AI-2-mediated signalling in bacteria

Authors

  • Catarina S. Pereira,

    1. Instituto Gulbenkian de Ciência, Oeiras, Portugal
    2. Instituto de Tecnologia Química e Biológica, Universidade Nova de Lisboa, Oeiras, Portugal
    Search for more papers by this author
  • Jessica A. Thompson,

    1. Instituto Gulbenkian de Ciência, Oeiras, Portugal
    2. Instituto de Tecnologia Química e Biológica, Universidade Nova de Lisboa, Oeiras, Portugal
    Search for more papers by this author
  • Karina B. Xavier

    Corresponding author
    1. Instituto de Tecnologia Química e Biológica, Universidade Nova de Lisboa, Oeiras, Portugal
    • Instituto Gulbenkian de Ciência, Oeiras, Portugal
    Search for more papers by this author

Correspondence: Karina B. Xavier, Instituto Gulbenkian de Ciência, 2780-156 Oeiras, Portugal. Tel.:+351 21 4464655; fax: +351 21 4407970; e-mail: kxavier@igc.gulbenkian.pt

Abstract

Success in nature depends upon an ability to perceive and adapt to the surrounding environment. Bacteria are not an exception; they recognize and constantly adjust to changing situations by sensing environmental and self-produced signals, altering gene expression accordingly. Autoinducer-2 (AI-2) is a signal molecule produced by LuxS, an enzyme found in many bacterial species and thus proposed to enable interspecies communication. Two classes of AI-2 receptors and many layers and interactions involved in downstream signalling have been identified so far. Although AI-2 has been implicated in the regulation of numerous niche-specific behaviours across the bacterial kingdom, interpretation of these results is complicated by the dual role of LuxS in signalling and the activated methyl cycle, a crucial central metabolic pathway. In this article, we present a comprehensive review of the discovery and early characterization of AI-2, current developments in signal detection, transduction and regulation, and the major studies investigating the phenotypes regulated by this molecule. The development of novel tools should help to resolve many of the remaining questions in the field; we highlight how these advances might be exploited in AI-2 quorum quenching, treatment of diseases, and the manipulation of beneficial behaviours caused by polyspecies communities.

Introduction

Bacteria are prevalent throughout the natural world, inhabiting diverse environments from thermal springs to the intensely acidic human stomach. Intrinsic to successful colonization and exploitation of such niches is their ability to sense these surroundings and modify behaviour accordingly. These modifications are dictated by the interplay between a complex network of signalling pathways: elicitation leads to changes in gene expression and induction of phenotypes optimal for growth in a given situation.

As the natural context for bacterial growth is rarely one of isolation, it is perhaps unsurprising that mechanisms for the detection of and response to other organisms are widespread amongst bacterial species. These mechanisms drive the regulation of phenotypes appropriate to a given social situation, including the induction of many cell density-dependent responses such as the synchronized production and secretion of virulence factors, bioluminescence, biofilm formation and changes in motility. Such behaviours are typically productive only when enough cells are working in unison, thus it is crucial that expression is limited solely to conditions in which they are beneficial. After nearly 30 years in development, the idea of bacteria as organisms capable of coordinated, synchronized gene expression and behaviour is now well accepted.

In 1964, Tomasz & Hotchkiss demonstrated that the induction of competence by Streptococcus pneumoniae was dependent upon a self-secreted ‘activator substance’ which accumulated with bacterial number. A few years later, Nealson et al. (1970) similarly reported cell density-dependent regulation of bacterial behaviour: the induction of luciferase expression and light production by the marine bacterium Vibrio fischeri. The term autoinduction was coined to describe the elicitation of phenotypic changes by bacteria in response to self-produced molecules called autoinducers. These molecules accumulate in the extracellular environment until a critical threshold concentration for detection is reached, after which downstream signalling and effector responses ensue.

Characterization of the autoinducer synthase and receptor proteins in V. fischeri (Engebrecht et al., 1983; Engebrecht & Silverman, 1987) enabled homologues of these proteins to be identified in diverse Gram-negative bacterial species including many pathogens (Gambello & Iglewski, 1991; Gambello et al., 1993; Piper et al., 1993; Pirhonen et al., 1993; Zhang et al., 1993; Fuqua et al., 1994). This means of inducing group behaviours, christened quorum sensing by Fuqua, Winans and Greenberg (and illustrated by the V. fischeri LuxIR quorum sensing paradigm in Box ), was clearly a widespread mechanism in bacteria (Fuqua et al., 1994).

Box 1: Quorum sensing in V. fischeri.

image

(a) The enzyme LuxI, synthesises an autoinducer, N-3-oxo-hexanoylhomoserine lactone (3OC6HSL), which diffuses out of the bacterial cell. (b) 3OC6HSL accumulates extracellularly with bacterial growth. (c) A critical concentration is reached at a particular cell density which then enables diffusion of 3OC6HSL back into the cell, where it binds and stabilises its cognate receptor, LuxR. (d) The change in conformation upon formation of the receptor-ligand complex enables LuxR to drive transcription of the lux operon, encoding enzymes required for luciferase expression as well as luxl. This creates a positive feedback loop: detection of signal promotes the synthesis of more signal, and the further induction of bioluminescence. 3OC6HSL is a member of the AHL (acyl homoserine lactone) class of autoinducers which comprise of a common homoserine lactone (HSL) ring coupled to a species-specific R group. AHLs are the predominant class of QS signals in Gram-negative bacteria.

The notion of bacteria communicating with each other through a chemical language ignited an explosion in research; an increasingly complex molecular lexicon has been discovered since (Bassler, 1999; Bassler & Losick, 2006). Many of these molecules are highly specific, produced and recognized by a single species. In the late 1990s, autoinducer-2 (AI-2) was shown to induce bioluminescence in Vibrio harveyi cultures; remarkably, AI-2 production was detected in diverse Gram-negative and -positive bacterial species (Bassler et al., 1997; Surette et al., 1999). Hence, AI-2 has the potential to enable interspecies communication (Bassler, 1999; Schauder & Bassler, 2001; Xavier & Bassler, 2003). Through integration with species-specific signalling, this molecule could also enable bacteria to distinguish between self and non-self, providing important information about the local species composition and the potential for behaviour to be fine-tuned to the particular social environment encountered.

In this article, we will provide a historical perspective of the discovery and characterization of AI-2, review the latest developments in signal detection, downstream signalling and responses, and give a critical appraisal of the arguments for and against AI-2 as an interspecies communication molecule. Given the pressing need for novel therapeutics in the face of rising antibiotic resistance, interference in quorum sensing and the consequent impact on hostile or beneficial bacterial behaviours provides an attractive alternative in the fight against bacterial disease. The current research strategies adopted and advances made towards the manipulation of AI-2 signalling will also be outlined, highlighting how continued research into the biology and chemistry of AI-2 can help to achieve this goal.

Discovery of a common language: a historical perspective on AI-2

The first indication of bacteria signalling between species emerged from the work of Greenberg et al. (1979) in which the bioluminescent response of V. harveyi was triggered by the simple addition of cell-free supernatant from cultures of other bacterial species. These bacteria were responding to something produced by the neighbouring marine species examined. Further light was shed in 1993, following Bassler et al.'s (1993) observation that V. harveyi mutant strains defective in acyl homoserine lactone (AHL) synthesis remained capable of quorum sensing-dependent gene activation. The possibility of a second quorum sensing system was proposed, the signal for which was called AI-2 (Bassler et al., 1994). An AHL-blind reporter strain was constructed which was bioluminescent only upon induction of the AI-2-dependent second system. This strain responded to culture fluids from several unrelated bacterial species. AI-2 production was evidently not restricted to a single bacterial species (Bassler et al., 1997), which was confirmed following the identification of the gene responsible for AI-2 activity, designated luxS, and the existence of homologues in several sequenced genomes (Surette et al., 1999). LuxS homologues can now be identified in 537 of the 1402 bacterial genomes currently sequenced [based on genomes from the KEGG database (http://www.genome.jp/kegg/) with homologues to the luxS gene of Salmonella Typhimurium with an e-value smaller than 10−12]. The detection of AI-2 activity in the extracellular media of every luxS-containing species studied promoted the idea that bacteria use AI-2 to communicate across the species barrier (Bassler, 1999; Surette et al., 1999).

Insight into AI-2 biosynthesis came 2 years later. Genome sequence analysis demonstrated that the Borrelia burgdorferi luxS gene was the third in a three-gene operon (Schauder et al., 2001; Winzer et al., 2002b). The two other genes, pfs and metK, encode enzymes involved in the activated methyl cycle (AMC), an important metabolic pathway for the recycling of S-adenosylmethionine (SAM), the major methyl donor in the cell (Miller & Duerre, 1968; Schauder et al., 2001). The release of the activated methyl group from SAM to an acceptor molecule gives rise to a toxic intermediate, S-adenosylhomocysteine (SAH), which is converted by Pfs to S-ribosylhomocysteine (SRH) (Fig. 1). Two independent groups showed that LuxS catalyzes the cleavage of SRH to homocysteine and a compound, at that time unidentified, with AI-2 activity (Schauder et al., 2001; Winzer et al., 2002b).

Figure 1.

The AMC and synthesis of autoinducers. (a) Transfer of a methyl group from SAM to an acceptor molecule generates the toxic intermediate, SAH. (b) Organisms that contain the SahH enzyme convert SAH directly to homocysteine. (c) Those with LuxS catalyse this reaction in two steps: Pfs first generates SRH and adenine from SAH, then LuxS converts SRH into homocysteine and DPD, the linear form of AI-2. (d) Homocysteine is regenerated into SAM via a series of reactions involving methionine. THF-CH3 and THPG stand for N-methyltetrahydrofolate and 5-methyltetrahydropteroyl glutamate, respectively. (e) LuxI-type proteins also use SAM and acyl carrier proteins (Acyl-ACP) as substrates for the production of AHL autoinducers. (f) Vibrio cholerae autoinducer CAI-1 is also synthesized from SAM and decanoyl Coenzyme A via a multistep reaction involving CsqA, a pyridoxal phosphatase (PLP)-aminotransferase like enzyme, and VC1059. Vibrio harveyi has an homologue to CqsA that uses octanoyl CoA yielding Z-3-aminoundec-2-en-4-one.

Though AI-2 could now be produced in vitro using purified LuxS and Pfs proteins with SAH as a precursor (Schauder et al., 2001; Winzer et al., 2002b), chemical characterization of the signal remained a challenge. It was later shown that the linear chemical product of LuxS is a very unstable molecule, 4,5-dihydroxy-2,3-pentanedione (DPD), providing a likely explanation for why classical chemical and biochemical methods were so unsuccessful in its identification (Fig. 1). DPD also spontaneously cyclizes into different isomers in solution, raising the question of which form was detected by AI-2-responding bacteria (Fig. 2) (Chen et al., 2002). It was only with the imaginative approach of exploiting the high affinity nature of the V. harveyi receptor, LuxP, to trap the corresponding ligand that this puzzle was solved. Crystallization of the LuxP-ligand complex yielded the first structure of a molecule with AI-2 activity, a cyclic, borated form of AI-2 called S-2-methyl-2,3,3,4-tetrahydroxytetrahydrofuran-borate (S-THMF-borate) (Chen et al., 2002). A further surprise came from the identification of a markedly different enantiomer of DPD, a non-borated R-2-methyl-2,3,3,4-tetrahydroxytetrahydrofuran (R-THMF) ligand, in the crystal structure of a second AI-2 receptor, LsrB, from Salmonella enterica ssp. enterica serovar Typhimurium (Fig. 2) (Miller et al., 2004). Although the signal synthase, corresponding biosynthetic pathway and chemical products are the same in every AI-2-producing bacterium tested thus far, these studies demonstrate that the molecule ultimately detected by these bacteria can differ. Despite this, AI-2-dependent interspecies communication remained a possibility as the different isomers of DPD are in equilibrium when in solution and are capable of rapid inter-conversion (Meijler et al., 2004; Miller et al., 2004). This was demonstrated experimentally: co-culture of V. harveyi and Escherichia coli, which detect S-THMF-borate and R-THMF (Fig. 2), respectively, leads to changes in AI-2-dependent gene expression in both organisms dependent upon the ability of each bacterium to synthesize and detect the signal (Xavier & Bassler, 2005b).

Figure 2.

AI-2 signalling molecules and receptors. (a) The LuxS product DPD spontaneously undergoes cyclization and hydration reactions in solution to form R-THMF (detected by Salmonella Typhimurium) and S-THMF. In the presence of boron, the latter molecule further reacts to form S-THMF-borate (detected by Vibrio harveyi). (b) These AI-2 ligands are coordinated by different interactions (hydrogen bonds or salt bridges, dashed yellow lines) with different amino acids residues in the binding sites of their respective receptors: LuxP containing S-THMF-borate from V. harveyi, and LsrB containing R-THMF from S. Typhimurium, although the structure of the receptors (c), shown by ribbon diagrams coloured in rainbow order from the N- to the C-terminus, reveals much similarity between these two proteins.

A language for debate: AI-2 in signalling and metabolism

As shown in the next sections, a clear role in quorum sensing has been demonstrated for AI-2 in certain bacterial species (Table 1). As LuxS is not only the AI-2 synthase but also an enzyme in the crucial central metabolic pathway, AMC (Fig. 1), it is equally possible that in other species this molecule is merely a metabolic by-product. Inactivation of luxS could result in changes in gene expression as a consequence of defective signalling, methionine metabolism, recycling of homocysteine or accumulation of intermediates of SAM metabolism. Therefore, to establish whether AI-2 is a bona fide signalling molecule or only a by-product of metabolism in a given species, it is extremely important to discriminate between these metabolic and signalling defects. One of the crucial experimental approaches to distinguish between the dual effects of luxS deletion is by signal complementation (Winzer et al., 2003; Vendeville et al., 2005; Hardie & Heurlier, 2008). As several organic synthesis methods for DPD have recently been developed (Meijler et al., 2004; De Keersmaecker et al., 2005; Frezza et al., 2005; Semmelhack et al., 2005; Smith et al., 2009; Kadirvel et al., 2010; Ascenso et al., 2011; reviewed in Yajima, 2011), and this chemical is now commercially available (Omm Scientific), signal complementation can be performed much more easily.

Table 1. Functions regulated by AI-2 signala
SpeciesFunctions regulated by AI-2AI-2 receptorReferences
  1. a

    Whole genome expression profiles such as microarrays analysis were not included in this table.

  2. b

    DPD supplemented to wild-type strain cultures.

  3. c

    This receptor is present in this bacteria but the connection to the regulated phenotypes remains to be demonstrated.

Actinobacillus pleuropneumoniae Biofilm formationb, adherence to host cells and growth in iron-limited mediumUnknownLi et al. (2011)
Actinomyces naeslundii and Streptococcus oralis Mutualistic biofilm formationUnknownRickard et al. (2006)
Aggregatibacter actinomycetemcomitans Biofilm formationLsrB and RbsBShao et al. (2007a,b)
Bacillus cereus Biofilm formationb LsrBc Auger et al. (2006)
Borrelia burgdorferi Increased expression of the outer surface lipoprotein VlsEb UnknownBabb et al. (2005)
Escherichia coli EHECChemotaxis towards AI-2, motility and HeLa cell attachmentLsrBc Bansal et al. (2008)
Escherichia coli K12Biofilm formation and motilityb LsrBc Xavier & Bassler (2005a), Gonzalez Barrios et al. (2006), Hegde et al. (2011)
AI-2 incorporation and chemotaxis towards AI-2LsrB
Haemophilus influenzae strain 86-028NPAI-2 incorporation and biofilm formationRbsBArmbruster et al. (2011)
Helicobacter pylori MotilityUnknownRader et al. (2007), Shen et al. (2010), Rader et al. (2011)
Moraxella catarrhalis Biofilm formation and antibiotic resistanceb UnknownArmbruster et al. (2010)
Mycobacterium avium Biofilm formationb UnknownGeier et al. (2008)
Pseudomonas aeruginosa Virulence factor productionUnknownDuan et al. ( 2003)
Salmonella enterica ssp. enterica serovar TyphimuriumPathogenicity island 1 gene expression and invasion into eukaryotic cellsLsrBc Taga et al. (2001, 2003), Miller et al. (2004), Choi et al. (2007, 2012)
AI-2 incorporationLsrB
Sinorhizobium meliloti AI-2 incorporationLsrBPereira et al. (2008)
Staphyloccocus aureus Capsular polysaccharide gene expression and survival rate in human blood and macrophagesUnknownZhao et al. (2010)
Staphylococcus epidermidis Expression of phenol-soluble modulin peptides, acetoin dehydrogenase, gluconokinase, bacterial apoptosis protein LrgB, nitrite extrusion protein and fructose PTS system subunitUnknownLi et al. (2008)
Streptococcus anginosus Susceptibility to antibioticsUnknownAhmed et al. (2007)
Streptococcus intermedius Haemolytic activity, biofilm formation and susceptibility to antibioticsUnknownAhmed et al. (2008, 2009)
Streptococcus gordonii Biofilm formationUnknownSaenz et al. (2012)
Streptococcus gordonii and Streptococcus oralis Mutualistic biofilm formationUnknownSaenz et al. (2012)
Streptococcus pneumoniae Biofilm formationUnknownVidal et al. (2011)
Vibrio cholerae Biofilms, protease and virulence factor production, and competenceLuxPJobling & Holmes (1997), Miller et al. (2002), Zhu et al. (2002), Hammer & Bassler (2003), Antonova & Hammer (2011)
Vibrio harveyi Bioluminescence, colony morphology, siderophore production, biofilm formation, type III secretion and metalloprotease productionLuxPBassler et al. (1993, 1994), Lilley & Bassler (2000), Chen et al. (2002), Mok et al. (2003), Henke & Bassler (2004ab), Waters & Bassler (2006)

In many cases, use of such preparations of pure DPD to chemically complement defects in luxS mutant bacteria has successfully demonstrated the role of AI-2 as a quorum sensing molecule. These studies as well as reports where AI-2 sensing mechanism has been elucidated are summarized in Table 1. The Supporting Information, Table S1 details further studies where defects were seen upon disruption of luxS but the link with altered signalling may not have been definitively proven.

Even if, in some species, LuxS is only an enzyme of the AMC, the resulting AI-2 could still be used as an information source by other species sharing the same environment, illustrating the complexity of determining how LuxS and AI-2 affect bacterial behaviour at the species and community level.

Beyond the debate regarding the classification of AI-2 as a signal or metabolite, the link between AI-2 production and SAM can have other implications. In all living organisms, SAM is the main methyl donor in many essential methylation reactions required for cell growth, development and chemotaxis. It is also the donor of other chemical groups necessary for many cellular processes such as synthesis of phospholipids and vitamins (Fontecave et al., 2004; Parveen & Cornell, 2011). As AI-2 is a product of SAM metabolism it is a good candidate to provide information on the metabolic status of the cell (Winans, 2002; Xavier & Bassler, 2003). Additionally, the wide use of SAM in methylation reactions results in a constant production of SRH, the substrate for the LuxS reaction. This ensures low-cost production of AI-2, again providing an economical readout of cell activity (Winans, 2002; Winzer et al., 2003; Xavier & Bassler, 2003; Keller & Surette, 2006).

Strikingly, the use of SAM as a precursor for autoinducer synthesis is not exclusive to AI-2. As shown in Fig. 1, SAM is also the substrate for the synthesis of AHLs and CAI-1 (cholerae autoinducer-1) despite their different chemical composition. It is perhaps not a coincidence that these three classes of autoinducers all derive from SAM and it is tempting to speculate that the reason for this common connection relates to the benefit of integrating central metabolism with quorum sensing.

How to listen: producing, secreting and sensing AI-2

Since the discovery of luxS, hundreds of studies have been published to further understand AI-2 signalling: mutants of luxS have been made in more than 50 bacterial strains (Table S1); microarray studies have been performed by several independent groups; and five LuxS protein structures have been determined. Here, while highlighting the progress made regarding AI-2 production, secretion and detection, we will also give emphasis to the open questions awaiting investigation.

The AI-2 synthase: LuxS

Determination of the structures of LuxS from Bacillus subtilis, Helicobacter pylori, Deinococcus radiodurans, Haemophilus influenzae and Streptococcus mutans has revealed that the synthase is a homodimeric metalloenzyme containing a novel αβ fold. This fold comprises an eight-stranded antiparallel β-sheet surrounded by multiple α-helices shared between the two proteins (Hilgers & Ludwig, 2001; Lewis et al., 2001; Ruzheinikov et al., 2001; Zhu et al., 2003; Rajan et al., 2005; Li et al., 2012). The shape and small size of the active site (being relatively inaccessible without conformational change) suggested the hydrolytic cleavage of a non-peptide molecule. This theory was confirmed by the identification of SRH as substrate (Hilgers & Ludwig, 2001). Zinc was initially proposed as the metal cofactor; however, enzymatic activity was subsequently found to be 10-fold higher with divalent iron (Zhu et al., 2003); this ion is coordinated by two histidines (His 54 and His 58, numbering follows that of LuxS from B. subtilis), a cysteine (Cys 126) and a water molecule. Glu 57 and Cys 84 are also important residues, as an acid and base, both essential for catalysis (Zhu et al., 2003; Rajan et al., 2005).

Recently, two different variants of LuxS were found in two naturally occurring strains of Campylobacter jejuni. One of the LuxS variants had 100-fold higher AI-2 activity than the other; sequence analysis revealed a point mutation (G92D) in the defective strain that, when reverted by site-directed mutagenesis, fully restored AI-2 production (Plummer et al., 2011). As Asp 92 is a conserved amino acid in the LuxS protein of a wide range of bacteria, it is likely that in these organisms this substitution would lead to a loss of AI-2 production. This point mutation could represent a revertible adaptive strategy for gain or loss of function, enabling an advantageous switch between high and low AI-2 production, according to the environmental requirements.

Post-translational modification in the form of phosphorylation of Thr 14 of the LuxS from Staphylococcus aureus has been observed; the corresponding increase in enzyme activity was thought to result from indirect changes in the conformation of the active site that stabilize the enzyme–substrate interaction. Interestingly, LuxS from Bifidobacterium longum is also Ser/Thr phosphorylated, yet the particular residue(s) affected are not known (Yuan et al., 2008; Cluzel et al., 2010). Thr 14 residue is conserved across many LuxS homologues; although phosphorylation of this residue could constitute a common mechanism for rapidly modifying LuxS activity across different species, further investigation is required for this supposition to be made firm.

Beeston et al. studied the regulation of luxS at the transcription level and observed constitutive expression of a luxS promoter fusion in S. Typhimurium. Levels of AI-2 correlated with changes in activity of the pfs promoter, suggesting that fluctuations in AI-2 synthesis were driven by Pfs and the availability of substrates from the AMC pathway rather than the levels of LuxS (Beeston & Surette, 2002). In contrast, in E. coli, expression of LuxS increases with bacterial growth (Hardie et al., 2003; Wang et al., 2005a; Xavier & Bassler, 2005a; Li et al., 2006). Additionally, in this bacterium luxS induction by glucose was observed, first using a luxS transcriptional fusion (Wang et al., 2005a) and then by qRT-PCR (Li et al., 2006). Activation by glucose was abolished by supplying cAMP to the culture and a crp deletion mutant clearly showed an increased transcription of luxS when compared to the wild type. As no binding of the cAMP-CRP to the luxS promoter region was observed in a gel motility assay, it was proposed that the negative regulation of luxS by cAMP and CRP was indirect (Wang et al., 2005a). A small RNA (sRNA), cyaR, was subsequently shown to inhibit luxS expression and AI-2 production by direct binding with complementary sequences in luxS mRNA, activating its degradation. As cyaR is positively regulated by the cAMP-CRP it is thus induced under conditions of low glucose, providing an explanation for the observed increase of LuxS in the presence of this carbohydrate (De Lay & Gottesman, 2009). A second sRNA, micA, affects the length and prevalence of luxS transcripts in an RNase III-dependent manner (Udekwu, 2010) but whether this regulation affects protein amounts, activity, or explains the observed growth dependence of LuxS expression is not yet known.

There is evidence that AI-2 can also be produced by a LuxS-independent pathway, as signal activity was reported from luxS-deficient extracts supplied with adenosine (Li et al., 2006) and from an E. coli luxS mutant carrying additional mutations that alter carbon fluxes (Tavender et al., 2008). Given these results, Tavender et al. (2008) suggested that AI-2 can be formed during spontaneous conversion of ribulose-5-phosphate, although the contribution and relevance of these potential alternative LuxS-independent pathways for AI-2 production remains to be demonstrated.

AI-2 export

As AI-2 is hydrophilic in nature, with low affinity for lipid binding, it is thought to be relatively membrane-impermeable, indicating that transport across this barrier is required (Kamaraju et al., 2011). The E. coli protein, YdgG, a member of a large family of putative membrane/transporter proteins, has been proposed as a potential AI-2 exporter (Herzberg et al., 2006; Rettner & Saier, 2010). However, because extracellular accumulation of AI-2 in the ydgG mutant, under conditions where uptake of AI-2 is inhibited, is only twofold lower than in the wild type, it is clear that the role of YdgG as an exporter is mild and other mechanisms for AI-2 export must exist. Thus, it is still unclear how AI-2 exits the bacterial cell in LuxS-containing species.

AI-2 receptors

To date, two classes of AI-2-specific receptors have been well characterized (Chen et al., 2002; Miller et al., 2004). Structures have been determined for members of both the LuxP family, found in Vibrio spp., and the LsrB family, first identified in S. Typhimurium but also present in other enteric bacteria and members of the Rhizobiaceae and Bacillaceae families (Pereira et al., 2009). The AI-2-binding capability of members of the ribose binding protein (RbsB) family, i.e. those from Aggregatibacter (Actinobacillus) actinomycetemcomitans and H. influenzae, suggests that these proteins may also act as AI-2 receptors (Shao et al., 2007b; Armbruster et al., 2011). Furthermore, responses to AI-2 have been detected in bacteria lacking LuxP-, LsrB- and RbsB-like receptors, so additional alternative receptors must exist (Table 1).

LuxP AI-2 receptor

LuxP is a periplasmic binding protein from the high-affinity substrate-binding protein family typically associated with ATP-binding cassette (ABC) uptake transport systems (Boos & Lucht, 1996; Bordignon et al., 2010). Unlike canonical high-affinity substrate-binding proteins, LuxP does not interact with a transport system but modulates the activity of a membrane-spanning sensor protein, LuxQ, upon binding of AI-2. As other two-component regulatory systems, LuxPQ regulates a phosphorylation signal transduction cascade that controls the downstream AI-2 quorum sensing regulon (Fig. 3). LuxP-like receptors have only been found in the Vibrionales, such as V. harveyi, Vibrio cholerae and Vibrio anguillarum (Bassler et al., 1994; Croxatto et al., 2004; Sun et al., 2004; Neiditch et al., 2005, 2006; Rezzonico & Duffy, 2008).

Figure 3.

Vibrio harveyi quorum sensing. (a) At low cell density, phosphate fluxes from the membrane receptors, LuxN, LuxPQ and CqsS, to LuxU and then LuxO. Accumulation of phosphorylated LuxO activates the transcription of five regulatory sRNAs, Qrr1-5, which in turn destabilize the luxR mRNA and inhibit expression of this high cell density master regulator. Conversely, Qrr1-5 promote expression of AphA, the low cell density regulator, which leads to changes in the regulation of more than 100 genes. (b) At high cell density, the autoinducers CAI-1, AI-2 and 3OH4CHSL accumulate in the surrounding media and bind to their corresponding receptors, CqsS, LuxPQ, and LuxN, respectively. Ligand binding promotes the phosphatase activity of these proteins, such that phosphate flow through the pathway is reversed. The resulting unphosphorylated LuxO does not induce the transcription of Qrr1-5; LuxR is produced which regulates the quorum sensing regulon of V. harveyi, whilst inhibiting the expression of AphA, such that a high cell density phenotype is induced. Vibrio cholerae has a similar circuit, although it does not have the 3OHC4HSL-LuxN circuit, contains four Qrrs and not five, and LuxR is called HapR. Note that the V. harveyi LuxR is not homologous to the LuxR in canonical LuxI/LuxR-type circuits (Box ).

The crystal structure of LuxP from V. harveyi revealed its ligand as the S-THMF DPD isomer complexed with boron (S-THMF-borate). S-THMF-borate is negatively charged: interaction with LuxP is stabilized through two positively charged arginine residues (Arg 215 and Arg 310) in the binding pocket. Ligand binding is promoted by further hydrogen bonds between AI-2 and Gln 77, Ser 79, Trp 82, Thr 266, and Asn 159 (Fig. 2b) (Chen et al., 2002; Miller et al., 2004).

Structures of LuxPQ complexes in the presence or absence of ligand have also been determined and a model to describe how ligand binding regulates phosphate flow through the phosphorelay cascade controlling quorum sensing-dependent genes was proposed (Neiditch et al., 2005, 2006). The relevance of these studies goes well beyond understanding signal transduction in quorum sensing (reviewed in Federle, 2009). Two-component systems such as the LuxPQ are widely used by bacteria to sense and respond to environmental stimuli. Many of these systems are well studied but in most cases the actual signals are not known (reviewed in Casino et al., 2010; Krell et al., 2010). Hence, structural studies on LuxPQ have served as an excellent model to comprehend, at the molecular level, how bacteria use two-component and phosphorelay systems for environmental adaptation.

LsrB AI-2 receptors

LsrB is also a high-affinity substrate-binding periplasmic protein. It interacts with the membrane components of an ABC transport system called Lsr (for LuxS regulated), which is involved in AI-2 uptake (Taga et al., 2001, 2003; Xavier & Bassler, 2005a; Pereira et al., 2008, 2009). Crystal structures of LsrB-ligand complexes have been solved for S. Typhimurium, Sinorhizobium meliloti and Yersinia pestis (Miller et al., 2004; Pereira et al., 2008; Kavanaugh et al., 2011); in all cases, receptors bind the R-THMF adduct of DPD. Sequence similarity between the three crystallized LsrB-like proteins is striking, with percentage identity between that of S. Typhimurium and that of Y. pestis, or S. meliloti, at 84% and 72%, respectively (Pereira et al., 2009).

A combination of biochemical, genetic and bioinformatic approaches identified further functional LsrB-like AI-2 receptors in the phylogenetically distinct families of the Enterobacteriaceae, the Rhizobiaceae, and the Bacillaceae (Pereira et al., 2009). This study revealed two distinct groups of potential orthologues to the S. Typhimurium LsrB protein: one with a high percentage identity (> 60%), and a second with identity below 36%. Strikingly, the six residues previously shown to form hydrogen bonds with the AI-2 ligand in the S. Typhimurium LsrB structure (Lys 35, Asp 116, Asp 166, Gln 167, Pro 220 and Ala 222, Fig. 2) were completely conserved in all the organisms belonging to the first group. These residues were predicted to be essential for ligand binding. Interestingly, all proteins of the second group diverged at residues Asp 166 and Asp 222, which were shown experimentally to be essential for AI-2 binding, suggesting that these proteins are not functional LsrB receptors. These studies provided criteria by which new LsrB-like receptors can be identified in the future (Pereira et al., 2009).

While the overall folding of LuxP and LsrB is noticeably similar (Fig. 2), the same cannot be said at the sequence level: identity is only 11%. This perhaps explains why attempts to discover novel classes of AI-2 receptors with either LuxP or LsrB as a reference have proven unsuccessful (Sun et al., 2004; Rezzonico & Duffy, 2008). As detection mechanisms must exist in all those bacteria with the ability to respond to AI-2 but have no characterized receptor, in light of the above, it is unlikely that the unidentified sensor proteins involved will be evolutionarily related to LuxP or LsrB.

There is a marked difference between the polarity of the LuxP and LsrB binding sites which is thought to be a major factor in constraining which AI-2 form is bound. Boron is essential for the AI-2 response in V. harveyi, whereas it inhibits the signalling elicited by S. Typhimurium (Chen et al., 2002; Miller et al., 2004). Miller et al. showed that high concentrations of boron in solution, like those found in marine environments, can shift the equilibrium between AI-2 isomers towards the S-THMF-borate, allowing induction of bioluminescence by V. harveyi. Conversely, boron depletion favours the accumulation of R-THMF and activation of the S. Typhimurium AI-2 system (Meijler et al., 2004; Miller et al., 2004). As the equilibrium state of AI-2 molecules reflects chemical environmental conditions, it could enable bacteria to detect both their biotic and abiotic context. For example, V. cholerae inhabits two distinct niches: the human intestine and the aquatic marine environment. The different concentration of boron in these two environments could enable V. cholerae to detect which environment it is in, whilst monitoring population density, via AI-2, and induce phenotypic changes accordingly (Hammer & Bassler, 2007; Federle, 2009).

RbsB receptor

The RbsB proteins of both A. actinomycetemcomitans and H. influenzae strain 86-028NP have been proposed as AI-2 receptors. These proteins have high homology, over 70% identity, with the periplasmic-binding component of the ribose ABC transporter (RbsB) of E. coli. Using a combination of luxS, rbsB and lsrB mutant strains of A. actinomycetemcomitans, along with signal complementation assays, it was shown that RbsB interacts with AI-2 and plays a role in the resulting phenotypic responses of this organism (James et al., 2006; Shao et al., 2007ab). Supporting the indication that LsrB and RbsB could both be involved in AI-2 recognition by A. actinomycetemcomitans, both proteins compete for AI-2 binding in vitro. Although H. influenzae isolate 86-028NP does not possess orthologues of the proteins of the Lsr system, it has a protein with high homology to the A. actinomycetemcomitans RbsB (86% percentage of identity). rbsB expression is upregulated by exogenously supplied AI-2 in H. influenzae strain 86-028NP and inactivation of RbsB reduced AI-2 uptake and impaired biofilm formation. This defect was similar to that observed in a luxS mutant, indicating that RbsB might also act as an AI-2 receptor in H. influenzae strain 86-028NP (Armbruster et al., 2011).

The next step towards a better understanding of the role of RbsB in AI-2 detection will be to determine its structure in complex with the signal, which could reveal whether RbsB binds the same AI-2 isomer as LsrB or an alternative rearrangement. Structural studies could also reveal the amino acid residues involved in AI-2 binding and open doors through novel bioinformatic studies to identify new RbsB-like AI-2 sensing proteins and clarify whether these proteins constitute a third class of AI-2 receptors.

Putting words into action: phenotypic responses to AI-2 signalling

LuxS/AI-2-dependent phenotypes have been studied across diverse bacteria. Due to the resulting wealth of literature, this section will focus first on studies describing AI-2 signalling in species containing LuxP- or LsrB-like receptors, then discuss those in which it was clearly demonstrated that AI-2 was acting as a signalling molecule even if the mechanisms of sensing and signal transduction are not yet known.

The role of AI-2 signalling in Vibrio spp

Several members of the Vibrio genus contain parallel quorum sensing systems: that involving AI-2 and LuxPQ; the CAI-1 and CqsS system discovered in V. cholerae and since identified in other Vibrio spp.; and the V. harveyi-specific AHL, [N-3-hydroxybutanoyl homoserine lactone (3OHC4HSL)], and LuxN system. Signalling downstream of all three converges upon a common phosphorelay cascade and response regulators (reviewed in Ng & Bassler, 2009). As a result, a synergistic role in the regulation of group behaviours has been attributed to AI-2 signalling, the precise mechanisms and phenotypic effects of which vary from species to species. Although this synergism makes it difficult to determine the particular contributions of a single autoinducer to the overall response, experiments with mutant V. harveyi, where signalling through the AHL-LuxN and CAI-1-CqsS systems had been abolished, showed that luciferase was still induced with increasing cell density. This response was AI-2-dependent, providing evidence of the individual contributions of this molecule (Bassler et al., 1994; Henke & Bassler, 2004b).

Signal transduction is most characterized in V. harveyi and V. cholerae (Fig. 3). LuxP, which is situated in the periplasm, binds borated-AI-2 and the resulting complex interacts with the cognate membrane histidine kinase, LuxQ. As the concentration of AI-2 changes, the enzymatic activity of this transmembrane protein shifts from kinase to phosphatase. Both the flux of phosphate through the downstream pathway and the activity of the quorum sensing master regulators, LuxR (V. harveyi), HapR (V. cholerae) and AphA (both species), are consequently altered (Bassler et al., 1994; Freeman & Bassler, 1999; Miller et al., 2002; Neiditch et al., 2005, 2006). When cell density and AI-2 concentrations are low, LuxQ acts as a histidine kinase (Fig. 3a). The cytoplasmic phosphotransferase protein, LuxU, is phosphorylated, then transferring the phosphate to a response regulator, LuxO. Phosphorylated LuxO in turn activates the transcription of multiple sRNAs named Qrr. These sRNAs interact with the chaperone Hfq and destabilize luxR, or hapR, mRNA in V. harveyi and V. cholerae, respectively (Lenz et al., 2004; Tu & Bassler, 2007; Svenningsen et al., 2009). Under these conditions, LuxR and HapR expression is prevented, whilst that of the low cell density regulator, AphA, is promoted (Rutherford et al., 2011). Conversely, when cell density and autoinducer availability increase, binding of the ligand to its cognate receptor switches LuxQ activity from that of kinase to phosphatase (Fig. 3b). The flow of phosphate through the pathway is reversed, leaving LuxO in an unphosphorylated state. LuxR and HapR proteins accumulate, whilst expression of AphA is inhibited (reviewed in Ng & Bassler, 2009). This promotes a shift between low and high cell density states; to further ensure the precise timing and rapid nature of this transition, Qrr and LuxO expression are also regulated by negative feedback loops (Tu & Bassler, 2007; Tu et al., 2008, 2010; Ng & Bassler, 2009).

The presence of distinct quorum sensing systems with common transduction pathways raises the question as to how and if the different autoinducers can be distinguished by and relate to distinct phenotypic outputs in V. harveyi and V. cholerae. Several studies have shown that various amounts and combinations of the three autoinducers trigger different phenotypic responses. This occurs due to differences in signal strength impact on LuxR protein production and promoter affinity. Given the different sources of these three autoinducers, it is possible that these microorganisms are able to monitor both numbers in and species composition of the surrounding community, modulating behaviours accordingly (Mok et al., 2003; Henke & Bassler, 2004b; Waters & Bassler, 2006; Long et al., 2009).

Extensive AI-2 quorum sensing regulons exist in these species: induction of LuxR in V. harveyi regulates the expression of nearly 100 genes. Phenotypes affected include the induction of bioluminescence, metalloprotease production, inhibition of siderophore production, type III secretion, and changes in colony morphology linked to changes in the extracellular polysaccharide matrix (Bassler et al., 1993, 1994; Lilley & Bassler, 2000; Mok et al., 2003; Henke & Bassler, 2004ab; Waters & Bassler, 2006). In V. cholerae, activation of HapR increases expression of the haemagglutinin protease and decreases biofilm formation, expression of virulence factors such as cholera toxin and the toxin co-regulated pilus (Jobling & Holmes, 1997; Miller et al., 2002; Zhu et al., 2002; Hammer & Bassler, 2003). AI-2 signalling also contributes to the induction of competence and has been studied in the context of mixed species biofilms, suggesting that AI-2 could promote interspecies lateral gene transfer (Antonova & Hammer, 2011).

The components of the LuxPQ AI-2 signal transduction pathway have homologues in other Vibrio spp. such as Vibrio parahaemolyticus, Vibrio vulnificus and V. anguillarum. Their quorum sensing systems have so far been shown to function in a similar fashion to those described above (Croxatto et al., 2004; Henke & Bassler, 2004a; Yildiz & Visick, 2009; Shao et al., 2011).

The role of AI-2 in bacteria with the Lsr system

One of the paradigms in quorum sensing is that autoinducers accumulate extracellularly with increasing population density. Bacteria with an LsrB-like receptor seem to contradict this paradigm: although AI-2 accumulates in the extracellular medium with early growth, once a certain cell density is reached these levels are depleted by import via the Lsr transport system. This transporter comprises a membrane channel formed by two transmembrane proteins, LsrC and LsrD, and an ATPase, LsrA, which provides the energy to AI-2 signal transport (Fig. 4) (Taga et al., 2001, 2003; Xavier & Bassler, 2005a).

Figure 4.

Current model for AI-2 detection, uptake and signalling via the Lsr system of Escherichia coli. (a) During early bacterial growth, little AI-2 is extracellular. LsrR repression is active and expression of the Lsr system is inhibited. (b) As growth ensues and extracellular AI-2 accumulates, initial uptake occurs and imported AI-2 is phosphorylated by LsrK. P-AI-2 binds LsrR, leading to de-repression of the lsr operon, lsrR and lsrK. (c) Lsr-dependent transport starts; AI-2 binds to the periplasmic binding protein, LsrB, and is internalized by the Lsr transporter formed by the two transmembrane proteins, LsrC and LsrD, and the ATP-binding protein LsrA. More P-AI-2 is produced and at high cell densities expression of the transporter is further induced. AI-2 import increases and extracellular levels of AI-2 are rapidly depleted as a result of this positive feedback loop. LsrG and LsrF degrade P-AI-2. It was recently proposed that the initial AI-2 uptake occurs via a PTS-dependent transport, but the permeases involved in this process are still unknown.

Although homologues of the Lsr transporter have been identified in many bacterial species, including Yersinia spp., Klebsiella pneumoniae, and even in Gram-positive Bacillus spp. (Pereira et al., 2009), the system is best characterized in S. Typhimurium, E. coli and S. meliloti (Taga et al., 2001; Wang et al., 2005a; Xavier & Bassler, 2005a; Pereira et al., 2008). Following internalization, AI-2 is phosphorylated by the kinase LsrK to produce phospho-AI-2 (P-AI-2). Genes encoding the Lsr transporter are in an operon regulated by the cAMP-CRP complex and the Lsr repressor (LsrR) (Taga et al., 2003; Wang et al., 2005a; Xavier & Bassler, 2005a). In the absence of P-AI-2, LsrR represses the transcription of the lsr operon. LsrR also represses the lsrRK operon, thus regulating its own expression and that of LsrK (Wang et al., 2005b). When P-AI-2 accumulates in the cell, it binds to LsrR (Xavier et al., 2007; Xue et al., 2009) and both operons are de-repressed. A positive feedback loop results, whereby uptake and phosphorylation of AI-2 promote expression of the transporter, which in turn drives more signal uptake and further induction of lsr. AI-2 is rapidly depleted from the extracellular medium (Fig. 4). Thus AI-2 induces its own internalization, phosphorylation and depletion. Downstream processing and eventual termination of the response involves two more enzymes encoded within the lsr operon, LsrF and LsrG (Taga et al., 2003; Xavier et al., 2007; Diaz et al., 2009; Marques et al., 2011). LsrG catalyses the isomerization of P-AI-2 to 3,4,4-trihydroxy-2-pentanone-5-phosphate (P-TPO): in its absence, bacterial cells accumulate P-AI-2 (Xavier et al., 2007; Marques et al., 2011). Based on structural and sequence similarities, LsrF is predicted to be an aldolase; although it was shown that an lsrF mutant accumulates P-AI-2, its function is yet to be confirmed (Taga et al., 2003; Diaz et al., 2009). In Salmonella, another gene, lsrE, is present in the operon but its function is also unknown (Taga et al., 2003).

Recent data from E. coli has added another piece to this puzzle, demonstrating that the phosphoenolpyruvate phosphotransferase system (PTS) is required for Lsr activation and AI-2 internalization (Pereira et al., 2012). The PTS catalyses transport across the periplasmic membrane of a large range of compounds concomitant with their phosphorylation (Postma et al., 1993). As PTS is essential for activation of lsr transcription but AI-2 internalization remains dependent upon LsrK- and P-AI-2-mediated alleviation of LsrR repression, Pereira et al. (2012) propose that initial internalization of AI-2 by E. coli is via a PTS-dependent mechanism. This would provide the first pool of AI-2, which is phosphorylated by basal levels of LsrK, relieves lsr repression and initiates Lsr-dependent AI-2 depletion from the extracellular environment (Fig. 4). Importantly, Lsr activation is inhibited in the presence of a constitutive non-phosphorylated form of PTS and thus AI-2 internalization is inherently associated to the phosphorylated levels of PTS. Given that the phosphorylation state of the PTS reflects the availability of certain substrates, and also the global metabolic status of the cell (Deutscher et al., 2006; Lazazzera, 2010), this could represent a mechanism through which bacteria hierarchically integrate quorum sensing with metabolic signals such that particular behaviours are only triggered in the appropriate context (Hardie et al., 2003; Pereira et al., 2012). Due to the conserved nature of the PTS, this regulatory mechanism is likely to be conserved in other Lsr-containing bacteria, supported by the repression of Lsr expression in a S. Typhimurium mutant strain lacking a fully functional PTS (Taga et al., 2003).

The physiological function of the Lsr system is not yet clear: why do these bacteria produce and export AI-2 to then incorporate and process it? It does not appear to be a carbon source, as neither S. Typhimurium nor S. meliloti are able to grow when AI-2 is used as the sole carbon source (Taga et al., 2001; Winzer et al., 2002a; Pereira et al., 2008). Bentley et al. propose a conventional quorum sensing role for the AI-2/Lsr system, where AI-2 (or P-AI-2) would regulate genes and functions other than those necessary for its internalization and processing. Signal processing would terminate this response (DeLisa et al., 2001; Ren et al., 2004b; Li et al., 2007). As the Lsr system imports self and non-self AI-2, the presence of lsr-containing bacteria, which deplete extracellular AI-2, could also impact upon the behaviours of neighbouring bacteria relying upon this molecule for signalling proposes. Thus another putative function for the Lsr is interference in AI-2-dependent quorum sensing (Xavier & Bassler, 2005b).

Supporting the hypothesis of Bentley et al. AI-2-dependent quorum sensing has been implicated in the induction of several important behaviours in E. coli. Addition of AI-2 to wild-type E. coli K12 stimulated biofilm formation and motility. This behaviour was dependent upon the presence of a functional LsrK enzyme, indicating that AI-2 signalling through the Lsr system was responsible for these phenotypes (Gonzalez Barrios et al., 2006). Further analysis of lsrR and lsrK mutant strains using microarrays increased the evidence for Lsr-dependent regulation of biofilm formation and motility (Li et al., 2007), although investigation into the phenotypes of luxS and lsrB mutant strains is required to fully confirm these conclusions.

AI-2 also appears to act as a chemoattractant for E. coli K12, a behaviour dependent upon both the L-serine receptor (Tsr) and LsrB (Hegde et al., 2011). As chemotaxis towards AI-2 was unaffected in the absence of LsrC, uptake of the molecule is seemingly dispensable for this process. Analogous to the maltose-binding protein and Tar interaction (Zhang et al., 1999), it is hypothesized that LsrB binding to AI-2 in the periplasm enables interaction with the periplasmic domain of Tsr; the resulting signalling elicits downstream chemotactic responses (Hegde et al., 2011). Concentration-dependent chemoattraction of enterohaemorrhagic E. coli (EHEC) by AI-2 has also been observed through the use of luxS mutant bacteria (Bansal et al., 2008). In this organism AI-2 also regulates pathogen motility and attachment to HeLa cells (Bansal et al., 2008). Similarly, luxS mutant enteropathogenic E. coli (EPEC) exhibit reduced motility compared to wild-type bacteria when in the presence of epithelial cells (Giron et al., 2002). These mutant bacteria give rise to reduced clinical illness in rabbits, associated with reduced adherence to the intestinal mucosa, demonstrating that AI-2 signalling may play a role in colonization and pathogenesis by EPEC (Zhu et al., 2007).

Overall, these studies show that in E. coli, luxS gene expression and AI-2 signalling are important for proper regulation of biofilm formation, motility, attachment to epithelial cells, and other crucial virulence traits.

Salmonella Typhimurium luxS mutants are impaired in several behaviours including biofilm formation, swarming and swimming, as shown by proteomic, transcriptomic and phenotypic analysis (Prouty et al., 2002; De Keersmaecker et al., 2005; Karavolos et al., 2008; Soni et al., 2008; Kint et al., 2009; Jesudhasan et al., 2010). LuxS-dependent regulation of flagellar phase variation could not be complemented with cell-free supernatants (Karavolos et al., 2008), and in many of the above cases, the ability of pure AI-2 to rescue mutant phenotypes was not reported (Soni et al., 2008; Kint et al., 2009; Jesudhasan et al., 2010). Curiously, although defects in biofilm formation were not rescued by exogenously supplied signal or constitutively expressed luxS, the introduction of luxS driven by its own promoter in a plasmid successfully restored biofilm formation (De Keersmaecker et al., 2005). This indicates that regulatory elements in the promoter region of luxS are important for biofilm formation, and that concentration and timing of AI-2 production (and therefore complementation with AI-2) are likely to be crucial for correct induction of signalling. This may partly explain the existence of contradictory studies regarding the role of AI-2 but does not account for the discrepancy between certain reports of S. Typhimurium virulence. While Perrett and colleagues could not identify a role for AI-2/luxS in this bacterium (Perrett et al., 2009), Choi et al. showed luxS mutant phenotypes such as defects in the induction of virulence genes and invasion of epithelial cells. These defects were restored by supplying AI-2 to the cultures and rescued in a luxS lsrR double mutation establishing a link between AI-2 regulation of pathogenicity and the Lsr system (Choi et al., 2007, 2012). The same group also showed that the luxS mutant is impaired in virulence in the mouse model of infection (Choi et al., 2007, 2012).

Few phenotypes have been conclusively attributed to AI-2/LsrB-dependent signalling in either E. coli or Salmonella aside from chemotaxis and signal removal from the extracellular environment. As a result, the possibility that the Lsr system might potentiate interference in the AI-2-regulated behaviours of neighbouring, non-self bacteria remains a reasonable hypothesis. Escherichia coli can demonstrably interfere with the AI-2-controlled behaviours of other bacterial species in an Lsr-dependent manner (Xavier & Bassler, 2005b). Furthermore, S. meliloti has no AI-2-producing capabilities, but AI-2 produced by others induces its Lsr expression and function. Like E. coli, S. meliloti can import and process signal produced by other bacteria, as demonstrated by co-culture with Erwinia carotovora (Pereira et al., 2008). Interference with AI-2 is proposed to confer competitive advantages upon Lsr-containing bacteria in polymicrobial communities through the disruption of signalling in nearby bacteria (Pereira et al., 2008).

LuxS and AI-2-dependent phenotypes have also been studied in other bacteria possessing Lsr-like receptors, Photorhabdus luminescens, for example. This organism has two distinct life cycle phases: as a symbiont in the gut of a nematode, and as a pathogen, when the bacterium, in association with the nematode, infects and kills a wide variety of insect larvae. Studies with this bacterium showed that luxS plays a role in the regulation of carbapenem production, a broad-spectrum antibiotic thought to prevent contamination of the insect by other microorganisms (Derzelle et al., 2002). Comparison of the expression profiles of wild-type and luxS mutant bacteria grown in the presence of enzymatically synthesized AI-2 showed that this molecule regulates more than 100 genes in P. luminescens, including the lsr operon. Amongst them are genes encoding outer membrane-associated proteins such as flagella and pili, those involved in virulence, and those that promote resistance to oxidative stress. Although restoration of mutant phenotypes was not confirmed, addition of AI-2 rescued the regulation of gene expression, indicating that phenotypes important for the early steps of insect invasion are AI-2- and Lsr-regulated (Krin et al., 2006).

Deletion of luxS in A. actinomycetemcomitans, an inhabitant of the oral commensal flora associated with severe oral cavity infections, reveals defects in leukotoxin production and growth under iron-limiting conditions, due to altered expression of several genes involved in iron transport and storage (Fong et al., 2001, 2003; James et al., 2006; Shao et al., 2007ab). Defects in biofilm formation were rescued by addition of partially purified AI-2 to luxS mutant strains, whereas expression of the enzyme SahH (which presumably bypasses the metabolic need of LuxS; Fig. 1) could not revert the luxS mutant phenotype. These results strongly suggest that LuxS function in this organism is signal-based rather than metabolic, strengthened by the observation that deletions in lsrB or the second AI-2 receptor gene of A. actinomycetemcomitans, rbsB, also lead to defects in biofilm formation (Shao et al., 2007a). Simultaneous inactivation of both receptors reduces biofilm formation to that seen in a luxS mutant strain. In summary, these two receptors contribute additively to the elicitation of AI-2-dependent responses necessary for biofilm assembly in A. actinomycetemcomitans (James et al., 2006; Shao et al., 2007ab).

AI-2 signalling and pathogenesis in other Gram-negative bacteria

AI-2 and luxS-dependent changes in bacterial behaviour, including those directly or indirectly associated with virulence, have been demonstrated for several species of Gram-negative bacteria in which the signal transduction machinery remains unknown. For example, in H. pylori, a gastric pathogen associated with stomach ulcers and cancer, biofilm formation is downregulated in a luxS-dependent manner. Conversely, transcription of flagellar genes and motility are upregulated in an AI-2-dependent manner (Cole et al., 2004; Rader et al., 2007; Shen et al., 2010). This molecule is also a chemorepellant for H. pylori, acting through the chemoreceptor TlpB. The effects of this can be seen clearly in the behaviour of luxS mutant bacteria where swimming is characterized by a reduced frequency of ‘stops’ between ‘runs’ (Rader et al., 2011). Importantly, motility, chemotaxis and LuxS are all required by H. pylori for full virulence in gerbils (Ottemann & Lowenthal, 2002; McGee et al., 2005; Osaki et al., 2006). However, a direct link between AI-2-dependent motility and virulence in vivo is yet to be shown.

In contrast to canonical LuxS enzymes, LuxS of H. pylori is not involved in the AMC but is necessary for cysteine production (Doherty et al., 2010). Motility defects in luxS mutant bacteria can, nevertheless, be attributed to defective AI-2 signalling, as exogenously added AI-2, but not cysteine (Doherty et al., 2010), complemented these impairments. This makes H. pylori a good candidate in which to explore the link between AI-2 signalling and virulence in vivo and offers potential for the development of AI-2 sequestering agents as novel therapeutics for the control of H. pylori-associated diseases.

The role of LuxS in B. burgdorferi, the causative agent of Lyme disease, is also slightly unusual. This organism does not have a full AMC as it does not have the enzymes to convert homocysteine (the other product of the LuxS enzyme) into methionine (Fig. 1). Thus, B. burgdorferi could be an organism in which LuxS and AI-2 signalling function can be analysed with fewer metabolic consequences (Babb et al., 2005; von Lackum et al., 2006; Riley et al., 2007). Addition of AI-2 to wild-type cultures of B. burgdorferi suggests a contributory role in the virulence of this organism due to changes in the expression of multiple proteins, including the outer surface lipoprotein, VlsE (Stevenson & Babb, 2002; Stevenson et al., 2003; von Lackum et al., 2006). This factor is involved in antigenic variation and thus immune evasion and the host response to infection (Eicken et al., 2002). Despite this potential link between AI-2 and virulence, luxS is dispensable for the natural colonization of tick vectors by B. burgdorferi and subsequent transmission to and infection of mice (Hubner et al., 2003; Blevins et al., 2004).

Deletion of luxS in Actinobacillus pleuropneumoniae, the causative agent of porcine pleuropneumonia, impairs its capabilities for biofilm formation, cell adherence, growth in iron-restricted conditions, and virulence in mice. It is, however, difficult to distinguish whether LuxS is functioning predominantly in signalling or metabolism in this species: chemical complementation rescues only the impairments in cell adherence and growth in low iron of luxS mutant bacteria but not biofilm formation. Attempts to transcomplement luxS mutant bacteria with wild-type AI-2-producing bacteria proved unsuccessful, even in adherence assays where cells are in close proximity; however, addition of AI-2 to wild-type cultures did increase their propensity to form biofilms (Li et al., 2011). These apparent inconsistencies can be attributed to the fact that, again, finely regulated levels of AI-2 are likely required for proper function of the system, which are simply not provided upon co-culture with wild-type bacteria.

AI-2 signalling and pathogenesis in Gram-positive bacteria

Despite the relative paucity of knowledge surrounding AI-2 detection and signal transduction in Gram-positive bacteria, responses by such organisms to the presence (or absence) of this molecule have been described. LuxS and AI-2 signalling are required for wild-type levels of resistance to antibiotics such as ampicillin by both Streptococcus anginosus and Streptococcus intermedius (Ahmed et al., 2007, 2009); biofilm production and haemolytic activity are also induced by AI-2 in the latter species (Ahmed et al., 2008, 2009; Pecharki et al., 2008). In S. pneumoniae, LuxS and AI-2 have also been shown to control biofilm formation, possibly through the regulation of the lytA, an autolysin previously implicated in biofilm assembly (Vidal et al., 2011).

Differing reports exist as to the function of AI-2 in S. aureus. Doherty and colleagues propose that the predominant role for LuxS in this species is metabolic, not signalling, for two reasons: chemical complementation did not restore the growth defect of luxS mutant bacteria in sulphur-limited conditions, not did luxS gene inactivation alter any of the specific virulence-associated traits analysed (Doherty et al., 2006). In contrast, Zhao et al. showed that expression of the cap gene cluster, involved in synthesis of capsular polysaccharide, and the two-component system genes kdpDE, were downregulated by LuxS and AI-2 in a dose-dependent manner. Additionally, it was observed that in S. aureus luxS mutant survival and growth were higher in human whole blood and monocytic cells and these phenotypes were restored to wild-type levels by AI-2 complementation (Zhao et al., 2010). Given that the cap system is upregulated during invasive S. aureus infections (Thakker et al., 1998; Voyich et al., 2005) and KdpDE upregulates cap gene expression by direct binding of KdpE to the promoter region of the cap operon, it was suggested that AI-2 inhibits capsule-associated virulence and this inhibition is via KdpDE (Zhao et al., 2010).

AI-2 regulates the transcription of many genes, including those encoding acetoin dehydrogenase, gluconokinase, LrgB, nitrite extrusion protein and a fructose PTS subunit in a second Staphylococcal species, Staphylococcus epidermidis. This opportunistic skin commensal is able to cause severe infections in the immunocompromised; analysis of culture filtrates also suggested that key virulence factors, the phenol-soluble modulin peptides, are under AI-2 control (Li et al., 2008).

AI-2 signalling in polyspecies communities

In many studies, monospecies experimental set-ups have been used to determine whether LuxS- and AI-2-dependent signalling regulate bacterial behaviour. Although these simplified systems are less variable and facilitate the analysis of individual players, bacteria rarely exist in such isolated contexts in nature. It may only be in polyspecies communities that the right context for proper induction of AI-2 signalling, and observation of the downstream effects or their benefits, is provided.

One of the first studies to address this issue made use of simple co-culture systems to demonstrate that AI-2 produced by one species can influence gene expression in others: signal produced by E. coli induced bioluminescence by V. harveyi, and reciprocally, V. harveyi induced the AI-2-regulated Lsr system of E. coli (Xavier & Bassler, 2005b). This proof of principle was given greater strength by the observation that E. coli also induced haemagglutinin protease production by V. cholerae; both species are potential inhabitants of the mammalian gut and so could encounter each other in a physiological context. As haemagglutinin protease facilitates V. cholerae detachment from the intestinal mucosa when V. cholerae cells reach high density, and is thought to be important for bacterial dissemination (Silva et al., 2006), the presence of E. coli could thus influence the dynamics and spread of cholera disease through changes in the availability of environmental AI-2.

AI-2 in members of the nasopharyngeal microflora

Similar approaches analysing the interplay between two species from a particular niche have subsequently been adopted to investigate interspecies signalling in medically relevant polymicrobial communities such as the nasopharygeal flora. Two members, H. influenzae and Moraxella catarrhalis, are frequently isolated from patients suffering with a common paediatric disease known as otitis media (OM). As chronic OM is associated with the formation of multispecies biofilms and increased antibiotic resistance (Maddocks & May, 1969; Hol et al., 1994), it is notable that these characteristics, as well as resistance to host clearance, are increased in M. catarrhalis when in the presence of H. influenzae (Armbruster et al., 2009, 2010). Although M. catarrhalis lacks the luxS gene and therefore does not produce AI-2, biofilm biomass and, consequently, resistance to antibiotic treatment, are increased upon addition of this molecule. Interestingly, increased loads of M. catarrhalis are recovered from rodents co-infected with AI-2-producing wild-type H. influenzae when compared with those obtained upon co-infection with the luxS mutant (Armbruster et al., 2010, 2011). These data indicate that through the provision of AI-2, H. influenzae promotes M. catarrhalis biofim formation and persistence in vivo during OM infections.

AI-2 in members of the oral cavity microflora

AI-2 signalling also promotes mutualistic biofilm formation between several species inhabiting the oral cavity. In Streptococcus gordonii, AI-2 restored the luxS mutant distinct biofilms architecture to that of the wild type (Saenz et al., 2012). This microorganism forms mutualistic biofilms with the periodontal pathogen Porphyromonas gingivalis; these were impaired when both organisms lacked the luxS gene but were restored when at least one of these organisms had the wild-type luxS gene (McNab et al., 2003). This same gene in Streptococcus oralis is likewise required for formation of biofilms with Actinomyces naeslundii, a luxS-deficient human commensal. Decreased biofilm formation between luxS mutant S. oralis and A. naeslundii could only be fully rescued by chemical complementation with a particular concentration, 0.8 nM, of AI-2 (Rickard et al., 2006). The mutualistic biofilm established between S. oralis and S. gordonii, also requires a similar AI-2 concentration (1 nM) to complement the dual-species luxS-luxS mutant biofilm (Saenz et al., 2012). These studies illustrate both the crucial role and, again, the exquisite sensitivity of interspecies AI-2 signalling in the development of mutualism between species of the oral microflora (Rickard et al., 2006; Saenz et al., 2012).

AI-2 in members of the oropharyngeal flora

Another species which, despite an inability to produce the molecule, responds to AI-2 produced by neighbouring bacteria is Pseudomonas aeruginosa. This major pathogen is a member of the oropharyngeal flora (OF) able to cause mortal pulmonary failure in patients with cystic fibrosis. The expression of a number of important virulence-associated proteins encoded by the genes rhlA (rhamnolipid biosynthesis), lasB (elastase), exoT (exotoxin), phzA1 and phzA2 (phenazine synthesis), and fliC (flagellar component) is upregulated in response to AI-2. These same genes were upregulated when P. aeruginosa was cocultured with non-pathogenic OF. Co-infection with this avirulent microflora also exacerbated the lung injury caused by P. aeruginosa in a rat model of infection. Interestingly, detectable levels of AI-2 were found in sputum samples recovered from cystic fibrosis patients, leading the authors to speculate that interspecies communication between the OF and P. aeruginosa, through AI-2, might promote the expression of virulence factors and pathogenesis by the latter species (Duan et al., 2003). This is one of the few published studies to explore the contribution of interspecies communication to bacterial pathogenicity in the context of the host environment. More investigation is required to fully establish whether a causal relationship between AI-2 production or signalling by the OF and virulence factor expression and pathogenesis by P. aeruginosa exists, and what the importance of this signalling molecule is in this natural polyspecies community.

Deciphering the message: the complexities of phenotypic analysis

An increasing number of tools with which to investigate AI-2 function at the species and polyspecies level are available; however, in many cases it remains unclear whether phenotypes observed in the absence of LuxS are a consequence of interruption of the AMC, signal disruption, or even a combination of both. The chemical properties, the potential different sensing mechanisms, integration with other quorum sensing pathways, environmental factors and cellular metabolism all add levels to the challenge of understanding the function of AI-2. Although successful chemical complementation of luxS mutant phenotypes with synthesized AI-2 potentially provides strong evidence that AI-2 is acting as a signal, the innate complexity of this system means that a lack of success does not conclusively prove the opposite. The studies outlined above highlight that timing of signal production, concentration, requirement for chemical modifications or additional environmental factors, and the percentage of the population responding to the signal can all contribute to the elicitation of productive signalling. These elements are often difficult to identify and appropriately reproduce in vitro, potentially explaining why the addition of AI-2 may have failed to restore behaviours to wild-type levels in some cases. Supplementation of wild-type bacteria with AI-2 circumvented this problem for A. pleuropneumoniae (Li et al., 2011). The physiological relevance of exposing cells to higher concentrations of AI-2 than they themselves produce is debatable; however, this situation could arise naturally in niches containing multiple AI-2-producing species, arguing the case for the use of this approach. The possibility that for certain bacteria and growth conditions, any potential AI-2 signalling complementation taking place could be masked by more severe metabolic defects should also be borne in mind. Expression of enzymes that prevent accumulation of intermediates resulting from luxS mutations, exemplified in the use of SahH to restore the methyl cycle in A. actinomycetemcomitans (Shao et al., 2007a), could solve this problem, although not without measurement of intracellular metabolites to ensure that this complementation is indeed working. The existence of multiple, convergent signalling pathways, as in Vibrio spp. may also mask the effects of luxS mutation and AI-2 signalling, especially as phenotypic assays are frequently qualitative and can disregard slight changes.

Recently, in many studies showing that deletions in luxS affected metabolism, the possibility that AI-2 is a signal has consequently been discarded. However, this conclusion is not necessarily appropriate when addition of AI-2 induces metabolic changes unrelated to its synthesis or processing. In light of the recent data associating the function of the Lsr in E. coli with the PTS (Pereira et al., 2012), the activity of which is intrinsically linked with cellular metabolic status, the possibility that AI-2 might regulate metabolic shifts should be considered.

Undoubtedly, identification of AI-2 receptors and the construction of corresponding mutant strains can provide robust evidence that potential AI-2-regulated behaviours are indeed responding to this signal. Progress has been made in the identification of receptors, many of which could now be tested for their involvement in the AI-2 responses described here (Table 1). On the other hand, in other species, even where AI-2 responses have been observed, the receptors remain a mystery. Janda and colleagues have recently developed a chemical probe that might lead to the identification of novel AI-2 receptors (Garner et al., 2011). This type of novel approach, in conjunction with established methods such as genetic screens and the use of bioinformatics, should speed developments in this area, and hopefully provide tools to clarify the role of AI-2 in individual organisms.

Interrupting/manipulating the conversation: future possibilities for exploiting AI-2 signalling

The amazing ability of bacteria to evolve and adapt, and their propensity for horizontal gene transfer, has led to the rapid acquisition and spread of multidrug resistance and the re-emergence of infectious diseases as one of the top five causes of human mortality worldwide. New therapeutic approaches are urgently required to combat this problem and much debate has surrounded bacterial quorum sensing as a potential target (Njoroge & Sperandio, 2009; Roy et al., 2011; Zhu & Li, 2012). These therapies would target the induction of group behaviours and could have a major impact on the control of bacterial communities deeply embedded in multispecies biofilms, which are frequently a major problem in chronic diseases such as that associated with cystic fibrosis or OM (Duan et al., 2003; Armbruster et al., 2010). From this perspective, AI-2 is of particular economic interest, as a single therapeutic could be developed with broad effects upon a multitude of species or diseases.

Multidisciplinary approaches and techniques are helping to target AI-2-dependent behaviours. The crystallization of receptor-inducer complexes has and will be of great benefit in determining signal structures, important residues for ligand binding, and how such interactions might be disrupted to prevent downstream signalling. In this field, the interference of AI-2-based signalling, known as quorum quenching, chemical and biological expertise has been brought together in the synthesis of antagonistic analogue molecules with the potential to target any of the steps in the signalling process.

Synthesis is the first step in the signalling pathway with the potential to be targeted; several LuxS inhibitors have been designed (Alfaro et al., 2004; Shen et al., 2006; Wnuk et al., 2009; Zhu & Li, 2012) but most progress has been made in developing organic synthesis routes to produce AI-2 antagonists and analogues (reviewed in Zhu & Li, 2012). Most approaches have targeted the C1 methyl group of this molecule with linear, branched or cyclic alkyl groups, and even a trifluoromethyl substituent. Interestingly, the effects of such molecules can often be either antagonistic or agonistic in different species (Frezza et al., 2007; Ganin et al., 2009; Lowery et al., 2009; Smith et al., 2009; Roy et al., 2010b; Gamby et al., 2012). Given the very different binding sites of the AI-2 receptors of V. harveyi and S. Typhimurium, this finding is perhaps not completely surprising.

More recently, DPD analogues with a new stereocenter and a methylene group at C5 (targeted due its vicinity to a cavity in both LsrB and LuxP ligand-receptor complexes) were also shown to elicit different responses in the two signalling systems tested: the Lsr of E. coli and the LuxPQ of V. harveyi (Rui et al., 2012). Analysis of both receptor structures was used to explain the results. In LsrB the C5 of the natural ligand is located in an unfilled region suggesting that there would be no significantly different interactions with the protein and the 5R or 5S isomers, an observation that was consistent with the results obtained experimentally. For LuxP the 5R isomer was a better agonist than the 5S and visual analysis of the crystal structure indicates that the 5S methyl is likely to come into steric conflict with the protein.

Janda et al. used a different approach based on the fact that crystallographic data of AI-2-receptor complexes show binding occurring exclusively with the closed forms of DPD (THMF forms in Fig. 2). Carbocyclic analogues of THMF, locked in the cyclic form, were synthesized (Tsuchikama et al., 2011); however, none of those tested exhibited activity in Vharveyi or S. Typhimurium assays. The authors suggested that these results showed that the open-closed equilibrium inherent within the structure of DPD was essential for biological activity. Perhaps a more reasonable hypothesis for the absence of activity of these analogues could result from the loss of the oxygen in the cyclopentane scaffold of the analogues and the coincident loss of hydrogen bonds between the ligand and the residues thought to be essential for ligand binding: with Arg 215 and Asn 159 in LuxP and Ala 222 in LsrB (Fig. 2) (Chen et al., 2002; Miller et al., 2004; Pereira et al., 2009). Taken together, synthesis and analysis of these DPD analogues provides further insight into AI-2 itself, and the potential for rationally designed therapeutics based on structural analysis of ligand-receptor complexes of new AI-2 quorum sensing modulators.

Several natural molecules have also been identified as potential AI-2-signalling inhibitors (reviewed in Zhu & Li, 2012), including brominated furanones extracted from the algae Delisea pulchra (Ren et al., 2001, 2004a) and cinnamaldehyde, although the latter also appears to be a more general inhibitor of Vibrio quorum sensing through inhibition of LuxR binding to DNA (Niu et al., 2006; Brackman et al., 2011).

Alternative means to disrupt quorum sensing have been investigated: quorum quenching of AHL-based signalling often exploits natural enzymes such as lactonases and acylases, which cleave the corresponding signal molecule such that it can no longer bind its cognate receptor (Dong et al., 2007; Amara et al., 2011). Similar strategies have been adopted for interference in AI-2 signalling based on the Lsr system (Fig. 4). lsrR mutant E. coli are capable of efficiently depleting and degrading signal and thus quenching the signalling of others (Xavier & Bassler, 2005b). Bentley, Sintim et al. focused instead on the potential of LsrK as a quorum quenching enzyme. Addition of purified LsrK to bacterial cultures disrupts the signalling of S. Typhimurium, E. coli and V. harveyi, whether in mono- or polyculture containing all three species (Roy et al., 2010a). The processing of extracellular AI-2 to P-AI-2 by the kinase adds a negative charge to this molecule, thought to interfere with its receptor binding capabilities.

A nanotechnological approach was used to develop a spatiotemporally controlled manner to deliver AI-2 signal to targeted bacteria. This system relies on an engineered biological nanofactory comprising an antibody, which functions to bind to the cell surface, and a module of three proteins: a helper protein to enable attachment to the antibody, and Pfs and LuxS to produce the signal (Fernandes et al., 2010). This sophisticated approach enables the manipulation of AI-2 signalling, with respect to signal concentration and timing, of a subpopulation of targeted bacteria or even eukaryotic cells (Fernandes et al., 2010; Hebert et al., 2010).

Concluding remarks

Bacterial niches are frequently shared by an amazing variety of bacterial species that often rely on each other to maintain their normal physiological functions. With this in mind, it seems unlikely that members of a multispecies population would be oblivious to their neighbours. Despite this, the study of interspecies signalling is still in its initial stage: AI-2 remains the best studied interspecies signal to date. Since its discovery, the field has exploded with a myriad of studies into the biology of this molecule in a vast range of species. Although LuxS is a metabolic enzyme, the disruption of which clearly has effects upon central metabolism, this does not preclude a role in bacterial signalling. After all, incorporating two functions, metabolism and signal synthesis, in one enzyme, could provide a means of coupling the production of AI-2 to the physiological status of the cell. This signal gives bacteria the potential to assess population numbers, of self and non-self, through integration of different quorum sensing systems, and combine this with information about nutrient availability, growth rate and other environmental cues. Notably, the synthesis of AHL autoinducers and CAI-1 are both associated with the same metabolic pathway as that involved in the synthesis of AI-2, and uptake of this latter molecule is also intrinsically linked, somehow, to metabolic status through the PTS. Perhaps integrating all these signalling mechanisms provides a global means through which bacteria can coordinate biotic and abiotic information to produce the optimum adaptive capability for the maximum range of situations. This adaptation often involves changes in metabolism, motility, extracellular polysaccharide synthesis, biofilm formation, and similar processes important in virulence. Targeting such behaviours through the use of chemical analogues or quorum quenching enzymes such as LsrK, which would block or manipulate signalling in mixed species communities, has many important potential applications. These are not only limited to the treatment of disease but could also enable the manipulation of beneficial behaviours in polyspecies communities such as that of the gut flora.

Many questions still surround AI-2: how is signal synthesis regulated; what are the means of AI-2 export; how is AI-2 detected in those bacteria where no receptor has yet been identified; what is the physiological role and benefit of the Lsr and AI-2 signalling interference; and to what extent is signal synthesis and response a population-wide behaviour? Further research, if carried out thoughtfully, making best use of the available tools to verify whether phenotypes are due to signalling or to metabolism, will shed light on the role of AI-2 in single and polyspecies environments and provide us with better means through which we can understand, exploit or inhibit bacterial behaviours for our own benefit.

Acknowledgements

We would like to thank the members of the Bacterial Signalling group, Stephen Miller and Rita Ventura for their reading and critical appraisal of this manuscript. Special thanks goes to Stephen Miller for his help with Fig. 2. Work in our laboratory is supported by Fundação para a Ciência e Tecnologia (FCT), Portugal grant PTDC/BIA-BCM/101585/2008 and an International Early Career Scientist grant from the Howard Hughes Medical Institute (HHMI 55007436). C.S.P. acknowledges FCT grant SFRH/BPD/78641/2011.

Ancillary