracellular infection. PLoS Pathog 8: e1002769.
  19. 19. Arnvig KB, Young DB (2009) Identification of small RNAs in Mycobacterium tuberculosis. Mol Microbiol 73: 397–408.
  20. 20. DiChiara JM, Contreras-Martinez LM, Livny J, Smith D, McDonough KA, et al. (2010) Multiple small RNAs identified in Mycobacterium bovis BCG are also expressed in Mycobacterium tuberculosis and Mycobacterium smegmatis. Nucleic Acids Res 38: 4067–4078.
  21. 21. Arnvig KB, Comas I, Thomson NR, Houghton J, Boshoff HI, et al. (2011) Sequence-based analysis uncovers an abundance of non-coding RNA in the total transcriptome of Mycobacterium tuberculosis. PLoS Pathog 7: e1002342.
  22. 22. Pellin D, Miotto P, Ambrosi A, Cirillo DM, Di Serio C (2012) A genome-wide identification analysis of small regulatory RNAs in Mycobacterium tuberculosis by RNA-Seq and conservation analysis. PLoS One 7: e32723.
  23. 23. Dornenburg JE, DeVita AM, Palumbo MJ, Wade JT (2010) Widespread antisense transcription in Escherichia coli. mBio 1: e00024–10.
  24. 24. Mitra A, Angamuthu K, Nagaraja V (2008) Genome-wide analysis of the intrinsic terminators of transcription across the genus Mycobacterium. Tuberculosis 88: 566–575.
  25. 25. Gardner PP, Barquist L, Bateman A, Nawrocki EP, Weinberg Z (2011) RNIE: genome-wide prediction of bacterial intrinsic terminators. Nucleic Acids Res 39: 5845–5852.
  26. 26. Liu JM, Livny J, Lawrence MS, Kimball MD, Waldor MK, et al. (2009) Experimental discovery of sRNAs in Vibrio cholerae by direct cloning, 5S/tRNA depletion and parallel sequencing. Nucleic Acids Res 37: e46.
  27. 27. Lee HJ, Hong SH (2012) Analysis of microRNA-size, small RNAs in Streptococcus mutans by deep sequencing. FEMS Microbiol Lett. 326: 131–136.
  28. 28. Mitschke J, Vioque A, Haas F, Hess WR, Muro-Pastor AM (2011) Dynamics of transcriptional start site selection during nitrogen stress-induced cell differentiation in Anabaena sp. PCC7120. Proc Natl Acad Sci U S A 108: 20130–20135.
  29. 29. Shinhara A, Matsui M, Hiraoka K, Nomura W, Hirano R, et al. (2011) Deep sequencing reveals as-yet-undiscovered small RNAs in Escherichia coli. BMC Genomics 12: 428.
  30. 30. Rasmussen S, Nielsen HB, Jarmer H (2009) The transcriptionally active regions in the genome of Bacillus subtilis. Mol Microbiol 73: 1043–1057.
  31. 31. Sharma CM, Hoffmann S, Darfeuille F, Reignier J, Findeiss S, et al. (2010) The primary transcriptome of the major human pathogen Helicobacter pylori. Nature 464: 250–255.
  32. 32. Szurmant H, Hoch JA (2010) Interaction fidelity in two-component signaling. Curr Opin Microbiol 13: 190–197.
  33. 33. Cole ST, Brosch R, Parkhill J, Garnier T, Churcher C, et al. (1998) Deciphering the biology of Mycobacterium tuberculosis from the complete genome sequence. Nature 393: 537–544.
  34. 34. Ewann F, Jackson M, Pethe K, Cooper A, Mielcarek N, et al. (2002) Transient requirement of the PrrA-PrrB two-component system for early intracellular multiplication of Mycobacterium tuberculosis. Infect Immun 70: 2256–2263.
  35. 35. Parish T, Smith DA, Kendall S, Casali N, Bancroft GJ, et al. (2003) Deletion of two-component regulatory systems increases the virulence of Mycobacterium tuberculosis. Infect Immun 71: 1134–1140.
  36. 36. Malhotra V, Sharma D, Ramanathan VD, Shakila H, Saini DK, et al. (2004) Disruption of response regulator gene, devR, leads to attenuation in virulence of Mycobacterium tuberculosis. FEMS Microbiol Lett 231: 237–245.
  37. 37. Göpel Y, Görke B (2012) Rewiring two-component signal transduction with small RNAs. Curr Opin Microbiol 15: 132–139.
  38. 38. Patek M (2007) Branched chain amino acids. Microbiol Monogr 5: 129–162 DOI: 10.1007/7171_2006_070.
  39. 39. Grundy FJ, Henkin TM (2006) From ribosome to riboswitch: control of gene expression in bacteria by RNA structural rearrangements. Crit Rev Biochem Mol Biol 41: 329–338.
  40. 40. Irnov, Kertsburg A, Winkler WC (2006) Genetic control by cis-acting regulatory RNAs in Bacillus subtilis: general principles and prospects for discovery. Cold Spring Harb Symp Quant Biol 71: 239–249.
  41. 41. Irnov I, Sharma CM, Vogel J, Winkler WC (2010) Identification of regulatory RNAs in Bacillus subtilis. Nucleic Acids Res 38: 6637–6651.
  42. 42. Zvi A, Ariel N, Fulkerson J, Sadoff JC, Shafferman A (2008) Whole genome identification of Mycobacterium tuberculosis vaccine candidates by comprehensive data mining and bioinformatic analyses. BMC Med Genomics 1: 18.
  43. 43. Peirs P, Lefèvre P, Boarbi S, Wang XM, Denis O, et al. (2005) Mycobacterium tuberculosis with disruption in genes encoding the phosphate binding proteins PstS1 and PstS2 is deficient in phosphate uptake and demonstrates reduced in vivo virulence. Infect Immun 73: 1898–1902.
  44. 44. Vanzembergh F, Peirs P, Lefevre P, Celio N, Mathys V, et al. (2010) Effects of PstS sub-units or PknD deficiency on the survival of Mycobacterium tuberculosis. Tuberculosis 90: 338–345.
  45. 45. Weber I, Fritz C, Ruttkowski S, Kreft A, Bange FC (2000) Anaerobic nitrate reductase (narGHJI) activity of Mycobacterium bovis BCG in vitro and its contribution to virulence in immunodeficient mice. Mol Microbiol 35: 1017–1025.
  46. 46. Fritz C, Maass S, Kreft A, Bange FC (2002) Dependence of Mycobacterium bovis BCG on anaerobic nitrate reductase for persistence is tissue specific. Infect Immun 70: 286–291.
  47. 47. Coates AR, Shinnick TM, Ellis RJ (1993) Chaperonin nomenclature. Mol Microbiol 8: 787.
  48. 48. Gonzalez-y-Merchand JA, Colston MJ, Cox RA (1996) The rRNA operons of Mycobacterium smegmatis and Mycobacterium tuberculosis: comparison of promoter elements and of neighbouring upstream genes. Microbiology 142: 667–674.
  49. 49. Gonzalez-y-Merchand JA, Colston MJ, Cox RA (1998) Roles of multiple promoters in transcription of ribosomal DNA: effects of growth conditions on precursor rRNA synthesis in mycobacteria. J Bacteriol 180: 5756–5761.
  50. 50. Verma A, Sampla AK, Tyagi JS (1999) Mycobacterium tuberculosis rrn promoters: differential usage and growth rate-dependent control. J Bacteriol 181: 4326–4333.
  51. 51. Audas TE, Jacob MD, Lee S (2012) Immobilization of proteins in the nucleolus by ribosomal intergenic spacer noncoding RNA. Mol Cell 45: 147–157.
  52. 52. Gottesman S, Storz G (2011) Bacterial small RNA regulators: versatile roles and rapidly evolving variations. Cold Spring Harb Perspect Biol 3: pii: a003798. doi: 10.1101/cshperspect.a003798.
  53. 53. Wolfe LM, Mahaffey SB, Kruh NA, Dobos KM (2010) Proteomic definition of the cell wall of Mycobacterium tuberculosis. J Proteome Res 9: 5816–5826.
  54. 54. Sachdeva P, Misra R, Tyagi AK, Singh Y (2010) The sigma factors of Mycobacterium tuberculosis regulation of the regulators. FEBS Journal 277: 605–626.
  55. 55. R Development Core Team (2011) R: A language and environment for statistical computing. R Foundation for Statistical Computing. Vienna, Austria, ISBN 3–900051–07–0, URL
  56. 56. Mitra A, Kesarwani AK, Pal D, Nagaraja V (2011) WebGeSTer DB–a transcription terminator database. Nucleic Acids Res 39(Database issue): D129–135.
  57. 57. Bonnet E, Wuyts J, Rouze P, Van de Peer Y (2004) Evidence that microRNA precursors, unlike other non-coding RNAs, have lower folding free energies than random sequences. Bioinformatics 20: 2911–2917.
  58. 58. Yao Z, Weinberg Z, Ruzzo WL (2006) CMfinder–a covariance model based RNA motif finding algorithm. Bioinformatics 22: 445–452.
  59. 59. Dudoit S, van der Laan MJ, Pollard KS (2004) Multiple testing. Part I. Single-step procedures for control of general type I error rates. Stat Appl Genet Mol Biol 3: Article13.
  60. 60. Holm S (1979) A simple sequentially rejective multiple test procedure. Scand J Stat 6: 65–70.
  61. 61. Dehó G, Zangrossi S, Sabbattini P, Sironi G, Ghisotti D (1992) Bacteriophage P4 immunity controlled by small RNAs via transcription termination. Mol Microbiol 6: 3415–3425.
  62. 62. Boorstein WR, Craig EA (1989) Primer extension analysis of RNA. Methods Enzymol 180: 347–369.
  63. 63. Huang da W, Sherman BT, Lempicki RA (2009) Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc 4: 44–57.
  64. 64. Kanehisa M, Goto S (2000) KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res 28: 27–30.
  65. 65. Hosack DA, Dennis G Jr, Sherman BT, Lane HC, Lempicki RA (2003) Identifying biological themes within lists of genes with EASE. Genome Biol 4: R70.