Living cells monitor their environment using a variety of signal-transduction systems, ranging from simple two-component systems in prokaryotes to highly complex signal-transduction networks in mammalian cells. Since environmental cues are always numerous, the ability to integrate multiple signals is indispensable if cells are to behave appropriately. However, the mechanisms and logic by which cells integrate environmental signals remain, by and large, poorly understood. Here we have quantitatively analyzed the integration of multiple autoinducer signals by the model quorum-sensing bacterium V. harveyi using single-cell fluorescence microscopy. Our studies reveal a unified response across the population, with moderate cell-to-cell variation. We find that signals from two distinct autoinducers, AI-1 and AI-2, are combined strictly additively in a single phosphorelay pathway, with each autoinducer contributing nearly equally to the total response. Moreover, the cell-to-cell variation in response is small enough that the entire population of cells can reliably distinguish at least three distinct conditions of external autoinducer concentration.
We used GFP under the control of the chromosomal sRNA Qrr4 promoter as a reporter of the activity of the quorum-sensing signaling pathway (). In all our strains, the GFP distribution was always single-peaked at all autoinducer concentrations, with cell-to-cell standard deviation no more than 40% of the mean, suggesting that populations of V. harveyi
cells respond coherently to autoinducer signals. By contrast, genes in some other bacterial systems are known to have bimodal (i.e., two-peaked) expression distributions. In many cases, bimodal gene expression is also hysteretic (i.e., cells remain for a long time in one state of expression), which constitutes a form of cellular “memory.” For example, bimodal distributions in gene expression enable sporulation and competence in B. subtilis
], stringent response in mycobacteria [34
], and induction of the lac
operon in Escherichia coli
]. In all these cases, bimodality and hysteresis are believed to provide advantages to the organism by enabling phenotypic diversity within isogenic populations. In general, hysteresis in gene expression requires some form of positive feedback. The lack of bimodality in our engineered strains of V. harveyi
is expected since there is no positive-feedback loop in the circuit controlling Qrr sRNA expression in these cells. Since our engineered strains lack both the downstream transcription factor LuxR and the autoinducer synthases, there exists the possibility that the sRNAs or LuxR could feed back positively to the synthases and produce a bistable circuit in wild-type cells. In quorum sensing, bistability has only been reported for a rewired LuxIR circuit in V. fischeri
]. In this case, the positive feedback and the resulting bistability and hysteresis occur at the population level and divide the entire population into two separate subpopulations, each with a unique phenotype. Our consistent observation of a narrowly peaked distribution of quorum-sensing responses strongly suggests that V. harveyi
cells respond in unison to the presence of autoinducer signals. For quorum-sensing cells, in contrast to bacteria undergoing competence, sporulation, or the stringent response, operating as a coherent population appears to be more important than maintaining phenotypic diversity.
An outstanding question is why V. harveyi
and related species use multiple autoinducer signals, but funnel all the information into a single pathway. We can envision two main possibilities (potentially in combination): The multiple autoinducers could reveal information about the community composition (e.g., which species are present and in what abundance), or the multiple autoinducers could reveal information about the development stage of the community (e.g., the growth stage of a biofilm). In support of the first possibility, the three autoinducers used by V. harveyi
have distinct ranges of species specificity: intraspecies for AI-1, within Vibrios for CAI-1, and across many species for AI-2 [7
]. Thus, different combinations of the three autoinducers could indicate different compositions of a bacterial community. In our experimental conditions, however, we found that cells could not distinguish between high AI-1/low AI-2 and high AI-2/low AI-1 (B and C). This result argues for the second possibility, namely that different combinations of autoinducers represent different stages of community development. For example, if a growing V. harveyi
community typically accumulates AI-2 before AI-1, then the signaling contour in C would always be traversed along the right edge, and cells could reliably interpret an intermediate signaling strength as a condition of high AI-2/low AI-1, since the opposite condition of high AI-1/low AI-2 would rarely, if ever, be encountered. In much of eukaryotic development (e.g., embryogenesis), the rate of development is fixed and driven by a clock [38
], obviating the need for a signal representing the stage of development. However, without the support of a surrounding organism, the rate of development of a bacterial community depends on unpredictable environmental conditions, such as nutrient availability, and therefore some means of determining the stage of development is required so that cells in the community can behave appropriately. Recent models of biofilm growth suggest that communities may be mixed at early stages, but that at later stages competition for nutrients by overgrowth of neighboring cells can result in large domains of cells descended from a single progenitor, and therefore composed of a single species [39
]. If so, generic signals such as AI-2 may be most informative at early stages of biofilm growth, while species-specific signals such as AI-1 may be reserved for later stages. We are currently exploring the order of accumulation of the V. harveyi
autoinducers AI-1, CAI-1, and AI-2 to test whether different autoinducer combinations could signal different stages of community development.
Given that the autoinducer signals are combined in one pathway in V. harveyi, why should the signals be combined additively, as we observe for AI-1 and AI-2? Simple alternatives would be for saturating autoinducer levels to be combined in “logic gates,” such as AND, in which both autoinducer signals would be required for a full response, or OR, in which either signal would be sufficient for a full response. However, these logic gates have only two possible output states: on or off. In contrast, the addition of the two autoinducer signals allows for more than two output states of the signaling pathway, and therefore potentially allows for more than two expression states of quorum-sensing regulated genes. Indeed, we discovered three distinct levels of signaling strength, represented by the heights of the plateaus in C. Moreover, the standard deviation of PQrr4-GFP expression across the population of cells was sufficiently small (B) so that the entire population can apparently distinguish the three distinct plateau heights. This means that, in principle, every cell in the population can distinguish three external autoinducer conditions: both autoinducers low, both autoinducers high, and a third condition in which one autoinducer is high and the other is low. The reliability with which cells can distinguish among these three conditions is increased by the equal spacing of the plateau heights as shown in C. Given a uniformly distributed input of autoinducer concentration and the observed level of noise (i.e., cell-to-cell variation in PQrr4-GFP expression), a significantly unequal spacing of the plateau heights would lead to overlapping distributions of PQrr4-GFP expression for the two more closely spaced plateaus. The implication is that noise might then cause some cells to misinterpret external conditions and regulate quorum-sensing genes inappropriately. The need for all cells to reliably distinguish among multiple autoinducer conditions may therefore explain not only the additivity of the quorum-sensing pathway, but also why the contributions of the AI-1 sensor LuxN and the AI-2 sensor LuxPQ to the total kinase activity are so nearly equal—equal kinase activities mean equally spaced plateau heights, which in turn mean that individual cells are less likely to confuse one autoinducer condition with another.
The existence of multiple quorum-sensing output states potentially underpins diverse patterns of quorum-sensing regulated gene expression. For example, in previous studies, the quorum-sensing circuit of V. harveyi
was found to act as an autoinducer “coincidence detector” (i.e., requiring both AI-1 and AI-2) for full induction of bioluminescence [19
]. Thus, in the present context, the three distinguishable levels of signaling output (indicated by Qrr4 promoter activity) appear to be collapsed by downstream signal-processing events to two levels of bioluminescence. More generally, the target genes of quorum sensing could be tuned to different signaling output levels so that only particular classes of genes are switched ON/OFF at early, middle, or late stages of community development. Alternatively, some genes could have graded expression between these different developmental stages. The requirement for multiple distinct output states might also explain our observation of a graded, rather than switch-like, response of the Qrr4 promoter. Specifically, our dose–response data are well described by a noncooperative, n
= 1 Hill function response to both autoinducers. Cooperativity would have resulted in an n
> 1 Hill function and therefore a more switch-like response of PQrr4
-GFP to autoinducers. During the signaling process, cooperativity could in principle have arisen from the binding of autoinducers to receptors, transfer of phosphate among the protein components in the phosphorelay, and/or binding of phosphorylated LuxO to DNA. Our results suggest that in fact all of these steps are noncooperative, despite the fact that the receptors are likely dimers [22
] and that LuxO may function as a tetramer or octamer [Tu KC, unpublished data]. Indeed, a graded noncooperative response of Qrr expression to autoinducers is essential for the existence of multiple, distinguishable quorum-sensing states, as a switch-like response of the Qrr expression would have allowed for only two states.
Based on a simple kinetic model for signaling (Equation 2
), we have argued that the kinase activities of LuxN and LuxPQ are regulated by autoinducers, whereas for most two-component receptors, it is still an open question whether the kinase or phosphatase or both activities are regulated by input stimuli. Previously, LuxN receptors have been successfully modeled as switching between two states: the ON (kinase dominant) and OFF (phosphatase dominant) states [41
]. Each receptor has intrinsic kinase and phosphatase rates depending only on the state in which the receptor exists. Extending this model to LuxPQ, the total cellular kinase activities KN
consist of a major contribution from those receptors in the ON state with little or no contribution from those in the OFF state. From the constraints set by additivity, we conclude that the phosphatase activities PN
are unregulated (i.e., receptors have the same phosphatase rates in both the ON and OFF states). Note that autoinducer concentrations only affect the thermal balance between ON and OFF states, and therefore the kinase and phosphatase activities are regulated only via the biasing of receptors between states (of course, the total kinase and phosphatase activities also depend on receptor concentrations). The low levels of PQrr4
-GFP expression with saturating AI-1 in the LuxN+
strain, saturating AI-2 in the LuxPQ+
strain, and saturating AI-1 plus AI-2 in the LuxN+
strain indicate that kinase rates in the OFF states are much smaller than those in the ON states for both LuxN and LuxPQ. By decreasing the fraction of receptors in the ON state, autoinducers reduce the total kinase activity of the quorum-sensing receptors in V. harveyi
. (See Text S1
for more details.)
Regulation of the kinase activities of LuxN and LuxPQ appears to be necessary to achieve three equally spaced levels of LuxO-P (Equation 3
). The requirement for kinase regulation in V. harveyi
quorum sensing therefore appears to stem from the need to combine multiple input signals into more than two distinguishable output levels of LuxO-P. One prediction from this analysis is that the sensor CqsS, which was not present in our strains, is likely to also have its kinase activity regulated by its autoinducer CAI-1. Moreover, CqsS is likely to contribute additively to total kinase activity and with a strength comparable to that of LuxN and LuxPQ, resulting in four maximally distinguishable levels of kinase activity and therefore four distinguishable autoinducer conditions.
The similarity of the responses to AI-1 and AI-2 is striking, not only in the amplitudes but also in the inhibition constants. We speculate that V. harveyi
usually encounters similar amounts of AI-1 and AI-2, and the responses of receptors have been optimized to match the natural dynamic range of autoinducer concentrations. It has been demonstrated that single mutations in the receptors LuxN and LuxPQ can result in dramatic changes in their inhibition constants [22
], so the similar values for AI-1 and AI-2 may represent an evolved optimum.
We also quantified the noise in PQrr4
-GFP expression in our three reporter strains. Noise is an inherent feature of signal transduction and gene expression both in prokaryotes and eukaryotes. Due to the low copy number of cellular components and the stochastic nature of biochemical reactions, fluctuations are inevitable. Large fluctuations might be deleterious for processes requiring precise control but beneficial for those providing phenotypic diversity. In quorum sensing, bacterial cells detect population cell density to coordinate their behavior on a community-wide scale. Low noise in quorum-sensing signal transduction might therefore benefit the population of cells by allowing all cells to behave correctly and in unison at each stage of community development. Indeed, we observed low noise in PQrr4
-GFP expression in all our strains. At all autoinducer concentrations the standard deviation over the mean was less than or close to 0.4 (). In other systems, the dominant source of cell-to-cell variation in gene expression has been attributed to extrinsic noise, e.g., differences among cells in concentrations of general purpose cellular components such as RNA polymerases and ribosomes [8
]. In the quorum-sensing circuit we have studied, the noise we observed is also likely due to extrinsic factors rather than to biochemical noise in phosphotransfer or transcription and translation of PQrr4
-GFP. The most likely source of the noise we observed is fluctuations in concentrations of the pathway components, such as the receptors LuxN and LuxPQ and the response regulator LuxO. The noisier response in LuxPQ pathway is very likely caused by variations in the copy number of the LuxPQ receptors, which suggests that there could be some additional regulation of receptor expression in the quorum-sensing circuit.