Sponges and Predators in the Small RNA World

ABSTRACT

Most noncoding small RNAs (sRNAs) that regulate gene expression do so by base-pairing with mRNAs, affecting their translation and/or stability. Regulators as evolutionarily distant as the trans-encoded sRNAs of bacteria and the microRNAs (miRNAs) of higher eukaryotes share the property of targeting short sequence segments that occur in multiple copies in bacterial and eukaryotic transcriptomes. This target promiscuity has major implications for sRNA function. On the one hand, it allows the sRNA to coordinately control several different targets and thus be at the center of regulatory networks. On the other hand, it allows the existence of target mimics or decoys that divert the sRNA/miRNA away from bona fide targets and thus serve as mechanisms to regulate the regulator. In addition, by competing for pairing with the same sRNA, bona fide targets establish a cross talk that can impact on each other’s expression levels. Here we review evidence that target mimicry and competition are important components of the regulatory architecture of bacterial sRNA networks.

LESSONS FROM EUKARYOTIC miRNAs: FROM SPONGES TO THE ceRNA HYPOTHESIS

MicroRNAs (miRNAs) are 20- to-24-nucleotide (nt)-long RNAs that guide Argonaute proteins to silence mRNA expression in animal and plant cells (1–3). Similarly to bacterial trans-encoded small RNAs (sRNAs), miRNAs act by establishing imperfect base-pair interactions with seed sequences that can be as short as 6 to 8 nt. Seeking ways to selectively control miRNA activity in vivo, a decade ago Ebert and coworkers engineered transcripts containing multiple tandemly arranged target sites for one or more miRNAs and had these constructs expressed at high levels in transfected mammalian cells (4). They found the exogenous RNAs to have the ability to sequester (“soak up”) the miRNAs, relieving the regulation of their natural targets. The authors termed the artificial transcripts “microRNA sponges.” At about the same time, a study on the mechanism responsible for inhibiting the activity of a miRNA (miR399) in plant cells identified an endogenous noncoding RNA, named IPS1, that could base-pair with miR399 and compete for its binding to the primary target (5). This indicated that a natural RNA could have sponge-like activity and that target site amplification was not required for this effect. Following these initial findings, several examples of miRNA target mimicry have been described involving different types of coding and noncoding RNAs (6, 7), including some of viral origin (8, 9). Particularly noteworthy is the case of the circular antisense RNA named CDR1as, highly expressed in human and mouse brain, which harbors as many as 74 potential target sites for the miR-7 miRNA and thus closely fulfills the original definition of a sponge (10). Recent evidence showed CDR1as to be a highly efficient miR-7 sponge in vivo: in cells lacking CDR1as, deregulation of miR-7 networks leads to profound defects in brain development and function (11).

The discovery of sponges opened new perspectives into the complexity of the miRNA targetome. Seitz argued that among the multitude of mRNA species that are potentially controlled by any given miRNA, only a tiny fraction produce clear phenotypic changes in response to miRNA regulation (which typically involves less than a 2-fold variation in mRNA levels) (12). He proposed that only the members of this minority are authentic miRNA targets; the remainder are “pseudotargets” whose miRNA binding sites have evolved to limit miRNA availability, so as to render regulation of the true targets more robust. Further elaboration on these ideas led to the so-called competitive endogenous RNA (ceRNA) hypothesis, which extends the sponge concept to the whole-transcriptome level (13). The ceRNA hypothesis views miRNA binding sites, also called miRNA response elements, as the letters of a language through which RNAs communicate with each other—that is, affect each other’s expression—through competition for shared miRNAs. The model conveys two innovative concepts: (i) that mRNAs are not merely a repository of protein-coding information but can act directly to regulate other mRNAs; and (ii) that portions of the transcriptome generally considered nonfunctional, such as pseudogene transcripts, play an active role by engaging in cross talk with their protein-coding counterparts (13).

Some experimental evidence, together with computational simulations, have challenged the central predicate of the ceRNA hypothesis by showing that effective cross talk would require a large excess of miRNA binding sites in the competing RNAs, possible only under artificial and/or unphysiological conditions (14–16). It is noteworthy that in these studies, as in the original formulation of the ceRNA hypothesis, miRNA partitioning among competitors is solely dictated by the equilibrium dissociation constant for each binding site and by the number of sites. The situation would drastically change if the miRNA were degraded upon binding to one of the competitors (8, 9). Introducing a channel of “stoichiometric decay” (17) might allow most current inconsistencies to be reconciled. We will see below that such a decay channel is a predominant feature of mechanisms regulating the activity of prokaryotic sRNAs.

TARGET MIMICS AND SPONGE-LIKE INHIBITORS OF sRNAs IN BACTERIA

Bacterial trans-encoded regulatory sRNAs differ from miRNAs in many respects, but a particularly relevant difference in the context of this review is that most, if not all, sRNAs are susceptible to cleavage upon pairing with an mRNA target (18–20). This brings a new variable into the sponging landscape: a competitive inhibitor of a bacterial sRNA might not need to stably “soak up” the sRNA; rather, it would suffice if it were efficient at capturing and promoting the destruction of the sRNA (provided that it is made in excess to the sRNA) (21). It is not a sponge, therefore, in the strict sense of the word, but rather something more like a “predator.” Since sRNA cleavage is not an obligate outcome of pairing (22), it seems possible that an RNA capable of base-pairing with the sRNA could function as a sponge and/or a predator depending on structural features of the sRNA-RNA duplex. Below we discuss representative examples of both types of mechanisms. We limit our coverage to RNA-RNA interactions. Classical and more recent examples of RNAs that regulate gene expression through sponging of regulatory proteins, namely, CsrA and Hfq, were reviewed recently (23).

Predatory Mimicry in the Regulation of Chitosugar Uptake

In Salmonella enterica and Escherichia coli, an ∼80-nt sRNA named ChiX (also named MicM) represses the synthesis of outer membrane chitoporin ChiP (also named ybfM) by base-pairing with a 12-nt sequence within the ribosome binding site of chiP mRNA and blocking translation of this mRNA (22, 24). Repression is relieved in the presence of the chitin-derived sugars chitobiose and chitotriose, consistent with the fact that ChiP is needed for the chitosugars to cross the outer membrane. The mechanism responsible for the relief of chiP repression was elucidated in 2009. Two parallel studies showed that growth in the presence of chitosugars induces the transcription of an RNA recognized as target by ChiX, which upon base-pairing with ChiX promotes cleavage and rapid degradation of the sRNA (24) (Fig. 1). This decoy target originates from the chbBCARFG operon, which encodes the components of the chitosugars’ inner membrane transport system (chbBCA), the operon’s transcriptional activator (ChbR), and two catabolic enzymes (ChbFG) (25). Thus, the ChiX predation mechanism effectively couples the outer membrane entry of chitosugars with their active transport across the inner membrane. The decoy sequence is encoded within the chbB-chbC intercistronic spacer; however, the actual form responsible for capturing ChiX is not known. ChiX cleavage correlates with the appearance, in wild-type but not in RNase E mutant cells, of an ∼400-nt RNA from the chbBCA portion of the transcript (24). This suggests that the decoy sequence acts as part of a much longer RNA that undergoes RNase E cleavage upon pairing with ChiX, generating the ∼400-nt intermediate. Nonetheless, the chbBC spacer sequence maintains its ability to capture ChiX and induce its cleavage even when removed from its natural context and expressed ectopically. By comparison, the chiP leader is much less effective at causing such cleavage (24). This is consistent with evidence indicating that ChiX action on chiP mRNA is, at least partially, catalytic (i.e., the sRNA is recycled a number of times before being degraded) (22). Thus, the ChiX-chiP and ChiX-chbBC RNA hybrids must differ in some features responsible for the differential fate of ChiX. Our initial proposal that destabilization of ChiX in the ChiX-chbBC hybrid results from the partial melting of a CG-rich stem-loop structure at the 3′ end of ChiX could not be experimentally confirmed. More-recent evidence suggests that binding affinity is the determining factor. The ChiX-chbBC RNA hybrid (19 bp) is predicted to be significantly more stable than the ChiX-chiP hybrid (12 bp). The 19-bp ChiX-chbBC RNA duplex is interrupted by two mismatches at adjacent positions. We found that eliminating these two mismatches by mutation—thus making the chbBC RNA fully complementary to ChiX over 21 consecutive nucleotides—stimulated ChiX cleavage even further (our unpublished observations). This raises the question as to why the wild-type sequence is not a perfect 21-bp match. The answer might be found in the observation that the pairing between ChiX and the chbBC spacer mRNA, while inactivating ChiX under inducing conditions (chbBC mRNA in excess), actually represses chbC gene expression when the operon is uninduced and transcribed at its basal level (ChiX in excess; Fig. 1) (26). A continuous 21-bp interaction is expected to make this repression even tighter. This might interfere with the inducibility of the entire network, as some low-level production of the transport system may be required when bacteria first encounter chitosugars to allow some of these molecules to leak into the cytoplasm and prime the induction cascade.

FIGURE 1.

Regulation of chitosugar uptake in Salmonella and E. coli. The chiP gene and the chbBCARFG operon encode proteins involved in the uptake and utilization of chitin-derived sugars. When no chitosugars are available, ChiP synthesis is prevented by constitutively made ChiX sRNA, which represses translation of chiP mRNA (made at a relatively high basal level), while the chbBCARF operon is repressed transcriptionally by the NagC repressor (not shown). ChiX further lowers the uninduced levels of the chb mRNA by pairing with a sequence in the chbB-chbC intercistronic region. In the presence of chitosugars, transcriptional activation of chbBCARF operon produces a large accumulation of the polycistronic mRNA. Now in excess over ChiX, this mRNA titrates out ChiX through base-pairing and promotes its degradation. ChiX depletion results in the derepression of the chiP mRNA.

sRNA Sponging by tRNA Spacer Sequences

The sRNAs RybB and RyhB control two major homeostatic networks in E. coli and S. enterica, the σ^E-dependent envelope stress response and iron homeostasis, respectively. The rybB gene is transcribed upon activation of alternative sigma factor σ^E triggered by folding defects in outer membrane proteins (OMPs) (27, 28). RybB downregulates some of the major OMPs, and in doing so, it suppresses the σ^E-activating signal and thus its own transcription. The ryhB gene is repressed by the Fur repressor and becomes derepressed when the intracellular iron is depleted. RyhB silences the expression of nonessential iron-binding proteins while upregulating iron uptake systems (29, 30). Thus, RyhB activity contributes to replenishing the intracellular iron pool, which, in turn, restores Fur-mediated repression of the ryhB gene. Recently, the RybB and RyhB regulons were chosen in the implementation of a new method for sRNA target identification based on high-throughput sequencing of transcripts copurified with the sRNA of interest (31). The method retrieved many of the previously known targets of RybB and RyhB, but in addition, it uncovered a peculiar new target that, surprisingly, copurified with either of the two sRNAs. The shared target is an ∼50-nt RNA originating from the 3′ external transcribed spacer (3′ ETS) of the tricistronic glyW-cysT-leuZ tRNA precursor. The 3′ETS^leuZ RNA, cleaved off the tRNA precursor by RNase E during tRNA maturation, contains sequences complementary to the pairing domain of both RybB and RyhB and can engage in a base-pair interaction with each of the two sRNAs (Fig. 2). The interaction does not promote the degradation of the sRNAs, and it actually stabilizes 3′ETS^leuZ RNA, suggesting that the RybB-3′ETS^leuZ and RyhB-3′ETS^leuZ hybrids are relatively long-lived (31). Thus, unlike what is observed in ChiX regulation (above), 3′ETS^leuZ sequesters the sRNAs as opposed to destroying them; that is, it acts as a true sponge. The sponging activity serves to absorb RybB and RyhB molecules made adventitiously due to noise in promoter activity (RybB) or incomplete Fur repression during normal growth. In doing so, 3′ETS^leuZ sets a threshold level of expression that each of the two sRNAs must attain to begin acting on the respective targets (Fig. 2). Above the threshold, 3′ETS^leuZ does not hamper further accumulation of either RybB or RyhB, presumably because the absorbing capacity of the sponge is saturated (31). One might then predict that full induction of either of the two sRNAs will free the 3′ETS^leuZ-bound fraction of the other sRNA, causing it to increase to a higher basal level. This suggests that the ability of 3′ETS^leuZ to target both sRNAs is designed to link iron homeostasis to the σ^E-dependent envelope stress response. Indeed, the use of a tRNA processing product as a sponge for RybB and RyhB might have evolved to allow the activities of the two sRNAs to be modulated as a function of the physiological state of the cell. The glyW-cysT-leuZ operon is predicted to be susceptible to the stringent control and the growth rate-dependent regulation. This implies that the sponging activity of 3′ETS^leuZ will be maximal under fast growth conditions but should drop abruptly if ppGpp levels increase or, more generally, growth slows down. Allowing the free fraction of RybB and RyhB to increase under these conditions could be important for “preadapting” the cells to an incoming stress. On one hand, (p)ppGpp is known to stimulate σ^E activity (directly by acting on σ^E promoter and indirectly by favoring the ability of σ^E to compete with σ⁷⁰ for binding to core RNA polymerase) (32) and is thought to be the primary signal responsible for induction of the σ^E response in stationary phase (33). On the other hand, iron limitation was reported to induce SpoT-dependent ppGpp accumulation (34). One could easily see how ppGpp-mediated relief of RybB and RyhB sponging would help the cells to more rapidly integrate the activities of these two sRNAs in the respective regulatory networks.

FIGURE 2.

sRNA sponging by a tRNA spacer sequence. The sRNAs RybB (blue) and RyhB (purple) are made in response to envelope stress or iron limitation, respectively. An ∼50-nt RNA, named 3′ETS^leuZ (red), released by RNase E cleavage of the glyW-cysT-leuZ tRNA precursor (top) can form stable base-pair interactions with both RybB and RyhB. This allows 3′ETS^leuZ to capture and sequester RybB and RyhB molecules that are made adventitiously (in the absence of any stress) due to transcriptional noise (left). Under inducing conditions (envelope stress or iron limitation), accumulation of either RybB or RyhB saturates the sponging capacity of 3′ETS^leuZ. This sets the threshold concentration (dotted line) that either of the two sRNAs must attain to begin performing its regulatory task: downregulation of OMPs for RybB (middle) or of nonessential iron-binding proteins for RyhB (right).

The SroC RNA: Target-Derived, but Not a Target Mimic

GcvB is a conserved, ∼200-nt sRNA that downregulates >40 different mRNAs, most of them encoding proteins involved in amino acid uptake (35, 36). Regulation is maximally exerted during exponential growth in nutrient-rich media. The precise role of this control is unclear, but it is probably aimed at balancing metabolic fluxes connected with the glycine cleavage system, whose main regulator, GcvA, activates the gcvB transcript (37). The glycine cleavage pathway is the major source of one-carbon units used, among others, in the synthesis of purines, thus directly linking amino acid and nucleotide metabolisms (38).

Among the GcvB-regulated mRNAs is the polycistronic transcript of the gltIJKL operon, which encodes the glutamate/aspartate ABC transporter. GcvB represses expression of the gltIJKL operon by pairing with a sequence within the leader region of the first cistron, gltI (35). As early as in 2003, Vogel and coworkers discovered that the gltIJKL operon also encoded an sRNA in the intercistronic region between gltI and gltJ (39). This sRNA, named SroC, is made from transcripts that don’t extend all the way to the end of the operon but terminate at a leaky Rho-independent terminator ∼150 nt downstream from the stop codon of the gltI gene. This ∼150-nt tail is clipped off by RNase E, and the released RNA fragment, SroC, associates with Hfq and is stably maintained in the cell, suggesting that it might have a function of its own (40). Indeed, this study revealed that SroC can bind to and inactivate GcvB, and thus drive a feedforward regulatory loop resulting in the derepression of the entire GcvB regulon (Fig. 3). SroC base-pairs simultaneously with two separate segments in the GcvB sequence, distant from each other, neither of which corresponds to the domain used by GcvB to pair with most of its targets. The interaction exposes GcvB to cleavage by RNase E, whereas SroC is not cleaved and can be recycled (40). The use of a specific binding domain implies that SroC does not need to compete with most GcvB targets for binding to GcvB. Combined with recycling, lack of competition is expected to make SroC a particularly effective GcvB predator even at low concentrations. Interestingly, GcvB-dependent regulation and its reversal by SroC appear to be entirely recapitulated in cells growing in LB medium. GcvB accumulates maximally during the exponential phase when SroC levels are low; then it declines rapidly when cells enter stationary phase (35). The decline coincides with a vast increase in the intracellular concentration of SroC (39). One might envision that the activity of the Rho-independent terminator that generates the SroC precursor in the first place is increased in stationary phase. The recent finding that transcription termination at the Rho-independent terminators of sRNA genes is enhanced under stress conditions and in stationary phase lends support to this possibility (41). Yet the physiological role of SroC in the GcvB network remains elusive. The link to glutamate, a molecule at the crossroads of key metabolic pathways, suggests that the action of SroC aims at rewiring the GcvB regulon in a manner more adapted to the metabolism of stationary phase and/or of stress conditions.

FIGURE 3.

Target-mediated derepression of the GcvB regulon. The sRNA GcvB downregulates several mRNAs encoding amino acid and small peptide transporters. Among these is the gltIJKL mRNA (left). Presence of a leaky Rho-independent transcription terminator in the spacer between gltI and gltJ causes a fraction of transcripts initiating at the gltI promoter to terminate prematurely in the spacer region (right). RNase E cleavage of the prematurely terminated transcripts generates SroC, an ∼150-nt RNA, which captures GcvB through a base-pairing interaction and destabilizes it. As a result, all of the GcvB targets become derepressed. Since the SroC precursor RNA itself is one of these targets, SroC activity drives a feedforward regulatory loop.

Some of the transcriptomic changes that result from the overexpression of the sroC gene in Salmonella persist in a ΔgcvB background and thus are likely to reflect regulation of genes outside the GcvB network. In particular, a handful of sRNAs were found downregulated in cells overproducing SroC. These effects were ascribed to SroC overaccumulation draining the Hfq pool (i.e., SroC acting as an Hfq sponge) and the decline in Hfq availability causing the destabilization of other Hfq-binding sRNAs (40). It turns out that for one of the downregulated sRNAs, Hfq depletion might not be the sole cause of destabilization. This is the case for MgrR, an sRNA that depends on the PhoP-PhoQ two-component system for expression (42). A recent study presented evidence that SroC can base-pair with MgrR and that this interaction stimulates MgrR turnover (43). Most importantly, by promoting MgrR decay, SroC alleviates the repression of MgrR’s primary target, the lipopolysaccharide-modifying enzyme EptB. The EptB-directed modification decreases the bacterial susceptibility to the antimicrobial peptide polymyxin B in both E. coli and Salmonella (42, 43). The PhoP-PhoQ two-component system is also known to contribute functions enhancing polymyxin B resistance (44). Hence the PhoP-PhoQ requirement for MgrR expression is counterintuitive, since in repressing EptB, MgrR is expected to decrease, not increase, polymyxin B resistance. The discovery that SroC can inactivate MgrR helps solve this conundrum. One may speculate that SroC production or action is somehow potentiated during polymyxin B exposure so as to prevent the EptB repression. This hypothesis remains to be tested.

Prophage-Encoded GcvB Sponges

Two 60-nt sRNAs with anti-GcvB activity, encoded in the genome of distinct prophages, have been identified in enteropathogenic E. coli O157:H7 (45). The two sRNAs, named AgvB1 and AgvB2, bind Hfq in vivo and in vitro and can base-pair with the main seed sequence of GcvB, the so-called R1 region. When overexpressed, AgvB1 alleviates the repression of at least one GcvB target (ddpA, the only one tested), presumably by competing with the dppA mRNA for pairing with GcvB. These findings, together with the observation that neither the phage-encoded sRNAs nor GcvB are destabilized upon formation of the hybrid, indicate that, unlike SroC, AgvB1 and AgvB2 sequester rather than inactivate GcvB (45). Deleting the agvB1 and agvB2 genes from the E. coli O157:H7 genome did not significantly affect the growth of the strain in laboratory media, but it had a fitness cost when bacteria were grown in mucus from the bovine terminal rectum, suggesting that the relief of GcvB repression confers a growth advantage in the host.

Glimpses of Pervasive Sponging from Global Snapshots

Two new methodologies—CLASH (cross-linking, ligation, and sequencing of hybrids [46]) and RIL-seq (RNA interaction by ligation and sequencing [47])—have made it possible to profile RNA-RNA interactions at transcriptome-wide levels (48). Both methods exploit high-throughput sequencing of RNA UV-cross-linked to a pertinent protein, affinity purified, and subjected to proximity ligation. The ligation step captures RNA molecules in the act of base-pairing at the time of RNA extraction, allowing their identification as chimeric reads in transcriptome sequencing output. RIL-seq and CLASH were recently implemented in E. coli with RNA cross-linked to Hfq (47) and RNase E (49), respectively. The results of both studies retrieve a global picture of the bacterial interactome of unprecedented depth. While confirming most of the known interactions, these studies unveil a plethora of putative new interactions, greatly expanding the networks of virtually all known sRNAs and identifying new ones. Included in the sRNA networks are not only mRNAs but also tRNAs, tRNA precursors, and sRNAs, suggesting that sRNA sponging by tRNA-associated sequences or other sRNAs is a pervasive phenomenon. Particularly striking is the expansion of the ChiX network, found to include not less than 20 unrelated mRNAs (47, 49), at least as many tRNAs (47, 49), and a handful of sRNAs (49). In the absence of a coherent biological framework to accommodate most of the new interactions, such unsuspected complexity is somewhat disconcerting. The CLASH data suggest that ChiX uses a different seed sequence to interact with most of the newly identified targets (49). Unlike the sequence used to base-pair with chiP and chbBC, and previously identified dpiBA mRNA (50), this second pairing region lies in a portion of the ChiX sequence that is not conserved between E. coli and Salmonella, raising doubt as to the existence of the same interactions in Salmonella. This would be surprising given that the vast majority of known sRNA networks are conserved between the two bacterial species. Thus, it seems important that the novel interactions be directly verified experimentally and characterized in quantitative terms. One such verification was performed in the RIL-seq study to confirm the discovery of a new sponge system (47). An sRNA reportedly processed from the 3′ end of the pspG mRNA (encoding phage shock protein G) and named PspH was shown to interact with, and destabilize, the sRNA Spot 42 (Spf). The latter is a key regulator of carbohydrate metabolism. Activated by growth on glucose, Spot 42 tightens catabolite repression by downregulating genes involved in uptake and catabolism of nonfavored carbon sources (51). The ability of PspH to destabilize Spot 42 (causing the derepression of Spot 42 targets) was assessed with PspH overexpressed from a plasmid, and it is currently unclear what might trigger this regulation under physiological conditions, in particular whether PspH accumulates during the phage shock response. We notice, however, that the sRNA encoded at the pspH locus in Salmonella (named STnc880) is transcribed from its own promoter and likely regulated independently of pspG (52). STnc880 is considerably smaller (79 nt) than what one can infer to be the PspH size from reference 47; however, it is highly conserved in the shared portion of the sequence and 100% identical to PspH in the region involved in the Spot 42 interaction.

BROADENING THE SPONGE CONCEPT: A SYSTEMS BIOLOGY PERSPECTIVE

A widespread routine for assessing the regulatory activity of an sRNA involves overexpressing the sRNA gene cloned in a multicopy plasmid in cells carrying a reporter gene fusion to a presumptive target and assessing a change in the reporter expression. We suspect that if one were to apply the exact same procedure but switching the roles, that is, overexpressing the target mRNA in the presence of cognate sRNA, one might observe a decrease in the sRNA’s regulatory activity on other targets. This is because the overproduced mRNA could sequester or promote cleavage of the cognate sRNA. In other words, many functional mRNAs that are targets of sRNA regulation might have the potential for acting as sRNA sponges. This capacity has interesting implications in the systems biology of sRNA networks. Indeed, one can imagine that a surge in concentration of a transcript at the interface of two networks, resulting from a regulatory change occurring in one network, could affect the linked network via a sponging activity (Fig. 4). For example, one predicts that all newly identified targets of ChiX (47, 49) will be derepressed upon activation of chitobiose operon transcription in E. coli. This type of mechanism was invoked to explain the silencing of the RprA sRNA network when E. coli cells enter stationary phase or during initial biofilm formation (53). Both conditions coincide with a strong transcriptional activation of the csgD gene, which encodes a positive regulator of curli and cellulose biosynthesis in E. coli and Salmonella. The csgD mRNA, a high-affinity target of RprA, was proposed to sequester RprA via the extensive base-pairing interaction and, in doing so, relieve the regulation of the other targets by this sRNA (53). Since csgD is also the target of additional sRNAs (e.g., OmrA and OmrB) (54), its sponging activity could conceivably extend to other networks.

FIGURE 4.

A sponging relay model. Depicted are two hypothetical sRNA networks (A and B) linked by an mRNA node (cyan-filled circle). (Top) The two sRNAs downregulate their respective targets. (Bottom) A transcriptional regulatory event leads to a large increase in the concentration of one of the mRNAs in network A (yellow-filled circle). The accumulated mRNA sequesters and destabilizes its cognate sRNA, resulting in the derepression of the entire network A, including the nodal mRNA. In turn, the latter acts as a sponge for cognate sRNA in network B, thus relieving, or attenuating (depicted here), the repression of the B network.

Other candidates for mRNA sponge activity are found in the regulation of quorum sensing in Vibrio species. Quorum sensing involves production and extracellular release of effector molecules, called autoinducers, which upon binding to specific receptors in the cell membrane signal the density state of the bacterial population, triggering a transcriptional response (55). At low cell densities (no autoinducers bound), phosphorelay-mediated activation of the transcription factor LuxO leads to the synthesis of five homologous sRNAs, the Qrr sRNAs, which repress several targets including the mRNA for master regulator LuxR. At high cell densities, autoinducer binding reverses the phosphate flow, inactivating LuxO and abolishing Qrr sRNA production. Three relevant Qrr-downregulated mRNAs are those encoding LuxO, LuxR, and the autoinducer synthase LuxM (55). Work by Bassler’s group recently showed that Qrr sRNAs form different base-pair interactions with each of the three mRNAs and that the nature of the interaction determines the fate of the sRNA-mRNA hybrid. Specifically, a representative Qrr sRNA, Qrr3, induces the degradation of luxR mRNA and is then recycled (catalytic degradation), forms a stable hybrid with luxO mRNA (sequestration), and induces the degradation of luxM mRNA while being degraded at the same time (coupled degradation) (56). Catalytic degradation produced the most robust regulatory response. Its use in luxR repression could be rationalized considering that high- to-low-cell-density transitions can be abrupt in nature (e.g., bacterial excretion from a host), requiring a rapid shutoff of the entire luxR regulon. Sequestration of luxO mRNA by Qrr3 was proposed to serve as a negative feedback mechanism buffering fluctuations in LuxO protein levels. Since Qrr3 is itself sequestered in the process, an alternative (or additional) interpretation is that luxO mRNA acts as a sponge to absorb surges in Qrr sRNA levels that result from transcriptional noise. Finally, the use of coupled degradation to regulate luxM allows this mRNA to turn into a regulator and contribute to the elimination of Qrr sRNAs at the onset of the high-cell-density program (56).

Overall, the above examples offer clues to better appreciate the evolutionary significance of sRNA-mediated regulation: sRNAs might have evolved not only to allow rapid responses, as usually thought, but also to achieve a continuous physiological balance between distinct cellular processes, with the target mRNA nodes acting as the connectors of communicating vessels (Fig. 4).

OUTLOOK

Since its discovery, sRNA-mediated regulation continues to attract the interest of a wide scientific audience and constitutes a fertile and dynamic field of research. Recent advances in the development of high-throughput methods have allowed a major leap forward in the characterization of the bacterial sRNA interactome. Still, in some ways, the recent breakthroughs epitomize the so-called knowledge paradox: as the body of knowledge increases, so does the area we become aware of not knowing. The scores of unsuspected sRNA interactions whose biological rationale remains unexplained highlight the gaps that still exist in our understanding of the inner workings of the bacterial cells. Filling these gaps will require validating the newly discovered interactions individually as well as applying the high-throughput screen to sRNA knockout or overproducing strains and to mutants of transcriptional regulators whose networks feed into sRNA networks (57, 58). This work, which could include proteomic and metabolomic analyses, seems an obligatory step to obtain a more complete picture of how sRNA activities are integrated in the physiological control of gene expression in bacterial cells.