Cell-cell communication or quorum sensing (QS) is a prevalent mechanism that coordinates gene expression in bacterial populations via diffusible chemical signals (Waters and Bassler, 2005; West et al., 2007a; Schuster et al., 2013; Hense and Schuster, 2015; Popat et al., 2015). These self-produced signal molecules accumulate with increasing cell density and trigger gene expression by activating a cognate receptor. A common theme is that QS primarily controls collective, cooperative traits such as biofilm formation, virulence, social motility, and nutrient digestion. Most of these group activities involve the secretion of extracellular products generally referred to as “public goods,” including exopolysaccharides, toxins, surfactants, and extracellular enzymes. At the transcriptome level, the response to QS-signaling can be dramatic, with hundreds of genes (about 5% of the genome) controlled by QS in the opportunistic pathogen Pseudomonas aeruginosa (Hentzer et al., 2003; Schuster et al., 2003; Wagner et al., 2003). Here, too, the set of genes encoding secreted factors is overrepresented compared to the genome at large (Gilbert et al., 2009). A similar overrepresentation of secretion is observed in the QS regulons of Vibrio and Erwinia species (Antunes et al., 2007; Barnard et al., 2007).

Theoretical arguments show that the benefit from a secreted public good usually increases with population density, justifying regulation by QS (Pai and You, 2009; Heilmann et al., 2015; Popat et al., 2015). At low density, a large fraction of the public good is lost to the environment by mass transfer, whereas at high density, environmental losses are reduced because the product benefits neighboring cells (Figure 1). These considerations are supported by a small number of empirical studies that show a positive density-dependent benefit for QS-controlled public goods, in one case a protease in P. aeruginosa (Darch et al., 2012), and in another case an antibiotic-degrading enzyme in a synthetic QS system (Pai et al., 2012).

Nevertheless, there are numerous examples of cell-associated products that are QS-controlled (Table 1). These products can be proteins or metabolites; they can be intracellular (cytoplasmic) or surface-attached. They have been designated private goods because their location or benefit is assumed to be exclusive to the producing cell (Figure 1). Perhaps the most prominent QS-controlled private good is P. aeruginosa nucleoside hydrolase (Nuh) (Heurlier et al., 2005; Mellbye and Schuster, 2011; Dandekar et al., 2012; Darch et al., 2012). It is a periplasmic enzyme that hydrolyzes adenosine and inosine, allowing the cell to grow on these nucleosides as the sole carbon (C) or nitrogen (N) source (Heurlier et al., 2005; Figure 2). For many cell-associated traits, it is not immediately clear why they are under the control of QS. A priori, their benefit appears non-collective and hence, should be independent of cell density.

In an attempt to explain the QS-control of private goods, some considered a potential function in the stabilization of co-regulated cooperative behaviors by public goods (Dandekar et al., 2012; Oslizlo et al., 2014; Majerczyk et al., 2016). The evolution and maintenance of cooperative behaviors is difficult to explain because they are subject to exploitation by cheaters that reap the benefits of cooperation without paying the costs (Hamilton, 1964; Hardin, 1968; Lehmann and Keller, 2006; West et al., 2007b). For example, in P. aeruginosa, QS-deficient cheater mutants can invade a population of QS-proficient cells that produce an extracellular protease required for growth on a proteinaceous C-source (Diggle et al., 2007; Sandoz et al., 2007; Dandekar et al., 2012; Pollitt et al., 2014; Asfahl et al., 2015). When the Nuh-dependent substrate adenosine is present as an additional C-source, however, the evolution of protease-deficient cheaters is constrained (Dandekar et al., 2012). This and other examples have emphasized the role of QS-regulated private goods in cheater control (Foster et al., 2004; Wilder et al., 2011; Dandekar et al., 2012; Oslizlo et al., 2014; Garcia-Contreras et al., 2015; Wang et al., 2015). In this article, we highlight the difference between cheater control by private goods as an adaptation (its “evolutionary purpose”) and as a by-product of selection for other benefits. Based on physiological and ecological considerations, we suggest several alternative functions for the QS control of private goods that all entail the optimization of a cell-density dependent net benefit. This benefit is realized either because there is a favoring ecological condition that correlates with high cell density, or because the private good itself is, directly or indirectly, linked to cooperative behavior. We discuss the implications of our analysis for the control of infections with novel antivirulence drugs and for the engineering of stable cooperative circuits in synthetic biology and biotechnology.

The maintenance of cooperation through co-regulation of multiple traits, generally referred to as pleiotropic constraint, has been initially described in the fruiting-body-forming amoeba Dictyostelium discoideum (Foster et al., 2004; Figure 3). The principal idea is that pleiotropic constraint stabilizes cooperation by linking the expression of a cooperative, public good to a non-cooperative, private good, often through common regulation. Private good production provides a benefit to a producer but not to a cheater in which common regulation is disrupted.

Several QS-controlled private goods have been investigated in this context (Figure 3). In addition to Nuh (Heurlier et al., 2005), these include catalase and superoxide dismutase in P. aeruginosa that detoxify reactive oxygen species (Hassett et al., 1999), an alternative cytochrome c oxidase in P. aeruginosa that provides resistance to self-produced hydrogen cyanide (Cunningham et al., 1997), and immunity proteins that protect against contact-dependent inhibition by Type VI secretion effectors in B. thailandensis (Majerczyk et al., 2016). Consequently, the growth advantage of QS-deficient cheaters in a cooperative environment decreases when private good expression is favored, i.e., in the presence of adenosine, oxidative stress, cyanide production, or Type VI-proficient cells (Dandekar et al., 2012; Garcia-Contreras et al., 2015; Wang et al., 2015; Majerczyk et al., 2016). Furthermore, the pleiotropic, QS-controlled suppression of costly functions in P. aeruginosa and Bacillus subtilis can also restrict the spread of certain QS-deficient mutants in co-culture (Wilder et al., 2011; Oslizlo et al., 2014). In all these cases, co-regulation of private goods appears to stabilize cooperation under specific culture conditions, but fundamentally, it is not clear whether these forms of cheater control represent an adaptation or a by-product of selection for other benefits.

To examine this question, we consider the concept of an evolutionarily stable strategy (ESS) derived from evolutionary game theory (see Box 1). Individual strategies are assigned a fitness pay-off or net benefit and compared. Net benefit is defined as the difference between benefit and cost. An ESS is a strategy that is resistant to invasion by an initially rare, alternative strategy. Thus, in order to determine whether the QS control of private goods for the purpose of cheater control is an ESS, we must consider alternative strategies. One such strategy is the QS-independent expression of the private good, uncoupled from common QS regulation (Figure 4). While QS-controlled expression of the private good disfavors cheaters, constitutive expression or no expression at all may confer a substantial net benefit to the producer under certain conditions. This is apparent if we consider the fact that QS restricts gene expression to a high cell density environment conducive to cooperation. For instance, when private good production is beneficial at low density, then constitutive expression invades QS-dependent expression. On the other hand, when private good production is beneficial neither at low nor high density, then no expression invades QS-dependent expression, as it saves costly private good production. The potential benefit from QS-independent expression can be further optimized if the expression of the private good is brought under the control of a dedicated regulatory system (Figure 4). Then the private good can be expressed only if it provides a benefit, irrespective of cell density.

These considerations challenge the notion that the QS-control of private goods has evolved and is maintained for the purpose of cheater control. Indeed, dynamic modeling shows that such pleiotropically constrained cooperation is susceptible to invasion by mutant strategies that independently express the private good (Box 1). Once the joint regulation of both public and private goods is interrupted, a defector can invade and cooperation is no longer stable. Consequently, QS-controlled private good expression can only be an ESS if it provides a direct fitness advantage that is superior to QS-independent expression. We propose several functions below that may satisfy this requirement.

To formulate insightful alternative hypotheses about the function and adaptive significance of private goods, a detailed mechanistic understanding is essential. We consider biochemical, physiological and ecological context. Importantly, demonstrating that a particular mechanism is well suited to a specific laboratory environment is different from claiming that this mechanism was selected for in the past due to its history of being adaptive. Thus, in vitro culturing experiments constitute a double-edged sword. They provide valuable insights into possible adaptations to specific selection pressures, but it is often not clear whether these selection pressures are relevant in the natural environment (Bailey and Bataillon, 2016).

By a common economical definition, private goods are both excludable and rivalrous, meaning that their benefits are only available to the producer, and that consumption by one necessarily prevents consumption by another (Kaul and Mendoza, 2003). In the microbial realm, a distinction between location and benefit is important, as is the appropriate designation of the relevant component involved in the behavior. Suppose an intracellular enzymatic pathway and its secreted metabolic endproduct; for example, the biosynthetic enzymes that produce exopolysaccharide (Vu et al., 2009). Here, the enzymes are the private goods, and the metabolic end-product is the public good. However, what we are interested in is not the enzymes per se but rather the benefit that the metabolic endproduct confers. Thus, the relevant good with respect to the resulting cooperative behavior is the extracellular metabolite. Because the enzyme is directly involved in the production of the metabolite, its connection to cooperative behavior and hence, regulation by QS, is obvious. As we will see below, this link may not be as obvious in other instances, such as central metabolic enzymes that help synthesize precursors of a secreted metabolite.

Once the relevant good has been identified, its classification as a strictly private or strictly public good may not always be clear-cut. Is a private good really exclusive to the producing cell or is partially shared by others? Economists typically call such goods with both public and private characteristics “mixed goods” (Pearce, 2002), whereby the balance between public and private benefits can vary. A mixed good that is partially shared by others can provide a cell-density-dependent benefit and would thus justify regulation by QS. The interplay between public and private aspects can also have remarkable effects on population dynamics and evolutionary stability. In a striking example from yeast, the products generated by a cell-associated enzyme, invertase, largely diffuse away; only 1% is captured by the producing cell (Gore et al., 2009). Yet, this small fraction is sufficient to allow a rare invertase-proficient strain to invade an invertase-negative cheater population. On the other hand, the density-dependent, cooperative nature of the activity also allows a rare cheater to invade an invertase-proficient population, resulting in an equilibrium of cheaters and co-operators (Gore et al., 2009).

Taken together, inferences about possible regulatory needs require a comprehensive mechanistic understanding of the phenotype in question which, as we will see below, is not always straightforward.

The QS-controlled private good that has arguably received the greatest attention is P. aeruginosa Nuh (Heurlier et al., 2005; Mellbye and Schuster, 2011; Dandekar et al., 2012; Darch et al., 2012). Nucleoside hydrolases are part of the nucleotide salvage pathway (Nyhan, 2014). This pathway permits the recycling of nucleosides and bases generated during the degradation of DNA and RNA, from both intracellular and extracellular sources. P. aeruginosa Nuh also promotes growth on adenosine and inosine as the sole C or N sources (Heurlier et al., 2005). Because QS induces Nuh expression about five-fold (Schuster et al., 2003), the growth of a QS-deficient mutant is impaired on these nucleosides (Heurlier et al., 2005). Nuh cleaves adenosine into adenine and ribose, and inosine into hypoxanthine and ribose (Heurlier et al., 2005, 2006; Figure 2). Adenine is a good N but a poor C-source in P. aeruginosa, although its immediate metabolic fate and possible conversion to hypoxanthine are not entirely clear (Heurlier et al., 2006). Hypoxanthine is degraded to glyoxylate and urea in several steps (Matsumoto et al., 1978). Glyoxylate can be converted to acetyl-CoA, which is oxidized in the TCA cycle to yield cellular energy and C-building blocks (Schuster et al., 2003; Heurlier et al., 2005). Ribose can be converted to ribose-5-phosphate by ribokinase for anabolic and catabolic reactions via the pentose phosphate pathway (Stover et al., 2000). We assume for now that these Nuh metabolites remain intracellular and that the treatment of Nuh as a private good is meaningful. We will examine this assumption in the next section.

Interestingly, Nuh is periplasmic in P. aeruginosa, whereas the three homologs in in E. coli are cytoplasmic (Petersen and Moller, 2001; Imperi et al., 2009a). The periplasmic location in P. aeruginosa might be a consequence of the poor nucleoside transport across the cytoplasmic membrane that lacks known nucleoside importers (West, 1996; Stover et al., 2000). In contrast to other Pseudomonads, P. aeruginosa also seems impaired in ribose uptake across the cytoplasmic membrane. It lacks the rbsD gene that has been shown to contribute to uptake (Oh et al., 1999). This property explains the slow growth rates on adenosine as the sole C-source (doubling times ~10 h) (Heurlier et al., 2005; Mellbye and Schuster, 2011; Dandekar et al., 2012) and indicates that adenosine, via ribose, does not serve as an ecologically relevant C-source in P. aeruginosa. It is instead the adenine moiety of adenosine that P. aeruginosa seems to be after, as an N-source or as a direct precursor for nucleic acid synthesis. Consistent with this notion, prolonged N-starvation increases Nuh activity (Fields, 2002; dissertation data not published elsewhere). The regulatory mechanism and its relationship to QS, however, are not clear. Regulation of nuh by the N-starvation sigma factor RpoN was proposed by Fields et al. but could not be confirmed experimentally (Schulz et al., 2015).

These mechanistic insights permit differentiated hypotheses about relevant ecophysiological context. Extracellular adenosine and other nucleosides are likely to be abundant when the environment contains damaged cells such as during infection or competition with other microbes (Patel et al., 2007; Gellatly and Hancock, 2013). These conditions could be a direct consequence of QS-induced secretion at high cell density: Nuh is part of a broad array of QS-controlled factors, including many extracellular enzymes and toxins, that can damage other cells to release nutrients. Here, Nuh would scavenge nucleosides for nucleic acid biosynthesis and as general N source. This property could then help control cheating by QS mutants, even though it may not have evolved for this purpose. A lasR mutant would have a growth disadvantage because it would not be able to utilize nucleosides to the same extent as the wild-type. This QS-specific effect might increase or decrease with N-limitation, depending on regulatory interactions.

In contrast, when selection is for adenosine as the C-source, as in Dandekar et al. (2012), then adaptations are predicted to be entirely different. Dandekar et al. reported that the evolution of lasR mutant cheaters in wild-type populations is suppressed when adenosine is provided in the growth medium along with casein as the sole C-sources (Dandekar et al., 2012). This outcome was attributed to pleiotropic control of both secreted proteases and cell-associated Nuh by QS. However, alternative interpretations are possible. Dandekar et al. only observed cheater control at very high concentrations of adenosine, when we expect Nuh activity to be maximal and subsequent cellular uptake to be rate-limiting. Under these conditions, QS-dependent differences in Nuh expression should have little effect on adenosine utilization and consequently on cheater control. To test this prediction empirically, we initiated co-cultures of the P. aeruginosa wild-type and a defined lasR mutant in medium containing either casein, adenosine, or both as the sole the C-sources, at concentrations equal to Dandekar et al. (Figure 5). A critical difference from Dandekar et al. is that these experiments were conducted with defined strains such that effects from undefined mutations that may arise during in vitro evolution of an all-wild-type population are excluded. We found that casein medium alone favors lasR mutant cheaters, as previously reported (Sandoz et al., 2007; Dandekar et al., 2012). In support of our predictions, we also found that adenosine medium alone does not favor either strain, and that the addition of adenosine to casein medium has no effect on lasR mutant enrichment compared with casein medium alone.

Given these results, how was the evolution of lasR mutants nevertheless suppressed in the original study? A second, intriguing observation provides insights: Successive rounds of subculturing selected for mutants with greatly improved growth on adenosine (Dandekar et al., 2012). These variants are expected to harbor mutations that increase ribose uptake, given that this function is impaired in P. aeruginosa. Such non-social adaptation to nutrient-limited environments has been reported in several studies (Morgan et al., 2012; Waite and Shou, 2012; Asfahl et al., 2015). Thus, P. aeruginosa may have entirely shifted C-source preferences from casein to the high levels of available adenosine that sustained growth throughout the growth cycle. Taken together, non-social adaptation to the specific growth environment, rather than pleiotropic control of cooperative growth, may have lead to the observed suppression of lasR mutants.

In the following section we categorize and discuss several plausible scenarios that explain the QS-dependent regulation of private goods (summarized in Table 1 and Figure 6). We consider relevant physiological and ecological context. In all cases, there is a cell-density dependent benefit to the expression of private goods, even though it is often not as obvious as that attained from public goods. Thus, these scenarios are in line with the principle that QS optimizes the net benefit attained from a product under its control (Pai and You, 2009). The benefit from the QS control of private goods is realized either because there is a specific ecological condition, generally a stressor, that is associated with high cell density, or because the private good itself is, directly or indirectly, associated with cooperation.

Here we have proposed different scenarios in which QS can serve to optimize a density-dependent benefit attained from private goods, providing an evolutionary “purpose” for their QS control. Taken together, we argue that that selection has brought traits under the control of QS whose benefits are associated with high cell-density environments (Popat et al., 2015). The association between benefit and cell density may be statistical (i.e., occur randomly but with certain probability, generally because relevant environmental conditions tend to correlate with cell density), or it may be deterministic (i.e., occur with certainty as a direct consequence of another action, generally because the private good itself is, directly or indirectly, linked to cooperation). Early anticipation of changes via QS adds a predictive element that may provide an advantage compared to a somewhat delayed response to the actual stimulus (Goo et al., 2012). For many traits, the optimal mode of control may be a combination of QS and other regulatory pathways. At low density, such co-regulation could mitigate some of the disadvantages from strict all-or-none QS-control. Cyanide resistance in P. aeruginosa, for example, is also directly induced by the presence of cyanide (Frangipani et al., 2014), which would even allow individual cells or a small group of cells to resist cyanide production by competing microbial species.

Using the ESS framework, we have also argued that QS control of private goods may not have evolved to stabilize public goods cooperation. This does not mean that pleiotropic cheater control does not happen or is not important; it only means that pleiotropic cheater control is a by-product for other benefits rather than an adaptation. It is important here to distinguish between the optimization and the stabilization of a behavior: The QS control of private goods has evolved because it optimizes a net benefit that is superior to other forms of regulation. The QS control of private goods is evolutionarily stable because of this optimization and because private goods - unlike public goods - preferentially benefit the producer. This includes products with a relatively small private share, as we have pointed out above.

The tractability of microbial model systems allows experimental verification of some of our predictions, in terms of both mechanism and evolutionary stability. Mechanisms can be tested according to the scenarios proposed above. As indicated, metabolic modeling could be useful in elucidating the role of central metabolism in optimizing secretion: Are the fluxes of central metabolites under quorate and non-quorate conditions in the model consistent with the experimentally observed regulation of the corresponding enzymes? Does the production of a given public good decrease when the fluxes through these reactions are restricted?

Evolutionary stability can be tested in co-culture competition and in vitro evolution experiments. Strains can be engineered such that they express a certain private good either constitutively or under QS control, and the fitness of the resulting strains can be compared. In vitro evolution under defined selective conditions can demonstrate the evolutionary stability of pleiotropic constraint and its susceptibility to invasion by mutant strategies. For example, using QS-controlled adenosine utilization in P. aeruginosa as a model, one could show that the QS control of Nuh breaks down and is replaced by constitutive expression if there is no cell density-dependent benefit to Nuh expression under the prevailing culture conditions.

Our analysis has implications for synthetic ecology (Dunham, 2007; Renda et al., 2014; Fredrickson, 2015). The propensity for cheating impedes the functionality of microbial communities engineered for the purpose of cooperative activities, for example biotechnological enzyme production. Our results suggest that the design of a regulatory constraint for the purpose of cheater control will have to be carefully considered. A mode of regulation that does not provide a fitness advantage over and could be invaded by an alternative mode of regulation would not be stable in the long-term.

Our analysis also has medical implications because the cooperative behaviors of many bacterial pathogens, such as QS, biofilm formation, or the secretion of toxins, are important virulence factors (Parsek and Singh, 2003; Rutherford and Bassler, 2012). Novel antivirulence strategies that target these behaviors have the potential to slow the evolution of resistance (Schuster et al., 2013; Allen et al., 2014; Ross-Gillespie et al., 2014). This beneficial effect is thought to be particularly relevant for cooperative behaviors because resistant cells would effectively behave as cooperators among a population of drug-sensitive cheaters in which cooperative virulence factor expression is suppressed (Andre and Godelle, 2005). This notion was confirmed experimentally by comparing the relative fitness of QS-proficient and QS-deficient strains of P. aeruginosa mimicking drug-resistant and drug-sensitive subpopulations, respectively (Mellbye and Schuster, 2011). QS-resistant mimics had an advantage only when growth was dependent on a cooperative rather than a non-cooperative trait under QS control. These results suggest that a clear functional analysis and distinction between public (cooperative) and private (non-cooperative) virulence factors is critical for treatment success.

Taken together, a deeper understanding of QS-controlled functions, and the evolutionary pressures that select for their maintenance, should improve our ability to predict evolutionary dynamics in natural environments, in pathogenic contexts, and in biotechnological applications.

MS designed the study, evaluated the literature, analyzed data and wrote the manuscript. DS performed experiments, reviewed the literature, and analyzed data. BH designed the study, evaluated the literature, and wrote the manuscript.