Search for MicroRNAs Expressed by Intracellular Bacterial Pathogens in Infected Mammalian Cells

MicroRNAs are expressed by all multicellular organisms and play a critical role as post-transcriptional regulators of gene expression. Moreover, different microRNA species are known to influence the progression of a range of different diseases, including cancer and microbial infections. A number of different human viruses also encode microRNAs that can attenuate cellular innate immune responses and promote viral replication, and a fungal pathogen that infects plants has recently been shown to express microRNAs in infected cells that repress host cell immune responses and promote fungal pathogenesis. Here, we have used deep sequencing of total expressed small RNAs, as well as small RNAs associated with the cellular RNA-induced silencing complex RISC, to search for microRNAs that are potentially expressed by intracellular bacterial pathogens and translocated into infected animal cells. In the case of Legionella and Chlamydia and the two mycobacterial species M. smegmatis and M. tuberculosis, we failed to detect any bacterial small RNAs that had the characteristics expected for authentic microRNAs, although large numbers of small RNAs of bacterial origin could be recovered. However, a third mycobacterial species, M. marinum, did express an ∼23-nt small RNA that was bound by RISC and derived from an RNA stem-loop with the characteristics expected for a pre-microRNA. While intracellular expression of this candidate bacterial microRNA was too low to effectively repress target mRNA species in infected cultured cells in vitro, artificial overexpression of this potential bacterial pre-microRNA did result in the efficient repression of a target mRNA. This bacterial small RNA therefore represents the first candidate microRNA of bacterial origin.


Introduction
RNA interference (RNAi), first reported in Caenorhabditis elegans in 1998 [1], involves the sequential cleavage of long double stranded RNAs (dsRNAs), often of exogenous or even viral origin, by the RNase III enzyme Dicer into ,22-bp small interfering RNA (siRNA) duplexes, bearing two 2-nt 39 overhangs, one strand of which is then loaded into the RNA-induced silencing complex RISC [2][3][4][5].The siRNA then guides RISC to complementary mRNAs, resulting in their inhibition.MicroRNAs (miRNAs) are similar in structure and function to siRNAs, at least in mammals, but are distinct in that they are encoded within the genome of the cell [6].Mammalian cells encode .1,000distinct miRNAs that are initially transcribed as part of one arm of an ,33-bp RNA stem-loop located in a long, capped, polyadenylated transcript referred to as a primary miRNA (pri-miRNA) [7].In animal cells, this stem-loop is bound by the nuclear RNase III enzyme Drosha and cleaved ,22 bp away from the loop to generate an ,60-nt RNA hairpin, bearing a ,2-nt 39 overhang, referred to as the precursor miRNA (pre-miRNA) intermediate.The pre-miRNA is exported to the cytoplasm where it is bound by Dicer, which cleaves ,22 bp from the base of the pre-miRNA, leaving a second 2-nt 39 overhang, to generate a short RNA duplex that is analogous in structure to an siRNA duplex [7].As in the case of siRNAs, one strand of this duplex is incorporated into RISC and guides RISC to complementary mRNA target sites, resulting in the post-transcriptional repression of mRNA function [6].Target recognition is primarily mediated by 59 nucleotides 2 through 8 of the miRNA, the so-called seed region, and the 59 end of a given miRNA is therefore usually highly discrete.In many cases, the other strand of the miRNA duplex, referred to as the passenger or star strand, can also be detected and can be annealed to the miRNA leaving the predicted 2-nt 39 overhangs.The implied existence of such a duplex intermediate provides strong support for the idea that a recovered 22-nt RNA is indeed a miRNA.
Several viruses have been shown to express high levels of virally encoded miRNAs in infected cells that are believed to facilitate viral replication, at least in part, by repressing host innate antiviral immune responses [8,9].Moreover, it was recently reported that small RNAs expressed by a plant fungus regulate cellular gene expression using the host cell's own RNAi machinery and thereby contribute to the pathogenicity of this fungus [10].In contrast, prokaryotes are not believed to express miRNAs, although they do express a wide array of small, non-coding RNAs (sRNAs) that regulate a diverse set of physiological processes inside the bacterial cell [11,12].For example, bacterial sRNAs form ribonucleoproteins that control cellular functions ranging from protein secretion to the recognition of foreign nucleic acids [13][14][15].Additionally, sRNAs can form sRNA/mRNA duplexes that rapidly alter mRNA stability and translation efficiency and thus allow the microbe to swiftly adapt gene expression in response to environmental stresses such as nutrient starvation, changes in pH or temperature [12].The ability of bacteria to rapidly adapt to such environmental stresses is of particular importance for the survival of pathogens combating the immune system.Accordingly, bacterial sRNAs are now recognized as critical regulators of bacterial virulence [16,17].In contrast to our extensive knowledge about sRNAs operating inside the bacterial cell, the question of whether bacterial pathogens also release sRNAs such as miRNAs to target host molecules has not, to our knowledge, been previously addressed.We therefore set out to explore if intracellular bacterial pathogens might have acquired the ability to express and translocate miRNAs, or miRNA precursor molecules, into the cytoplasm of infected cells in order to subvert the host cell RNAi machinery and promote bacterial replication and pathogenesis.In particular, as intracellular bacteria are located in the cytoplasm, we hypothesized that they might transcribe and then secrete small RNA hairpins, functionally analogous to pre-miRNAs, that might then be subjected to Dicer cleavage and incorporation into RISC.If this was indeed the case, then we would expect to recover high levels of such bacterial miRNAs, and possibly also the cognate passenger strand, by deep sequencing.In the present study, we analyzed the small RNA expression profile in cells infected by several distinct varieties of intracellular bacteria (Chlamydia, Legionella and Mycobacteria) and asked if these cells expressed small RNAs of bacterial origin that are loaded into the host RISC and are able to regulate mRNA expression.

Chlamydia trachomatis infection
For infections with Chlamydia trachomatis strain LGV-L2 434/ Bu, we used HeLa CCL2 cells.Infections were carried out in six 150-mm diameter tissue culture dishes by adding a suspension of C. trachomatis LGV-L2 434/Bu elementary bodies (EBs) at an MOI of 20.Prior to infection, confluent HeLa monolayers were washed once with PBS, then a suspension of EBs with the appropriate number of infectious units was added to the cells in 5 ml of culture medium and rocked gently for 30 min at room temperature.Cells were then washed once with PBS, replenished with 30 ml of fresh culture medium and returned to the tissue culture incubator.At 42 h post infection, cells were inspected under a microscope (Olympus CK Inverted Tissue Culture Microscope) to confirm that all cells were infected.Just before preparation of lysates for immunoprecipitation, or TRIzol lysis and small RNA sequencing, cells were washed twice with ice-cold PBS.

Legionella pneumophila infection
The wildtype Legionella pneumophila strain CR39 was cultured in ACES-buffered yeast extract broth supplemented with FeNO 3 and cysteine at 37uC, as described [18].Post-exponential-phase cultures of L. pneumophila were washed in PBS, resuspended in tissue culture media and added to adherent RAW264.7 cells in 10cm culture dishes at an MOI of 0.1.Cells were harvested at 48 hpi and lysed in TRIzol.

Mycobacterial infection
Mycobacterium marinum strain M [19] was cultured from stocks in 25-ml flasks with 10 ml of complete 7H9 medium for 36-48 h, incubated at 33uC/5%CO 2 .Mycobacterial cultures were washed with PBS, resuspended in DMEM plus 10% FBS, prepared as a single cell M. marinum suspension by passage through a tuberculin syringe, counted and adjusted for the MOI of 10 used for the in vitro infections.
Mycobacterium smegmatis from stocks [20] was cultured in two 14-ml snap cap tubes with 5 ml of complete 7H9 medium for 24-36 h and incubated with agitation at 37uC.Then, a single cell suspension was prepared, counted and adjusted for an MOI of 10.
Frozen stocks of Mycobacterium tuberculosis were cultured in complete 7H9 medium at 37uC/5%CO 2 until their concentration reached an OD of 1.3.Then, cultures were spun, washed twice with 50 ml of PBS, resuspended in 4 ml, sonicated to give a single cell suspension and then diluted to give an MOI of 10.
To measure M. marinum MM-H RNA expression after different periods of infection, 12-well plates were inoculated with one ml of a RAW264.7 cell suspension (3.5610 5 cells/ml).After 24 h, the cells were infected with 500 ml of a single cell M. marinum suspension (MOI of 10) and incubated at 33uC for 12 h.Then 500 ml of streptomycin (200 mg/ml) were added to all the infected wells.For the first time point (12 h), media were removed and 500 ml TRizol was added for 5 min.The lysate was then added to an Eppendorf tube and stored at 280uC.The rest of the wells were incubated at 33uC and harvested as described for the previous timepoint at 24, 48 and 72 h post infection.
For small RNA deep sequencing, two 100-mm dishes of RAW264.7 cells at 70% confluency were infected using a single cell suspension of either M. marinum or M. smegmatis at an MOI of 10.After 24 h, the medium was removed, 4 ml of TRizol were added and the cells lysed for 20 min at room temperature with rocking agitation.For M. tuberculosis, 2610 7 THP-1 cells treated with PMA were infected with M. tuberculosis at an MOI of 10.
For RISC immunoprecipitation assays, nine dishes of RAW264.7 cells were infected with M. marinum and another nine with M. smegmatis.Nine dishes of PMA-treated THP-1 cells were infected with M. tuberculosis.After 24 h of infection, all culture medium was removed, each plate was washed with 10 ml of PBS and the cells then treated with 1 ml of lysis buffer (see below) and incubated with rocking agitation.Cells were scraped and liquid and cell debris collected, centrifuged at 4uC for 15 min at 13,000 g and then filtered through cellulose acetate membrane micro-columns.

In vivo infection
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health.All animal studies were approved by the Institutional Animal Care and Use Committee (IACUC) of Duke University (protocol A065-13-03).Animals were observed daily by staff and weighed once weekly.Infected animals were euthanized if they had lost more than 15% of their peak body weight.Euthanasia was by isofluorane overdose, followed by organ removal.
Specific pathogen-free C57BL/6 mice were obtained from Jackson Laboratories (Bar Harbor, ME).For each experiment, a vial of stock bacilli (3.6610 8 ) was aerosolized as described previously [21].The number of viable organisms in each organ sample was determined by plating serial dilutions of the lung homogenates.Initial infection dose (determined by plating lung tissue 24 h post-infection) was 250 CFU and CFU at harvest (in other mice infected at the same time) was 4.3610 5 6SD 1.9610 5 .Lung tissue for RNA analysis was homogenized in 4.5 ml RNAlater (Life Technologies) and then a 36 volume of TRIzol was added for RNA extraction.

Deep sequencing
Total RNA was extracted using TRIzol (Life Technologies) and sequentially ligated to 39 and 59 adapters and then reversetranscribed using a TruSeq Small RNA Sample Prep Kit (Illumina) and SSIII (Life Technologies) [22].Total length of the adapters used was 125 nt.cDNAs were amplified by PCR.The number of PCR cycles was determined by pilot experiments to find a point at which amplicons showed logarithmic amplification.PCR products between 140 bp and 170 bp in length, which corresponds to original RNAs of between 15 and 45 nt, were then isolated by excision from a polyacrylamide gel.Deep-sequencing of the PCR products was performed using a HiSeq2500 machine (Illumina) for 50 base pair single read.
For RNA immunoprecipitation (RIP), cells were collected and lysed using lysis buffer (50 mM HEPES, 150 mM KCl, 2 mM EDTA, 1 mM NaF, 0.5% NP40, 0.5 mM DTT).Lysates were incubated with Dynabeads Protein G (Life Technologies) conjugated with an Ago-specific monoclonal antibody (ab57113, Abcam) [23].The beads were repeatedly washed with PBS and then TRIzol was added to extract bound RNA, which was subjected to small RNA deep-sequencing as described above.

Bioinformatics
Sequence reads were mapped to bacterial genomes (GenBank accession number: AM884176 for C. trachomatis, NC_002942 for L. pneumophila, NC_010612 for M. marinum, NC_008596 for M. smegmatis and AL123456 for M. tuberculosis) and the host genome (hg19 for human genome and mm9 for mouse genome) by Bowtie and analyzed by SAMtools and DARIO [24,25].RNA secondary structures were predicted by mfold [26].Sequence data are available at the Sequence Read Archive, http://www.ncbi.nlm.nih.gov/sra(SRP042180).

qRT-PCR
To confirm expression of a bacterial small RNA, total RNA was extracted with TRIzol, then reverse transcribed to cDNA and quantified by qPCR using a Custom TaqMan small RNA Assay (Life Technologies).

Luciferase assay
Oligonucleotides bearing two copies of a perfect target sequence for the putative M. marinum miRNA MM-H were cloned into the 39 UTR of the RLuc gene in a lentiviral vector.Another lentiviral vector, which contains the firefly luciferase (FLuc) gene, was used as a transduction control, as described previously [27].Packaged virus was transduced into RAW264.7 cells 12 h after infection with M. marinum.Dual luciferase assays were performed 24 h after transduction (i.e., 36 h after infection with M. marinum) using a Dual-Luciferase Reporter Assay System (Promega), according to the manufacturer's protocol.
The potential pre-miRNA sequence for MM-H was cloned into the pSUPER vector where it is expressed under the control of an H1 pol III promoter [28].The two 39 C residues in this pre-miRNA sequence were mutated to five successive T residues in order to terminate Pol III transcription after the second encoded U residue.An oligonucleotide bearing two perfect target sequences for MM-H was then cloned into the 39 UTR of the RLuc indicator gene in the psiCHECK2 vector (Promega), which also encodes an FLuc gene as an internal control.The expression vector and luciferase reporter vector were co-transfected into 293T cells using polyethylenimine.Dual luciferase assays were then performed 24 h after transfection, as described above.

Chlamydia trachomatis
C. trachomatis is an obligate intracellular bacterium that causes a major sexually transmitted disease as well as a potentially severe eye infection referred to as Chlamydia conjunctivitis or trachoma [29].HeLa cells were infected with C. trachomatis at a multiplicity of infection (MOI) of 20 and total RNA isolated 42 h after infection.Deep-sequencing analysis of the small RNA transcriptome in the 15-to 43-nt size range (total RNA-seq) yielded ,11 million reads which could be computationally aligned to either the human or bacterial genome, with 24.9% of reads derived from the bacterial genome (Fig. 1A).We next conducted RNA immunoprecipitation (RIP), using an antibody specific for the essential Argonaute (Ago) component of RISC [23], to enrich for small RNAs associated with the cellular RISC.Deep-sequencing analyses for these RISC-associated small RNAs (RIP-seq) yielded ,18 million reads.When compared to the total RNA-seq data, the RIP-seq procedure substantially increased the percentage of reads that derived from known human miRNAs, from 16.6% to 81%.Consistent with their predominantly miRNA origin, human small RNAs recovered by RIP-seq showed a size of 2262 nt, as predicted for authentic miRNAs (Fig. 1B).In contrast, reads derived from the bacterial genome declined from 24.9% of the observed reads in the total RNA-seq data to only 1.3% in the RIPseq data (Fig. 1A) and these residual reads were not clustered at the 2262-nt size (Fig. 1B).
Table 1 shows the 10 most abundant reads derived from the C. trachomatis genome, here referred to as CT-A through CT-J.Recovery of all of these 10 small RNAs was enhanced by RISC immunoprecipitation, which suggests these small RNAs are associated with RISC.The two most abundant bacterial small RNAs were derived from bacterial tRNAs (CT-A and CT-B) and were too large to be miRNAs, at 36 nt and 32 nt respectively.In fact, only CT-E, CT-H and CT-J showed a size typical for a miRNA.We modeled the RNA secondary structures adopted by these three small RNAs together with flanking bacterial genomic sequences, and CT-E and CT-H can indeed form possible RNA stem-loop structures (Table 1 and Fig. S1), as expected for pre-miRNAs.However, these small RNAs are derived from both the stem and loop regions of the predicted stem-loops.In contrast, canonical miRNAs are invariably entirely derived from one arm of the pre-miRNA stem [6,7].Moreover, it has been reported that, in order for a miRNA to be functionally relevant, it has to contribute .0.1% of the total miRNA pool in a cell [30].Yet, none of the observed bacterial small RNAs account for more than 0.09% of the total miRNA pool (Table 1) and they are therefore unlikely to exert a significant phenotypic effect, even if they are loaded into RISC.Taken together, these data argue that C. trachomatis does not express miRNAs in infected human cells.

Legionella pneumophila
L. pneumophila is a Gram-negative bacterium commonly found in soil and fresh water sources where it resides and replicates inside free-living amoeba.Although L. pneumophila is primarily a pathogen of protozoa, aerosolized and inhaled bacteria can infect alveolar macrophages and cause opportunistic infections in humans and animal models [31].The ability of L. pneumophila to successfully infect a broad spectrum of host species is likely founded in Legionella's capacity to inject ,300 distinct bacterial proteins into the host cytoplasm [32].In order to determine whether L. pneumophila is also able to generate and transfer bacterial pre-miRNAs into host cells, we infected the murine macrophage-like cell line RAW264.7 with L. pneumophila at an MOI of 0.1 and isolated total RNA at 48 h post infection.Deepsequencing analysis of the total small RNA population (15-to 43nt size range) yielded ,14 million reads which could be aligned to either the mouse or bacterial genome.Of these, 9.5% of were of L. pneumophila origin (Fig. 2A).Analysis of RISC-associated small RNAs, by immunoprecipitation using a pan-Ago antibody, yielded ,20 million reads and, as expected, increased the percentage of reads that align to known murine miRNAs from 37.6% to 77.4% of the total small RNA population.Moreover, the murine reads obtained by RIP-seq were almost all in the expected 2262-nt size window (Fig. 2B).In contrast, small RNA reads derived from the L. pneumophila genome decreased from 9.5% of the total small RNA reads to 1.4% of the RISC-associated small RNA reads (Fig. 2A) and were not clustered at the predicted 2262-nt size expected for authentic miRNAs (Fig. 2B).The 10 most abundant small RNA reads of L. pneumophila origin, termed LP-A through LP-J, are shown in Table 2 and all 10 were significantly increased in the RISC-associated small RNA library.However, 7 out of 10 were derived from bacterial tRNAs (Table 2).Of the three remaining reads, LP-D, together with flanking bacterial genomic sequences, is predicted to have the potential to fold into an RNA stem-loop structure (Fig. S2).Yet, the length of this small RNA, at 15 nt, is too short for an authentic miRNA [6] and all 10 of the predominant L. pneumophila small RNAs not only lie outside the expected size range for a functional miRNA but are also present at levels too low to be functionally relevant [30] (Table 2).Taken together, these data argue that L. pneumophila does not express miRNAs of bacterial origin in infected murine cells.

Mycobacterium marinum
M. marinum, a mycobacterium that is a natural intracellular pathogen of ectotherms [33], is the closest animal relative of the human Mycobacterium tuberculosis complex [19].Mouse RAW264.7 cells were infected with M. marinum at an MOI of 10 and total RNA isolated 24 h after infection.Deep-sequencing analysis of the total small RNA population yielded ,9 million reads which could be aligned to either the mouse or M. marinum genome.Of these, 10.4% were of bacterial origin (Fig. 3A).Deep sequencing of RISC-associated small RNAs by Pan-Ago RIP-seq yielded ,16 million reads and 6.3% of these were of bacterial origin.Unlike the murine RISC-associated small RNA reads, which showed the expected 2262-nt size, we did not observe a significant enrichment of bacterial reads of around 22 nt, and these instead clustered at ,18 nt (Fig. 3B).The 10 most common small RNA reads of M. marinum origin, termed MM-A to MM-J, are listed in Table 3.Among these, MM-H looked like a promising miRNA candidate.This small RNA is primarily 23 nt in length and is derived from a pre-miRNA-like stem-loop structure of canonical appearance, derived from a bacterial non-coding region, in which the MM-H sequence occupies one side of the predicted stem (Table 3 and Fig. 3C).Moreover, deep sequencing identified a small number of reads derived from the opposite side of the stem and offset by 2 nt at both 39 ends, as expected for a miRNA passenger strand [34,35] (Fig. 3C and Table S1).Also, the MM-H small RNA possessed a highly discrete 59 end yet showed length variation at the 39 end (Table S1), as would be predicted for an authentic miRNA, where the seed sequence located at the 59 end is the primary determinant of mRNA target recognition [6].This finding, which is unique to MM-H among all the various bacterial small RNAs described in this report, strongly suggests that the 59 end of the MM-H small RNA results from a specific processing step rather than from random RNA degradation.
M. marinum is a facultative intracellular parasite that is able to survive both inside or outside host cells and we wondered whether production of the MM-H small RNA, and particularly the precise definition of the 59 end of MM-H, might be dependent on the mammalian cell miRNA processing machinery.We therefore conducted small RNA deep-sequencing analyses of broth-grown M. marinum.A limited number of small RNAs derived from this predicted bacterial RNA stem-loop were indeed recovered from the broth-grown bacteria.However, these small RNAs clearly did not possess a discrete 59 end, as seen upon RIP-seq of RISCassociated bacterial small RNAs in mouse cells infected with M. marinum and no reads sharing the same 59 end as the dominant intracellular form of MM-H were recovered (Fig. 3D).This result indicates that the MM-H bacterial small RNA requires the host cell RNA processing machinery for its biogenesis.
While quantitative RT-PCR (qRT-PCR) with MM-H-specific primers and probes confirmed MM-H expression by 24 h after infection of RAW264.7 cells by M. marinum, the level of expression did not increase further over the next 72 h (Fig. 3E).We next asked if MM-H was able to repress mRNA function using a Renilla luciferase (RLuc)-based indicator construct bearing two perfectly complementary target sites for MM-H in the RLuc 39 UTR.This RLuc indicator cassette was transduced into RAW264.7 cells, either infected or uninfected with M. marinum, using a lentiviral vector and RLuc activity measured 24 h after transduction [27].However, bacterial infection did not detectably suppress expression of the RLuc reporter gene (Fig. 3F).
We considered the possibility that the level of expression of MM-H in infected RAW264.7 cells, at ,0.18% of the RISCassociated small RNA pool, might be too low to be active in gene repression (Table 3) [30].To test this idea, we co-transfected 293T cells with the MM-H RLuc reporter vector together with a second vector in which the MM-H RNA, and flanking bacterial sequences sufficient to form a pre-miRNA-like stem-loop (Fig. 3C), were transcribed from a human Pol III promoter [28].Exogenous overexpression of this putative bacterial pre-miRNA species in fact resulted in a dramatic reduction in the activity of the MM-Hspecific RLuc reporter construct (Fig. 3G).Therefore, MM-H not only has many features typical of an authentic miRNA but also is able to effectively repress a target mRNA bearing perfectly complementary target sites when ectopically expressed in the form of the pre-miRNA shown in Fig. 3C.Table 3. Top 10 small RNA reads of M. marinum.

Mycobacterium smegmatis
M. smegmatis is a non-pathogenic mycobacterium defined as a fast-growing mycobacterial species [20].Murine RAW264.7 cells were infected with M. smegmatis, at an MOI of 10 and total RNA isolated 36 h after infection.Deep-sequencing analysis of the total small RNA population yielded ,15 million reads, of which 24.6% were of M. smegmatis origin (Fig. 4A).RIP-seq using an antibody specific for the Ago component of RISC resulted in ,21 million computationally assignable reads and, as expected, enriched the proportion of reads that were of host cell miRNA origin from 21% to 66.2%.In contrast, the percentage of reads that map to the M. smegmatis genome declined from 24.6% to 7.8% of the total library.Again, murine small RNA reads in the RIP-seq library showed an obvious peak size of 2262-nt (Fig. 4B), while the bacterial RISC-associated small RNAs were predominantly , 19 nt in size.However, a secondary peak with a length of 21 to 24 nt could be observed for the M. smegmatis-derived small RNAs.
The top ten most prevalent M. smegmatis-derived small RNA species, designated MS-A through MS-J, were all enriched by RISC immunoprecipitation and several of these have the potential to form RNA stem-loop structure (Table 4).Yet, these bacterial small RNAs do not map to one arm of the predicted RNA stemloops, as is invariably the case with authentic miRNAs [6,7], and their length is consistently at or below 18 nt, too short for real miRNAs (Table 4 and Fig. S4).
We did not find M. smegmatis small RNA reads with homology to the M. marinum small RNA among the 10 most prevalent reads.However, the sequence of MM-H, and its flanking sequences, is relatively well conserved in several Mycobacterial species (Fig. 3H).Indeed, by deep-sequencing, we did find small RNAs from the homologous sequence in M. smegmatis derived from both the 59 and 39 arms of the cognate predicted RNA stemloop structure (Fig. 4C and Table S2).However, the predicted seed sequence of MM-H [6] is not conserved in M. smegmatis, and the most prevalent small RNA that maps to this region in M. smegmatis was not enriched by RISC IP.Moreover, the expression level of this small RNA (889 reads for the most abundant sequence, which is equivalent to 0.004% of the total cellular small RNA population) is extremely low (Table S2).

Mycobacterium tuberculosis
To investigate M. tuberculosis, the mycobacterial pathogen of greatest importance to human health [36], we performed infections with M. tuberculosis of both human cell lines and of a mouse in vivo model.Human THP-1 cells were infected with M. tuberculosis at an MOI of 10 and total RNA isolated 36 h after infection.Total RNA-seq yielded ,25 million reads that could be computationally assigned to either the human or M. tuberculosis genome, while RIP-seq using an antibody specific for cellular Ago proteins resulted in ,31 million assignable reads.The percentage of reads that map to known human miRNAs increased from 38.1% of the total small RNA library to 58.6% of the RISCassociated small RNA library, as expected (Fig. 5A).Twenty-seven percent of reads mapped to the M. tuberculosis genome by total RNA-seq and this deceased sharply to 3.8% of the RISCassociated small RNAs.The length distribution of the human reads showed an obvious peak at around 2262-nt, while the bacterial RISC-associated small RNAs clustered below 19 nt in size, too small for authentic miRNAs (Fig. 5B) Of the 10 most prevalent bacterial small RNAs recovered from M. tuberculosis infected cells, here termed MT-A to MT-J, only one, MT-F, falls within the predicted length limits for an authentic miRNA (Table 5).This small RNA, together with its flanking genomic region, does have the potential to form an RNA stemloop structure.However, the MT-F small RNA extends over the terminal loop into both arms of the stem, which is not consistent with its origin as a Dicer cleavage product (Fig. S5).Although the MM-H sequence from M. marinum is relatively well conserved in the M. tuberculosis genome (Fig. 3H), no small RNAs from this region were detected in M. tuberculosis-infected cells.
We also performed deep sequencing of small RNAs derived from an M. tuberculosis in vivo infection model [21] as a longer, persistent infection could result in the accumulation of bacterial small RNAs in infected cells.We infected mice with M. tuberculosis and isolated total RNA from the lungs 6 weeks after infection.Total RNA-seq for the sample yielded ,21 million reads, of which 2.6% were of M. tuberculosis origin (Fig. 5C).Mouse reads peaked around 2262-nt, while most bacterial reads were again shorter than 19 nt (Fig. 5D).Five of the 10 most prevalent bacterial reads recovered in this in vivo experiment overlapped with the top 10 reads observed in the infected THP-1 cells in culture (Tables 5 and 6).Yet, none of these has the characteristics expected for a miRNA; they are too short and the region from which small RNAs derive in the predicted RNA stemloop structures is inconsistent with their origin as Dicer cleavage products [6] (Fig. S5).Small RNAs derived from the M. tuberculosis genomic sequence with homology to the M. marinum of MM-H small RNA (Fig. 3H) were not detected.

Discussion
Previous work has identified a number of virally encoded miRNAs that enhance virus replication by down-regulating host cell mRNAs encoding proteins with antiviral potential [8,9].In at least one case, a viral miRNA has been shown to significantly enhance viral pathogenesis in vivo [37].Similarly, a plant fungal pathogen has recently been shown to express miRNAs in infected plants that repress the host innate immune response and promote fungal pathogenicity [10].We therefore hypothesized that intracellular bacteria, such as C. trachomatis, L. pneumophila or Mycobacterium spp., which are known to translocate bacterial proteins into the host cell [32,[38][39][40], might also encode miRNAs that could attenuate the ability of the host cell to resist bacterial invasion.
Because intracellular bacteria actively translocate microbial products into the host cell cytoplasm [32], we hypothesized that the most likely mechanism for bacterial miRNA production would involve the expression of a pre-miRNA-like hairpin RNA by the bacterium, either directly or as a result of RNA processing, that would be translocated into the cytoplasm and processed by the host cell Dicer protein to generate a miRNA duplex intermediate, one or both strands of which would then be loaded into RISC.In this way, the bacterium would be able to take advantage of the precise processing capacity of Dicer to generate a highly discrete 59 end [6].This would generate to a single miRNA seed sequence, located at positions 2 to 8 of the miRNA, and lead to targeting of a specific population of cellular mRNAs.Moreover, Dicer processing is thought to enhance miRNA loading into RISC [41].
If this hypothesis is correct, then small RNA sequencing of mammalian cells infected by intracellular bacterial pathogens should lead to the recovery of small RNAs of bacterial origin, ,2262 nt in length, that derive from one arm of a predicted ,60nt hairpin RNA.Ideally, we would also observe small RNA reads derived from the other arm of the same pre-miRNA hairpin, representing the passenger strand, offset by the predicted 2-nt 39 overhangs [6].The putative bacterial miRNA would be expected to have a discrete 59 end, but not necessarily a discrete 39 end, and to be associated with RISC.Finally, we would expect expression of this bacterial small RNA in infected cells to result in the repression of mRNAs bearing complementary target sequences, including mRNAs that are expressed from miRNA indicator constructs.
To address whether intracellular bacterial pathogens indeed express miRNAs in infected cells, we performed deep sequencing of small (15-43 nt long) RNAs expressed in mammalian cells infected by C. trachomatis and L. pneumophila, as well as three mycobacterial species, M. marinum, M. smegmatis and M. tuberculosis.All five of these bacteria gave rise to substantial levels of small bacterial RNAs, as determined by total small RNA deep sequencing, ranging from ,27% of the total small RNA population in M. tuberculosis infected THP-1 cells to ,9.5% of Table 4. Top 10 small RNA reads of M. smegmatis.While the majority of small bacterial RNAs are therefore clearly not RISC-associated, we did identify several bacterial small RNAs that were enriched, sometimes to a remarkable extent, upon deep sequencing of RISC-associated-as opposed to total-small RNAs from infected cells (Tables 1, 2, 3, 4, 5, 6).However, these small RNAs are almost all unlikely to be authentic miRNAs.In particular, the large majority were not of the expected 2262 nt size expected for authentic miRNAs, they were not predicted to fold into a canonical pre-miRNA-like stem-loop when flanking genomic RNAs were included (Figs.S1, S2, S3, S4, S5), they did not have a discrete 59 end (data not shown) and they were generally expressed at levels (#0.1% of the total miRNA pool) that are too low to be functionally relevant [6,30].
Some interesting aspects of these data are nevertheless worth discussing.For example, it is striking that tRNA fragments were a major source for bacterial small RNAs in C. trachomatis and, especially, L. pneumophila but were not observed in any mycobacterial small RNA library (Tables 1, 2, 3, 4, 5, 6).RISCassociated human small RNAs of tRNA origin have been reported previously and were proposed to arise from Dicer processing of tRNA cloverleaf structures that had ''collapsed'' to form a simple RNA hairpin [42].However, the recovered bacterial tRNA fragments were invariably $32 nt in length and are therefore too large to result from Dicer processing of bacterial tRNAs.
We note that some of the mycobacterial small RNAs recovered are actually expressed at quite high levels (Tables 3-6), thus potentially suggesting that they might be functionally relevant.We also note that several of the bacterial small RNAs that were recovered at high levels from M. tuberculosis-infected human THP-1 cells were also among the most highly expressed small bacterial RNAs recovered from the lungs of M. tuberculosisinfected mice (Tables 5 and 6), thus demonstrating a remarkable level of reproducibility and again suggesting a potential functional relevance.Despite these intriguing findings, the overall conclusion is that, with one single possible exception, none of the bacterial small RNAs reported here has the characteristics of an authentic miRNA.That exception is the M. marinum MM-H small RNA.In particular, we note that: 1) All the MM-H RNA reads derive from one arm of a predicted pre-miRNA-like hairpin (Fig. 3C and Table S1).2) We obtained 12 reads that derive from the opposite side of the predicted MM-H stem-loop and that are offset by 2 nt from the MM-H small RNA, consistent with their origin as a miRNA passenger strand and strongly suggesting that MM-H is generated by Dicer cleavage of the proposed RNA hairpin (Fig. 3C and Table S1).
3) The 59 end of MM-H is highly discrete when recovered from M. marinum-infected cells (Table S1) but was ''ragged'' when small RNAs from broth-grown M. marinum were sequenced (Fig. 3D).Indeed, no small RNAs with the same 59 start site as the MM-H small RNA recovered from infected cells were identified in the broth-grown M. marinum culture.This again argues that a host cell-specific RNA processing factor, presumably Dicer, was required to generate MM-H in infected cells.4) MM-H reads were enriched by 18-fold as a percentage of the total small RNA population in the RISC-associated versus total small RNA library, thus suggesting that the MM-H RNA is loaded into RISC.5) The major MM-H variant represented ,0.18% of the RISCassociated miRNA pool in infected cells (Table 3), thus suggesting a level of expression that at least has the potential to exert a phenotypic effect [30].The predicted stem-loop structures and origin of these small RNAs is depicted in Fig. S5.doi:10.1371/journal.pone.0106434.t006 Despite this latter prediction, we were not, in fact, able to detect repression of an RLuc-based MM-H indicator plasmid in M. marinum-infected RAW264.7 cells (Fig. 3F).We considered that MM-H is actually expressed in RAW264.7 cells at a level that is very close to the threshold of 0.1% of the total miRNA pool below which miRNAs are invariably ineffective [30].To test this idea, we therefore expressed the pre-miRNA-like RNA hairpin for MM-H (Fig. 3C) using an RNA Pol III-based expression vector and indeed saw a dramatic and specific inhibition of the same RLucbased indicator plasmid (Fig. 3G).Therefore, it is clear that the pre-MM-H hairpin can indeed serve as a substrate for Dicer cleavage and that the mature MM-H small RNA has the potential to load into RISC and act as a repressor of target mRNA expression.However, it remains unclear whether MM-H is an authentic bacterial miRNA generated from a pre-miRNA hairpin that is secreted by M. marinum into the infected cell cytoplasm or simply a bacterial RNA fragment that is fortuitously generated by Dicer cleavage of a stable bacterial RNA hairpin that is inadvertently released into the cell cytoplasm from bacteria that have lysed.
In favor of the idea that MM-H is not a true bacterial miRNA are the inability to detect MM-H function in M. marinum-infected RAW264.7 cells (Fig. 3F) and the fact that MM-H is not conserved, in terms of sequence or expression, in two other related mycobacterial species, M. smegmatis and M. tuberculosis (Fig. 3H, Table S2).Nevertheless, it remains possible that MM-H may accumulate to higher, functionally relevant levels in vivo that greatly exceed the low level of MM-H expression seen in culture (Fig. 3E).It is worth noting that among the vacuole-bound pathogens analyzed in this report, the pathogenic mycobacteria are thought to be unique in their ability to escape into the cytoplasm [43][44][45], which might greatly facilitate the release of a pre-miRNA precursor into the cellular compartment where Dicer is active.The identification of at least one candidate bacterial miRNA suggests that other intracellular pathogens, e.g., Shigella, Rickettsia, Salmonella and Listeria, are also worth examining for the potential expression of bacterial miRNAs.

Supporting Information
Figure S1 Predicted secondary structures for small RNA of C. trachomatis.Predicted RNA secondary structure and origin of small RNAs of C. trachomatis listed in Table 1.The structures were predicted by mfold.The large black arrow and the white arrow indicate the 59 and 39 ends of the recovered small RNA, respectively.Small arrows indicate possible passenger strands, if detected.(PDF) Figure S2 Predicted secondary structure for small RNAs of L. pneumophila.Predicted secondary structures and origin of small RNAs of L. pneumophila, listed in Table 2.The structures were predicted by mfold.The large black arrow and the white arrow indicate the 59 and 39 ends of the recovered small RNA, respectively.(PDF) Figure S3 Predicted secondary structure for small RNAs of M. marinum.Predicted secondary structure and origin of several small RNAs of M. marinum, listed in Table 3 5 and 6.The large black arrow and the white arrow indicate the 59 and 39 ends of the recovered small RNA, respectively.(PDF) Table S1 Small RNAs derived from the M. marinum MM-H stem-loop structure.

(PDF)
Table S2 Small RNAs derived from the MM-H homology region of M. smegmatis.(PDF)

Figure 1 .
Figure 1.Deep-sequencing of small RNAs in C. trachomatisinfected cells.Results of deep-sequencing for C. trachomatis-infected HeLa cells.A) Gene annotation based on sequencing reads aligned to either human or bacterial genome.Values indicate the percentage of reads of each gene type in either the total or RISC IP from small RNA libraries.B) Length distribution of reads.The X-axis shows the length of small RNAs (nucleotide, nt) and the Y-axis shows the percentage of reads of each length in the total or RISC IP library.doi:10.1371/journal.pone.0106434.g001

Figure 3 .
Figure 3. Deep-sequencing of small RNAs in M. marinum infected cells and characteristics of the MM-H miRNA candidate.A and B) Results of deep-sequencing for M. marinum.The figures were generated as described in Fig. 1.C) Predicted RNA secondary structure of a possible MM-H precursor, including flanking sequences.The bold line indicates the putative mature miRNA and the thin line indicates a possible passenger strand.D) 59 end starting positions of small RNAs from the structure shown in panel a, found in infected cell lines (total RNA-seq and RIP-seq) and in broth-grown bacteria.The Y-axis shows the absolute read count of each small RNA.Large and small arrows indicate the starting positions of the putative mature miRNA and possible passenger strand shown in panel A. E) Time-course analysis of the expression level of the MM-H RNA measured by qRT-PCR.Relative expression levels relative to the 24-h time point were normalized to the host cell U6 RNA level are indicated.Data shown represent the average of two experiments.F) Inhibitory activity of the putative M. marinum MM-H miRNA.An RLuc-based indicator, with or without two copies of a perfectly complementary target sequence for the putative MM-H miRNA inserted into the 39 UTR, was transduced into RAW264.7 cells infected or uninfected with M. marinum.Relative RLuc activity was measured at 24 h post-transduction and then normalized to uninfected control and to the FLuc internal control, which is present in a second lentiviral vector transduced simultaneously.A representative experiment is shown.G) Similar to panel F, except in this case a Pol III-based expression vector encoding the putative MM-H pre-miRNA shown in panel C was co-transfected into 293T cells along with an RLuc-based MM-H indicator plasmid and an FLuc-based internal control.Average of three experiments with SD indicated.H) Sequence homology of the MM-H region in various mycobacteria.The predominant M. marinum MM-H sequence is indicated by a bold line.Nucleotides that differ from M. marinum are highlighted.doi:10.1371/journal.pone.0106434.g003

a
The predicted stem-loop structure and origin of several small RNAs is depicted in Fig.S3.b Predicted stem-loop structure and origin of this small RNA is depicted in Fig.3C.doi:10.1371/journal.pone.0106434.t003

Figure 4 .
Figure 4. Deep-sequencing of small RNAs in M. smegmatis-infected cells.A and B) Results of small RNA deep-sequencing for M. smegmatisinfected RAW264.7 cells.These figures were generated as described in Fig. 1.C) Predicted RNA secondary structure of the MM-H homology region of M. smegmatis.Bold line indicates the putative mature miRNA and the thin line a possible passenger strand found by deep-sequencing.doi:10.1371/journal.pone.0106434.g004

a
The predicted stem-loop structures and origin of these small RNAs is depicted in Fig.S4.doi:10.1371/journal.pone.0106434.t004thesmall RNAs in L. pneumophila-infected RAW264.7 cells.However, for all five bacterial species the percentage of reads of bacterial origin dropped substantially when RISC-associated small RNAs were deep sequenced.This contrasts with authentic cellular miRNAs, which greatly increased as a percentage of the reads obtained upon sequencing of RISC-associated small RNAs (Figs. 1, 2, 3, 4, 5).

Figure 5 .
Figure 5. Deep-sequencing of small RNAs in M. tuberculosis infected cells and mice.These figures were generated as described in Fig. 1, for in vitro infection of THP-1 cells (A and B) and in vivo infection of mice (C and D).doi:10.1371/journal.pone.0106434.g005

a
The predicted stem-loop structures and origin of these small RNAs is depicted in Fig.S5.doi:10.1371/journal.pone.0106434.t005

a
Exact identity to a small RNA obtained in vitro is indicated.b Figure S3Predicted secondary structure for small RNAs of M. marinum.Predicted secondary structure and origin of several small RNAs of M. marinum, listed in Table 3.The large black arrow and the white arrow indicate the 59 and 39 ends of the recovered small RNA, respectively.Small arrows indicate possible passenger strands, if detected.(PDF) Figure S4 Predicted secondary structure for small RNAs of M. smegmatis.Predicted secondary structures and origin of small RNAs of M. smegmatis, listed in Table 4.The large black arrow and the white arrow indicate the 59 and 39 ends of the recovered small RNA, respectively.Small arrows indicate possible passenger strands, if detected.(PDF)FigureS5Predicted secondary structures for small RNAs of M. tuberculosis.Predicted secondary structures and origin of small RNAs of M. tuberculosis, listed in Tables5 and 6.The large black arrow and the white arrow indicate the 59 and 39 ends of the recovered small RNA, respectively.(PDF)

Table 5 .
Top 10small RNA reads of M. tuberculosis in culture.

Table 6 .
Top 10small RNA reads of M. tuberculosis in vivo.