Identification and characterization of small RNAs in Yersinia pestis.

Yersinia pestis, the etiologic agent of plague, is closely related to Yersinia pseudotuberculosis evolutionarily but has a very different mode of infection. The RNA-binding regulatory protein, Hfq, mediates regulation by small RNAs (sRNAs) and is required for virulence of both Y. pestis and Y. pseudotuberculosis. Moreover, Hfq is required for growth of Y. pestis, but not Y. pseudotuberculosis, at 37°C. Together, these observations suggest that sRNAs play important roles in the virulence and survival of Y. pestis, and that regulation by sRNAs may account for some of the differences between Y. pestis and Y. pseudotuberculosis. We have used a deep sequencing approach to identify 31 sRNAs in Y. pestis. The majority of these sRNAs are not conserved outside the Yersiniae. Expression of the sRNAs was confirmed by Northern analysis and we developed deep sequencing approaches to map 5' and 3' ends of many sRNAs simultaneously. Expression of the majority of the sRNAs we identified is dependent upon Hfq. We also observed temperature-dependent effects on the expression of many sRNAs, and differences in expression patterns between Y. pestis and Y. pseudotuberculosis. Thus, our data suggest that regulation by sRNAs plays an important role in the lifestyle switch from flea to mammalian host, and that regulation by sRNAs may contribute to the phenotypic differences between Y. pestis and Y. pseudotuberculosis.

Introduction

Yersinia pestis, the etiologic agent of plague, continues to pose a threat to human health both naturally and as a bioweapon. Y. pestis is closely related to Yersinia pseudotuberculosis. Both species are human pathogens and are believed to have evolved from each other ~1,500–20,000 y ago;, ~75% of their genes share ≥ 97% nucleotide identity. Despite their genetic similarity, the diseases caused by these two organisms vary greatly. Y. pestis infects both mammalian and arthropod hosts and is typically transmitted to humans through the bite of an infected flea. In humans, infection by Y. pestis usually manifests itself as bubonic and pneumonic plague. In contrast, Y. pseudotuberculosis is an enteropathogen that causes a gastrointestinal disease transmitted by the fecal-oral route.

Small RNAs (sRNAs) serve as important components of many regulatory circuits in bacteria. sRNAs are typically non-coding RNA molecules of < 500 nt, transcribed from intergenic regions. The majority of sRNAs characterized to date have been shown to downregulate gene expression at the post-transcriptional level by base-pairing with target mRNAs. These pairing interactions result in changes in transcription attenuation, translation initiation or mRNA stability.

Hfq is an RNA-binding protein that is required for the stability and/or regulatory function of many sRNAs. Hfq is also required for the virulence of many pathogenic bacteria, including Y. pestis and Y. pseudotuberculosis, suggesting that sRNAs are key regulators of virulence genes in these species. Moreover, Hfq is required for efficient biofilm formation and gut blockage in the flea, important processes for transmission to a mammalian host, and for the growth of some Y. pestis strains, but not Y. pseudotuberculosis, at 37°C. Strikingly, Hfq amino acid sequence is 100% identical across eight sequenced strains of Y. pestis and four of Y. pseudotuberculosis, suggesting that regulation by sRNAs rather than Hfq itself contributes to the difference in viability of Y. pestis and Y. pseudotuberculosis hfq mutants at 37°C.

Most studies of sRNAs have focused on Escherichia coli, for which > 100 sRNAs have been identified,- and deep sequencing approaches have led to the identification of similar numbers in other bacterial species.- Although all bacterial species are expected to express numerous sRNAs, conservation of sRNAs is generally poor, so the specific sRNA pool differs widely between species. Until recently, very few sRNAs had been identified in any Yersinia species. A recent study utilized a deep sequencing approach to identify 150 putative sRNAs in Y. pseudotuberculosis. The majority of the putative sRNAs identified are conserved in Y. pestis, but the expression and dependence on Hfq of five sRNAs (seven were tested) differs between the closely related species. The authors also showed that deletion of specific sRNAs in Y. pseudotuberculosis leads to attenuation of the pathogen in a mouse model of infection and that the inactivation of an sRNA in Y. pestis reduces virulence in a mouse model of pneumonic plague.

In this work, we utilized a deep sequencing approach to identify putative sRNAs expressed in Y. pestis. We confirmed expression of 31 sRNAs by Northern analysis, of which only 17 match previously identified putative sRNAs. We developed genome-scale 5′ and 3′ RACE (rapid amplification of cDNA ends) approaches to map the 5′ and 3′ ends of most of the sRNAs. We observed a wide variety of expression patterns that depend upon temperature and the presence of Hfq. All of the sRNAs we identified are conserved in Y. pseudotuberculosis (most are 100% identical), most are conserved in other Yersinia species, but fewer than half are conserved in E. coli. We observed detectable expression in Y. pseudotuberculosis for all but one of the sRNAs, but the temperature- and Hfq-dependent expression patterns of many sRNAs differed between Y. pestis and Y. pseudotuberculosis. Thus, our data suggest that differences in sRNA expression may contribute to the differences in Y. pestis and Y. pseudotuberculosis biology.

Identification of putative sRNAs in Y. pestis

To identify novel Y. pestis sRNAs, we purified RNA from Y. pestis KIM6+ grown at 37°C and constructed a cDNA library for Illumina sequencing. Following sequencing and mapping of reads to the reference genome, we identified genomic regions with contiguous sequence reads that partially or fully overlap an intergenic region (JCVI genome annotation), with at least one position having > 500 mapped sequence reads. Thus, we generated a list of 50 putative sRNAs with a high level of confidence. We excluded repetitive sequence, although we noted that many sequences mapped to repetitive sequence partially overlapping predicted transposases. Fully antisense RNAs could not be identified due to the lack of strand information in the sequencing data.

Validation and characterization of sRNAs by northern blot

To confirm the presence of the putative sRNAs, and to determine their expression profiles in Y. pestis and Y. pseudotuberculosis, we isolated RNA from both species at 28°C and 37°C in hfq+ cells, isogenic Δhfq mutants and Δhfq strains complemented with a multi-copy plasmid that encodes hfq. We then performed Northern analysis with radiolabeled oligonucleotides designed to probe each sRNA. The deep sequencing data did not provide strand information so sRNAs were first probed on the plus strand, and any that could not be detected were then probed on the minus strand. This approach confirmed that 32 of the 50 putative sRNAs are expressed at a detectable level and have a size consistent with that of an sRNA (< 500 nt). All Northern-confirmed sRNAs are listed in Table 1 and shown in . Representative examples of the northern blots are shown in Figure 1A. Confirmed sRNAs were assigned “Ysr” (Yersinia sRNA) names, in accordance with previously identified sRNAs in Y. pseudotuberculosis. One putative sRNA is in fact a protein-coding mRNA for the gene rmf (see below). Of the 31 confirmed sRNAs, 14 have not been described previously, and of the remaining 17, only five have been detected by a method other than deep sequencing.

Table 1. List of validated sRNAs

sRNA namea	Start coordinateb	End coordinateb	Gene context
Ysr151/RnpB	125535	125684	y0115 > / < / < y0117
Ysr88/152	143205	143274	y0129 (gltD) > / > /y0130 >
Ysr148/153/GlmZ	421034	421106, 421107	< y0376/ > /y0377 (hemY) >
Ysr7/154/MicA	999263	999189	y0887 (gshA) > / < /y0888 >
Ysr155/RyfD	1021180	1021317	y0913 > / > /y0914 (clpB) >
Ysr156/Ffs	1179191	1179276	< y1045/ > / < y1046 (hha)
Ysr145/157	1229994	1230059, 1230060	< y1090/ > / < y1091
Ysr158	1341830	1341894	y1196 (ubiF) > / > / < yt030 (tRNA-Gln)
Ysr159/CyaR	1527324	1527218	< y1378/ < / < y1379
Ysr23/160	1593088, 1593089	1593141, 1593147, 1593149	y1436 > / > / < y1437
Ysr161	1624496	1624402, 1624433	< y1465 (nanT)/ < /y1466 (cysP) >
Ysr163	1902373	1902412	< y1720/ > /y1722 >
Ysr164	1916071, 1916098	1915948	y1731 > / < /y1732 (ypeR) >
Ysr165	1961869	1961985	y1782 > / > /y1783 >
Ysr11/166/FnrS	2195768	2195887	y1994 (zntB) > / > / < y1995
Ysr167	2363578	2363674	< y2138/ > /y2139 (rstA) >
Ysr73/169	2457081	2457002	< y2228/ < / < y2229
Ysr170	2550744	2550868, 2550869	y2316 > / > /y2317 >
Ysr171	2815750	2815620	y2553 (manZ) > / < /y2554 >
Ysr172	2841229	2841287, 2841312	< y2579/ > / < y2580
Ysr174	3097172	3097124	< y2796 (serS)/ < / < y2797
Ysr65/175	3111993	3111905	< y2808/ < / < y2809 (clpA)
Ysr177	3330179	3330055	< y3023 (moaA)/ < /y3024 >
Ysr179/CsrB	3457464	3457752	< y3145/ < / < y3146 (syd)
Ysr45/180/GcvB	3467030	3467076	< y3154 (gcvA)/ > /y3155 >
Ysr149/181	3501524, 3501536	3501452, 3501453	< y3181 (aas)/ < / < y3182
Ysr182/6S RNA	3632001, 3632002	3632114	y3299 > / > /y3300 >
Ysr183/SroG	3904566	3904637	< y3520 (ribB)/ < /y3522 >
Ysr185/Spot 42	4230754	4231124	< y3808/ < /y3809 (engB) >
Ysr186/CsrC	4232923	4233049	< y3810/ > / < y3811 (polA)
Ysr146.2/187	4314248	4314315	< y3871/ > /y3872 (ompR) >

For sRNAs identified by Koo et al., the existing name is indicated first, followed by our systematic name. Names of E. coli homologs are also indicated where appropriate, e.g., Ysr148/153/GlmZ was identified by Koo et al. as Ysr148, our systematic name is Ysr153, and it is homologous to E. coli GlmZ. Underlined names indicated sRNAs that have not been identified previously in Yersinia species.

Bold text indicates ends identified using Deep RACE techniques. Other end coordinates are estimates derived manually from the RNA-seq data.

Figure 1. (A) Verification of Ysr expression by northern blot analysis. All northern blots are shown in and duplicate northern blots for most sRNAs are shown in . A northern blot for hfq mRNA from a corresponding set of RNA samples is also shown (note that this blot is reproduced as part of ). Corresponding 5S rRNA northern blots for four of the same membranes are shown in . (B) Clusters of sRNA based on k-means clustering of sRNA expression data. Rows correspond to individual RNAs, while columns correspond to conditions for which expression was measured by northern blot. Shading indicates the relative expression of sRNAs for each strain/condition. Expression numbers are indicated as a percentage of the level for condition in which the RNA level is highest. (C) Northern analysis of Ysr170, Ysr172 and Ysr179/CsrB in additional strains of Y. pestis and Y. pseudotuberculosis. Duplicate northern blots are shown in . Note that Y. pseudotuberculosis PTB52c is the same strain as used in (A).

We observed a remarkable variety of expression patterns with respect to temperature, dependence upon hfq and species. We used an unsupervised learning algorithm to group the sRNAs into seven clusters, based on their expression patterns (Fig. 1B). These clusters highlight expression patterns that are common to multiple sRNAs. Clusters 1 and 2 consist largely of sRNAs that are constitutively expressed in both species, regardless of temperature or the presence of hfq (Ysr155/RyfD, Ysr156/Ffs, Ysr161, Ysr163, Ysr177, Ysr182/6S RNA, Ysr183/SroG, Ysr146.2/187, Ysr151/RnpB, Ysr88/152, Ysr73/169, Ysr65/175 and Ysr186/CsrC). Cluster 3 consists of sRNAs that are expressed similarly in both species but whose expression is dependent upon the presence of hfq (Ysr145/157, Ysr159/CyaR, Ysr164 and Ysr149/181). Cluster 4 consists of sRNAs that are expressed in both species but whose expression is dependent upon the presence of hfq only in Y. pestis (Ysr151/RnpB, Ysr148/153/GlmZ, Ysr7/154/MicA and Ysr158; the protein-coding RNA, Ysr173/rmf, also fell in Cluster 4). Cluster 5 consists of RNAs that are expressed in both species, whose expression in Y. pestis is dependent upon the presence of hfq, and whose expression is higher at 37°C than 28°C (Ysr167, Ysr170, Ysr171 and Ysr174). We also observed sRNAs whose expression increased in the absence of hfq in Y. pseudotuberculosis but not Y. pestis (Ysr23/160 and Ysr165 from Cluster 6), whose expression in both species increased in the absence of hfq at 28°C but decreased in the absence of hfq at 37°C (Ysr179/CsrB from Cluster 6) or whose expression was only detectable in Y. pestis (Ysr172, the sole member of Cluster 7).

Interestingly, for sRNAs in Cluster 5, as well as Ysr158, Ysr173/rmf (Cluster 4) and Ysr165 (Cluster 6), deletion of hfq in Y. pseudotuberculosis had no substantial effect on sRNA levels whereas expression of hfq from a multi-copy plasmid in the Δhfq strain resulted in a substantial decrease in sRNA levels relative to hfq+ (Fig. 1B). This result was observed for independent biological replicates and may be due to aberrant effects of Hfq overexpression. Consistent with this, probing the same membranes with radiolabeled oligonucleotide specific to hfq mRNA revealed that hfq is grossly overexpressed in plasmid-complemented Y. pseudotuberculosis but not Y. pestis (Fig. 1A; ). The hfq northern blot data also indicated that hfq transcript levels are substantially lower at 37°C than at 28°C in both Y. pestis and Y. pseudotuberculosis, suggesting that varying Hfq levels may contribute to temperature-dependent changes in expression of some sRNAs.

Effects of hfq and temperature on sRNA levels in other Y. pestis and Y. pseudotuberculosis strains

To determine whether the different effects of hfq and temperature on sRNA expression between Y. pestis and Y. pseudotuberculosis are species-specific rather than strain-specific, we measured expression of three sRNAs, Ysr170, Ysr172 and Ysr179/CsrB, in another Y. pestis strain (CO92) and three other Y. pseudotuberculosis strains (PTB51c, PTB57c and PTB54c; Fig. 1C). The effect of temperature on sRNA expression was consistent across all strains. Specifically, in both Y. pestis and Y. pseudotuberculosis, expression of Ysr170 is higher at 37°C than at 28°C, whereas expression of Ysr179 is higher at 28°C than at 37°C (expression of Ysr172 is unaffected by temperature). There are species-specific differences in expression patterns for two of the sRNAs that are conserved across all strains tested: the temperature dependence of Ysr179 expression is greater in Y. pseudotuberculosis than in Y. pestis (Fig. 1A and C), and Ysr172 is only expressed in Y. pestis. In contrast, the effect of the Δhfq mutation in Y. pestis was not completely consistent between the KIM and CO92 strains. Specifically, Ysr170 expression is less dependent upon hfq in Y. pestis CO92 than in KIM, Ysr172 expression is not dependent upon hfq in Y. pestis CO92 (fully dependent in KIM), and hfq suppresses Ysr179 expression at 28°C in Y. pestis KIM but not CO92. Thus, some of the differences in sRNA expression between Y. pestis and Y. pseudotuberculosis are likely due to strain-specific effects. Consistent with this, our previous study revealed differences in the growth dependence on hfq between Y. pestis KIM and Y. pestis CO92. Nevertheless, there are clear differences in sRNA expression patterns between Y. pestis and Y. pseudotuberculosis that are conserved in all strains tested.

We have previously shown that hfq is important for growth of Y. pestis but not Y. pseudotuberculosis, when cells are cultured at 37°C but not at 28°C. Hence, any sRNAs that show expression differences between 28‒37°C, between wild-type and Δhfq cells or between Y. pestis and Y. pseudotuberculosis, are potentially associated with the unique biology of Y. pestis. As described above, we observed many temperature-dependent differences in sRNA expression. In some cases, the effect of deleting hfq was specific to one temperature, e.g., Ysr179/CsrB (Fig. 1), indicating complex interactions between temperature and hfq dependence. Differences in expression of sRNAs in Y. pestis between 28‒37°C could contribute to differences in gene expression for bacteria in a flea vector and bacteria in a mammalian host. Furthermore, differences in sRNA expression patterns between Y. pestis and Y. pseudotuberculosis could contribute to the physiological differences between these species. This is particularly likely for Ysr172, which is undetectable in Y. pseudotuberculosis (Fig. 1A and C).

Mapping of sRNA ends using Deep RACE

We developed genome-scale 5′ and 3′ RACE approaches to precisely determine the ends of the sRNAs confirmed by Northern analysis (Fig. 2A and B) because our deep sequencing data do not allow for such precise mapping. These methods combine conventional RACE with deep sequencing using the Ion Torrent platform (any deep sequencing platform would suffice). In addition to allowing for simultaneous analysis of many RNAs, these methods produce multiple sequence reads for each individual sRNA (e.g., 1,945 sequence reads for Ysr23/160). This allows us to identify multiple 5′ ends and to accurately determine the relative abundance of each. We propose that these methods be named “Deep 5′ RACE” and “Deep 3′ RACE.” Using these methods, we successfully mapped the 5′ ends of 18 sRNAs (and rmf mRNA) and the 3′ ends of 28 sRNAs (and rmf mRNA). The major 5′ and 3′ ends for these sRNAs are listed in Table 1 and raw data are provided in . Four representative examples are shown in Figure 2C‒F. For the sRNAs for which we mapped both unique 5′ and 3′ ends, the median length is 84 nt. In most cases we detected unique ends, but some RNAs have multiple 5′ ends, e.g., Ysr149/181 (Fig. 2E; Table 1). In addition, many sRNAs have multiple 3′ ends clustered around a single location, e.g., Ysr148/153/GlmZ (Fig. 2C), Ysr149/181 (Fig. 2E), Ysr17/154/MicA (Fig. 2F; Table 1). Most sRNAs are located entirely within intergenic regions but some overlap the ends of adjacent genes, e.g., Ysr165 (Fig. 2D; Table 1). Our Deep 5′ RACE method is very similar to a previously described method, “Deep-RACE.” To the best of our knowledge, no method equivalent to Deep 3′ RACE has been described previously. Given the increasing availability of deep sequencing, we anticipate that these methods will become widespread for the large-scale identification of RNA 5′ and 3′ ends.

Figure 2. Representative examples of 5′ and 3′ Deep RACE data. (A) Schematic for the Deep 5′ RACE method. (B) Schematic for the Deep 3′ RACE method. (C) Deep 5′ RACE data (blue) and Deep 3′ RACE data (green) for four selected sRNAs. Flanking annotated genes are indicated by gray boxes.

Y. pestis sRNAs fall into multiple classes based on overlap with annotated genes

Mapping of 5′ and 3′ ends revealed multiple classes of sRNA. The major class is intergenic, i.e., no overlap with annotated genes. Based on our knowledge of equivalent sRNAs in other bacterial species, we anticipate that the majority of intergenic sRNAs function as regulators by base-pairing with distally encoded mRNAs. In contrast to the intergenic sRNAs, seven of the sRNAs overlap an annotated protein-coding gene. In some cases, this may be an artifact of incorrect gene annotation. However, several of the overlapped genes have well-described functions. Two of the sRNAs, Ysr167 and Ysr171, overlap annotated genes in the antisense orientation. These sRNAs may be responsible for regulation of the overlapping gene, as has been observed previously for antisense RNAs in other species. Four of the sRNAs, Ysr88/152, Ysr155/RyfD, Ysr161 and Ysr165, overlap the 5′ end of an annotated gene, in the sense orientation. These sRNAs may include riboswitches which can generate short RNAs at the start of genes by promoting transcription attenuation or RNA processing. One sRNA, Ysr73/169, overlaps the 3′ end of a gene in the sense orientation. This, and other sRNAs that overlap annotated genes in the sense orientation, may be processed fragments of the mRNAs that they overlap.

We determined whether any of the sRNAs might be protein-coding. Specifically, we translated sRNAs in silico and searched for open reading frames of > 30 amino acids that have significant sequence identity with proteins annotated in other species. We identified one RNA, initially named Ysr173, which encodes a homolog of E. coli Rmf, ribosome modulation factor. We note that rmf is not annotated for Y. pestis but is annotated for Y. pseudotuberculosis. Other sRNAs might also be protein-coding, as has been observed for some sRNAs in E. coli. This is particularly true for Y. pestis, which has a less well-annotated genome as compared with E. coli. Indeed, we found many differences between the gene annotations available for Y. pestis KIM from different databases. Strikingly, rmf mRNA levels in Y. pestis (but not Y. pseudotuberculosis) are dependent upon the presence of hfq (Fig. 1B; ), indicating that the effects of Hfq specific to Y. pestis are not limited to non-coding RNAs. rmf abundance may be controlled by direct association of Hfq. Alternatively, rmf may be regulated by an sRNA in an Hfq-dependent manner.

Sequence conservation of sRNAs between Y. pestis, Y. pseudotuberculosis, Y. enterocolitica and E. coli

We used BLAST to search for sequence conservation between each of the sRNAs and the genomes of Y. pseudotuberculosis, Y. enterocolitica and E. coli. Specifically, we searched using the sequence beginning 100 bp upstream and ending 100 bp downstream of the coordinates identified in the initial deep sequencing experiment. A summary of this analysis is shown in Table 2. For sequences with matches in any of these species, we performed alignments using ClustalW (). In all cases, sRNA sequences in Y. pestis and Y. pseudotuberculosis were extremely similar, as expected due to the high sequence identity between these species. In most cases, nucleotide identity was 100%, including for Ysr172 which is not detectably expressed in Y. pseudotuberculosis (Fig. 1). All but two of the sRNAs are conserved in Y. enterocolitica, suggesting that they represent a core set of sRNAs for this genus. Only 14 of the sRNAs are conserved in E. coli. Given that most of the E. coli homologs have been characterized, conservation of sRNAs between E. coli and Y. pestis provides insight into the function of these sRNAs in Y. pestis. Several were only partially conserved, suggesting that their functions have diverged between the two species. Three Y. pestis sRNAs that did not generate a BLAST match in E. coli are located in the same gene context (i.e., same synteny with flanking genes) as known E. coli sRNAs, e.g., Ysr185/Spot 42. We propose that these sRNAs are shared between the two species but have diverged extensively with respect to their mRNA targets. As described previously for MicF (one of the 50 putative sRNAs that failed the northern blot analysis), some sRNAs conserved between Y. pestis and E. coli are conserved only over short stretches of sequence that are known to be required for base-pairing with targets in E. coli. We propose that these sRNAs share a “core” set of mRNA targets across the Enterobacteriaceae that rely on the conserved sequence for base-pairing, but also have species-specific mRNA targets. In a few cases, e.g., Ysr165, Ysr172, sequence identity with E. coli was found only for the sequence flanking the sRNA ends. Thus, the sequence similarity is unlikely to reflect functional conservation of the sRNA.

Table 2. Summary of alignments of nucleotide sequences of identified RNAs

	Y. pseudotuberculosisb		Y. enterocolitica		E. coli
sRNAa	% identityc	% coveraged	% identityc	% coveraged	% identityc	% coveraged
Ysr151/RnpB	99	100	92	100	89	85
Ysr88/152	100	100	83	64	N/A	N/A
Ysr148/153/GlmZ	100	79	72	61	N/A	N/A
Ysr7/154/MicA	99	100	86	99	80	33
Ysr155/RyfD	100	100	91	100	79	93
Ysr156/Ffs	99	100	92	72	84	40
Ysr145/157	99	100	79	100	N/A	N/A
Ysr158	99	100	69	40	N/A	N/A
Ysr159/CyaR	99	100	83	73	N/A	N/A
Ysr23/160	98	100	73	52	N/A	N/A
Ysr161	93	100	82	55	N/A	N/A
Ysr163	82	89	N/A	N/A	N/A	N/A
Ysr164	99	96	78	38	N/A	N/A
Ysr165	99	100	90	52	80	30
Ysr11/166/FnrS	100	100	87	57	83	41
Ysr167	99	100	79	82	N/A	N/A
Ysr73/169	99	100	76	100	N/A	N/A
Ysr170	100	100	N/A	N/A	N/A	N/A
Ysr171	100	100	84	69	79	42
Ysr172	100	78	87	22	83	25
Ysr173/rmf	100	76	89	75	74	72
Ysr174	100	100	82	54	81	25
Ysr65/175	100	66	93	36	N/A	N/A
Ysr177	99	100	79	100	N/A	N/A
Ysr179/CsrB	100	85	87	85	N/A	N/A
Ysr45/180/GcvB	100	100	94	75	80	75
Ysr149/181	100	100	91	68	N/A	N/A
Ysr182/6S RNA	99	100	89	79	78	78
Ysr183/SroG	99	100	92	58	70	52
Ysr185/Spot 42	96	100	76	98	N/A	N/A
Ysr186/CsrC	99	83	89	82	87	40
Ysr146.2/187	99	100	90	68	N/A	N/A

a Underlined RNAs were predicted in a bioinformatic study.

b The Y. pseudotuberculosis 32953 strain was used as a reference genome since no reference genome sequence exists for PTB52c.

c % identity of sRNA between Y. pestis and representative organism.

d % coverage of BLAST search of sRNA 100 bp upstream and downstream.

Given that almost all of the sRNAs and their flanking sequence are > 95% identical between Y. pestis and Y. pseudotuberculosis, what accounts for the differences in expression patterns? One possibility is that Hfq functions differently in the two species. Hfq has been implicated previously in promoting stability of many sRNAs in other bacterial species. However, Hfq is 100% identical at the amino acid sequence level between Y. pestis and Y. pseudotuberculosis, expression of hfq mRNA is similar between the two species (Fig. 1A) and the majority of sRNAs are 100% identical between the two species. Hence, it is unlikely that differential binding of Hfq alone contributes to the differences in expression patterns. Transcription of the sRNAs may be regulated differently between the two species, although any such differences would be due to trans-acting factors since the sequences upstream of the sRNAs are also well-conserved between the two species. Our preferred model is that mRNA target availability determines the stability of the sRNAs. Thus, differences in the abundance of mRNA targets for each sRNA between Y. pestis and Y. pseudotuberculosis would alter the dynamics of base-pairing, in turn, altering the susceptibility of sRNAs to degradation. Consistent with this hypothesis, pairing of sRNAs to their target mRNAs has been shown previously to promote sRNA degradation. Differences in mRNA target abundance could also impact the size of the available pool of Hfq, which may be limiting.

A previous study of sRNAs in E. coli identified likely homologs in Y. pestis based on sequence conservation. Eight of these predictions are consistent with the sRNAs we identified (Table 2). However, we did not detect RNA-seq signal for several predicted sRNAs. This is likely due to the fact that the equivalent E. coli sRNAs are expressed under stress conditions, e.g., RprA expression is induced during stationary phase. Therefore, we propose that many sRNAs were not detected in our study due to condition-specific expression patterns.

Comparison to another sRNA study in a Yersinia species

A recent study used deep sequencing to identify 150 putative sRNAs in Y. pseudotuberculosis. Surprisingly, only 17 of the 31 confirmed sRNAs that we identified are shared with the list of putative sRNAs identified by Koo et al. Thus, there are substantial disparities between the two studies. We propose the following explanations for these disparities: (1) six of the putative sRNAs identified by Koo et al. are not conserved in Y. pestis; (2) some of the putative sRNAs identified by Koo et al. may have escaped detection in our study by virtue of generating an insufficient number of sequence reads in the initial deep sequencing analysis. This is likely to be the case for tmRNA, for which we detect > 100 sequence reads but fewer than 500, the cut-off we used. However, for many of the sRNAs identified by Koo et al., we detect few or no sequence reads in the corresponding location in Y. pestis; (3) in a few cases, putative sRNAs identified by Koo et al. were specific to a particular growth phase that differs from the conditions in our study. These sRNAs would likely be missed by our approach; (4) only 29 of the putative 150 sRNAs identified by Koo et al. were successfully validated by Northern analysis and/or RACE. Hence, it is possible that many of the remaining 121 candidates would be below our detection threshold by Northern analysis or are false positives. Consistent with this, 20 of the 49 putative sRNAs tested were undetectable by Northern analysis. Furthermore, the 49 putative sRNAs selected for Northern analysis by Koo et al. have considerably higher expression levels, based on the deep sequencing data, than those that were not tested (median number of sequence counts ~7-fold higher for those tested by Northern analysis). This greatly increases the likelihood that the untested putative sRNAs would be below the detection threshold of Northern analysis or are false positives; (5) it is possible that there is a much larger pool of sRNAs and the two studies have each identified a subset of that pool. Consistent with this, deep sequencing has been used to identify > 500 putative sRNAs in Vibrio cholerae and > 300 putative sRNAs in E. coli; (6) Koo et al. identified sRNAs in Y. pseudotuberculosis whereas we identified sRNAs in Y. pestis. Although the DNA sequences corresponding to almost all these putative sRNAs are highly conserved between the two species, we have shown that sRNA expression patterns vary considerably between the two species. Hence, it is likely that many of the putative sRNAs identified by Koo et al. are well expressed in Y. pseudotuberculosis but would be below the limit of detection, either by deep sequencing or Northern analysis, in Y. pestis.

Our northern blot data confirm the existence in Y. pestis of 13 of the 101 putative sRNAs identified by Koo et al. but not tested by northern blot in that study. Furthermore, our data provide the first characterization of the expression of these 13 sRNAs in Y. pestis and Y. pseudotuberculosis. The 50 putative sRNAs we identified in Y. pestis by deep sequencing include relatively few false positives, as evidenced by the high success rate when testing by northern blot. This is likely due to the stringent cut-off used for assignment as a putative sRNA. In contrast, the 150 putative sRNAs identified by Koo et al. likely include many false positives. It is also likely that the list of putative sRNAs identified by Koo et al. has many fewer false negatives than our list. By careful comparison of the data sets from both studies, it may be possible to prioritize additional putative sRNAs for validation.

Conclusions

In summary, we have identified 32 sRNAs in Y. pestis, of which 14 are novel, and 11 of these 14 have no known E. coli homolog. Based on the patterns of sRNA expression and on the differences between sRNA expression in Y. pestis and Y. pseudotuberculosis, we propose that many of these sRNAs contribute to the unique biology of Y. pestis, and may play important roles in virulence.

Materials and Methods

Strains and growth conditions

Strains of Y. pestis used in this study were KIM6+ (Pgm+ pCD1− pMT1+ pPst+), KIM6+ Δhfq:cat, KIM6+ Δhfq-multi-copy complementation:kan (Δhfq-C), CO92 and CO92 Δhfq:cat. Strains of Y. pseudotuberculosis used were PTB52c WT (pYV−; serotype IB; YP-HPI+), PTB52c Δhfq:cat, PTB52c Δhfq-multi-copy complementation:kan (Δhfq-C), PTB51c (pYV−; serotype IB; YP-HPI+), PTB57c (pYV−; serotype III; YP-HPI-) and PTB54c (pYV−; serotype III; YP-HPI-). Construction of mutant strains and growth conditions used in this study have been described previously. All strains of Y. pestis and Y. pseudotuberculosis were grown in brain heart infusion (BHI) media. To obtain cells for RNA isolation, 5 ml of BHI was diluted 1:5 with overnight culture and cells were grown for 4 h at 28°C or 37°C. Cells were then harvested at 4°C and stored at -80°C.

RNA isolation and purification

Cells were resuspended in 1 ml TRIzol (Invitrogen), incubated at room temperature for 5 min and centrifuged at 12,000 × g for 10 min at 4°C (all subsequent centrifugations were performed at this temperature). The supernatant was transferred to a new microfuge tube and 200 µl chloroform:isoamyl alcohol (24:1 ratio) was added. The sample was shaken vigorously for 15 sec, incubated at room temperature for 3 min and centrifuged at 12,000 × g for 15 min. The aqueous phase was transferred to a new microfuge tube where 500 µl isopropanol was added and incubated for 10 min at room temperature. Following centrifugation at 12,000 × g for 10 min, the supernatant was decanted and washed with 1 ml ice-cold 75% ethanol and centrifuged at 7,600 × g for 5 min. The supernatant was decanted, and the residual supernatant was removed by pipette. The RNA pellet was air-dried, then resuspended in 30 µl of RNase-free water.

Resulting RNA was treated by DNase I (New England Biolabs) to remove any remaining DNA. A total of 10 µl DNase I was utilized in a final volume of 500 µl and incubated for 1 h at 37°C at which point 600 µl of isopropanol was added and precipitated overnight at -80°C. Following precipitation, the RNA was centrifuged at 12,000 × g for 20 min. The supernatant was decanted and washed with 1 ml ice-cold 75% ethanol and centrifuged at 7,600 × g for 5 min. The supernatant was decanted, and the residual supernatant was removed by pipet. The RNA pellet was air-dried, then resuspended in 30 µl of RNase-free water. RNA was used for initial deep sequencing screening and replicate RNA samples were pooled, quantitated and aliquoted into microfuge tubes for use in Northern analysis.

Initial deep sequencing and characterization of sRNAs

Isolated RNA from Y. pestis KIM6+ grown at 37°C was separated on a 6% denaturing polyacrylamide gel and RNA below 400 nt was cut from the gel. The resulting RNA was electro-eluted using dialysis tubing following washing with 1 ml of 0.1 X TBE. Electro-elution was run at 100 V for 30 min. The resulting TBE was ethanol precipitated and the resulting RNA was examined on an agarose gel to ensure there was no residual rRNA. A cDNA library was constructed using the Illumina RNA-seq kit, following the manufacturer’s instructions except that DNA was gel-purified from 200 bp and above rather than 300 bp before the PCR amplification step. This modification increased the likelihood of identifying sRNAs.

The DNA library was sequenced using an Illumina Genome Analyzer II (Harvard Medical School). Reads were mapped to the Y. pestis KIM genome using Bowtie with default settings. Sequences were piled up to determine the number of sequence reads that mapped to each nucleotide of the genome. Putative sRNAs were identified as regions of contiguous sequence that partially or fully overlap an intergenic region (JCVI genome annotation) with at least one position with > 500 mapped sequence reads. Fully antisense RNAs could not be identified due to the lack of strand information in the sequencing data.

Northern transfer and hybridization

For Northern analysis, a total of 15 µg of RNA was separated on a 1.5% formaldehyde MOPS gel and transferred to a nylon membrane by capillary action. Hybridization was performed using 60-mer oligonucleotides () that were γ32P-ATP end-labeled with T4 polynucleotide kinase (Fermentas) for 1 h at 37°C. Membranes were hybridized at 42°C for 2 h in Amersham Rapid-hyb Buffer (GE Life Sciences) and were washed as per manufacturer’s protocol. Densitometric quantitation of Northern Blots was performed using ImageQuant software and Sum Above Background calculations. Percent values indicated in Figure 1B are normalized to the condition with the highest signal. Four blots were reprobed with an oligonucleotide specific to 5S rRNA, as a loading control ().

Deep 5′ and 3′ RACE and computational analysis

RACE experiments were performed using RNA from Y. pestis KIM6+ RNA isolated at 37°C and the FirstChoice® RLM-RACE Kit (Ambion). For 5′ RACE, we used a modified 5′ RLM-RACE Protocol. To start, total of 8 μg RNA was treated with tobacco acid pyrophosphatase (TAP) at 37°C for 1 h, followed by ligation of the 5′ RACE adaptor 37°C for 1 h. The resulting RNA was then reverse transcribed according to the manufacturer’s protocol. PCR was then performed on the resulting cDNA using a primer containing the Ion Torrent sequence for 5′ RACE amplification and primers specific for each sRNA confirmed by Northern analysis (). For some sRNAs, the PCR had to be re-amplified using a portion of the original reaction and the same primers, which then resulted in successful amplification of bands for all sRNAs. Resulting PCR products were purified using the QIAquick PCR Purification Kit and eluted in “Low TE” (10 mM Tris, 0.1 mM EDTA). All products were then pooled together and sent for Ion Torrent deep sequencing.

For 3′ RACE, we utilized the miScript Reverse Transcription Kit (Qiagen) to perform reverse transcription on Y. pestis KIM6+ RNA isolated at 37°C according to the manufacturer’s protocol. PCR was performed on the resulting cDNA using the primers containing Ion Torrent sequence for 3′ RACE (universal primer) and each sRNA. Resulting PCR products were purified using the QIAquick PCR Purification Kit, eluted in “Low TE,” pooled and sent for Ion Torrent deep sequencing using a 314 chip (Wadsworth Center Applied Genomic Technologies Core Facility).

Any sequences lacking the expected 5′ or 3′ adaptor sequences were removed. We then extracted non-adaptor sequence from the remaining reads and mapped then to the Y. pestis KIM genome using BWA with default parameters. Reads were assumed to be associated with an sRNA if they were located within 1 kbp of the predicted location and were located on the predicted strand.

Clustering analysis of Ysrs verified by northern blot analysis

To assign expression patterns of Ysrs into groups, we used k-means clustering to partition the sRNAs into k = 9 clusters, selecting k by minimizing the value of the Kelley penalty. These nine groups were manually adjusted to seven for clarity.

Conservation analysis

Sequences from 100 bp upstream to 100 bp downstream of each sRNA (coordinates derived from the original Illumina sequencing data) were used to search against the Y. pseudotuberculosis 32953, Y. enterocolitica 8081 and E. coli K-12 (MG1655) strains using BLAST with the default parameters. BLAST matches were realigned using ClustalW.

Click here for additional data file.