University of Southern California Dissertations and Theses

Quorum sensing interaction networks in bacterial communities

(USC Thesis Other)

Quorum sensing interaction networks in bacterial communities

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content i Quorum sensing interaction networks in bacterial communities by Kalinga Pavan Thushara Silva A Dissertation presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (PHYSICS) May 2019 Copyright 2019 Kalinga Pavan Thushara Silva ii Acknowledgements I would like to thank my Ph.D. advisor James Boedicker, my lab members and my Ph.D. committee members, Mohamed El Naggar, Rosa Di Felice, Vitaly Kresin and Kenneth Nealson for all the support given to me throughout my Ph.D. I am thankful to all my family members and friends; especially my father Tudor, my mother Susila and my sister Ruvini for giving me advice and giving me unconditional support whenever I needed it. I finally thank my wonderful wife Ruvindi for being by my side and taking care of me. iii Table of Contents Acknowledgements ......................................................................................................................... ii List of Figures ............................................................................................................................... vii List of Tables ................................................................................................................................. xi Abstract ........................................................................................................................................... 1 Chapter 1: Introduction ................................................................................................................... 3 1.1: Quorum sensing crosstalk and quorum quenching in bacterial communities ..................... 3 1.2: A brief overview the thesis ................................................................................................... 6 1.3: Summary of the chapters ...................................................................................................... 8 Chapter 2: Signal destruction tunes the zone of activation in spatially distributed signaling networks .......................................................................................................................................... 9 2.1: Abstract ................................................................................................................................ 9 2.2: Introduction .......................................................................................................................... 9 2.3: Materials and Methods ....................................................................................................... 11 2.3.1: Bacterial strains and plasmid ....................................................................................... 11 2.3.2: Culturing conditions .................................................................................................... 12 2.3.3: Interference assay ........................................................................................................ 12 2.3.4: Microscopy measurements .......................................................................................... 12 2.3.5: Growth measurements ................................................................................................. 13 2.3.6: Plate-reader measurements .......................................................................................... 13 2.4: Results and Discussion ....................................................................................................... 14 2.4.1: Quantifying signal exchange in spatially distributed microbial networks ................... 14 2.4.2: AiiA inhibits quorum sensing in a synthetic microbial community ............................ 16 iv 2.4.3: Signal destruction leads to finite and tunable zones of activation in spatial networks 17 2.4.4: Modeling the impact of signal destruction on signal propagation ............................... 18 2.5 Conclusions ......................................................................................................................... 22 2.6: Supplementary Information ................................................................................................ 24 2.6.1: Plasmids ....................................................................................................................... 24 2.6.2: Calculation of the number of proteins per cell for the production rate of AiiA .......... 31 2.6.3: Calculating the fluorescence in a mixed culture using the plate reader ....................... 31 2.6.4: Testing the model for effects of the growth interactions, diffusion coefficients, degradation rates and internalization of the signal ................................................................ 34 Chapter 3: Quantifying the strength of quorum sensing crosstalk within microbial communities ....................................................................................................................................................... 36 3.1: Abstract .............................................................................................................................. 36 3.2: Author Summary ................................................................................................................ 36 3.3: Introduction ........................................................................................................................ 37 3.4: Results ................................................................................................................................ 40 3.4.1: Detecting Quorum Sensing Crosstalk Using the Plate-Assay Setup. .......................... 40 3.4.2: Measuring the scaling of crosstalk delay with community composition. .................... 42 3.4.4: Robustness of signal transduction to crosstalk. ........................................................... 46 3.4.5: Measuring the crosstalk potential of wild isolates. ...................................................... 48 3.5: Discussion .......................................................................................................................... 50 3.6: Materials and methods ....................................................................................................... 53 3.6.1: Bacterial strains and plasmids...................................................................................... 53 3.6.2: Culturing conditions. ................................................................................................... 53 3.6.3: Plate assay. ................................................................................................................... 54 3.6.4: Microscopy measurements. ......................................................................................... 54 3.6.5: Growth measurements. ................................................................................................ 54 v 3.6.6: Plate-reader measurements. ......................................................................................... 54 3.7: Supplementary Information ................................................................................................ 55 3.7.1: Calculating the probabilities of each AHL binding to the LuxR receptor using Boltzmann weights................................................................................................................. 58 3.7.2: Testing the model for effects of the growth interactions and AHL internalization. .... 68 Chapter 4: Disruption of microbial communication yields a 2D percolation transition ............... 73 4.1: Abstract .............................................................................................................................. 73 4.2: Introduction ........................................................................................................................ 73 4.3: Experiments on quorum sensing bacteria. .......................................................................... 75 4.3.1: Experimental setup ...................................................................................................... 75 4.3.2: Experimental results .................................................................................................... 76 4.4: Mathematical model ........................................................................................................... 82 4.4.1: Reaction-diffusion model ............................................................................................ 82 4.4.2: Modelling results ......................................................................................................... 83 4.5: Shifting the transition point by changing the production rate of AiiA. .............................. 85 4.6: Methods .............................................................................................................................. 89 4.6.1: Bacterial strains............................................................................................................ 89 4.6.2: Bacterial growth conditions ......................................................................................... 91 4.6.3: Bacterial strains............................................................................................................ 91 4.6.4: Microscopic measurements and image analysis .......................................................... 91 4.7: Supplementary information ................................................................................................ 92 4.7.1: Approximating a system with degraders onto an effective degrader-free system. ...... 97 4.7.2: Mapping basal production rate onto site occupation probability in the degrader-free limit. ....................................................................................................................................... 99 4.7.3: Consequences for observed critical exponents as degrader density is varied. ........... 101 vi Chapter 5: A neural network model predicts community-level signaling states in a diverse microbial community .................................................................................................................. 114 5.1: Abstract ............................................................................................................................ 114 5.2: Introduction ...................................................................................................................... 114 5.2.1: Quantifying pairwise crosstalk between strains of Bacillus subtilis.......................... 118 5.2.2: Pairwise weights predicts quorums sensing activation patterns in the 5-strain community. .......................................................................................................................... 123 5.2.3: The network state depends on the inoculation ratio of strains. .................................. 125 5.2.4: Using ComX perturbations to switch community-level QS activation states. .......... 127 5.3: Discussion ........................................................................................................................ 128 5.4: Materials and methods. .................................................................................................... 130 5.4.1: Bacterial strains and growth media and conditions. .................................................. 130 5.4.2: β-galactosidase assay. ................................................................................................ 131 5.4.3: Growth measurements. .............................................................................................. 132 5.4.4: Mathematical modelling and simulations. ................................................................. 132 5.5: Supplementary information. ............................................................................................. 134 Chapter 6: Concluding remarks .................................................................................................. 145 6:1: Impact of my work ........................................................................................................... 145 6:1:1 Robust method of quantifying quorum sensing interactions ...................................... 145 6:1:2 The plate based approach enabled us to understand how quorum sensing is influenced by quorum sensing crosstalk or quenching within diverse communities of cells ................ 146 6:1:3 Neural network approach to make accurate predictions and controlling quorum sensing activation in well-mixed bacterial populations .................................................................... 146 6:1:4 Criticality observed in disrupting large-scaled communication through quenching .. 147 6:2: Future directions ............................................................................................................... 147 References ................................................................................................................................... 150 vii List of Figures Figure 2.1: The sender receiver plate assay to measure the spatial propagation. ......................... 15 Figure 2.2: The degrader influences quorum sensing activation .................................................. 17 Figure 2.3: The degrader tunes the activation zone ...................................................................... 18 Figure 2.4: Theoritical comparisons with the experimental data .................................................. 22 Figure S2.1: The map for the plasmid in the sender.. ................................................................... 24 Figure S2.2: The map for the plasmid in the degrader. ................................................................ 25 Figure S2.3: The map for the plasmid in the receiver................................................................... 25 Figure S2.4: The change in fluorescent intensity per cell of the receivers as a result of coculture with the sender strain or the sender strain and the interferer strain. ............................................. 26 Figure S2.5: Analysis of growth of the activation for the case of no interference. ...................... 26 Figure S2.6: Sensitivity of results to the activation threshold value............................................. 27 Figure S2.7: Addition of the interferer strain did not significantly change the doubling time of the sender or the receiver. ................................................................................................................... 27 Figure S2.8: The plate assay experiment shown here for normalized interference level 0.9 shows that the activation of the receivers stop at a distance 2.3 mm. ...................................................... 28 Figure S2.9: Control experiment compares the activation dynamics of the sender and receiver alone or when paired with the wild type (WT) host that does not produce the AiiA enzymes ... 29 Figure S2.10: Simulation results for the activation time vs. distance for the receivers ................ 29 Figure S2.11: The model predictions made for the changes in the fluorescent intensity per receiver cell, with and without the AiiA species. ......................................................................... 30 Figure S2.12: Regulated vs constitutive production of AiiA in the interference strain. ............... 30 Figure S2.13: The parameters that dictate the dynamics of the activation zone ........................... 33 Figure S2.14: In simulations changing the production rate of the AHL and the degradation rate of the AHLs by the interferers........................................................................................................... 34 Figure S2.15: Change in activation dynamics as a result of growth interactions between species or AHL internalization .................................................................................................................. 35 Figure 3.1: An experimental assay to quantify crosstalk between bacterial quorum sensing signals. .......................................................................................................................................... 39 viii Figure 3.2: The assay captures a range of crosstalk behaviors. Schematic of the crosstalk assay. ....................................................................................................................................................... 41 Figure 3.3: The dependence of activation dynamics on the number of interactor cells. .............. 43 Figure 3.4: Robustness of the activation of gene expression to crosstalk .................................... 47 Figure 3.5: Testing the model and experiments with natural isolates........................................... 49 Figure S3.1: Plasmid maps. .......................................................................................................... 55 Figure S3.2: Plate reader data for detecting the fluorescent changes in the receivers. ................. 56 Figure S3.3: Sensitivity of results to the activation threshold value............................................ 57 Figure S3.4: Control experiments ................................................................................................. 57 Figure S3.5: Growth times of the strains and mixtures. ............................................................... 58 Figure S3.6: A statistical mechanical model of two different AHL signals binding to a signal receptor. ........................................................................................................................................ 59 Figure S3.7: The energy and multiplicity of each unbound and bound state. .............................. 59 Figure S3.8: The Boltzmann weights for each state. .................................................................... 60 Figure S3.9: Fitting the experimental data to obtain interaction weights. .................................... 63 Figure S3.10: Testing the effects of the growth rates of the interactors on crosstalk ................... 63 Figure S3.11: Testing the effects of the production rates of the interacting AHLs on crosstalk. . 64 Figure S3.12: Testing the effects of the diffusion coefficient of the AHLs of interactors on crosstalk. ....................................................................................................................................... 64 Figure S3.13: Testing the effects of the degradation coefficient of the interacting AHLs on crosstalk. ....................................................................................................................................... 65 Figure S3.14: The influence of the non-cognate AHL binding energy on crosstalk. ................... 66 Figure S3.15: The influence of the weights of the AHLs on crosstalk ......................................... 66 Figure S3.16: Simulated activity profiles of the receivers (per receptor) under well mixed conditions in the presence of signal coming from both a sender strain and an interactor strain .. 66 Figure S3.17: The influence of feedback between interactors and senders on activation of the receivers. ....................................................................................................................................... 67 Figure S3.18: The effect on the growth of the senders and receivers due to P. aeruginosa ........ 68 Figure S3.19: The impact on crosstalk with non-quorum sensing interactions. ........................... 69 Figure S3.20: Simulating the impact of Pseudomonas aeruginosa growth influences on quorum sensing activation in the receiver strain. ....................................................................................... 70 ix Figure S3.21: Observing crosstalk and signal degradation in a well-mixed setup ....................... 71 Figure S3.22: Robustness of the network to interference when there is a large excess of inhibitory interactors ..................................................................................................................... 71 Figure S3.23: Spatial distribution of cells after 16 h. ................................................................... 72 Figure 4.1: Experimental assay to measure spatial patterns of cellular communication in the presence of interference ................................................................................................................ 76 Figure 4.2: Experimental demonstration of a degrader-dependent percolation transition............ 79 Figure 4.3: Scaling quantities from experiments. ......................................................................... 81 Figure 4.4: Simulation of quorum sensing activation in the sender and degrader community. ... 84 Figure 4.5: Changing the production rate of AiiA shifts the critical density ................................ 86 Figure S4.1: Senders mixed with degraders ................................................................................. 92 Figure S4.2: Choosing the threshold to determine quorum sensing activation ............................ 93 Figure S4.3: 20x microscopy images of the senders mixed with a dummy strain at high density. ....................................................................................................................................................... 93 Figure S4.4: Senders maintain uniform coverage of the plate at higher ratios of degraders ........ 94 Figure S4.5: Determining the amount of senders needed to uniformly cover the whole plate ..... 95 Figure S4.6: Changing the threshold slightly to observe changes to the amount of active pixels 96 Figure S4.7: A linear relationship is observed in the critical region for the number of activated sites and the amount of degraders. .............................................................................................. 103 Figure S4.8: Determining the critical amount of degraders (nAiiAc) ........................................... 104 Figure S4.9: Data range to calculate the critical exponents. ....................................................... 104 Figure S4.10: Sensitivity of the critical exponents to range of data used in fits.. ....................... 104 Figure S4.11: Sensitivity of the critical exponents to the threshold. .......................................... 105 Figure S4.12: Windowing the experimental images (2x2) and measuring the scaling properties. ..................................................................................................................................................... 106 Figure S4.13: Calculating the critical exponents of the transition from reaction-diffusion simulations for the average cluster size. ..................................................................................... 107 Figure S4.14: Calculating the critical exponents of the transition from reaction-diffusion simulations for the order parameter.. .......................................................................................... 108 Figure S4.15: Calculating the critical exponents of the transition from reaction-diffusion simulations for the correlation length. ........................................................................................ 109 x Figure S4.16: Finite-size scaling of the simulation results. ........................................................ 110 Figure S4.17: Critical exponents diverges from the classical values for simulations of smaller systems. ....................................................................................................................................... 111 Figure S4.18: A strain with a lower production rate of the degradative enzyme. ...................... 112 Figure S4.19: Well mixed simulations of the changes in the AHL concentration with respect to the amount of degraders. ............................................................................................................. 113 Figure 5.1: A neural network simplifies the complexity of QS crosstalk in diverse communities. ..................................................................................................................................................... 117 Figure 5.2: Experimental setup oto measure QS activity. .......................................................... 119 Figure 5.3: Pairwise measurements.. .......................................................................................... 122 Figure 5.4: Experimental validations of the model predictions when more than two signals are present. activation patterns for the community. .......................................................................... 124 Figure 5.5: Initial loading ratio of the producers dictates the final QS activation state. ............ 126 Figure 5.6: Perturbing the system to change the QS activation state.......................................... 128 Figure S5.1: Fluorescence per cell vs time for the testers when mixed with their cognate supernatants................................................................................................................................. 134 Figure S5.2: The fold changes in the LacZ expression of each tester A, B, C, D and E respectively ................................................................................................................................. 135 Figure S5.3: Doubling times for the producers and testers, see methods for further details ...... 136 Figure S5.4: The root mean squared error (indicated with the color intensity) was minimized for changing values of θ and f .......................................................................................................... 137 Figure S5.5: The best fit curves for f and θ. ............................................................................... 138 Figure S5.6: The quorum sensing activation landscape for the tester A when mixed with the supernatant of producer A and the supernatant of producer A, B, C, D and E respectively ...... 138 Figure S5.7: The quorum sensing activation landscape for the tester B when mixed with the supernatant of producer B and the supernatant of producer A, B, C, D and E respectively....... 139 Figure S5.8: The quorum sensing activation landscape for the tester C when mixed with the supernatant of producer C and the supernatant of producer A, B, C, D and E respectively....... 139 Figure S5.9: The quorum sensing activation landscape for the tester D when mixed with the supernatant of producer D and the supernatant of producer A, B, C, D and E respectively. ..... 140 xi Figure S5.10: The quorum sensing activation landscape for the tester E when mixed with the supernatant of producer E and the supernatant of producer A, B, C, D and E respectively ....... 140 Figure S5.11: The overall pairwise quorum sensing interaction map. In the left, the green arrow represents activation, the red flat head represents inhibition. ..................................................... 141 Figure S5.12 : LacZ fold change comparisons of the 5 supernatant case from simulations and experiments. Here the values indicated are the fold change in LacZ when increasing SE from simulations (left) and experiments (right)................................................................................... 142 Figure S5.13: Simulation results of the ComX concentration normalized by Ɵ over time ........ 142 Figure S5.14: Simulation results of the ComX concentration normalized by Ɵ over time for P A ..................................................................................................................................................... 143 Figure S5.15: The comparisons of the pairwise crosstalk of our results with Ansaldi et al. ..... 144 List of Tables Table 2.1: The values of the parameters found in Equations 1-3. ................................................ 20 Table 3.1: The weight parameters ................................................................................................. 47 Table 3.2: The bacterial strains used in this study. ....................................................................... 53 Table S3.1: The parameters used in the simulations..................................................................... 62 Table 4.1: Values of the critical exponents. .................................................................................. 85 Table 4.2: Primer sequences used for making the plasmids. ........................................................ 90 Table S4.1: The values of the parameters used for the modeling. .............................................. 102 Table S4.2: The values of the critical exponents from full and windowed experimental images ..................................................................................................................................................... 106 Table S4.3: The values of the critical exponents from the finite scaled simulation results ........ 111 Table S4.4: The values of the critical exponents for the case of the degrader with lower production rate ............................................................................................................................ 112 Table 5.1: The best fit values for fi and θi. ................................................................................. 121 Table S5.1: Parameter values used in the study for the modeling. ............................................. 136 1 Abstract Quorum sensing is a process in which microbes exchange molecular signals to coordinate cellular activity within populations of cells. Although typically viewed as a mechanism for a single bacterial species to measure their own cell density, recent work has shown that quorum sensing enables cells to gain information about multiple facets of the local environment, including physical and chemical conditions as well as local community composition. Often in these environments, signaling activity is influenced by the activity of neighboring cells. Some cells produce signals capable of binding to receptors in other cell types, resulting in signal crosstalk between two cell types. Other neighboring cells produce enzymes that degrade signal in the local environment, a process known as quorum quenching. Although previous studies characterized several examples of quorum sensing crosstalk and quorum quenching, less work has been done to understand the role of such cell to cell interactions in modulating quorum sensing activation within communities of quorum sensing bacteria. Therefore, my major goals were to broaden our understanding of the importance of quorum sensing crosstalk and quenching on regulating quorum sensing activation within diverse communities of bacteria, to understand the biophysics of cell signaling with microbial communities. In this thesis, theoretical and experimental tools from both physics and quantitative biology, including thermodynamics, neural networks, synthetic gene circuits, and reaction- diffusion models, were used to find the rules that govern quorum sensing activation within diverse communities of cells. I observed that the spatial propagation of quorum sensing activation was robust to crosstalk but was sensitive to quorum quenching. The sensitivity to quorum quenching was utilized to observe that a critical amount of quorum quenching was required to shut down the large-scaled quorum sensing activation of a spatially dispersed bacterial colony. I confirmed that the breakdown of such large-scaled quorum sensing coordination followed a percolation transition. In a well-mixed group of quorum sensing bacteria I was able to quantify pairwise crosstalk to predict and control the quorum sensing outcomes within a five-strain community of Bacillus subtilis. With my work, we have been able to broaden 2 our understanding of the role of quorum sensing and quorum quenching in communities of bacteria. Such knowledge can be utilized to devise better methods to prevent quorum sensing regulated bacterial infections and to improve the implementation of cell signaling networks in synthetic microbial communities. 3 Chapter 1: Introduction 1.1: Quorum sensing crosstalk and quorum quenching in bacterial communities Bacteria are complex microorganisms generally living in highly diverse and crowded environments [1,2]. Microbial communities can be found in the ocean, soil and even in the human body. Communication within these cellular networks regulates the activity of each individual cell, as well as the emergent functions that are the result of coordination within the community. Quorum sensing is one major communication mechanism bacterial cells use to coordinate collective behaviors such as biofilm formation, antibiotic production and bacterial infection. Quorum sensing is the process whereby bacteria produce and secrete small chemical molecules known as autoinducers. Over time as cell density increases, the extracellular autoinducer concentration increases. When the autoinducer concentration exceeds a critical threshold, the bacterial population modulates its gene expression profile to activate high cell density behaviors [3–5]. Quorum sensing was first observed in the early 1970s by Nealson et al [6], where the marine bacteria Vibrio fischeri activated bioluminescence in late-exponential phase. Extensive work done in the late 20 th century led to the discovery that such quorum sensing processes are regulated by a variety of small molecule signals, including acyl homoserine lactones (AHL) produced by many Gram-negative bacteria and oligo peptides (autoinducing peptides-AIP) produced by many Gram-positive bacteria [7–9]. Distinct AHLs have been identified in different Gram-negative bacteria, with each having subtle changes such as different lengths of the carbon chain, while for Gram-positive bacteria the AIPs produced by each species have significant differences but within the same genus, each strain would produce similar AIPs. For instance, the Agr signal and ComX signal produced by Staphylococcus and Bacillus respectively, were significantly different, but the ComX signals produced by Bacillus subtilis and Bacillus mojavensis, although different, were structurally similar [10,11]. Even though quorum sensing is historically viewed as a process of a single species regulating its own gene expression, numerous reports have shown cell-cell signaling interactions between species contributed to regulation of 4 quorum sensing phenotypes [11–15]. Two main types of cell-cell interactions that influence quorum sensing activation are quorum sensing crosstalk and quorum quenching [16–18]. Crosstalk in general can be defined when two species, talking in two languages interpret each other’s conversations, even though they do not speak the same languages. Similarly, in the bacterial quorum sensing world, crosstalk implies that two bacterial species producing two different types of autoinducers exchange these signals and influence quorum sensing activation. In 1997, Bassler et al. [19] observed such crosstalk between Vibrio cholera, Vibrio parahaemolyticus and Vibrio harveyi when the autoinducers of by V. cholera and V. parahaemolytic influenced the quorum sensing activation of V. harveyi. Quorum quenching occurs when a species produces an enzyme that is capable of degrading autoinducers, therefore depleting the autoinducer concentration and inhibiting quorum sensing activation. For instance, in 1999, Dong et al. identified an enzyme known as AiiA, commonly found in Bacillus subtilis, that can interfere with quorum sensing activation by cleaving the lactone ring of acyl-homoserine lactones [20]. A main goal of this work was to understand how crosstalk and quorum quenching influence the spatial and temporal dynamics of quorum sensing activation within bacterial communities. In the recent years, many studies have identified several examples of crosstalk and quenching in bacterial communities. For instance, Wu et al. [14] characterized the QS pairwise interactions of the AHLs 3-oxo-C6-HSL and 3-oxo-C12-HSL on the LuxI/R and LasI/R QS systems. Scott et al. [16] conducted a similar but extensive study on the pairwise effects of the AHLs 3-oxo-C6 HSL, 3-oxo-C8-HSL and 3-oxo-C12-HSL on the LuxR, LasR, RpaR and TraR QS systems to understand how to construct higher-level genetic circuitry for the use in microbial consortia. Additionally, there have been many studies looking at the functionality of quorum quenching molecules in degrading AHLs [21–23]. Although certain crosstalk and quenching mechanisms were identified, there is a limitation in the efforts to understand the role of crosstalk and quenching in regulating quorum sensing expression in a community setup, largely due to the complexities that arise from quorum sensing activation being interdependent on species composition, spatial position, heterogeneity, growth and the surrounding environment [1,24,25]. Using certain concepts in Physics and Biology, such as thermodynamics, probability theory, reaction-diffusion modeling, synthetic biology, systems biology and neural networks, I tried to 5 understand the role of quorum sensing crosstalk and quorum quenching in governing quorum sensing activation in a community context. Furthermore, robust quantification of crosstalk and quenching was another objective of my study. This could potentially lead to strategies to ultimately control quorum sensing activation within a bacterial network. The importance of setting such goals was motivated by the variety of quorum sensing systems seen in diverse communities of cells. To date, approximately 105 quorum sensing systems have been identified [8,9] to produce 56 distinct variations of AHLs, 231 variants of AIPs and 45 variants of quorum quenching molecules [26]. Additionally, a number of non-AIP or non-AHL based quorum sensing systems and their autoinducers have been recognized such as the PQS, CAI-1, CSFs quorum sensing systems [27–29]. Furthermore, a global quorum sensing system, luxS, has been identified in both Gram-negative and Gram-positive bacteria. With such a large number of quorum sensing signals, quorum sensing crosstalk and quenching play a major role in shaping the behavior of the community. For example, out of the 300-500 bacterial species present in the human gut, about 10 species have been identified to participate in quorum sensing [30] and Thompson et al. [31] showed that an exogenous supply of autoinducers can change the whole composition and ratio of the species that exist in the gut. Additionally, supplementing quorum quenching enzymes has been known to reduce quorum sensing regulated virulence factors and hence reduce bacterial infections [20,32,33]. As stated above, such examples indicate why it is important to understand the role of crosstalk and quenching in quorum sensing activation. Quantifying these cell-cell interactions will enable prediction of quorum sensing activity within complex communities of cells. As quorum sensing has been known to be linked with bacterial infections and pathogenicity, understanding the role of quorum sensing crosstalk and quenching can potentially lead to better strategies in fighting against bacterial infections. The traditional use of antibiotics is a straight forward approach to fighting bacterial infections, but in general the use of antibiotics may also kill bacteria that are beneficial to human health [34,35]. Furthermore, bacteria are able to become resistant to antibiotics such that administering antibiotics might not be an effective strategy to reduce bacterial infections [36]. Therefore, it is crucial that we devise alternative methods, such as inhibiting virulence factors by controlling quorum sensing, to reduce the pathogenicity of bacteria. Although I do not describe specific methods for such strategies, I discuss the role of 6 quorum sensing crosstalk and quenching within a population and quantify these cell-cell interactions, which enabled us to predict and confirm how we can control quorum sensing activity within a given population of cells. Such approaches can be potentially used to construct and devise schemes to fight bacterial infections. Quantifying quorum sensing crosstalk and quenching can additionally be beneficial for systems biology as one of the major interest is to build bio-synthetic circuits using quorum sensing genes [16,37,38]. Quorum sensing crosstalk is largely considered as a problem when it comes to building these synthetic circuits as it is a factor that is often ignored and contributes to unwanted levels of gene expression [39]. Through robust quantification, I believe it will be potentially useful in making accurate predictions of the outputs of bio-synthetic circuits and improving their efficiencies. 1.2: A brief overview the thesis In this thesis the primary aim was to understand the role of quorum sensing crosstalk and quenching on quorum sensing activation in networks of bacteria. The study was additionally focused on quantifying crosstalk and quenching such that quorum sensing activation patterns can be predicted in complex communities to determine if it is possible to systematically control these complex activation patterns through the manipulation of these cell-cell interactions. To quantify quorum sensing crosstalk and quenching, I developed a spatio-temporal assay which uses a LuxI/R sender and a LuxR receiver [38,40]. In this assay, we measured the quorum sensing activation of the receivers when they are distributed on a plate which has a sender colony at the center. We interfere with the normal propagation of quorum sensing activation by introducing a strain which either produces a non-cognate autoinducer molecule or a quorum quenching enzyme. I was able to quantify the impact of crosstalk by introducing a weight parameter, which was extracted by comparing simulation results of a reaction-diffusion model with experimental results. These weights enabled us to demonstrate that theoretical predictions were consistent with the experimental observations. The results from the simulations and experiments confirmed that the spatial propagation of quorum sensing activation was largely robust to quorum sensing crosstalk, but it was strongly impacted by quorum quenching. 7 The strong impact of quorum quenching was utilized to disrupt the quorum sensing activation in a long distance spreading distribution of senders. Here, the senders were spread onto a substrate and I looked at how the quorum sensing activation behaves when the amount of quenchers increased. I observed that with the introduction of the quenchers, there are regions of inactivity within the distribution of senders and with increasing the amount of quenchers, the sizes of the inactive regions increased. There was a critical amount of quenchers at which the large connectivity of the quorum sensing active senders was completely destroyed. There were unique scaling properties within the structures of the quorum sensing active clusters that enabled us to deduce that the transition from large-scale quorum sensing activation to small-scale quorum sensing activation followed a certain biophysical rule known as a percolation transition. Due to the universality observed in a percolation transition, it has been a well-studied and attractive topic for many disciplines and fields, and it has been observed in numerous abiotic contexts [41–43]. Adding to a recent report [44], here we have proved that such a deep physical rule can exist within the world of bacteria. So far I have discussed how quorum sensing crosstalk and quorum quenching impact the spatio-temporal properties of quorum sensing activation. In communities of bacteria, cells are mixed well and there can be a multitude of quorum sensing signals in the local environment. In this general setup, to achieve the goal of understanding the role of cell-cell interactions, it is important to know how diverse communities of cells utilize mixtures of quorum sensing signals to regulate quorum sensing activation. In the human brain, neurons are faced with a similar situation where there are around a 100 billion neurons and each neuron having tens to thousands of synaptic signaling inputs such that a certain output will be generated [45,46]. Neural networks have been used to predict the output of a certain neuron by implementing weights that represent the strength and sign of a given interaction [47]. I implemented such a neural network model to understand if it is possible to predict and control quorum sensing activity in a Bacillus subtilis community which produced 5 variants of the ComX autoinducer. The pairwise weights between each autoinducers and each strain were quantified, and these weights were used to make accurate predictions of the quorum sensing activation patterns of the full 5-strain community. Further simulations and experiments concluded that, within a community of cells that undergo quorum sensing crosstalk, the quorum sensing activation state depends on multiple factors such as the ratio of the autoinducers and the loading ratio of the strains. Additionally, through time 8 sensitive external perturbations, I was able to control the quorum sensing activation state of strains within a community. This work shows that quorum sensing crosstalk and quorum quenching will impact the patterns of quorum sensing activity within a community of cells. The results described in this thesis are novel and can be of potential use to many fields of study as discussed previously. Throughout the manuscript, these studies will be described in detail and the conclusions will be presented in Chapter 6. Moreover, in Chapter 6 I will additionally discuss potential future directions of this work. 1.3: Summary of the chapters In Chapter 2, I describe how the spatial activation of quorum sensing in the LuxI/R system can be tuned using a quorum quenching strain which produces the degradative enzyme AiiA. I describe how we can quantify the quorum sensing activation distance with respect to the amount of interfering strain that was used. In Chaper 3, I discuss how to quantify quorum sensing crosstalk between the LuxI/R, LasI and RhlI quorum sensing systems when species are distributed spatially. I describe how we can quantify such interactions using weights and showed that the quorum sensing activation times are sensitive to the species composition and type of AHL present. In Chapter 4, I discuss the observation of a percolation transition when a critical amount quenchers (named as AiiA interferers) are introduced to LuxI/R quorum sensing species, which have spanned quorum sensing activation at a large distance. In Chapter 5, a neural network model is implemented to simplify and predict the quorum sensing crosstalk between five Bacillus subtilis strains producing five variants of the ComX autoinducer. In this study we were able to deduce that in a given mixture of quorum sensing species, the quorum sensing activation depended on autoinducer concentration, species concentration and time sensitive external perturbations. In Chapter 6, I explore the impact of my work and future directions. 9 Chapter 2: Signal destruction tunes the zone of activation in spatially distributed signaling networks This work appears as published in Biophysical journal. Volume 112, issue 5, p1037-1044, March 14, 2017. 2.1: Abstract Diverse microbial communities coordinate group behaviors through signal exchange, such as the exchange of acyl-homoserine lactones (AHLs) by Gram-negative bacteria. Cellular communication is prone to interference by neighboring microbes. One mechanism of interference is signal destruction through the production of an enzyme that cleaves the signaling molecule. Here we examine the ability of one such interference enzyme, AiiA, to modulate signal propagation in a spatially distributed system of bacteria. We have developed an experimental assay to measure signal transduction and implement a theoretical model of signaling dynamics to predict how the system responds to interference. We show that titration of an interfering strain into a signaling network tunes the spatial range of activation over the centimeter length scale, quantifying the robustness of the signaling network to signal destruction and demonstrating the ability to program systems-level responses of spatially heterogeneous cellular networks. 2.2: Introduction Microbes use small molecule signals to coordinate group behaviors. For example, many Gram-negative bacteria exchange acyl-homoserine lactones (AHLs) to regulate quorum sensing and related emergent behaviors such as biofilm formation, collective motility, bioluminescence, and virulence [48–51]. Activation of quorum sensing involves the production of a signal by a synthase, for example LuxI used here which was isolated from Aliivibrio fischeri. The signals exit the cell either by passive diffusion through the membrane or active transport [52], and can be detected in neighboring cells by specific receptors, here the receptor LuxR. Binding of the signal to the receptor leads to global changes in gene expression. The signals enable bacteria to relay information about their physical, chemical, and biological environment [7,53–56]. 10 Communities that coordinate behavior through such mechanisms often have a fitness advantage [57,58]. Not surprisingly, other bacteria have evolved mechanisms to interfere with the signaling process. Bacteria exhibit various signal interference mechanisms [1,3]; collectively, these mechanisms that interfere with bacterial cell communication have been termed as quorum quenching [1,23,59]. The two main mechanisms of quorum quenching found in natural communities are signal crosstalk and signal destruction. Signal crosstalk between related signals can result in competitive binding to the receptors [1,15,24], and some bacteria produce compounds whose only known function is receptor antagonism [1,23,59]. Signal destruction often involves the production of an enzyme that chemically modifies or cleaves signaling molecules. The influence of signal destruction on signaling propagation through space has not been systematically explored. Here we develop an experimental assay to quantify multispecies pattern formation. The assay enables direct measurement of the robustness of signal exchange to the introduction of a variable amount of a signal degrading strain. Since initial work by Basu et al. demonstrating the ability to program pattern formation through signal exchange [37], several studies have examined many aspects of bacterial pattern formation. Experimental work has examined scale invariant patterns [60], the impact of signal removal on long distance signaling [61], tuning the concentration dependence of the bacterial response to signal [62], multispecies pattern formation that was programmable and robust to environmental perturbations [63], the control of patterns formation coupled to motility [64], and the ability of signal degradation to impact wildtype behaviors such as biofilm formation and swarming [33]. Theoretical advances have also deepened our understanding of the sensitivity of patterns to model parameters [65]. Here we build upon these previous studies to determine what level of interference can be tolerated by a microbial network exchanging AHLs, and predict and experimentally validate changes in the signaling patterns in the presence of interference. Interference by signal destroying enzymes is a potential route to control bacterial community function in industrial and biomedical applications. For instance, signal destruction can be used to limit cell-cell communication in microbes, which has been directly linked with pathogenicity [1,23,59]. The ability to precisely program and control emergent behaviors of 11 microbial networks through molecular-level manipulation requires a thorough, quantitative understanding of signal exchange and interference in mixed communities. Although previous studies (13, 15, 16,19) have identified many natural systems with signal interference, no quantitative measurements have been taken to understand the robustness of signal exchange in spatially heterogeneous systems in the presence of interference. Here we developed an experimental assay to examine the influence of a signal destroying enzyme, the lactonase AiiA, on the spatiotemporal dynamics of signal exchange in a multi-strain synthetic signaling network. The assay measures the robustness of signaling to interference. We find that AHL signaling in the presence of an interference strain producing the enzyme AiiA results in a limited zone of activation. Using a theoretical model, we show that the size of this zone is tunable by adjusting the number of interfering cells and signal producing cells. These results demonstrate that signal interference limits the spatial range of activation in the vicinity of a signal producing colony and can be implemented to predictably program the response of multi-strain bacterial communities. 2.3: Materials and Methods 2.3.1: Bacterial strains and plasmid The sender strain is Escherichia coli (NEB 5-alpha) with the plasmid ptD103IuxI sfGFP(Kn) (Figure S2.1) from [66] . This strain produces the signal N-(3-oxohexanoyl)-HSL and the cognate receptor LuxR. The GFP reporter gene activates in response to quorum sensing activation. The interfering strain consists of an Escherichia coli host harboring the plasmid ptD103aiiA(Cm), see Figure S2.2A and [66]. The receiver strain is Escherichia coli (NEB 5-alpha) with the plasmid pTD103LuxR RFP, shown in Figure S2.3. In order to construct the receiver strain, the plasmid pTD103luxI sfGFP(Kn) was mutagenized by inserting mCherry in position of sfGFP, using Gibson assembly kit (New England Biolabs). The receiver strain plasmid expresses mCherry fluorescent protein in response to quorum sensing activation. 12 2.3.2: Culturing conditions Bacterial strains were taken from frozen stocks and grown in a 14 mL Falcon tube with 5 mL of LB broth with appropriate antibiotics for plasmid maintenance. Cultures were grown in a shaker at 220 RPM and 37 o C. After 16 hours of growth, the culture in the late log phase was taken out and 1 mL of the inoculum was transferred into a micro-centrifuge tube. The tube was centrifuged at 15000 rpm for 1 min and the supernatant was discarded. The cells were then resuspended by vortexing in LB broth without antibiotics. Late log phase cultures were used such that quorum sensing of the sender strain was activated before measurement in the plate assay. 2.5% LB agar was used in the plate assay. During time lapse experiments the cells were imaged at 37 o C. 2.3.3: Interference assay 10 µL of the receiver strain was mixed with the interference strain in a micro-centrifuge tube. The mixture was loaded onto an LB agar plate by pipetting, and cells were distributed evenly using 4 mm diameter sterilized glass beads. After 10 minutes, 1 µL of the sender strain was pipetted to the middle of the plate. The plate was imaged using a fluorescent microscope for time-lapse experiments. 2.3.4: Microscopy measurements The images were taken using a Nikon eclipse TI fluorescent microscope. Time lapse experiments were carried out at 37 o C using a temperature controlled chamber. Samples were imaged with GFP and RFP illumination at a magnification of 20x. RFP images were taken every 15 minutes for 16 hours at 30 different distances from the sender colony, extending in two directions outward from the sender colony. Activation times were calculated at each position. Exposure time at each position was 1 s for RFP and 500 ms for GFP. No significant photobleaching was observed. Each image taken was saved in .tiff format and analyzed using a custom Matlab code. A low threshold was applied to the RFP images to identify the location of receiver cells within each 13 image. An upper threshold was used to identify the receivers that had activated quorum sensing. For each time point and position, the fraction of cellular pixels above the quorum sensing activation threshold was calculated. If the fraction of activated pixels exceeded 10%, that position was included as part of the activated region. 2.3.5: Growth measurements To obtain the growth curves the cells were grown in LB media with appropriate antibiotics as mentioned above. After 16 hours growth we diluted 5 µl of the cells in 4995 µl of LB. Dilutions of the culture were plated on selective media to measure cell density over time. To test the growth influences on the receivers and senders when in coculture with the interferes, we diluted 5 µl of the senders (or receivers) with 5 µl of the interferers in 4990 µl of pure LB broth and the previous procedure was repeated by selecting for the senders and receivers using proper antibiotic plates. 2.3.6: Plate-reader measurements Tecan Infinite m200 Pro plate reader was used to measure growth rates and fluorescence in well-mixed conditions. Cells were grown in LB media as mentioned above. Cells were then diluted 1000 fold in LB media containing antibiotics, and cultured for an additional 3 hours. After three hours of growth, 200 µl of these early log-phase cells were loaded into a flat bottom 96-well plate. The plate was inserted into the plate reader set to 37 o C and the optical density and fluorescent intensity were measured every 15 minutes for 15 hours. Optical density measurements were carried out at a wavelength of 600 nm. For GFP measurements, a wavelength of 485 nm was used for excitation and a wavelength of 515 nm was used for emission. For mCherry fluorescence measurements, a wavelength of 590 nm was used for excitation and a wavelength of 650 nm was used for emission. In the supplementary materials we introduce a method to analyze the fluorescence of a single species from a mixed culture. 14 2.4: Results and Discussion 2.4.1: Quantifying signal exchange in spatially distributed microbial networks An assay was developed to examine the dynamics of signal exchange in multistrain bacterial communities. The assay is based on previous reports that used solid agar plates to measure the response to the diffusive exchange of signaling molecules (19,20,21,22). As shown in Figure 2.1A, our system consists of a 2.5 % LB plate with a receiver strain evenly distributed on top of the plate. The receiver strain does not produce the signal (AHL); however, it produces the receptor LuxR and increases the expression of a fluorescent reporter gene, mCherry, in response to AHL signal. A sender strain containing the synthase gene luxI produces the AHL signal and is pipetted onto the center of the plate. The sender strain makes the receptor LuxR, and as in the natural quorum sensing system the synthase is positively regulated by AHL. The sender strain produces green fluorescent protein (GFP). Control measurements reported in Figure S4 show that the fluorescence intensity of the receiver strain increases upon coculture with the sender strain. E. coli strains contained plasmids based on constructs reported in [66], see Figures S2.1-2.3. Similar sender/receiver systems have been developed and investigated in previous work [37,69–75]. Upon adding the sender strain to the center of the plate, the sender strain excretes AHL signal, which diffuses into the lawn of receiver strains to activate expression of a fluorescent reporter gene (mCherry). As shown in Figure 2.1B, the receiver cells closest to the sender cells are activated first and over time, this region of activation expands outwards. Fluorescent images, shown in Figure 2.1B, represent the stitching together of images from several microscope positions taken at 20X magnification such that activation can be observed at the single-cell level over centimeter length scales. The dynamics of signal exchange in the system can be quantified by measuring the time at which receiver cells activate at a specific distance from the center of the sender cells. We set a threshold pixel intensity to indicate this transition from the basal to activated rate of fluorescence expression, see Figure S2.5. Quorum sensing activation at a given position is defined as 10% of the pixels belonging to cells being above this threshold. In Figure S2.6 we show that the activation dynamics are not strongly dependent on the value of the 15 activation threshold used in analysis. Figure 2.1C shows the dynamics of activation at distances of up to 12 mm from the sender colony, the size of the plate used in the assay. Figure 2.1: a) The two strain signaling network consists of a sender strain (GFP) producing AHL signal and a receiver strain (mCherry) containing an AHL-regulated fluorescent reporter gene. In the plate assay, the receivers are evenly distributed on an LB-agar plate, and a colony of sender strains is added to the middle of the plate. b) Time-lapse fluorescent imaging captures the activation of the receiver cells as AHL diffuses outward from the sender colony. The green fluorescence shows the location of the sender strain. c) Activation time of the receivers as a 16 function of distance from the sender colony. The data points have been fitted to exponential curve. 2.4.2: AiiA inhibits quorum sensing in a synthetic microbial community To extend the signaling assay to the examination of interference between strains, we first validated the interference capabilities of a strain producing the enzyme AiiA, a lactonase originally isolated from Bacillus thuringiensis [1,23,59]. A plasmid containing aiiA, from [66], was transferred to E. coli to create the interference strain and added to the sender and receiver community, as depicted in Figure 2.2A. Figure 2.2B shows the activation of quorum sensing for the sender strain alone and in coculture with the AiiA interference strain. Here, the fluorescence intensity of the sender strain was measured in a 96-well plate reader under well-mixed conditions. Coculture with the interference strain reduced the level of quorum sensing activation, indicating that the AiiA producing strain interfered with AHL signal transduction. We did not observe significant growth influences that can be caused by the interfering strain on the senders or receivers, see Figure S2.7. Next we both experimentally and theoretically examine the influence of AiiA-mediated 17 interference in spatially distributed networks of cells. Figure 2.2: a) To quantify the impact of interference on signal exchange, a third strain is added to the system that produces the enzyme AiiA, a lactonase that cleaves AHL signal. b) Addition of the interference strain to the sender and receiver community reduces the fluorescent intensity per cell of the receiver strain. Here, the negative control (green) shows that the addition of the wild type (WT) yields a similar curve to the case of no interference. Errorbars show standard deviation of 3 replicate measurements. 2.4.3: Signal destruction leads to finite and tunable zones of activation in spatial networks The interference strain was added to the plate-based assay to quantify the influence of signal destruction on the signaling dynamics. The interference strain was added to the lawn of receiver cells, as depicted in Figure 2.3A. Different amounts of the interfering strains were titrated into the receiver cells to measure the robustness of signal propagation to a variable level of interference. Figure 2.3B shows the impact of different levels of interference to signaling dynamics. The no interference data, from Figure 2.1C, is shown for reference. As the level of interference increased, we observed a slight delay in quorum sensing activation. Additionally, the system only activated up to a finite distance. For example, at a normalized level of interference of 0.9, 18 receiver cells at a distance of 2.3 mm activated at about 5.5 hours. The experiment continued to run for a total time of 16 hours and activation beyond 2.3 mm was not observed, see Figure S2.8. The radius of the activation zone was inversely proportional to the level of interference. As seen in Figure S2.7, there are no significant growth influences between the strains. To confirm that this effect was due to production of the AiiA enzyme, in control experiments a wildtype strain that does not produce AiiA was substituted for the interference strain (Figure S2.9). Figure 2.3: a) Introducing the AiiA cells in the plate assay setup. b) The activation time v. distance from the sender plot for variable levels of interference. The number of interference cells in the assay is normalized to the number of cells in 100 µL of cells in a culture with an absorbance of 1.0 at 600 nm. Curves show an exponential fit to show qualitative trends in the data. Experiments were run for 16 hours, much longer than the time needed to observe the maximum radius of activation. 2.4.4: Modeling the impact of signal destruction on signal propagation A model for signal exchange in the presence of interference was derived to further explore the ability of signal destruction to limit the zone of activation. Our model, based on previously reported models from [66,76] accounts for the growth of the sender strain, the receiver strain, and the interference strain. All three strains follow the logistic growth equation: 𝜕𝑛 𝜕𝑡 = 𝜇 𝑛 (1 − 𝑛 𝑇𝑜𝑡𝑎𝑙 𝑠 ) (2.1) 19 where, 𝜇 is the growth rate constant, n is the cell density, 𝑛 𝑇𝑜𝑡𝑎𝑙 is the total cell density (senders + receivers+ interferers) and s is the maximum cell density in the system. All three strains use the same E. coli host and were shown to have similar growth characteristics, see Figure S2.7. Figure S2.7 shows that there are no growth influences on the sender and the receivers when cocultured with the interferer. Previously, several models have been used to describe quorum sensing signaling [66,76– 78], including signaling in the presence of signal destruction [66]. The dynamics of the AHL signal, [AHL], by the sender strain can be described by the following equation: 𝜕 [𝐴𝐻𝐿 ] 𝜕𝑡 = 𝐷 𝐴𝐻𝐿 𝛻 2 [𝐴𝐻𝐿 ] + 𝑛 𝐴𝐻𝐿 ( 𝜌 𝐴𝐻𝐿 [𝐴𝐻𝐿 ] 𝑚 1 [𝐴𝐻𝐿 ] 𝑚 1 +𝜃 1 𝑚 1 + 𝜌 𝑏 1 ) − 𝛾 1 [𝐴𝐻𝐿 ] − 𝛾 2 [𝐴𝑖𝑖𝐴 ] [𝐴𝐻𝐿 ] 𝑚 2 [𝐴𝐻𝐿 ] 𝑚 2 +𝜃 2 𝑚 2 , (2.2) where, DAHL is the diffusion coefficient for the signal, 𝑛 𝐴𝐻𝐿 is the cell density of the senders, 𝜌 𝐴𝐻𝐿 is the maximum production rate of the AHL per cell, 𝜌 𝑏 1 is the basal level of AHL production, 𝜃 𝑖 represents the concentration of the signal at half maximum activity, mi represents the cooperative coefficients, 𝛾 1 is the basal rate of signal degradation in media [76,79], 𝛾 2 is the rate of signal degradation due to the AiiA, and [𝐴𝑖𝑖𝐴 ] is the concentration of AiiA signal. The Hill’s function represents a threshold amount of signal needed to activate quorum sensing up to a maximum production rate. The change in the amount of AiiA is described by, 𝜕 [𝐴𝑖𝑖𝐴 ] 𝜕𝑡 = 𝑛 𝐴𝑖𝑖𝐴 (𝜌 𝐴𝑖𝑖𝐴 [𝐴𝐻𝐿 ] 𝑚 3 [𝐴𝐻𝐿 ] 𝑚 3 +𝜃 3 𝑚 3 + 𝜌 𝑏 2 ) − 𝛾 3 [𝐴𝑖𝑖𝐴 ], (2.3) where, 𝜌 𝐴𝑖𝑖𝐴 is the maximum production rate of the AiiA per cell, 𝜌 𝑏 2 is the basal level of AiiA production, 𝑛 𝐴𝑖𝑖𝐴 is the cell density of the AiiA cells, and 𝛾 3 is the basal rate of AiiA degradation. The values of all the parameters are given in Table 2.1. Given the time scale of our simulations, we assume that the degradation rate of the AiiA will be low and set it to the degradation rate of the AHLs. In the SI we estimate the number of AiiA molecules per cell, and conclude that the production rates used give a reasonable level of protein per cell. In the simulations, we solve the equations using the finite difference method. The interference strain is evenly distributed in the simulation grid, at various loading densities. 20 Activation occurs at a distance X when the concentration of AHL signal at that point reaches the threshold concentration. Growth rate constants were experimentally obtained, as shown in Figure S2.7A. Additional parameters were reported in previous studies. Parameter Value 𝜇 1.50 hrs -1 (experimentally calculated) s 10 9 cells per ml DAHL 1.764 mm 2 hrs- 1 [77] 𝜌 𝐴𝐻𝐿 2.3 x 10 -9 nmol hrs -1 per cell [76] 𝜌 𝑏 1 2.3 x 10 -10 nmol hrs -1 per cell [76] 𝜃 1 70 nM [76] 𝜃 2 70 nM (32) m1 2.5 (32) m2 2.5 (32) 𝛾 1 0.005545 hrs -1 [76] 𝛾 2 0.01 hrs -1 [66] 𝜌 𝐴𝑖𝑖𝐴 2.3 x 10 -9 nmol hrs -1 per cell 𝜌 𝑏 2 2.3 x 10 -10 nmol hrs -1 per cell 𝛾 3 0.005545 hrs -1 𝜃 3 70 nM [76] m3 2.5 (32) Table 2.1: The values of the parameters found in Equations 1-3. As shown in Figure 2.4A, simulations of the dynamics of activation in the plate-based assay qualitatively matched activation profiles from the experimental measurements shown in Figure 2.1. Next the interference strain was incorporated into the simulations. As shown in Figure 2.4B, model predictions for the plate-based assay are in close agreement with experimental results (see Figure S2.10 for simulation predictions of the well-mixed system). As the percentage of interference strain increased, the activation zone shrank from 10 mm down to 2 mm, demonstrating that the amount of interfering cells within the system tunes the radius of 21 activation. As in experiments, simulations were run for 16 hours, much longer than the time needed to reach steady-state activation profiles. In these experiments expression of aiiA is regulated by the AHL quorum sensing signal, although simulations and experiments shown in Figure S2.11 demonstrate that strains without AHL-dependent aiiA expression exhibit similar activation dynamics. In simulations and experiments we varied the number of cells in the sender colony and observed that when the number of cells in the sender strain was increased or decreased by a factor of 5, the activation range curve shifted (Figure 2.4C). Together the ratio of sender to interference strains dictates the spatial extent of activation. To further explore the parameters that dictate the dynamics of the activation zone, in simulations we varied multiple model parameters including the diffusion coefficient and the degradation of AHL due to the media (Figures S2.12 and S2.13). We observe that the major contributors to the activation zone are the diffusion of the AHLs and the degradation of the AHL by the interferer. In Figure S2.14 we demonstrate that the potential growth interactions between the strains and internalization of the signal by non-receiver strains also influence spatiotemporal activation patterns. 22 Figure 2.4: a) A model using Equation 1-3 simulates receiver cell activation in the plate-based assay (top) and the experimental comparison is shown below. Here the normalized interfernce level is 0.9. b) The maximum activated radius of the receivers vs. the normalized amount of interference cells. The experimental measurements (blue dots) agreed with model predictions (red line). c) Theoretical predictions and experimental data points for the activation radius when the number of sender cells is increased by 5X (blue line and data points) or decreased to 0.2X (green line and data points). Error bars show the standard deviation of three replicate measurements. 2.5 Conclusions We have demonstrated that signal interference with a bacterial communication pathway through signal destruction creates a finite region of activation around a colony of signal emitting cells. Similar spatial activity patterns have been programmed using external chemical gradients or synthetic signaling circuits [37,62,64,80]. Here a plate-based experimental assay was 23 implemented to quantify the influence of interference on signal transduction in a spatially heterogeneous system. The assay measurements were supported by a reaction-diffusion simulation of signal dynamics, demonstrating the utility of such models in predicting signal exchange in systems with complex geometries that could not easily be analyzed experimentally. Both the theoretical and experimental results revealed the capability of tuning the system- level parameters of a multi-strain bacterial community. By adjusting the amount of interference strain, the zone of activation was tunable from 2 to 10 mm. Theoretical predictions and experimental results shown in Figure 2.4C demonstrate that parameters such as the number of sender cells also modulated the radius of activation. Similar theoretical models of signaling have explored the network parameters that regulate bacterial pattern formation [60,65]. Such strategies for adjusting the dynamics of multispecies interactions might prove useful for programming the activity of synthetic communities. Recent interest in engineering microbial communities to perform complex tasks [69,74,81], such as nanoscale synthesis and in biomedical applications [37,82–84], call for new experimental techniques to control the network-level behavior of microbial communities [2,82,85,86]. Although characterization of diverse communities often focuses on well-mixed systems, non-linearities in the interactions can result in unexpected outcomes, particularly in spatially heterogeneous systems [70,87,88]. The plate-based assay developed here may prove useful to quantify the outputs of spatially distributed, multispecies networks [89]. Such approaches identify control parameters, such as the level of signal destruction shown here, that strongly determine global behaviors. These experiments, in conjunction with theoretical models, will enable us to understand and predict the parameters that determine community behavior in multispecies bacterial communities. Our results suggest that in natural systems even low levels of interference will modulate the dynamics of signal propagation, and hence the functional outputs of the community. Many natural systems are spatially heterogeneous, and the consequences of specific spatial distributions could be predicted in simulations. It remains unclear to what extent interference constrains signaling to local areas. Recent studies have examined quorum sensing that occurs locally within small populations [90,91], but there are examples of long distance coordination of cellular behaviors known to be coordinated by signal exchange [52,92] . Given the number of 24 quorum quenching and signal interference mechanisms present in natural communities (13, 16, 17), future work should examine the contribution of different types of interference beyond signal destruction, such as receptor blocking through competitive binding, in shaping global signaling responses. 2.6: Supplementary Information 2.6.1: Plasmids Plasmids to create the sender and interferer were obtained from Addgene. The plasmids were inserted into E. coli Dh5alpha. The plasmids in the sender and interferer were based on study [66]. Figure S2.1: The map for the plasmid in the sender. luxI encodes for the production of the AHL 3-oxo-c6-HSL. luxR encodes for the production of the receptor protein LuxR. Once the AHL binds to the LuxR protein it could bind to the pluxI promoter and activate the expression of the genes luxI, luxR and sfGFP. 25 Figure S2.2: The map for the plasmid in the Interferer a) regulated under the pluxI promoter b) constitutive under the lacUV5 promoter. For the regulated interferer strain, the AHLs diffusing from the sender will bind to the LuxR protein, and activate the pluxI promoter and express the aiiA gene while for the constitutive interferer strain the aiiA gene will be produced constitutively. The aiiA gene will encode for the production of the degradative enzyme AiiA. Figure S2.3: The map for the plasmid in the receiver. The pluxI promoter will activate due to the diffusing AHLs from the sender. Once activated, LuxR and the mCherry fluorescent protein will be produced. We constructed the receiver, based on the ptD103 plasmid. The luxI and sfGFP genes were deleted from the ptD103luxI sfGFP plasmid and the mCherry sequence was added in place of the sfGFP gene. 26 Figure S2.4: The change in fluorescent intensity per cell of the receivers as a result of coculture with the sender strain or the sender strain and the interferer strain. Fluorescent intensity was measured in the plate reader at 37 o C as described in the main text. Figure S2.5: Analysis of growth of the activation for the case of no interference. At the threshold (10% of the maximal pixel value for the microscope), we see clear activation (white) of the receivers at a distance of 2 mm (bottom edge of the frame) from the senders. The images here 27 are taken from the microscope and the size of each frame is 750 µm. The receiver activates from bottom to top. Figure S2.6: Sensitivity of results to the activation threshold value. a) The pixel intensity of the receivers over time at a distance 2 mm from the sender and in a negative control of receivers without added sender. b) The calculated activation times when we change the threshold was raised or lowered by 10%. c) Upon applying the threshold, the number of pixels above threshold was counted for receiver cells 2 mm from the sender and in the no sender negative control. Figure S2.7: Addition of the interferer strain did not significantly change the doubling time of the sender or the receiver. a) Growth curves for the sender, receiver, and interferer strain alone and in coculture. In coculture experiments, selective plating enabling tracking of 28 individual species. b) Fits to the linear doubling times of the sender, receiver, and interference strains. The errorbars show standard deviation from four replicates. A standard unpaired t test was used to calculate the two tailed p value. Comparisons between the doubling times of the senders or receivers with and without addition of the interferer had p > 0.5. Figure S2.8: The plate assay experiment shown here for normalized interference level 0.9 shows that the activation of the receivers stop at a distance 2.3 mm. The experiments were done for 16 hours, and no propagation in the activation was observed. The dark areas within activated regions are the result of variability in the initial distribution of receiver cells 29 Figure S2.9: Control experiment compares the activation dynamics of the sender and receiver alone (black dots) or when paired with the wild type (WT) host that does not produce the AiiA enzymes (blue X’s). Plate-assay setup as described in the main text. Figure S2.10: Simulation results for the activation time vs. distance for the receivers 30 Figure S2.11: The model predictions made for the changes in the fluorescent intensity per receiver cell, with and without the AiiA species. Here we obtain the concentration profiles of the AHLs in a well-mixed setup and convert it to the fluorescent intensity per cell by multiplying with an arbitrary value. Figure S2.12: Regulated vs constitutive production of AiiA in the interference strain. a) Plate reader experiments shows that the constitutive interferer (purple) yields a similar response to the regulated interferer (red). b) Simulations confirms that the plate assay setup will yield a 31 similar response to the regulated case. To simulate the constitutive production of AiiA, we delete the Hill’s coefficient appearing in equation (2.3) such that the AiiA strain will constitutively produce the AiiA at a constant rate of production of 2.3 x 10 -9 nM hrs -1 per cell. 2.6.2: Calculation of the number of proteins per cell for the production rate of AiiA In this section we will justify that the production rate assumed for AiiA (2.3 x 10 -10 nmol hrs -1 per cell) in the modelling is reasonable. The system is simulated for 10 hours. At t=10 hours, the amount of AiiA per cell can be estimated as: 2.3 x 10 -10 nmol hrs -1 per cell x 10 hours= 2.3 x 10 -9 nmol per cell or 2.3 x 10 -18 mol per cell. In molecules, AiiA per cell would be: 2.3 x 10 -18 mol per cell x 6.022 × 10 23 molecules/mol ~ 10 6 molecules per cell. Hence, we estimate on the order of 10 6 AiiA proteins produced per cell at t=10 hours in the simulations, which is within the range of the standard number of proteins per bacterial cell [93], particularly for a production from a gene encoded on a plasmid. 2.6.3: Calculating the fluorescence in a mixed culture using the plate reader In this section our objective is to calculate the fluorescence of a species in a mixed culture. In the mixed culture, there will be two species, one with a particular fluorescence (ex: RFP) and the other strain will not have the same fluorescence. We experimentally measure the total fluorescence of the mixture using a plate-reader. We define the following terms, Fluorescence readout from the plate reader = Fl; Number of cells = n; Optical density readout from the plate reader = O; Background fluorescence per cell= α; Fluorescence per protein = ω; Number of fluorescent protein per cell = f; 32 Absorbance of light per cell = b; Let’s consider a general case where species A produces the fluorescence (ex:RFP) and species B doesn’t produce any or the same fluorescence as species A. There will be a background fluorescence due to the light scattering off the edges of the cell. First we will consider the total fluorescence of the mixed culture; Fltotal = n A f ω + n A αA +nB αB+ Flm (2.4) Where, Flm is the background fluorescence of the media. The corresponding optical density will be; Ototal = n A b +nB b + Om (2.5) Where, Om is the background fluorescence of the media. If species B is grown in a single culture the fluorescence of species B will be; FlB = nB αB+ Flm (2.6) The corresponding optical density of species B will be; OB = nB b + Om (2.7) Since we use E. coli (DH5alpha, NEB10beta) cells for all of the different species in our study, we assume that the background fluorescence per cell in all the cells is the same (αA = αB = α1). Now, the total fluorescence of the mixed culture for all the cells, Fl total −Fl 𝑚 O total −O 𝑚 = n A f ω+n A α 1 +n B α 1 n A β+n B β (2.8) The total fluorescence of species B in the single culture, Fl B −Fl 𝑚 O B −O 𝑚 = n B α 1 n B β (2.9) Now lets define ∆A = equation (2.8) - (2.9); ∆A = n A f ω+n A α 1 +n B α 1 n A β+n B β − α 1 β = n A f ω+n A α 1 +n B α 1 − (n A +n B )α 1 (n A +n B ) β 33 = n A f ω (n A +n B )β (2.10) Now from (2.5); (n A + n B )β = O total − O m (2.11) ∅ 𝐴 = n A f ω , is the total fluorescence of the species A in the mixture. Hence by using equations (2.10) and (2.11), ∅ 𝐴 = (∆A) (O total − O m ) (2.12) Hence, by taking the fluorescent readout (F) and the optical density readout (O) of the mixed sample and the single cultured B sample we can calculate the total fluorescence of the species A in the mixture. Finally we deduce the total fluorescence per cell = ∅ 𝐴 (O total −O m ) (2.13) Figure S2.13: To understand the parameters that dictate the dynamics of the activation zone we changed a) the sender AHL diffusion coefficient b) the sender AHL degradation rate by the media (𝜸 𝟏 ). In our experiment these values are constants. It can be seen that the degradation by the media is not a major contributor to the inhibition zone since even at zero degradation (𝜸 𝟏 =0, yellow line), we see no deviations from 𝜸 𝟏 (red line). If the AHLs are degraded by the media at a higher rate, the activation radius for lower normalized interference values decreases, but above a certain 𝜸 𝟏 value, the radius for higher normalized interference seems to be affected as well. This is due to the effect from 𝜸 𝟏 being comparable to the effect 34 from the interferer. Additionally, we changed c) the growth rate of the signal sender and observed no shift in the plot. This is due to the fact that the senders are already active when they are loaded into the plate such that levels of activation and degradation change proportionally. Figure S2.14: In simulations changing a) the production rate of the AHL, b) the degradation rate of the AHLs by the interferers. Compared to known quorum sensing production rates, the production rates used here are very large to have any significant effect on the plot. The plot is more sensitive towards the degradation rate of the interferers on the other hand, as seen in b), doubling or halving the degradation rate shifts the plot significantly. 2.6.4: Testing the model for effects of the growth interactions, diffusion coefficients, degradation rates and internalization of the signal To test the possible effect of growth influences on the sender-receiver system caused by the interferer we considered a competitive Lotka–Volterra equation [94], 𝜕𝑛 𝜕𝑡 = 𝜇 𝑛 (1 − 𝑛 𝑇𝑜𝑡𝑎𝑙 𝑠 ) −𝛼 1 𝑛 𝑛 𝑖𝑛𝑡𝑒𝑟𝑓𝑒𝑟𝑒𝑟 (2.14) where, 𝛼 1 is the growth effect the interfering species has on the sender-receiver system. To test the effect of signal internalization we modified equation (2.3) such that, 𝜕 [𝐴𝐻𝐿 ] 𝜕𝑡 = 𝐷 𝐴𝐻𝐿 𝛻 2 [𝐴𝐻𝐿 ] + 𝑛 𝐴𝐻𝐿 ( 𝜌 𝐴𝐻𝐿 [𝐴𝐻𝐿 ] 𝑚 1 [𝐴𝐻𝐿 ] 𝑚 1 +𝜃 1 𝑚 1 + 𝜌 𝑏 1 ) − 𝛾 1 [𝐴𝐻𝐿 ] − 𝛼 2 𝑛 𝑖𝑛𝑡𝑒𝑟𝑓𝑒𝑟𝑒𝑟 [𝐴𝐻𝐿 ] 𝑚 4 [𝐴𝐻𝐿 ] 𝑚 4 +𝜃 4 𝑚 4 (2.15) 35 where, 𝛼 2 is the AHL internalizing rate per interfering cell which is modulated by the number of AHLs present in the vicinity of the interferer. Figure S2.15: Change in activation dynamics as a result of growth interactions between species a) or AHL internalization b) To simulate these conditions, we used equations (2.14) and (2.15). For both conditions, we assume that there is no AHL degradation by the interference strain, the influence of the interferer on the receiver is solely due to either growth influences or AHL internalization. The red curve represent the usual case when the interferer degrades the AHL, plotted here for comparison. For b) we used, 𝒎 𝟒 = 2.5 and 𝜽 𝟒 = 𝟕𝟎 𝒏𝑴 . 36 Chapter 3: Quantifying the strength of quorum sensing crosstalk within microbial communities This work appears as published in PLOS Computational Biology. Volume 13, issue 10, e1005809, October 19, 2017. 3.1: Abstract In multispecies microbial communities, the exchange of signals such as acyl-homoserine lactones (AHL) enables communication within and between species of Gram-negative bacteria. This process, commonly known as quorum sensing, aids in the regulation of genes crucial for the survival of species within heterogeneous populations of microbes. Although signal exchange was studied extensively in well-mixed environments, less is known about the consequences of crosstalk in spatially distributed mixtures of species. Here, signaling dynamics were measured in a spatially distributed system containing multiple strains utilizing homologous signaling systems. Crosstalk between strains containing the lux, las and rhl AHL-receptor circuits was quantified. In a distributed population of microbes, the impact of community composition on spatio-temporal dynamics was characterized and compared to simulation results using a modified reaction-diffusion model. After introducing a single term to account for crosstalk between each pair of signals, the model was able to reproduce the activation patterns observed in experiments. We quantified the robustness of signal propagation in the presence of interacting signals, finding that signaling dynamics are largely robust to interference. The ability of several wild isolates to participate in AHL-mediated signaling was investigated, revealing distinct signatures of crosstalk for each species. Our results present a route to characterize crosstalk between species and predict systems-level signaling dynamics in multispecies communities. 3.2: Author Summary In nature, bacteria are commonly found in spatially heterogeneous mixtures. Within these environments, multiple species communicate using chemical signals, and crosstalk often governs the activities of microbial populations, including interactions with the host system, forming biofilms, and bioluminescence. Understanding such bacterial interactions is essential to control 37 and prevent these population-level behaviors regulated by signal exchange. Additionally, quantifying bacterial crosstalk will help improve the robustness of synthetic cellular networks that utilize signal exchange. Although cellular signaling is understood in well-mixed systems with one signal, we lack a detailed understanding of signaling in spatially distributed cellular networks or networks with multiple signals. We created an experimental system to observe and quantify microbial crosstalk between three bacterial languages. A mathematical model was implemented to predict the consequences of the exchange of multiple signals within cellular networks and good agreement between the experimental results and theoretical predictions was observed. In the mathematical model, a single parameter was sufficient to account for crosstalk between bacterial species. These experimental and theoretical tools enable us to better understand and predict how signaling influences the behavior of both natural and synthetic microbial communities. 3.3: Introduction Microbes communicate with each other in order to coordinate behavior and gene expression through a process known as quorum sensing. Several Gram-negative bacteria use acyl-homoserine lactones (AHLs) as a signal to communicate [3,51,52,95–97]. These signaling systems typically consist of a synthase, such as luxI, which produces a variant of AHL, and a receptor, such as luxR, which binds to AHLs. The receptor enacts global changes in gene expression in response to high concentrations of AHLs. Over 150 quorum sensing systems have been characterized [98,99], with most species containing one or a few signaling pathways. Each system typically produces one dominant version of AHL [98,99], and 56 different AHLs have been identified to date [98]. Variant versions of AHL involve changes in the length of the carbon chain extending from the lactone ring and chemical modifications of this carbon chain such as the addition of carbonyl groups [96,98]. Variation in the chemical structure of AHLs impacts both affinity for the receptor and the regulatory response [3,98,99]. Several examples of crosstalk between signaling microbes, in which signal produced from one species binds to the receptor of a second species, have been reported [3,15,24,40,95,100–102]. For example, Chromobacterium violaceum, a pathogenic Gram-negative bacteria that produces a shorter chained AHL, activates gene expression in Vibrio 38 harveyi, a Gram-negative marine bacteria that produces a longer chained AHL [40]. Each such pairing of AHL and receptor inhibits or promotes the activation of gene expression to a variable degree. Multispecies communities collectively produce complex mixtures of signals and the activation of gene expression within the community is influenced by crosstalk between different AHL variants. Quantifying AHL-mediated crosstalk will help us build a predictive understanding of the signaling dynamics within heterogeneous microbial populations, potentially enabling us to control the activation of gene expression in natural and synthetic microbial communities [16,39,40,95–97,103–105]. Here we develop and implement a signaling assay using sender, receiver, and interactor strains to measure signaling dynamics in populations containing multiple AHLs, see Figure 3.1A. This crosstalk assay assesses the robustness of signal exchange within mixed microbial communities. Here we use a sender strain using the AHL 3-oxo-C6 HSL and multiple interactor strains producing signals including 3-oxo-C12 HSL made by the synthase LasI and C4-HSL made by the synthase RhlI. Our work expands the scope of prior studies that focused on signal exchange in well-mixed environments, where diffusion of AHLs plays a minimal role [14,16,24,39,66,106,107]. Initial work done by Canton et al. [106] quantitatively explored the ability of multiple AHLs to bind to one receptor in a plate-reader. Wu et al. [14] used a microfluidic and flow-cytometry approach to measure the interactions between the lux and las circuits. Other studies such as McClean et al. [15] observed the response of the signaling system in Chromoacterium violaceum to purified variants of AHL. Although diffusion plays a role here, there was no significant quantitative work done to capture the spatial effects. In another study Dilanji et al. [25] looked at influences of a diffusive AHL wave front produced by an exogenously added chemical on agar plates. As an alternate to adding AHLs exogenously, we added a sender E. coli colony to the middle of the plate capable of synthesizing AHL molecules, as shown in Figure 3.1B and similar to previous experiments [108]. Our assay incorporates an interactor strain to determine the robustness of AHL signaling to crosstalk from several AHL signals, including signals produced by wild bacterial isolates. The exchange of multiple AHLs has been reported [69,102], but these studies minimized crosstalk by using AHLs that will weakly interact with each other. Consequently, AHL crosstalk in natural contexts, where multiple signals are exchanged in spatially structured communities, is 39 not well understood. Our study experimentally measured the effects of a neighboring interactor strain on the time scale of signaling between sender and receiver strains. The interactor strain produces a non-cognate AHL signal that influences the ability of a receiver strain to respond to the cognate AHL signal emitted from the sender strain. The consequences of such crosstalk were examined over length scales much larger than individual cells. A mathematical model was derived and compared to experimental results to predict complex AHL-receptor interactions occurring in microbial populations in nature. Figure 3.1: An experimental assay to quantify crosstalk between bacterial quorum sensing signals. A. The crosstalk assay measures the consequences of introducing a variable amount of an interactor strain into a quorum sensing network containing a sender and receiver strain. B. The sender strain is placed in the middle of an agar pad, surrounded by a uniform mixture of receiver and interactor strains. C. As the signal from the sender colony diffuses outward, the receiver strain produces a red fluorescent protein in response to a threshold level of signal. D. The spatial dynamics of gene expression in the receivers are compared in the presence and 40 absence of the interactor strain. Hypothetical curves show how the interactor strain shifts the activation curve, either promoting activation (excitatory crosstalk) or repressing activation (inhibitory crosstalk). 3.4: Results 3.4.1: Detecting Quorum Sensing Crosstalk Using the Plate-Assay Setup. Our experimental setup was based on a LuxI/LuxR sender-receiver type plate assay [25,68,102,106,108,109] with the addition of an interacting strain. In this setup, the sender strain produces the AHL 3-oxo-C6 HSL [66], which then binds to the receptor protein LuxR and activates gene expression in the senders when there is a sufficient amount of AHLs present. Activated gene expression in the senders results in elevated GFP production, see Figure 3.1A. The receivers have the capability of producing the LuxR receptor protein, and in the presence of a sufficient amount of AHLs, they will activate gene expression of RFP. The interactors constitutively produce a non-cognate AHL corresponding to the lasI or rhlI synthase genes. The plasmids encoding these constructs are shown in Figure S3.1. Coculture of the receiver and the sender resulted in a threefold increase in fluorescence and the introduction of the interactors produced an equivalent or reduced level of fluorescence, see Figure S3.2. Receiver cells containing a LuxR regulated fluorescent reporter were distributed on a 20 mm diameter LB agar pad containing a sender strain colony in the middle of the plate, see Figure 3.1B. The sender contains a plasmid expressing the LuxRI circuit, see Figure S3.1. The activation times of the gene expression in the receivers were measured with respect to the distance from the sender colony. By adding an interacting strain producing an additional AHL into the lawn of receiver cells, the shift in the activation profile quantifies crosstalk between the interactor strain and the sender/receiver system. As shown in Figure 3.1C, in the plate assay receiver cells adjacent to the senders express RFP around 5.25 hours and the receiver cells located at larger distances activated RFP after 6-8 hours. To quantify this propagation of activation within the plate, RFP fluorescent images were used to calculate the time it takes to activate RFP expression in the receivers at multiple 41 distances from the sender colony. Activation of RFP is defined as when at least 10% of the pixels belonging to cells have a pixel intensity greater than a threshold value. Activated cells displayed a clear increase in fluorescence intensity (Figure S3.2) and the measured activation times were not sensitive to small changes in the threshold intensity used in image analysis, see Figure S3.3. As mentioned previously [40], there are two main types of AHL crosstalk mechanisms between microbial species as shown in Figure 3.1D. In the context of crosstalk, the senders will produce the cognate AHL for the receivers while the interactors will produce a non-cognate signal variant, which binds to the LuxR receptor. For excitatory crosstalk, the interacting species are promoting the activation of the receivers while for inhibitory crosstalk, the interacting species are repressing the activation. The introduction of an interactor species to the sender-receiver setup should shift the activation times of the receivers. As an initial positive control shown in Figure 3.2, sender cells containing luxI were added to the lawn of receivers to verify a decrease in activation time, by 0.75 hr, for an interactor strain producing the cognate signal. In a negative control where the interactor strain is the wild type E. coli host strain not producing any AHL signal, the activation time is unchanged. This result indicates that the space taken up by interactor cells did not affect the response of the receiver cells (Figure S3.4). Figure 3.2: The assay captures a range of crosstalk behaviors. Schematic of the crosstalk assay. A kymograph of fluorescence expression in the receiver cells at a position 2 mm from the sender colony for five different conditions. The squares show fluorescent images taken between 42 4 and 6 hours. The top two lines demonstrate that the addition of an interactor strain that does not produce any signal, WT, did not change the activation time. The addition of an interactor strain containing a signal synthase gene (bottom three conditions) shifted the time of activation. We tested crosstalk from non-cognate signals by introducing an E. coli strain containing the synthase genes rhlI or lasI, producing C4-HSL and 3-oxo-C12-HSL as their principle products, respectively [14,40]. We chose these two AHL circuits based on the evidence of both excitatory and inhibitory crosstalk with the LuxI/R system from previous studies [14,39,106,110]. As shown here, when the E. coli LasI strain was introduced as the interactor strain, the receivers activated RFP earlier compared to the no crosstalk control. When the E. coli RhlI strain was introduced as the interactor strain, the receivers activated RFP later. These initial tests confirmed that both inhibitory and excitatory crosstalk could be observed in our assay. As seen in Figure S3.5, the introduction of these interactor strains does not influence the growth of the sender or receiver strains, supporting the conclusion that the observed effect was due to the crosstalk among the AHLs and receptor. 3.4.2: Measuring the scaling of crosstalk delay with community composition. The delay in the activation of the receiver strain as a function of the number of interactor cells was measured by varying the amount of interactor strain loaded onto the plate. The amount of interactor strain added to the plate is captured as the interactor to receiver ratio, which is defined as the ratio of the number of interactors cells loaded on the plate assay to the number of receiver cells loaded on the plate assay. The number of receivers was always kept constant at 10 8 cells. As shown in Figure 3.3, the shift in the activation time was proportional to the amount of interacting cells. The activation curves are shown for the cases of excitatory crosstalk (Figure 3.3A), with LasI as the interactor strain, and inhibitory crosstalk (Figure 3.3B), with RhlI as the interactor strain. These experiments quantify how the propagation of the activation front depends on the community composition, both in terms of the types of signals produced and the relative amount of each interactor strain in the environment. 43 Figure 3.3: The dependence of activation dynamics on the number of interactor cells. A. The addition of the LasI strain as the interactor strain reduces the activation time compared to case of no crosstalk. B. The addition of the RhlI strain as the interactor increases activation times compared to the no crosstalk case. In both cases, the shift in activation time was proportional to the amount of interactor strain added, as defined by the ratio of interactor to receiver (see text for details). Experimental data are from three independent measurements. The plots in the background show the trend of the data and are solely meant to guide the eye. 3.4.3: Modeling quorum sensing crosstalk with a reaction-diffusion equation. In this section, we build a mathematical model to explore the correlations of the micro level binding of a signal to a receptor, and the macro level spatiotemporal patterns of gene expression in a system incorporating quorum sensing crosstalk. In previous work done by [66– 68,107,111,112], the authors have implemented a logistic growth equation (3.1) and a reaction- diffusion model (3.2) to simulate signal production from growing cells. The logistic growth equation considers the transient behavior of the cell density 𝑛 𝑖 , which is growing at a rate of µ per cell. As the media has a finite amount of resources, the total cell density (senders + receivers + interactors), 𝑛 𝑇 , will approach a saturated cell density of s. 𝜕 𝑛 𝑖 𝜕𝑡 = µ 𝑛 𝑖 (1 − 𝑛 𝑇 𝑠 ) (3.1) 44 Initially the senders produce the AHLs at a basal rate of 𝜌 𝑏 per cell. The sender AHLs, with a concentration of cs, diffuse away from the cells with a diffusion coefficient of 𝐷 𝑐 and are degraded by the media at a rate of 𝑑 𝑎 . 𝜌 accounts for an increase in AHL production in the presence of signal. For the senders, the activity (A) defines how this activated rate of signal production, due to changes in production of the synthase protein, depends on the concentrations of multiple AHLs. 𝜕 𝑐 𝑠 𝜕𝑡 = 𝐷 𝑐 𝛻 2 𝑐 𝑠 + 𝑛 𝑠 (𝜌 𝐴 + 𝜌 𝑏 ) − 𝑑 𝑎 𝑐 𝑠 (3.2) In the presence of an interacting AHL, the transcriptional activity will be modulated due to the binding of AHLs to the LuxR receptors. Each signal has a variable influence on the activity, both in the ability to bind to a receptor and the downstream influence of such binding on the expression of quorum sensing controlled genes. The ability of an AHL to bind to the LuxR receptor depends on the binding energy and the local concentration of each AHL. In simulations, we considered the probability of an AHL to bind to a receptor and introduced a weight to account for the downstream influence of each AHL variant on gene expression. Therefore, the activity takes the form of, 𝐴 = 𝑔 (∑ 𝑃 (𝑐 𝑖 )𝑤 𝑖 𝑗 𝑖 =0 ) (3.3) where, g is the number of receptors per cell, 𝑖 is the index to describe the type of AHL, 𝑐 𝑖 is the concentration of the i th AHL, 𝑃 (𝑐 𝑖 )is the probability of an AHL binding to the receptor, 𝑤 𝑖 is the weight parameter and j is the total number of interacting AHLs. The probability of binding accounts for differences in the binding affinity of each signal variant to the receptor, as well as the competition for multiple signals to bind to the same receptor. The number of receptors (g) changes from a basal level of 100 to 600, as gene expression levels increase due to signal accumulation. To model this smooth transition we used a Hill’s function, see Table S3.1. It is only physical to have zero or positive levels of transcriptional activity, therefore the weights should also be greater than or equal to zero. The weight parameter (𝑤 𝑖 ) relates the number of AHL-receptor complexes to the extent of gene regulation, with large positive weights indicating that complexes formed by that AHL lead to strong upregulation of quorum sensing regulated genes while weights close to zero lead to inhibition of these genes. The weight is determined by the affinity of the bound receptor for 45 the promoter region of quorum sensing regulated genes, the efficiencies of transcription and translation of quorum sensing regulated genes, and the rate of dissociation for the AHL-receptor complex. A deterministic Boltzmann weight approach was applied to calculate the receptor binding probabilities from AHL concentrations and receptor binding energies [113], see the mathematical model in section 3.7.1 and Figures S3.6-S3.8. Parameter values given in Table S3.1 were measured in control experiments or obtained from previous experimental studies [2,105,111,114,115]. Since the interactor strain did not produce any receptors, signal production was constitutive and did not incorporate positive feedback from AHL level. The activity (A) for the interactors was zero and signal production occurred at a basal level, 𝜕 𝑐 𝑖𝑛𝑡 𝜕𝑡 = 𝐷 𝑐 𝛻 2 𝑐 𝑖𝑛𝑡 + 𝑛 𝑖𝑛𝑡 𝜌 − 𝑑 𝑎 𝑐 𝑖𝑛𝑡 (3.4) The constitutive production rate was assumed to have the same value as the maximum production rate of the senders. These equations were solved using the finite difference method. The model predictions were obtained considering the transient behavior of the AHL concentrations. In simulations, we considered two concentric circles; the inner circle has a radius of 1 mm while the outer circle has a radius of 10 mm. The cell densities are governed by equation (3.1), the dynamics of the signals of the senders are governed by equation (3.2) and (3.3), and the interactors by equation (3.4). The initial conditions for simulations were chosen to mirror experimental conditions. As in experiments, initially 10 7 sender cells were added to the inner circle. The inner circle was assumed to have an initial AHL concentration of 70 nM. The outer circle has a variable mixture of receivers and interactors distributed evenly in space. In all cases, there were 10 8 receivers cells. The initial concentration of the interactor AHL in the outer circle was 70 nM [31]. In simulations, the amount of interactor strain was adjusted, as specified by the interactor to receiver ratio. Based on the diffusive gradients of signals created by the senders and the interacting species, the activity of the receivers was calculated using, 𝐴 𝑅 = 𝑔 (𝑃 (𝑐 𝑆 )𝑤 𝑠 + 𝑃 (𝑐 𝑖𝑛𝑡 )𝑤 𝑖𝑛𝑡 ) (3.5) 46 where, 𝑃 (𝑐 𝑆 ) is the probability that the AHLs from the sender will bind to the receptor, 𝑃 (𝑐 𝑖𝑛𝑡 ) is the probability that the AHLs from the interactor will bind to the receptor, 𝑤 𝑠 is the weight associated with the sender AHL and 𝑤 𝑖𝑛𝑡 is the weight associated with the interactor AHL. The activity of the receivers modulates the production of the fluorescent gene reporter (RFP), as the reporter gene is transcribed by a promoter regulated by signal bound receptor. The level of activity of the receivers acts as an indicator of changes in gene expression resulting from the crosstalk. Therefore, we define a threshold activity level for the activation of gene expression and used equation (3.5) to track whether the activity level of the receivers exceeded this threshold. In simulations, the threshold activity was taken to be half of the maximum activity level when there is no crosstalk. 3.4.4: Robustness of signal transduction to crosstalk. To obtain signal weights of the sender AHL and the two interacting AHLs, simulation results were fit using the experimental data from Figure 3.3. The weight parameter (w1) for the signal 3-oxo-C6-HSL binding to the LuxR receptor was fit using the experimental data for no crosstalk. Non-linear least square fitting method was used for this purpose, see Figure S3.9a. Additional weight terms are needed to account for each interacting AHL. To identify the weight parameters of the interacting AHLs, experimental plots shown in Figure 3.3 were fit for the case of 0.9 ratio of interactor to receiver using the non-linear least square fitting (Figure S3.9). Calculated weights for signals produced by LasI and RhlI interactor strains are shown in Table 3.1. Using these weights, activation curves for the ratio of interactor to receiver of 0.2 and 0.5 were simulated, as shown in Figure 3.4A and 3.4B, revealing a scaling similar to experimental data shown in Figure 3.3. Synthase Type of AHL Weight LuxI 3-oxo-C6-HSL 𝑤 𝐿𝑢𝑥 = 0.701 ± 0.016 LasI 3-oxo-C12 HSL 𝑤 𝐿𝑎𝑠 = 0.400 ± 0.013 RhlI C4-HSL 𝑤 𝑅 ℎ𝑙 = 0.002 ± 0.001 47 Table 3.1: The weight parameters derived from the comparison of the model simulations and the experimental results. These values represent the regulatory response of each type of AHL bound to the LuxR receptor protein. The error represents the standard deviation for the weight values. Figure 3.4: Robustness of the activation of gene expression to crosstalk. A,B. Theoretical predictions for the response of the receiver cells in the presence of excitatory crosstalk or inhibitory crosstalk. The experimental data from Figure 2.3 are shown for comparison. C. Comparisons between the experimental measurements of the activation of gene expression in the plate assay to predictions made using the reaction diffusion model. Lines show the predicted change in the activation time at multiple distances from the sender colony as a function of the amount of interactor strain added to the plate. Predictions were made using the experimentally calculated crosstalk weights for the RhlI and LasI interactor strains. Data points show experimental measurements at selected distances from Figure 2.3. 48 The validated model of crosstalk has enabled the exploration of the robustness of signal propagation. Signal propagation in the presence of variable levels of crosstalk was simulated for both the LasI and RhlI interactor strains. Figure 3.4C shows the predicted delay in the activation of the receiver strain at distances of 0, 2, 5, 7 and 10 mm for ratio of interactor to receiver values between 0 and 1. Data points show experimental measurements of the activation of the RFP response at those distances and crosstalk levels, revealing a good agreement with model predictions. Using the model, we predicted the sensitivity of signaling dynamics to changes in model parameters, including cell growth rate (Figure S3.10), signal production rate (Figure S3.11), diffusion coefficient (Figure S3.12), and the signal degradation rate (Figure S3.13). Activation times strongly depended on the diffusion coefficient, signal production rate, and signal degradation rate, as together these parameters set the concentration profile of the signal. Growth rate did not affect the activation times, likely because cells were loaded onto the plate at a density near saturation, so few divisions took place during the experiment. Additionally, we observe that the crosstalk is highly correlated to the binding energy and the weight parameter, see Figure S3.14 and Figure S3.15. The activity of the receiver strain for variable concentrations of the signals made by the sender strain and the interactor strain is also plotted in Figure S3.16. The model predicts that the activity of the senders are unaffected by signal exchange with the interactors, see Figure S3.17. 3.4.5: Measuring the crosstalk potential of wild isolates. The assay also enables measurements of crosstalk with wild isolates. As an initial test, crosstalk with wild type Pseudomonas aeruginosa was measured, as shown in Figure 3.5A. The presence of Pseudomonas aeruginosa at 0.9 ratio of interactor to receiver delays activation by several hours. Because the las and rhl genes are derived from Pseudomonas aeruginosa, the weight parameters extracted from the E. coli interactor strains were used to predict the expected delay in activation as a result of crosstalk with these two systems. We simulated interactions with a hypothetical interactor strain containing both the las and rhl. Here the activity for the receivers will be, 𝐴 𝑅 = 𝑔 (𝑃 (𝑐 𝐿𝑢𝑥 )𝑤 𝐿𝑢𝑥 + 𝑃 (𝑐 𝐿𝑎𝑠 )𝑤 𝐿𝑎𝑠 + 𝑃 (𝑐 𝑅 ℎ𝑙 )𝑤 𝑅 ℎ𝑙 ). (3.6) 49 In Figure 3.5A, we observed that the trend of the simulated activation curve is similar to the experimental results, showing delayed activation and a shallower activation curve across the plate. The predicted delay was shorter than the experimentally measured delay by approximately 2 hours. Additional simulations are performed to determine if a delay in growth of the sender strain or the influence of the sender strain AHL on AHL production in P. aeruginosa might contribute to the additional delay in activation. A reduction in the sender strain growth rate, when cocultured with P. aeruginosa, was confirmed in growth measurements, see Figure S3.18. Figure 3.5A shows the prediction of the model that incorporates growth influences. Although QS activation was delayed, growth interactions alone were not sufficient to reproduce the 2 hour delay in activation, see Figure S3.19, Figure S3.20, and section 3.7.2. Figure 3.5: Testing the model and experiments with natural isolates. A. The comparison of simulation results to the experimental results with Pseudomonas aeruginosa as the interactor strain. Data from the LasI and RhlI interactor strains were used to predict the combined influence of the LasI and RhlI quorum sensing circuits in P. aeruginosa. The LasI + RhlI + growth influences line adds the experimentally measured reduction in the growth rate of E. coli in the presence of P. aeruginosa to the model, see Figures S3.18-S3.20. B. The plate assay was used to measure the interference potential of four wild bacterial isolates at 0.9 ratio of interactor to receiver. Lines, shown to guide the eye, are exponential fits to the data. 50 Next, the crosstalk potential of four additional wild isolates was tested. The four were added to the plate assay at 0.9 ratio of interactor to receiver. 16S rRNA sequencing identified the wild type species as Aeromonas hydrophila, Aeromoans veronii, Pantoea agglomerans, and Pantoea vagans. The ability of these strains to produce AHL has been reported previously [116– 121]. In Figure 3.5B, we observed that when A. veronii or P. vagans were added to the lawn of receiver cells, the activation of the RFP in the receivers was earlier as compared to no interactor strain. A. hydrophila and P. agglomerans both delayed activation. The extent of crosstalk was different for each species, suggesting that the activation of genetic expression in diverse communities is likely influenced by crosstalk of variable strength from multiple species. 3.5: Discussion Our results give new insights into signaling within mixed communities of bacteria. Adapting an approach used in previous studies [25,63,68,108,122,123], we created a sender- receiver type plate assay to quantify the activation of gene expression due to AHL-mediated signaling in the presence of multiple signal producing strains. The assay measured the robustness of specific signaling networks to interference by a strain producing a non-cognate signaling molecule. When comparing the spatial reaction-diffusion based assay to well-mixed systems, we found that the spatial assay is able to differentiate an interactor strain that produces a non- cognate signal from a strain that produces a signal destroying enzyme, see Figure S3.21. Although crosstalk between quorum sensing networks has been previously reported [11,13– 15,24,40], our titration of the interacting strain revealed the sensitivity of signal-mediated gene expression in a spatially distributed network to interference. As shown in Figure 3.4C, at distances 2 mm or less, the fold change is less than 10% even for crosstalk ratios of 1. At distances of 10 mm, the activation time has changed by approximately 10% at only 20% ratio of interactor to receiver. These numbers suggest that quorum sensing based genetic activity is largely robust to interference and that any abundant species should activate its quorum sensing network in a typical system. Our results are specific to LasI and RhlI interference with LuxRI QS system, and it is yet unclear if some AHL systems would have evolved differing levels of robustness or sensitivity to particular non-cognate signals. 51 The model demonstrates the robustness of the AHL network to interference is in part due to differences in the binding energies of cognate and non-cognate signals. Figure S3.14 shows that as the binding energy of the non-cognate signal weakens, crosstalk from the non-cognate signal has little effect on gene regulation. The ratio of interactors to receiving cells also influences robustness. As shown in Figure S3.22, when interactor cells greatly outnumber the receiver cells, robustness is lost and the expression of LuxR/I regulated genes is delayed by several hours. A third factor affecting robustness is the interaction weight for the non-cognate signal, as shown in Figure S3.15. Future work should further characterize the range of interaction weights present in real systems. A better understanding of the robustness of signal exchange in mixed populations would be beneficial to the implementation of quorum sensing gene circuits in synthetic microbial communities [16,63,96,103,105,106]. Our experimental measurements aided in the development of a detailed model to predict AHL-based signaling dynamics in mixed populations. The model accounts for crosstalk between strains using a single parameter called the weight that we calculate from experiments for a given receptor for each combination of signals. This weight accounts for the downstream regulatory consequences of a receptor binding to the AHLs, and would be related to fundamental processes such as receptor dimerization, interactions between the receptor, DNA, and RNA polymerase, and the transcription and translation of the AHL-regulated genes. We found that a model using a single weight value was in good agreement with experimental activation dynamics covering over 1 cm of space with variable amounts of interference. This close agreement between the model and experiments suggests that the model can be implemented to examine quorum sensing crosstalk in more complex and realistic contexts, such as in the presence of more than two strains, when cells are heterogeneously distributed in space, or even when transport dynamics are spatially dependent [66,67,108]. Since the experiments in these contexts would be challenging, our assay and model provide a straightforward path towards predicting signaling dynamics in complex conditions. In addition to predicting the dynamics in complex conditions, as mentioned above, the model has enabled an exploration of how robustness to interference might emerge by adjusting the parameters that regulate the response to signal exchange. Robustness can be achieved if the receptor has evolved to bind the non-cognate signal much more weakly than the cognate signal. The difference in the receptor binding energies between the non-cognate and cognate signals needed for robustness is influenced by the number of interactor cells and the 52 influence of the non-cognate signal on gene expression, as captured in the weight term. Some receptors may have evolved a sufficient amount of binding discrepancy based on interactor strains and non-cognate signals typically encountered. Our analysis of signal interference with wild species revealed a wide variety of crosstalk patterns within natural populations. We found both excitatory and inhibitory crosstalk within our isolates and a variable extent of crosstalk with the luxRI quorum sensing system. Previous measurements have also shown that non-cognate AHLs can interact with receptors such as LuxR to varying degrees [14,40]. Because we use the wild isolate directly instead of purified signal, cell free supernatant, or synthetic producer strains, we capture both direct and indirect signaling interactions with the interacting strain. Examples of indirect interactions include modulation of growth rate and gene regulatory pathways, and the evolving spatial distributions of the interactor. We attempted in the case of the interaction with Pseudomonas aeruginosa to specify the source of these indirect signaling interactions by independently accounting for the influence of each AHL signal produced by the interactor strain and growth influences of the interactor strain on the receiver cells. Although the model predicted an increased delay in activation due to growth effects, as shown in Figure 3.5A, there are still additional currently unknown interactions that further delay activation. We speculate that the QscR receptor residing in Pseudomonas aeruginosa might be absorbing the sender AHLs and contributing to this delay [124], although other non-AHL based regulatory interactions between species likely contribute to signaling dynamics. Future efforts should attempt to disentangle the direct and indirect interactions that influence signal transduction to improve our ability to predict signaling dynamics in real populations. In addition, for some ecological niches, growth dynamics and cell movement can affect the AHL gradients in unexpected ways and these factors should be incorporated to any future work to understand signaling dynamics in complex environments [125,126]. Using the assay to broadly sample interactions between known AHL signal-receptor system, such as luxRI, and wild signal producers should yield new insights into patterns of crosstalk within real environments and their consequences in ecosystem level regulation of quorum sensing. 53 3.6: Materials and methods 3.6.1: Bacterial strains and plasmids. In Table 2, we have represented the details of the bacterial strains used in this study. The host strain used for the sender, receiver and interactors are Escherichia coli NEB 5-alpha. The major QS signals are 3-oxo-C6 HSL for the sender strain, 3-oxo-C12 HSL for the LasI interactor strain, and C4-HSL for the RhlI interactor strain, see Figure S3.1 for further details. The plasmids were either obtained from Addgene [66] or constructed using Gibson assembly (New England Biolabs). Species or strain Plasmid Obtained from Escherichia coli sender ptD103LuxI sfGFP [66] E. coli receiver ptD103LuxR RFP [114] E. coli LasI interactor pZE2501Las this study E. coli RhlI interactor pZE2501Rhl this study Pseuodmonas aeruginosa [127] Aermonas hydrophila [128] Aermonas veronii [128] Pantoea agglomerans this study Pantoea vagans this study Table 3.2: The bacterial strains used in this study. 3.6.2: Culturing conditions. The bacterial strains were inoculated from frozen stocks in a 12 ml Falcon tube with 5 mL of LB broth with appropriate antibiotics. The inoculum was grown in a shaker at 220 RPM at 37 o C for 16 hours. Cells were resuspended in fresh media to remove signal in the supernatant. Late log phase cultures were used such that quorum sensing of the sender strain was activated before measurement in the plate assay. 54 3.6.3: Plate assay. The plate assay was setup as described in Silva et al [114]. The interactor strain was mixed with 100 µl of the receivers in a 1.5 mL centrifuge tube and spread onto the top of 2.5% LB agar plates using sterile 4 mm glass beads. Figure S3.23 shows that the spatial distribution of cells remained mixed during the assay. 3.6.4: Microscopy measurements. A Nikon eclipse TI fluorescent microscope was used for image acquisition. Experiments were done at 37 o C using a temperature controlled chamber. Samples were imaged at a magnification of 20x. To record the RFP activation in the receiver cells, RFP images were taken every 15 minutes for 16 hours at 30 different distances from the sender colony. Activation times were calculated at each position. Exposure times were 1s for RFP and 500 ms for GFP. No significant photobleaching was observed. Each image taken was saved in .tiff format and analyzed using a custom Matlab code. A low threshold was applied to the RFP images to identify the location of the receiver cells within each image. An upper threshold was used to identify the receivers that had activated RFP. For each time point and position, the fraction of cellular pixels above the RFP activation threshold was calculated. If the fraction of activated pixels exceeded 10%, that position was included as part of the activated region, see Figure S3.3. 3.6.5: Growth measurements. To obtain growth curves, overnight cultures were diluted 1 to 1000 in LB media and selective plating was performed to measure cell density over time. To obtain growth curves from mixtures of strains, each strain had a unique resistance marker and was plated on the appropriate selection plate. 3.6.6: Plate-reader measurements. 55 Tecan Infinite m200 Pro plate reader was used to measure growth rates and fluorescence activation in well-mixed conditions. Cells were grown to late log phase, diluted 1000 fold in pure LB media, and cultured for an additional 3 hours. After three hours of growth, 200 µl of these early log-phase cells were loaded into a flat bottom 96-well plate. The plate was inserted into the plate reader set to 37 o C and the optical density and fluorescent intensity were measured every 15 minutes for 16 hours. Optical density measurements were carried out at a wavelength of 600 nm. For GFP measurements, a wavelength of 485 nm was used for excitation and a wavelength of 515 nm was used for emission. For RFP fluorescence measurements, a wavelength of 590 nm was used for excitation and a wavelength of 650 nm was used for emission. 3.7: Supplementary Information Figure S3.1: Plasmid maps. Maps of the plasmids used to construct the A. sender, B. receiver, and (C, D) interactor strains. Sender strain plasmids are from [66]. The quorum sensing 56 promoter pluxI drives genetic expression in both the sender and receiver while expression is constitutive in the interactors by the lacUV5 promoter. Figure S3.2: Plate reader data for detecting the fluorescent changes in the receivers. We have senders (50 µl) + receivers (100 µl) in red, senders (50 µl) + LasI (50 µl) + receivers (100 µl) in blue, senders (50 µl) + RhlI (50 µl) + receivers (100 µl) in magenta and the receivers (100 µl) in yellow. The fluorescent intensity per cell is calculated in these multistrain mixtures as in [114]. The errorbars represents the standard deviation from three sets of replicates. 57 Figure S3.3: Sensitivity of results to the activation threshold value. A. The pixel intensity of the receivers over time at a distance 2 mm from the sender and in a negative control of receivers without added sender. B. Lines show exponential fits through the experimental data when reanalyzed using a threshold value (t) that was raised or lowered by 10%. Figure S3.4: Control experiments. Activation time vs Distance plots of the no crosstalk case (red), the wild type (WT) acting as the interactor strain at 0.9 ratio of interactors to receivers 58 (black), and additional senders acting as the interactors at 0.2 ratio of interactor to receiver (green). The lines show the trend of the experimental data obtained from three trials. Figure S3.5: Growth times of the strains and mixtures. The doubling time of the senders (27.3 ± 8.5 min), receivers (27.9 ± 5.1 min), LasI (26.3 ± 0.7 min), RhlI (29.8 ± 2.0 min), senders cultured with LasI (27.3 ± 0.5 min) , receivers cultured with LasI (28.5 ± 3.2 min) , senders cultured with RhlI (27.7 ± 0.6 min) and receivers cultured with RhlI (27.8 ± 5.1 min). The errorbars represent the standard deviation from three sets of replicates. 3.7.1: Calculating the probabilities of each AHL binding to the LuxR receptor using Boltzmann weights. For simplicity, we will consider a situation where there are two different types of AHLs binding to a single receptor, see Figure S3.6. 59 The energy of the unbound state depends on the unbound AHLs floating in the total solution. We assume that a single AHL has an energy of 𝜀 𝑠𝑜𝑙 in solution. Therefore, the energy of the unbound state is the total number of unbound AHLs multiplied by 𝜀 𝑠𝑜𝑙 . The AHL binds to a receptor with a binding energy of 𝜀 𝑏 . The bound AHL will change the number of unbound AHLs by one. Figure S3.6 depicts the possible states of two different types of AHLs binding to a single receptor. Figure S3.6: A statistical mechanical model of two different AHL signals binding to a signal receptor. Using the states discussed here, we find the energy and multiplicity of each signal being bound to a receptor by implementing the Boltzmann statistics as shown in Figure S3.7. Energy Multiplicity Figure S3.7: The energy and multiplicity of each unbound and bound state. 60 Here, 𝜀 𝑠𝑜𝑙 is the energy for an AHL in solution, 𝜀 𝑏 1 is the binding energy of AHL type 1, 𝜀 𝑏 2 is the binding energy of AHL type 2, Ω is the number of available sites, AHL 1 is the number of AHL type 1 and AHL 2 is the number of AHL type 2. As discussed in [113], to calculate the probability of a particular state occurring, we need to consider the Boltzmann weights (these are different from the weight terms in the Activity) of each particular scenario separately, see Figure S3.8. Multiplicity Boltzmann weights Figure S3.8: The Boltzmann weights for each state. Here, 𝜷 = 𝟏 𝒌 𝑩 𝑻 , where kB is the Boltzmann constant and T is the temperature. In Figure S3.8, we simplified the multiplicity by considering that Ω >>AHL; Ω! (Ω-AHL))! ~ Ω 𝐴𝐻𝐿 . This is a valid assumption to make considering the fact that the number of available states will be much higher than the number of AHLs present in the system. Now that we have the Boltzmann weights of each state, calculating the probabilities is straightforward. The probability of AHL type 1 binding to the receptor is then, P(𝐴𝐻𝐿 1 )= Ω (𝐴𝐻𝐿 1 +𝐴𝐻𝐿 2 −1) (𝐴𝐻𝐿 1 −1)!𝐴𝐻𝐿 2 ! 𝑒 −𝛽 (((𝐴𝐻𝐿 1 −1)+𝐴𝐻 𝐿 2 )𝜀 𝑠𝑜𝑙 +𝜀 𝑏 1 ) Ω (𝐴𝐻𝐿 1 +𝐴𝐻𝐿 2 ) 𝐴𝐻𝐿 1 !𝐴𝐻𝐿 2 ! 𝑒 −(𝛽 (𝐴𝐻𝐿 1 +𝐴𝐻𝐿 2 )𝜀 𝑠𝑜𝑙 ) + Ω (𝐴𝐻𝐿 1 +𝐴𝐻 𝐿 2 −1) (𝐴𝐻𝐿 1 −1)!𝐴𝐻𝐿 2 ! 𝑒 −𝛽 ((𝐴𝐻𝐿 1 −1)+𝐴𝐻𝐿 2 )𝜀 𝑠𝑜𝑙 +𝜀 𝑏 1 ) + Ω (𝐴𝐻𝐿 1 +𝐴𝐻 𝐿 2 −1) 𝐴𝐻𝐿 1 !(𝐴𝐻𝐿 2 −1)! 𝑒 −𝛽 ((𝐴𝐻𝐿 1 +(𝐴𝐻𝐿 2 −1))𝜀 𝑠𝑜𝑙 +𝜀 𝑏 2 ) 61 (3.7) We could further simplify this equation by multiplying the denominator and numerator by 𝐴𝐻𝐿 1 !𝐴𝐻𝐿 2 ! Ω (𝐴𝐻𝐿 1 +𝐴𝐻𝐿 2 ) ; (3.8) If we consider that the volume of a box is 𝑉 𝑏𝑜𝑥 , then the total volume of the system is Ω𝑉 𝑏𝑜𝑥 . The AHL concentration could be written as, 𝑐 𝐴𝐻𝐿 = 𝐴𝐻𝐿 Ω𝑉 𝑏𝑜𝑥 . We could also define a local reference concentration which corresponds to having all sites in the lattice occupied as 𝑐 0 = 1 𝑉 𝑏𝑜𝑥 . Hence, (3.9) Analogous equations were derived for the probability of AHL type 2 binding to the receptor. Using the same procedure, we calculate the probability of the unbound AHL as, (3.10) P(unbound)= 1 1+ 𝑐 𝐴𝐻𝐿 1 𝑐 0 𝑒 𝛽 (𝜀 𝑠𝑜𝑙 −𝜀 𝑏 1 ) + 𝑐 𝐴𝐻𝐿 2 𝑐 0 𝑒 𝛽 (𝜀 𝑠𝑜𝑙 −𝜀 𝑏 2 ) P(𝐴𝐻𝐿 1 )= 𝐴𝐻𝐿 1 Ω 𝑒 −𝛽 (((𝐴𝐻𝐿 1 −1)+𝐴𝐻𝐿 2 )𝜀 𝑠𝑜𝑙 +𝜀 𝑏 1 ) 𝑒 −(𝛽 (𝐴𝐻𝐿 1 +𝐴𝐻 𝐿 2 )𝜀 𝑠𝑜𝑙 ) + 𝐴𝐻𝐿 1 Ω 𝑒 −𝛽 ((𝐴𝐻𝐿 1 −1)+𝐴𝐻𝐿 2 )𝜀 𝑠𝑜𝑙 +𝜀 𝑏 1 ) + 𝐴𝐻𝐿 2 Ω 𝑒 −𝛽 ((𝐴𝐻 𝐿 1 +(𝐴𝐻𝐿 2 −1))𝜀 𝑠𝑜𝑙 +𝜀 𝑏 2 ) P(𝑐 𝐴𝐻𝐿 1 )= 𝑐 𝐴𝐻𝐿 1 𝑐 0 𝑒 𝛽 (𝜀 𝑠𝑜𝑙 −𝜀 𝑏 1 ) 1+ 𝑐 𝐴𝐻𝐿 1 𝑐 0 𝑒 𝛽 (𝜀 𝑠𝑜𝑙 −𝜀 𝑏 1 ) + 𝑐 𝐴𝐻𝐿 2 𝑐 0 𝑒 𝛽 (𝜀 𝑠𝑜𝑙 −𝜀 𝑏 2 ) 62 Parameter Value References 𝜇 1.50 (±0.02) hrs -1 (experimentally calculated) S 10 9 cells DAHL 1.764 mm 2 hrs -1 [2] 𝜌 2.3 x 10 -9 nM hrs -1 per cell [111] 𝜌 𝑏 2.3 x 10 -10 nM hrs -1 per cell [111] 𝑑 𝑎 0.005545 hrs -1 [111] 𝑔 100 at non active state to 600 at activation [105] 𝑚 𝑔 2.5 [114] 𝜃 𝑔 70 nM [114] 𝑇 300K C0 1 M 𝜀 𝑠𝑜𝑙 − 𝜀 𝑙𝑢𝑥 = ∆𝜀 𝑙𝑢𝑥 12.991 KJ mol -1 [115] 𝜀 𝑠𝑜𝑙 − 𝜀 𝑙𝑎𝑠 = ∆𝜀 𝑙𝑎𝑠 11.426 KJ mol -1 [115] 𝜀 𝑠𝑜𝑙 − 𝜀 𝑟 ℎ𝑙 = ∆𝜀 𝑟 ℎ𝑙 9.615 KJ mol -1 [115] Table S3.1: The parameters used in the simulations. To model the change in g over time, we use a Hill’s function with, g =100 + 500 [𝒄 𝒔 ] 𝒎 𝒈 [𝒄 𝒔 ] 𝒎 𝒈 +𝜽 𝒈 𝒎 𝒈 , so g starts with 100 and increases to 600 at activation. Here Cs is the AHL concentration of the senders. 63 Figure S3.9: Fitting the experimental data to obtain interaction weights. A. The experimental data points in the case of no crosstalk are plotted in red. The blue line shows the simulation result using the best-fit value for the interaction weight. The best fit weight was 0.701± 0.016 for the AHLs produced by the LuxI binding to the LuxR receptor. B. Crosstalk with the LasI interactor was fit for the ratio of interactor to receiver level of 0.9. The black data points are from experiments and the pink line shows simulation results using the best-fit interaction weight of 0.400 ± 0.013. C. Crosstalk with the RhlI interactor was fit for the ratio of interactor to receiver level of 0.9. The green data points are from experiments and the yellow line shows simulation results using the best-fit interaction weight of 0.002 ± 0.001. Figure S3.10: Testing the effects of the growth rates of the interactors on crosstalk. Simulations were run using equations from the main text and parameters found in Table 3.1 and 64 Figure S3.11: Testing the effects of the production rates of the interacting AHLs on crosstalk. For A. the RhlI interactor at 0.9 ratio of interactor to receiver and B. the LasI interactor at 0.9 ratio of interactor to receiver. The inset in (B) show results for 0.2 ratio of interactor to receiver. Simulations were run using equations from the main text and parameters found in Table 3.1 and Table S3.1. Figure S3.12: Testing the effects of the diffusion coefficient of the AHLs of interactors on crosstalk. For A. RhlI interactor at 0.9 ratio of interactor to receiver and B. the LasI interactor at 65 0.9 ratio of interactor to receiver. Simulations were run using equations from the main text and parameters found in Table 3.1 and Table S3.1. We observe that the crosstalk can be strongly influenced by the diffusion coefficient. Figure S3.13: Testing the effects of the degradation coefficient of the interacting AHLs on crosstalk. For A. RhlI interactor at 0.9 ratio of interactor to receiver and B. the LasI interactor at 0.9 ratio of interactor to receiver. The inset in B is for 0.2 ratio of interactor to receiver. We observe that the crosstalk is sensitive to the degradation coefficient. Simulations were run using equations from the main text and parameters found in Table 3.1 and Table S3.1. 66 Figure S3.14: The influence of the non-cognate AHL binding energy on crosstalk. The binding energy of the interactor is ∆𝜺 𝒊𝒏𝒕 . In simulations, we tested for various ∆𝜺 𝒊𝒏𝒕 for the case of A. inhibitory crosstalk and B. excitatory crosstalk. Here the ratio of interactor to receiver was kept at 0.9. Figure S3.15: The influence of the weights of the AHLs on crosstalk. The weight parameter of the interactor is changed in simulations for the case of 0.9 ratio of interactor to receiver. In the right, we have the activation time vs. distance projection of the plot on the left for some selected weight values. We observe that at a weight of 0.3 the response of the receivers coincide with the case of no interactors. Figure S3.16: Simulated activity profiles of the receivers (per receptor) under well mixed conditions in the presence of signal coming from both a sender strain and an interactor strain. Plots show the activity of the receiver in the presence of combinations of the sender and interactor signals between 0 and 100 nM. (Left) the interactor strain is the same as the sender, 67 having the LuxI synthase. (Center) the interactor strain has the LasI synthase. (Right) the interactor strain has the RhlR synthase. AT = 210 is the threshold level of activity required for activation. The right plot shows a non-zero level of activation even in the case of no sender cells and an inhibitory interactor strain, suggesting that inhibitory crosstalk could be mistaken for excitatory crosstalk in well mixed experiments. Figure S3.17: The influence of feedback between interactors and senders on activation of the receivers. Simulation results of when the senders have feedback from the interactor (left=RhlI interactor, right=LasI interactor) compared to when it does not have feedback. To simulate the no feedback condition, the activity of the sender was considered to be only dependent on the sender AHL. 68 Figure S3.18: The effect on the growth of the senders and receivers due to P. aeruginosa. The doubling time of the senders (27.3 ± 8.5 min), receivers (27.9 ± 5.1 min), senders cultured with P. aeruginosa (35.1 ± 2.7 min), and receivers cultured with P. aeruginosa (38.2 ± 1.4 min). The errorbars represents the standard deviation from three sets of replicates. 3.7.2: Testing the model for effects of the growth interactions and AHL internalization. To test the possible effect of growth influences on the sender-receiver system, caused by the interactor, we considered a competitive Lotka–Volterra model [94,128], 𝜕𝑛 𝜕𝑡 = 𝜇 𝑛 (1 − 𝑛 𝑇𝑜𝑡𝑎𝑙 𝑠 ) −𝛼 1 𝑛 𝑛 𝑖𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑜𝑟 (3.11) where, 𝛼 1 is the growth effect the interactor species has on the sender-receiver system. To test the effect of signal internalization we used, 𝜕 𝑐 𝑠 𝜕𝑡 = 𝐷 𝑐 𝛻 2 𝑐 𝑠 + 𝑛 𝑠 (𝜌 𝐴 + 𝜌 𝑏 ) − 𝑑 𝑎 𝑐 𝑠 − 𝛼 2 𝑛 𝑖𝑛𝑡 [𝑐 𝑠 ] 𝑚 4 [𝑐 𝑠 ] 𝑚 4 +𝜃 4 𝑚 4 (3.12) 69 where, 𝛼 2 is the AHL internalizing rate per interactor cell which is modulated by the number of AHLs present in the vicinity of the interactor. Based on [114] we used, 𝑚 4 = 2.5 and 𝜃 4 = 70 𝑛𝑀 . Figure S3.19: The impact on crosstalk with non-quorum sensing interactions. A. The interactors have a growth influence on the senders and receivers. Simulations were run using equation (11), based on analysis from [114]. The interactors are at 0.2 ratio of interactor to receiver. B. The interactors are capable of internalizing the AHLs produced by the senders. Simulations were run using equation (3.12), based on analysis from [114]. 70 Figure S3.20: Simulating the impact of Pseudomonas aeruginosa growth influences on quorum sensing activation in the receiver strain. We introduce a growth influence on the receivers and senders, caused by the interactor strain with both las and rhl systems in simulations. Increasing 𝛂 𝟏 (growth effect parameter), increased the activation times and decreased the activation radius. To yield a similar time shift as observed in experiments (brown) with Pseudomonas aeruginosa as the interactor strain, the model would require additional factors such as AHL internalization or non-quorum sensing interspecies regulatory interactions. 71 Figure S3.21: Observing crosstalk and signal degradation in a well-mixed setup. We simulate a well-mixed system with either inhibitory crosstalk, excitatory crosstalk, or interference by a species producing an AHL degradative enzyme based on [114]. We observe similar responses for the cases of inhibitory crosstalk and signal destruction, indicating complications of distinguishing between these two mechanisms under well mixed conditions. Figure S3.22: Robustness of the network to interference when there is a large excess of inhibitory interactors. Simulation data showing the activation time vs. distance from the sender for increasing level of the rhl interactor. Here the lines represent different levels of ratio of interactors to receivers. The activation time of the receivers shifts to around 11-17 hours from 5- 7 hours when there is 5 times more interactors than receivers demonstrating diminishing robustness. 72 Figure S3.23: Spatial distribution of cells after 16 h. Distribution of cells in the assay after 16 hours. Two E. coli strains, one producing GFP and the other producing RFP, were uniformly distributed in the assay. Fluorescent images show the distribution of these strains after 16 hours at 37 o C. The left image shows the GFP channel and the right image shows an overlay of the GFP and RFP channels. After growth on the plate, over lengths scales of around 10 microns some regions appear enriched in a single cell type. Over larger lengths, the two cells remaining uniformly mixed. 73 Chapter 4: Disruption of microbial communication yields a 2D percolation transition This work is currently accepted for publication at Physical Review E. 4.1: Abstract Bacteria communicate with each other to coordinate macro-scale behaviors including pathogenesis, biofilm formation and antibiotic production. Empirical evidence suggests that bacteria are capable of communicating at length scales far exceeding the size of individual cells. Several mechanisms of signal interference have been observed in nature, and how interference influences macro-scale activity within microbial populations is unclear. Here we examined the exchange of quorum sensing signals to coordinate microbial activity over long distances in the presence of a variable amount of interference through a neighboring signal-degrading strain. As the level of interference increased, communication over large distances was disrupted and at a critical amount of interference, large-scale communication was suppressed. We explored this transition in experiments and reaction-diffusion models, and confirmed that this transition is a 2D percolation transition. These results demonstrate the utility of applying physical models to emergence in complex biological networks to probe robustness and universal quantitative features. 4.2: Introduction Recent years have seen increasingly clear evidence that the traditional view of bacteria as `loner’ organisms must be revised [40,51,58,95,125]. In natural habitats, diverse microbial species regularly interact with each other and their environment, and such interactions frequently have a profound effect on the resulting microbial phenotypes [14,129,130]. For example, a commonly found interaction is signal interference, and one commonly found interference mechanism is where signals produced by one type of species are degraded by a different species, which in turn will affect cellular activity and gene expression [40]. From a physics perspective, such multispecies communities offer a novel context in which to discover and characterize emergent, collective phenomena in strongly interacting many- body networks [12,44,131–133]. One particularly notable and relevant example of such 74 collective behavior is percolation, where networks display an abrupt transition in large-scale global connectivity at a critical threshold of a varying parameter, typically occupation probability. Percolation is an attractive topic to physicists for several reasons, the most important one arguably being its universality – the fact that systems which are quite dissimilar microscopically display equivalent quantitative scaling behavior near the transition [41–43]. While percolation has been observed and analyzed in numerous abiotic contexts, other than a recent report [44] there is relatively little work exploring its implications for biological systems, particularly community-level ecosystems. It remains a challenge to identify suitable ‘model’ community-level networks in which to explore potential biological realizations of percolation. An ideal system should be simultaneously experimentally and theoretically tractable, while also being capable of yielding qualitative and quantitative insight into the workings of real microbial communities in situ. Microbial communication networks are a particularly strong candidate for such a system. Bacteria use small chemical signals known as autoinducers (AI) to communicate with their neighbors. The basic process of communication, known as quorum sensing [3,17,51,52,96], involves cells secreting, sensing, and responding to a threshold concentration of AI molecules, thereby regulating density-dependent gene expression. Quorum sensing (QS) regulates several collective, group-level behaviors, including pathogenesis [97,104], biofilm formation [134], antibiotic production [40,135,136], and bioluminescence [52]. Empirical evidence suggests that these behaviors can span large distances [12,52,114] . In this work, we examine the robustness of such communication networks to interference by analyzing spatial patterns of activation in a model bacterial community consisting of a sender cell, which produces and responds to AI signals, and a degrader cell, which produces an enzyme capable of destroying the signal, as shown in Figure 4.1(a) and similar to strains used in previous studies [38,107]. This model community allows a controlled, systematic study of interactions that occur naturally within microbial ecosystems [20,21]. For example, Bacillus subtilis, a Gram-positive bacteria commonly found in soil, secretes the degradative enzyme AiiA used in the study. In its natural environment, B. subtilis interacts with many soil strains that use AHL-based quorum sensing, including species of Pseudomonas, Serratia, and Eneterobacter [22]. We explore through both experiments and theoretical models how the ratio 75 of these two strains influences the ability of the senders to communicate over long distances. Our primary result is the discovery of a percolation transition that leads to a sudden breakdown in long-range communication at a critical strain ratio. We observe universal critical scaling exponents at the transition that are independent of the specific properties of individual strains, and show that the specific transition point can be shifted in a predictable way by selective genetic modification of strains. 4.3: Experiments on quorum sensing bacteria. 4.3.1: Experimental setup An experimental assay was developed to test the capability of microbes to coordinate activity in spatially structured environments. As shown in Figure 4.1(b), a specified ratio of sender and degrader cells were mixed together and applied to the top of an LB agar plate. Cells emit diffusible signals and spread out via cell division, but are not motile. Sender cells are a strain of Escherichia coli containing the LuxRI quorum sensing system and a GFP reporter gene for quorum sensing activation. Senders increase the expression of the GFP reporter gene in response to an increase in the concentration of a diffusible signaling molecule, here acyl- homoserine lactones (AHL). The response function of GFP production is non-linear, exhibiting a sharp increase in cellular fluorescence when the amount of signal exceeds a threshold concentration [112,114]. Degrader cells are a strain of E. coli constitutively expressing the lactonase enzyme AiiA [38]. AiiA remains inside the degrader cells and degrades AHL quorum sensing signals that diffuse into the cell, including the 3-oxo-C6-HSL produced by the sender cells [114,137]. 76 Figure 4.1: Experimental assay to measure spatial patterns of cellular communication in the presence of interference. (a) The sender (yellow) produces an acyl-homoserine lactone (AHL) signal, 3-oxo-C6-HSL, while the degrader (red) produces the AHL degradative enzyme AiiA constitutively. When the senders are in monoculture, quorum sensing is uniformly activated by all cells across the plate. When degraders are introduced, AHLs are degraded resulting in regions where sender cells do not activate quorum sensing. (b) Senders and degraders are mixed at a ratio of 10 9 : n (units of cells), where n varied from 0 to 2 x 10 9 cells. The mixture is loaded into a small petri dish with LB agar. 4.3.2: Experimental results In the presence of degraders, both the degrader and sender cells influence the local concentration of signal. In regions where, by chance, degrader cells are more abundant, AHLs will be depleted and potentially dip below the threshold concentration needed for activation. Consequently, senders will not activate GFP expression and will appear dark in fluorescent images. In Figure 4.2(a), we have a sender strain, distributed with the degrader strain, at a ratio 77 of 10 9 :6 x 10 8 (sender: degrader). Here, we observe that in the regions where degraders are more abundant, senders have not activated GFP (Figure S4.1). To determine whether a sender has activated quorum sensing or not, we define a fluorescence intensity threshold for activation, which was 25% of the maximum detected GFP value (Figure S4.2). The validity of the threshold we chose and the robustness of our results to slight changes in threshold will be discussed later. The activation of quorum sensing in the sender cells was monitored using fluorescence microscopy. The sender cells that activate quorum sensing express high levels of GFP and appear green in the fluorescent microscopy images shown in Figure 4.2(a), Figure 4.2(b), whereas sender cells that did not activate quorum sensing are dark and not visible. Plates are imaged at 20x magnification 16 hours after inoculation. For each image the field of view spans an area of 750 µm x 630 µm, and a minimum of 16 fields of view were imaged for each experimental condition. As shown in Figure 4.2(b), when no degrader cells are added to the plate, all the senders activate quorum sensing and produce green fluorescence throughout the plate. On plates containing a mixture of degrader and sender cells, dark regions were observed that represent regions where the quorum sensing signal did not accumulate above the threshold concentration required for activation. The sizes of these inactivated regions were larger in plates containing more degrader cells, and activation was completely inhibited when more than 10 9 cells of degrader were spread onto the plate. As seen in Figure S4.3, introducing a dummy strain that is the same strain of E. coli, but which does not contain the plasmid necessary for producing the signal degrading enzyme, did not result in dark patches in the fluorescence image, even when the dummy strain was loaded 3 times higher than the sender strain. This demonstrates that inhibition of quorum sensing activity is due to the depletion of the signal concentration, and not caused by competition between the two strains for resources or space. This is further validated with Figure S4.4, where we introduce the sender strain, which produces RFP constitutively, and observe that even for high amounts of degrader, the senders seem to occupy most of the plate. For all experiments, we load 10 9 senders, as it gives a near uniform distribution of the cells on the plate (see Figure S4.5). 78 The transition from large connected regions of activation to segregated patches of activation observed in Figure 4.2(b) appears similar to a percolation transition. In percolating systems, regions of activity have a sharp transition in connectivity above a critical value of control parameters [138,139]. Here the control parameter is the amount of degrader cells in the population. The connectivity of activated regions is quantified by measuring percolation events. A percolation event occurs when at least one continuous region of activation spans the entire system. In our experiments, percolation is defined as a continuous region of fluorescence intensity above a threshold value that touches opposite sides of the image, a region of space 750 µm x 630 µm. As discussed above, this threshold (thr) was selected to be 25% of the maximum fluorescent intensity detected by the microscope (thr = 148) (Figure S4.2). By applying this threshold, we obtain the amount of pixels above this threshold, and we observe that the amount of active pixels are largely robust when the threshold is changed by up to 20% from the value we select (Figure S4.6). We quantified percolation in each fluorescent image. Figure 4.2(c) shows the percentage of images that exhibited percolation (percolation frequency) for each amount of degrader cells tested. At low numbers of degraders, there will always be a percolation event. Once the density of degraders reaches a critical value near 3x10 8 cells, the percolation frequency rapidly decreases towards zero. 79 Figure 4.2: Experimental demonstration of a degrader-dependent percolation transition. (a) Here we represent the 20x microscopic GFP image, of the sender cells (left) in the presence of degraders at a ratio of 10 9 :6x10 8 (sender: degrader). The GFP is due to quorum sensing activation. The dark regions in the green fluorescent image is due to the senders not activating quorum sensing due to the presence of degraders, see Figure S4.1. We measured the GFP pixel intensity across the blue line, represented in the fluorescent image, and observed that the fluorescence dips below the threshold value at certain points on the line (see Figure S4.2 for further details on the threshold). (b) The 20x fluorescent microscopy images reveal the regions of active (green) and inactive senders (dark) as we increase the amount of degraders. (c) The number of percolation events in our images was measured as a function of the amount of degraders added the plate. Here the data is taken from three replicate plates, with each plate having a minimum of 16 images. The error bars represent the standard error over replicates. Percolation has been observed in many diverse physical systems, including dielectric breakdown [140], porous transport [141], and even the epidemiology of infectious diseases [142]. The properties of such percolation networks are known to follow characteristic scaling laws. We measured values for critical exponents associated with the scaling laws for three well known quantities from 2D-percolation theory [43,143]: the average cluster size <s>, the order parameter P∞ and the correlation length ξ. For this analysis, pixels in the fluorescent images are assigned a value of active or “occupied” if the fluorescent intensity is above the 80 threshold (thr), or conversely a value of inactive or “empty” if the fluorescent intensity is below threshold. If an active pixel is touching another active pixel in either of the 8 neighboring positions, we define these two pixels as being connected. If a connection of pixels spreads from one edge of the image to its opposite edge, we define this collection of active pixels as a spanning cluster. To obtain the average cluster size <s>, we measured the areas of all the clusters of activated pixels, excluding pixels in the spanning cluster, and took the average of the measurements. The order parameter, P∞, is defined as the probability that an activated pixel belongs to the spanning cluster, and was determined by taking the ratio of the number of active pixels in the spanning cluster to the total number of active pixels in all clusters. Finally, the correlation length ξ measures the length scale over which activation is correlated; in other words, if a given pixel is activated, it is expected that, on average, pixels within a distance ξ of that pixel would also be activated. To obtain ξ, we considered the largest non-spanning cluster, and calculated the average distance between two pixels within this cluster. These three quantities are plotted in Figure 4.3. In the classical theory these quantities are generally plotted against the occupation probability [143]. In section and Figure S4.7, we show that at the transition, the occupation probability scales essentially linearly with the amount of degraders. As a result of this linear relationship, all quantities are expected to have the same universal scaling behavior with respect to the amount of degraders: 〈𝒔 〉 ∝ |𝒏 𝑨𝒊𝒊𝑨 − 𝒏 𝑨𝒊𝒊𝑨 𝒄 | −𝜸 , 𝑷 ∞ ∝ |𝒏 𝑨𝒊𝒊𝑨 − 𝒏 𝑨𝒊𝒊𝑨 𝒄 | 𝜷 , 𝝃 ∝ |𝒏 𝑨𝒊𝒊𝑨 − 𝒏 𝑨𝒊𝒊𝑨 𝒄 | −𝝂 , where, 𝒏 𝑨𝒊𝒊𝑨 is the amount of degraders, 𝒏 𝑨𝒊𝒊𝑨 𝒄 is the critical amount of degraders at the percolation threshold, and 𝜸 , 𝜷 and 𝝂 are the critical exponents for the transitions of average cluster size, order parameter and correlation length respectively. We determined 𝒏 𝑨𝒊𝒊𝑨 𝒄 to be 6.5 x 10 8 by considering the 𝒏 𝑨𝒊𝒊𝑨 at which we observe maximal values of the average cluster size and correlation length [143] (Figure S4.8). We find that the experimentally measured exponents are consistent with the expected values for a standard 2D percolation transition [43,143], as shown in Figure 4.3 and Table 4.1. To calculate the critical exponents, we generated a log-log transform of the scaling quantities near the critical point, see Figure S4.9, and calculated the gradient of a linear fit via least squares regression, with uncertainties determined via the standard error of the slope, as shown in Table 4.1. In section , we further justify these results via additional tests of the robustness of the critical 81 exponents to changing thresholds. The calculated critical exponents were not sensitive to the exact number of points fit nor the 20% changes in the threshold value used to classify activated pixels, (see Figure S4.10 and S4.11). Such scaling rules should be independent of the scale over which they are observed, and to demonstrate this scale-free nature of our results, we reanalyzed the data for a different sized imaging window and found the calculated exponents were similar (see Figure S4.12, Table S4.2). Figure 4.3: Scaling quantities from experiments. (a) The average cluster size <s> , (b) the order parameter P∞ , (c) and the correlation length ξ calculated from experimental data. The error bars represent the standard error from 3 replicate plates and a minimum of 16 images per plate. In (d), (e), (f), we have the log-log transform of the points in the critical region for <s>, P∞ and ξ respectively, marked as closed blue circles. The slope of the linear fits to these points yields the critical exponents. The blue line represents the linear fit (least square fit) with the experimentally obtained exponent while the red line represents the prediction from classical percolation theory, values listed in Table 4.1. The error bars in d-f are determined via propagation of the log of the uncertainty. 82 4.4: Mathematical model 4.4.1: Reaction-diffusion model To obtain a deeper understanding of the transition, we can compare these experimental results to numerical simulations of the underlying biochemical kinetics. We developed a model that captured the dynamics of signal exchange and cell growth. The growth of the cells is modeled with a logistic growth equation [12,114], 𝜕𝑛 𝜕𝑡 = 𝜇 𝑛 (1 − 𝑛 𝑡 𝑠 ) (4.1) Here, the total cell number ( n t), equal to sender cells plus degrader cells at a particular location in space, grows exponentially with rate 𝜇 up to a saturated cell number ( s), at which point growth will cease. The dynamics of the signals produced by the senders and the degradative enzymes produced by the degraders can be described by reaction diffusion equations, 𝜕 [𝐴𝐻𝐿 ] 𝜕𝑡 = 𝐷 𝐴𝐻𝐿 𝛻 2 [𝐴𝐻𝐿 ] + 𝑛 𝐴𝐻𝐿 ( 𝜌 𝐴𝐻𝐿 [𝐴𝐻𝐿 ] 𝑚 1 [𝐴𝐻𝐿 ] 𝑚 1 +𝜃 1 𝑚 1 + 𝜌 𝑏 1 ) − 𝛾 1 [𝐴𝐻𝐿 ] − 𝛾 2 [𝐴𝑖𝑖𝐴 ] [𝐴𝐻𝐿 ] 𝑚 2 [𝐴𝐻𝐿 ] 𝑚 2 +𝜃 2 𝑚 2 (4.2) The AHLs are initially produced at a lower basal rate of 𝜌 𝑏 1 , by an 𝑛 𝐴𝐻𝐿 amount of sender cells, and the AHL concentration ([AHL]) increases over time. AHLs will diffuse into the media with a diffusion coefficient of 𝐷 𝐴𝐻𝐿 . Once the concentration builds up to a critical threshold value of 𝜃 1 , the production rate of AHLs increases by 𝜌 𝐴𝐻𝐿 . This switching from the lower to higher production rate is represented by a Hill function with a Hill coefficient of m1. The AHLs are degraded in the media at a rate of 𝛾 1 , and by the degradative enzyme at a rate 𝛾 2 . The degrader enzyme concentration is [AiiA] and the amount of signals degraded is regulated by the presence of the signal. We represent this by an additional Hill function with a Hill coefficient of m2, where the AHL concentration needs to accumulate beyond a critical number to get a considerable amount of degradation [38,114]. Note that, unlike the AHLs, the degrader enzymes do not freely diffuse throughout the medium, but are localized within the degrader cells. The 83 degraders will constitutively produce the degradative enzyme at a rate of 𝜌 𝐴𝑖𝑖𝐴 within the degrader cells, a protein which decays at a rate of 𝛾 3 . 𝜕 [𝐴𝑖𝑖𝐴 ] 𝜕𝑡 = 𝑛 𝐴𝑖𝑖𝐴 𝜌 𝐴𝑖𝑖𝐴 − 𝛾 3 [𝐴𝑖𝑖𝐴 ] (4.3) In simulations, sender and degrader cells are initially distributed randomly within an area of 0.49 mm 2 . In this area there are 22500 pixels (150 x 150) and each pixel has a size of approximately 25 µm 2 . Each pixel can hold approximately 25 cells. Sender cells are fixed and distributed homogenously while the degraders will change and be distributed randomly. Cells do not move over time, but spread to adjacent regions of the plate due to cell growth. Signal is produced and degraded over time, reaching a steady-state AHL concentration profile within 16 hours. Regions of space where the signal concentration is above the quorum sensing threshold (𝜃 1 ) are plotted, as shown in Figure 4.4(a). These would correspond to regions of high fluorescent intensity in experiments. All the values for the parameters are taken from Silva et al. [114], and reported in Table S4.1. 4.4.2: Modelling results As in experiments, we first obtain the percolation frequency as a function of the amount of degraders, and, as shown in Figure 4.4(b), observe that at 10 8 cells the percolation frequency decreases, similar to experimental observations. Subsequently, the same image processing algorithms that we apply to calculate <s>, P∞ and ξ from experimental activation maps are applied to simulation outputs, allowing us to plot <s>, P∞ and ξ from simulations and calculate the critical exponents, as shown in Figure 4.4(c), Table 4.1 and Figure S4.13-S4.15. Via finite- size-scaling analysis, similar to the windowing we perform on experimental images, we observe that as system size increases, the exponents systematically converge onto the expected classical value for an infinite system (Figure S4.16, S4.17 and Table S4.3). 84 Figure 4.4: Simulation of quorum sensing activation in the sender and degrader community. (a) The AHL concentration of the senders from simulations are marked in green, with green indicating signal concentration that exceeded the threshold needed to activate quorum sensing. (b) The frequency of percolation events from the simulations. (c) The average cluster size <s>, the order parameter P∞ and the correlation length ξ obtained from simulations. The error bars represent the standard error over 100 simulations. 85 Critical exponents For the average cluster size (𝜸 𝐏 ) For the order parameter (β 𝐏 ) For the correlation length (ν) 2D percolation universality class 2.39 0.14 1.33 From experiments 2.28 ± 0.36 0.17± 0.06 1.36± 0.12 From reaction- diffusion simulations 2.42± 0.16 0.16± 0.03 1.35 ± 0.06 Table 4.1: Values of the critical exponents. The values here represent the slope of the log-log linear fit. The linear fits were performed using least square fitting method, and the uncertainty here is the standard error (SE) of the slope represented as SE= √∑ (𝒚𝒊 −𝒚𝒊̂) 𝟐 (𝒏 −𝟐 ) √∑(𝒙𝒊 −𝒙̅) 𝟐 , where, 𝒚𝒊 is the dependent variable value for point i, 𝒚𝒊̂ is the dependent variable value for point i from the least square fit, 𝒙𝒊 is the independent variable value for point i, 𝒙̅ is the mean of the independent variable, and n is the number of observations. 4.5: Shifting the transition point by changing the production rate of AiiA. The behavior of the scaling quantities near criticality and the values obtained from experiments and theory revealed that when a signal degrading strain is added to a spatial cellular network, the breakdown of effective communication follows a sharp transition whose spatial distribution of activity follows a percolation transition. The senders’ capability to coordinate activity over large distances via the exchange of quorum sensing signals fails if the amount of degrader cells exceeds approximately 3x10 8 cells. The critical amount of degrader cells leading to this transition is a tunable parameter. In simulations, we observe that changing the production rate of degradation enzyme per cell shifted the critical number of degraders required to disrupt long distance communication, as shown in Figure 4.5(a). To validate this observation experimentally, we constructed a new degrader strain with tunable expression of the aiiA gene (see Figure S4.18a and S4.18b). By introducing 86 isopropyl β-D-1-thiogalactopyranoside (IPTG) at different concentrations, the expression level of aiiA is controlled. When the senders are mixed with the new degrader strain in a plate reader, we observe that the steady state fluorescence decreases with increasing IPTG, confirming that production of the degradative enzyme is regulated by the inducer (Figure S4.18c). In Figure 4.5(b), similar to the simulation results, we observe that the critical degrader amount can be shifted in experiments by increasing the production rate of the degradative enzyme by introducing 1 mM of IPTG to the plate. We calculated the critical exponents for the case of increased AiiA production and once again obtain similar values to the classical theory (Table S4.4). In Figure 4.5, we also have shown the universality in the nature of the transition by changing the coordinates in the log (Amount of degraders) axis by reducing it to (log (Amount of degraders) - log (Amount of degraders)c )/ (log (Amount of degraders)c), where log (Amount of degraders)c is the point at which the percolation frequency goes to zero. Figure 4.5: Changing the production rate of AiiA shifts the critical density. The Percolation frequency (Perc. freq.) (a) from simulations and (b), from experiments for two different production rates of the AiiA enzyme. For both the simulations and experiments, the blue points represents the case when the normal degraders are used. In the simulations, the red data points represent the case of when the production rate of the degrader enzyme is reduced to 0.6 times its original. In experiments, the red data points represents the case when the new degraders with the plasmid pZSaiiA_lac_O1 was used. The IPTG level was kept at 1.0 mM here such that the 87 production rate of the degradative enzyme was decreased (see Figure S4.18). The error bars for the simulations represent the standard error over 100 iterations. The error bars for the experiments represent the standard error over 3 sets of experiments. 4.6: Discussion With the recent advances in systems and molecular biology, many new studies are being conducted to understand the complex behaviors of microbes that arise due to cellular and spatial heterogeneity [12,13,38,107,114,144–146]. Several studies have shown that many cellular systems exhibit emergent phenomena previously studied in non-biological contexts and that can be explained using physical and statistical mechanical rules [147–150]. For instance, Wioland et al. discuss the capability of bacterial vortices, whose lattices are hydrodynamically coupled, to spontaneously form into distinct patterns characterized by ferro- and antiferromagnets [147]. Rudge et al. observed the emergence of striking fractal patterns in growing colonies of E. coli [150]. Physical models such as the Hopfield model have also been implemented to predict how the capacity for groups of cells to make decisions scales with the number of cell types [146]. In the spirit of these observations, here we have identified that disrupting the long-range communication of a quorum sensing microbe through interference by a degrader strain yields complex physical behavior characteristic of a percolation transition. Critical transitions in cellular distributions have been previously reported. For example, Ratzke et al. reported that agar concentration tuned the distribution of cells from one large cluster to small patchy clusters [87]. However, this phenomena was not uanalyzed as a percolation transition. More recently, a percolation transition was reported for electrochemical signaling in biofilm of Bacillus subtilis [44] . As was observed in the quorum sensing network, the connectivity of active cells sharply changed at a critical point and the cluster size of active cells scaled as a power law. These studies emphasize the utility in applying physical models to study biological networks and their emergent behaviors, an approach that is expected to continue to yield insights into the fundamental capabilities of these systems and potentially reveal new strategies for robust control of system function. In our specific systems, it was particularly surprising that the spatial distribution of quorum sensing activity was extremely sensitive to the number of degrader cells. Although in well mixed simulations (Figure S4.19), we observed a similar ratio of degraders to senders of 88 1:2 was needed to turn off the senders. The long distance coordination of quorum sensing activity was robust to small amounts of interference by degraders, but as the number of degrader cells approached the number of sender cells there was a sharp change in the connectivity of activated regions. The cellular network transitioned from large connected regions of activity to smaller patches of active cells over a factor of 10 change in cell number, and this change followed a classic 2D percolation transition. These results demonstrate that cellular networks can undergo critical changes in behavior as a function of the ratio of cell types. In our previous work [114], we analyzed how signal degrading cells influence the spatial range over which a colony of sender cells can activate a circular pattern of quorum sensing up to a maximum radius that depended on the number of degrader cells. Here, we examined spatial patterns of activation that naturally emerge from well mixed populations of sender and degrader strains, and revealed scale-free patterns of quorum sensing activation that fit the definition of a percolation transition. Since changes in signaling states, including quorum sensing activation, are important for many coordinated functions of cellular networks, including virulence and biofilm formation [96,151], small changes in the composition of a community might be a strategy to control the global function of microbial communities. Successful implementation of such a strategy requires knowledge of the critical point of the transition and how it scales with system parameters, such as enzyme activity, protein expression, and growth dynamics. In our system, additional theoretical analysis in the Supplementary Text 1 determined the major factors that dictate the transition point of percolation were the degradation rate of the AHLs by the AiiA enzyme, the degradation of the AiiA enzyme by the media, the production rate of the AHLs and the production rate of the AiiA enzyme. Here we demonstrated that the critical point can indeed be tuned by changing the production rate of the signal degrading enzyme. When analyzing our results as a 2D percolation transition, the critical exponents were calculated in both experiments and theoretical simulations [43,143]. The slight changes in the exponent values from the experiments compared with the classical theory can be attributed to the inefficiencies in measurements caused by subtle inhomogeneity in the cell distributions, and to the fact that we are using a finite system as opposed to an infinite system [41,42]. Through additional image analysis such as windowing and finite scaling, we observed that the values of 89 the critical exponents are robust universal quantities, with fairly small deviation from the infinite classical values even as the system size gets smaller. Future work in this direction should investigate signaling dynamics within more complex environments, such as those found in situ, to see if there are instances where the qualitative network behavior can indeed be described by ordinary percolation. Since quorum sensing enables bacteria to exchange information regarding the local environment [3,40,52], the consequences of breaking down such communication by interference raises questions about how cells have evolved to coordinate activity in the presence of interference. The patterns that emerged in our experiments were stable for many hours, although how the activity of cellular networks evolves over much longer time periods and the consequences of being far-from- equilibrium warrant further study. The critical behavior observed here is yet another example of biological systems operating near critical conditions [148,149,152–158]. Examination of natural signaling networks that have been tuned to function in the vicinity of a percolation transition should help to understand the importance of criticality in such biological contexts. 4.6: Methods 4.6.1: Bacterial strains The sender strain is Escherichia coli (NEB 5-alpha) with the plasmid pTD103IuxI sfGFP(Kn) obtained from Prindle et al. [107] Here, luxI, luxR and sfGFP are under the luxIP quorum sensing promoter. When quorum sensing has been activated, the senders will turn on a green fluorescent response. The degrader strain is Escherichia coli (NEB 5-alpha) with the plasmid pTD103aiiA_lac (Figure S4.18a). This plasmid was constructed by mutagenesis, using NEB Q5 ® site-directed mutagenesis kit, by replacing the luxIp promoter in the plasmid pTD103aiiA obtained from Prindle et al. [107] with the lacUV5 constitutive promoter. The degradative enzyme, AiiA, is constitutively produced under lacUV5 promoter. 90 The strain with the lower AiiA production rate was constructed by Gibson assembly, using the NEB Gibson assembly ® master mix, by replacing the yfp gene sequence in the plasmid pZS25O1+11-YFP, obtained from Garcia et al. [159], with the aiiA gene sequence (Figure S4.18b). The aiiA gene sequence was obtained from pTD103aiiA_lac. Since the new plasmid pZS25O1+11_AiiA has the O1 repressor binding site, the expression level of the AiiA enzyme can be controlled by isopropyl β-D-1-thiogalactopyranoside (IPTG). Primers used for the mutagenesis and Gibson assembly are given in Table 4.2. The constitutive red fluorescent sender strain was constructed by electroporation of the plasmid pTD103IuxI sfGFP(Spec) into the Escherichia coli TK140 pZS25O1+11-mCherry obtained from Garcia et al. [159]. For the dummy strain in Figure S4.3 we used the strain, Escherichia coli TK140 pZS25O1+11-mCherry obtained from Garcia et al. [159]. Name Function Sequence AiiA LacUV5 F The forward primer to replace the luxIP promoter from pTD103aiiA with the lacUV5 vector to construct the plasmid pTD103aiiA_lac. GGCTCGTATAATGTGTGGGAATTCATT AAAGAGGAGAAAG AiiA LacUV5 R The reverse primer of the above. GGAAGCATAAAGTGTAAAGCTTATGT TAAGTAATTGTATTC pZS_YFP_1F The forward primer used to obtain the plasmid backbone, excluding the YFP, from the plasmid pZS25O1+11-YFP for the construction of the plasmid pZS25O1+11_AiiA. GAA AAT TAC GCC CTT GCA GCG TAA AAG CTT AAT TAG CTG A pZS_YFP_1R The reverse primer of the above. CTG GGA TGA AAT AAA GTT TCT TTA CTG TCA TTG CGG TAC CTT TCT C pZE_AiiA_F The forward primer used. GGA GAA AGG TAC CGC AAT GAC AGT AAA GAA ACT TTA TTT CAT CCC AG pZE_AiiA_R The reverse primer of the above. GAC TCA GCT AAT TAA GCT TTT ACG CTG CAA GGG CGT AAT TTT C Table 4.2: Primer sequences used for making the plasmids. 91 4.6.2: Bacterial growth conditions Similar to Silva et al. [12], the cells were grown in 5ml of Lysogenic broth (LB) which were in 14mL Falcon tubes with appropriate antibiotics for plasmid maintenance. The cultures were put into a shaker at 200 RPM and 37 o C. Cells were grown in liquid culture for 16 hours. 4.6.3: Bacterial strains In experiments, sender culture was transferred to a micro centrifuge tube and centrifuged it at 15000 RPM for 1min. The supernatant was discarded, cells were re-suspended in fresh media, cell number was adjusted to 10 9 sender cells, and a variable amount of degrader cells were added. The final volume of culture was 1 mL, such that the volume of culture loaded onto the LB-agar plate was the same in all experiments. The 1 mL mixture with 10 9 : n amount of sender: degrader, was pipetted on to a 3 mm LB-agar Petri dish (Figure 4.1(b)). This plate was then incubated at 37 o C for 16 hours. 4.6.4: Microscopic measurements and image analysis After 16 hours growth the plate was loaded into a Nikon eclipse TI fluorescent microscope and GFP fluorescent images were obtained at a magnification of 20x. Exposure time was 500 ms. The images were analyzed in Matlab 2016b using custom analysis algorithms. For each plate, a minimum of 16 different regions were imaged and analyzed. In these codes, we first set the same threshold value (thr) for fluorescence intensity to identify quorum sensing activated regions. Each pixel had an intensity value of 1 or 0 representing quorum sensing activation or non-activation respectively. Next, each activated cluster is identified and assigned a number. If an activated pixel is being occupied by a neighboring pixel (including diagonal neighbors), the two pixels are considered to be in the same activated cluster. After the activated clusters are labeled, we looked at whether active clusters spanned from top to bottom or left to right to determine if for each image, a percolation transition occurred. Next properties such as, areas of clusters and mean distance between two active pixels 92 within the same cluster was determined to obtain the scaling properties mentioned in the manuscript. 4.7: Supplementary information Figure S4.1: Senders mixed with degraders. The images in the left are 20x fluorescent microscopy images of the senders, when mixed with the degraders at a ratio of 10 9 :6x10 8 (senders: degraders). The senders used here are capable of constitutively producing a red fluorescent protein, under the lacUV5 promoter, and at quorum sensing activation the senders turn on a green fluorescent response, regulated by the plux promoter. Here we have a green fluorescent image (left), which indicates quorum sensing activation. The red fluorescent image (middle), represents that the senders are distributed over the whole plate. The red florescent pixel intensity (right), represents the distribution of the senders across the blue line, shown in Figure 4.2a in the manuscript. The dark regions in the green fluorescent image is due to the senders not activating quorum sensing due to the presence of degraders. 93 Figure S4.2: Choosing the threshold to determine quorum sensing activation. At 20x microscopy, we imaged the senders with respect to time for the case of zero-degraders and indicated the average fluorescent pixel intensity vs time. Here the fluorescent pixel intensity (y- axis) is the raw fluorescent intensity obtained from the microscope subtracted by the background intensity. The error bars are the standard error calculated from 6 different measurements taken from a 6 different random regions imaged from the same plate over time. The threshold (thr) was chosen to be 25% of the maximum fluorescent intensity detected. thr was the same for all images for data analysis. Figure S4.3: 20x microscopy images of the senders mixed with a dummy strain at high density. Here the dummy strain does not produce any degradative enzymes. The dummy strain constitutively produces a red fluorescent response under the promoter lacUV5. When mixed in a 1:1 ratio (left 2 images), the senders activate quorum sensing across the entire plate. This result demonstrates that competition for space or resources does not create the dark regions observed in 94 experimental images. This Figure was taken from Silva et al. PLOS Computational Biology, 13(10): e1005809, (2017). The right figure show the sender and the dummy strains mixed at a 1:3 ratio, and the senders still activate quorum sensing across the entire image. We observe that the percolation frequency for both cases is 1. (Note that the senders used here are different from the ones used in Figure S4.1, as they are not producing red fluorescence.) Figure S4.4: Senders maintain uniform coverage of the plate at higher ratios of degraders. Here we show RFP and GFP 20x fluorescent images of the senders while increasing the amount of degraders by 0, 5x10 8 , 8x10 8 and 10 9 . The red fluorescence, under the lacUV5 promoter, is 95 constitutively produced by the senders. Upon quorum sensing activation, senders produce green fluorescent protein, which is regulated by the plux promoter. We calculate the amount of GFP pixels divided by the amount of RFP pixels (bottom, left) and observe that as the amount of degraders increase, the amount of senders with GFP activation decreases. As we increase the amount of degraders, the amount of RFP pixels does not significantly change (bottom, right). For the fluorescent images we apply thresholds to calculate the amount of pixels. For green fluorescence, the same threshold as in Fig. S2 was applied. For red fluorescence, a threshold value 25% of the maximum detected intensity was used. The error bars represent the standard error from 6 different imaged regions taken from the same experiment. Figure S4.5: Determining the amount of senders needed to uniformly cover the whole plate. Here we distribute 10 6 , 10 7 , 10 8 and 10 9 senders (from left to right), and observe that the total number of green fluorescent pixels spreads over the whole plate when the amount of senders are 10 9 . 96 Figure S4.6: Changing the threshold slightly to observe changes to the amount of active pixels. a. Here we have thresholded fluorescent images for when we have 10 8 degraders mixed 97 with 10 9 senders. We changed the threshold by adding or subtracting 10%, 20%, 30%, 50% and 90% to the threshold value for this image and b. calculated the amount of activated pixels (pixels above the threshold). For this image, up to a change of 20% does not have a significant effect on the amount of active pixels. c. We changed the threshold for 50 images by adding or subtracting 10% and 20% from the threshold value and calculated the amount of active pixels. The first pie chart represents that all the amount of active pixels remained the same for all the images (100%) when the threshold was changed by 10%. The second pie chart represents that 1 out of the 50 images had a change (2%) in the amount of active pixels for a 20% change. 4.7.1: Approximating a system with degraders onto an effective degrader-free system. The equations that follow are extensively explained in the theory section of the main manuscript. All parameter values are given by Table S4.1. [𝐴𝐻𝐿 ] – AHL concentration 𝐷 𝐴𝐻𝐿 – Coefficient of diffusion of the AHL through the media 𝛾 1 − Rate of degradation of the AHL by the media 𝑛 𝐴𝐻𝐿 − Amount of senders 𝜌 𝑏 1 − Basal rate of AHL production 𝜌 𝐴𝐻𝐿 − Induced rate of AHL production 𝑚 1 − Hill’s coefficient 𝜃 1 − AHL threshold for QS activation [𝐴𝑖𝑖𝐴 ]- Concentration of the degrader 𝛾 2 - Rate of degradation of the AHL by AiiA 𝑚 2 − Hill’s coefficient 𝜃 2 − Threshold for high degradation by AiiA 𝜌 𝐴𝑖𝑖𝐴 - Production rate of AiiA 98 𝑛 𝐴𝑖𝑖𝐴 - Amount of degrader cells 𝛾 3 - Degradation rate of the AiiA Consider the steady-state conditions for the system of reaction-diffusion equations: 𝜕 [𝐴𝐻𝐿 ] 𝜕𝑡 = 𝐷 𝐴𝐻𝐿 ∇ 2 [𝐴𝐻𝐿 ] − 𝛾 1 [𝐴𝐻𝐿 ] + 𝑛 𝐴𝐻𝐿 (𝜌 𝑏 1 + 𝜌 𝐴𝐻𝐿 [𝐴𝐻𝐿 ] 𝑚 1 [𝐴𝐻𝐿 ] 𝑚 1 +𝜃 1 𝑚 1 ) − 𝛾 2 [𝐴𝑖𝑖𝐴 ] [𝐴𝐻𝐿 ] 𝑚 2 [𝐴𝐻𝐿 ] 𝑚 2 +𝜃 2 𝑚 2 = 0 (4.4) 𝜕 [𝐴𝑖𝑖𝐴 ] 𝜕𝑡 = 𝑛 𝐴𝑖𝑖𝐴 𝜌 𝐴𝑖𝑖𝐴 − 𝛾 3 [𝐴𝑖𝑖𝐴 ] = 0 (4.5) We can solve (4.4) for [AiiA] and plug the solution back into (4.5). This yields an effective single equation: 𝜕 [𝐴𝐻𝐿 ] 𝜕𝑡 = 𝐷 𝐴𝐻𝐿 ∇ 2 [𝐴𝐻𝐿 ] − 𝛾 1 [𝐴𝐻𝐿 ] + 𝑛 𝐴𝐻𝐿 (𝜌 𝑏 1 + 𝜌 𝐴𝐻𝐿 [𝐴𝐻𝐿 ] 𝑚 1 [𝐴𝐻𝐿 ] 𝑚 1 +𝜃 𝑚 1 ) − 𝛾 2 𝛾 3 𝑛 𝐴𝑖𝑖𝐴 𝜌 𝐴𝑖𝑖𝐴 [𝐴𝐻𝐿 ] 𝑚 2 [𝐴𝐻𝐿 ] 𝑚 2 +𝜃 2 𝑚 2 = 0 (4.6) This can be interpreted as the steady-state condition for an equivalent degrader-free system with an ‘effective’ basal production rate 𝜌 𝑏 1 𝑒𝑓𝑓 : 0 = 𝐷 𝐴𝐻𝐿 ∇ 2 [𝐴𝐻𝐿 ] − 𝛾 1 [𝐴𝐻𝐿 ] + 𝑛 𝐴𝐻𝐿 (𝜌 𝑏 1 𝑒𝑓𝑓 + 𝜌 𝐴𝐻𝐿 ) = 0 (4.7) 𝜌 𝑏 1 𝑒𝑓𝑓 = 𝜌 𝑏 1 − 𝛾 2 𝜌 𝐴𝑖𝑖𝐴 𝑛 𝐴𝑖𝑖𝐴 𝛾 3 𝑛 𝐴𝐻𝐿 (4.8) In our experiments, our only variable control parameter is the amount of degraders n AiiA. We want to analyze the onset of the percolation transition as we vary this parameter. To the extent that the steady state approximation is valid, the dependence of this percolation behavior on nAiiA, at a fixed nAHL, will be entirely encoded by how nAiiA enters into 𝜌 𝑏 1 𝑒𝑓𝑓 . More specifically, our percolation transition is characterized by scaling relations for the average cluster size ⟨s⟩, the probability of being in the giant connected component P∞, and the correlation length ξ: 99 〈𝑠 〉 ∝ |𝑝 (𝜌 𝑏 1 𝑒𝑓𝑓 ) − 𝑝 𝑐 | −𝛾 (4.9) 𝑃 ∞ ∝ |𝑝 (𝜌 𝑏 1 𝑒𝑓𝑓 ) − 𝑝 𝑐 | −𝛽 (4.10) 𝜉 ∝ |𝑝 (𝜌 𝑏 1 𝑒𝑓𝑓 ) − 𝑝 𝑐 | −𝜈 (4.11) and the scaling exponents have universal values (in 2D) of γ = 43/18, β = 5/36, ν = 4/3. The challenge, then, is in determining the mapping p( 𝜌 𝑏 1 𝑒𝑓𝑓 ) that allows us to regulate the percolation control parameter p via the experimental ‘knob’ nAiiA.. For convenience, we will just analyze the onset of the transition in the degrader free regime, then map the system with degraders onto the equivalent degrader free system to generalize. 4.7.2: Mapping basal production rate onto site occupation probability in the degrader-free limit. In the degrader free limit, as we vary the basal rate 𝜌 𝑏 1 , we can now ask the following question: for a given 𝜌 𝑏 1 , what is the probability p that a given site is activated? For simplicity, let us take the limit that the number of lattice sites N → ∞, and ask the slightly simpler question: what is the probability that the lattice site at the center is activated, pcenter? It is clear that by symmetry, in the infinite ‘bulk’ limit, p = pcenter for all lattice sites. This ansatz amounts to neglecting any surface boundary effects that modify p away from the interior of the lattice. Now, let us consider the space of possible ways that the center can be activated. This space is, for all practical purposes, infinite, since there are an essentially infinite number of ways that some sites can be activated while others remain inactivated. However, for simplicity, we will make the following approximation: we assume that if there is any activation in the system, the center site must activate. Heuristically, this assumption can be justified by the argument that the center site has more nearest-neighbors, next-nearest neighbors, and so forth than any other site. Therefore, if any other lattice site in the system has a high enough density of neighbors to induce activation, then the center site has a high likelihood of having at least that much density, and probably a greater amount. Nevertheless, it should be noted that this is only a leading- order approximation, and strictly speaking, there is a non-zero likelihood of having inhomogeneous ‘islands of activation’ that do not reach the center. However, these higher-order corrections are 100 unlikely to change the qualitative form of p(𝜌 𝑏 1 ). Future work should consider incorporating these effects to more rigorously define the mapping p(𝜌 𝑏 1 ). With this assumption, then, we find that the problem has been greatly simplified. In particular, we can just take all cells to be at their basal activity level 𝜌 𝑏 1 , and then ask the question: with this minimal activity, what is the probability that the [AHL] at the center lattice site is above the threshold concentration θ? This probability will serve as our approximation for p(𝜌 𝑏 1 ) in the bulk limit. We can determine this distribution in the following manner. Consider an infinitesimal shell of radius dr located between r and r+dr. Let us ask the question, what is the expected distribution of AHL signals due to senders in this shell, pexpected(𝜌 𝑏 1 ; r)? By radial symmetry, we know that any cell in this region will contribute an equal number of signaling molecules to the center, [AHLcontributed](r), since they are all at an equal distance r. Furthermore, we know that [AHLcontributed](r) will be linearly proportional to 𝜌 𝑏 1 . We can summarize this with the statement [𝐴𝐻𝐿 𝑐𝑜𝑛𝑡𝑟𝑖𝑏𝑢𝑡𝑒𝑑 ](𝑟 ) = 𝜌 𝑏 1 𝑓 (𝑟 ) (4.12) where f(r) is a radially symmetric function. Thus, the expected distribution of signals is simply equal to the expected distribution of the number of cells in the infinitesimal shell, multiplied by the number of signals contributed per cell 𝑝 𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑 (𝑛 𝐴 𝐻 𝐿 ; 𝑟 ) = 𝑝 𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑 (𝑛 𝑐𝑒𝑙𝑙𝑠 ; 𝑟 ) × [𝐴𝐻𝐿 𝑐𝑜𝑛𝑡𝑟𝑖𝑏𝑢𝑡𝑒𝑑 ](𝑟 ) (4.13) Since the areal amount of sender cells is nAHL/ATotal, where ATotal is the total system area, the number of cells in the infinitesimal shell is a Poisson-distributed random variable with mean 𝑚 ̅ 𝑐𝑒𝑙𝑙𝑠 , 𝑚 ̅ 𝑐𝑒𝑙𝑙𝑠 (𝑟 ) = 𝑛 𝐴𝐻𝐿 𝐴 𝑇𝑜𝑡𝑎𝑙 × 2𝜋𝑟 𝑑𝑟 (4.14) and thus, the contributed number of signaling molecules at the center is likewise a Poisson distributed random variable with mean 𝑚 ̅ 𝐴𝐻𝐿 , 𝑚 ̅ 𝐴𝐻𝐿 (𝑟 ) = [𝐴𝐻𝐿 𝑐𝑜𝑛𝑡𝑟𝑖𝑏𝑢𝑡𝑒𝑑 ](𝑟 ) × 𝑛 𝐴𝐻𝐿 𝐴 𝑇𝑜𝑡𝑎𝑙 × 2𝜋𝑟 𝑑𝑟 (4.15) 101 = 𝜌 𝑏 1 𝑓 (𝑟 ) × 𝑛 𝐴𝐻𝐿 𝐴 𝑇𝑜𝑡𝑎𝑙 × 2𝜋𝑟 𝑑𝑟 (4.16) Integrating, the total concentration at the center is the sum of many independent Poisson distributed random variables, so it itself is a Poisson distributed random variable with mean 𝑚 ̅ 𝑡𝑜𝑡𝑎𝑙 , 𝑚 ̅ 𝑡𝑜𝑡𝑎𝑙 = 𝜌 𝑏 1 𝑛 𝐴𝐻𝐿 𝐴 𝑇𝑜𝑡𝑎𝑙 × ∫ 𝑓 (𝑟 ) × 2𝜋𝑟 𝑑𝑟 ∝ 𝜌 𝑏 1 (4.17) The last proportionality is a demonstration of the remarkable fact that the proportionality turns out to be simply linear in 𝜌 𝑏 1 , with no need to explicitly evaluate the integral! In other words, if we parameterize 𝑚 ̅ 𝐴𝐻𝐿 = α 𝜌 𝑏 1 , p(𝜌 𝑏 1 ) ≈ pcenter simplifies to just the probability that a Poisson distributed random variable 𝜌 𝑏 1 , with mean 𝑚 ̅ 𝑡𝑜𝑡𝑎𝑙 , is greater than or equal to θ: 𝑝 (𝜌 𝑏 1 ) ≈ 𝑝 (𝑛 > 𝜃 ; 𝑚 ̅ 𝑡𝑜𝑡𝑎𝑙 = 𝛼 𝜌 𝑏 1 ) = 𝑃 (𝜃 , 𝛼 𝜌 𝑏 1 ) (4.18) where P(s, x) is the regularized gamma distribution with shape parameter θ and scale parameter α 𝜌 𝑏 1 . Taylor expanding, this function can be approximated as: 𝑃 (𝜃 , 𝛼 𝜌 𝑏 1 ) ≈ 𝛼 𝜃 𝜌 𝑏 1 (4.19) In other words, to a first approximation, the occupation probability is linearly proportional to 𝜌 𝑏 1 , or more generally if we allow for degraders, 𝜌 𝑏 1 𝑒𝑓𝑓 . 4.7.3: Consequences for observed critical exponents as degrader density is varied. Let us now return to our direct experimental situation, where the only variable control parameter is nAiiA, which is of course related to 𝜌 𝑏 1 𝑒𝑓𝑓 by (4.7). Combining (4.7) and (4.18), we find that   𝑝 ≈ 𝛼 𝜃 (𝜌 𝑏 1 − 𝛾 2 𝜌 𝐴𝑖𝑖𝐴 𝑛 𝐴𝑖𝑖𝐴 𝛾 3 𝑛 𝐴𝐻𝐿 ) (4.20) If we plug (4.19) into our scaling relations (4.8-4.10), we find that the exponents in terms of nAiiA are exactly the same as those in terms of p: 〈𝑠 〉 ∝ |𝑝 (𝜌 𝑏 1 𝑒𝑓𝑓 ) − 𝑝 𝑐 | −𝛾 ∝ |𝑛 𝐴𝑖𝑖𝐴 − 𝑛 𝐴𝑖𝑖𝐴 𝑐 | −𝛾 (4.21) 102 𝑃 ∞ ∝ |𝑝 (𝜌 𝑏 1 𝑒𝑓𝑓 ) − 𝑝 𝑐 | −𝛽 ∝ |𝑛 𝐴𝑖𝑖𝐴 − 𝑛 𝐴𝑖𝑖𝐴 𝑐 | −𝛽 (4.22) 𝜉 ∝ |𝑝 (𝜌 𝑏 1 𝑒𝑓𝑓 ) − 𝑝 𝑐 | −𝜈 ∝ |𝑛 𝐴𝑖𝑖𝐴 − 𝑛 𝐴𝑖𝑖𝐴 𝑐 | −𝜈 (4.23) where nAiiAc is defined by setting (17) equal to pc:   𝑝 𝑐 = 𝛼 𝜃 (𝜌 𝑏 1 − 𝛾 2 𝜌 𝐴𝑖𝑖𝐴 𝑛 𝐴𝑖𝑖𝐴 𝑐 𝛾 3 𝑛 𝐴𝐻𝐿 ) (4.24) As a result, when we run simulations and experiments varying only nAiiA, and observe only ⟨s⟩, P∞, and ξ, a log-log regression should give the same universal exponent values that we would expect if we could somehow explicitly vary p itself! These exponents should be universal and independent of specific parameter values. Parameter Value 𝜇 1.50 hrs -1 S 10 9 cells DAHL 1.764 mm 2 hrs- 1 𝜌 𝐴𝐻𝐿 2.3 x 10 -9 nM hrs -1 per cell 𝜌 𝑏 1 2.3 x 10 -10 nM hrs -1 per cell 𝜃 1 70 nM m1 2.5 𝛾 1 0.005545 hrs -1 𝛾 2 0.01 hrs -1 𝜌 𝐴𝑖𝑖𝐴 2.3 x 10 -9 nM hrs -1 per cell 𝛾 3 0.005545 hrs -1 𝜃 2 70 nM m2 2.5 Table S4.1: The values of the parameters used for the modeling. All of these parameters are obtained from references [12,111,114]. 103 Figure S4.7: A linear relationship is observed in the critical region for the number of activated sites and the amount of degraders. The normalized number of activated sites vs amount of degrader (a) from experiments and (b) from simulations. Activated sites here represents the green fluorescent pixels above the threshold (thr). The number of activated sites were normalized by the amount of the activated sites for the case of amount of degrader = 0. A linear relationship between the activated sites and the amount of degraders in the critical region is observed, confirming the validity of the relationship derived in equation (4.19) of the supporting documents. 104 Figure S4.8: Determining the critical amount of degraders (nAiiAc). Here we have the average cluster size vs amount of degrader (left) and the correlation length vs amount of degraders (right). We observe that at the amount of degrader = 6.5 x 10 8 , the average cluster size and the correlation length goes to its maximum value. Hence, we chose n AiiAc = 6.5 x 10 8 . The error bars represent the standard error from multiple replicates. Figure S4.9: Data range to calculate the critical exponents. Data point near the percolation transition were used to calculate the critical exponents. Range of amount of degraders used to fit for the critical exponents is shown in red. Figure S4.10: Sensitivity of the critical exponents to range of data used in fits. The critical exponents were calculated using varying numbers of data points on either side of the critical 105 transition. In Figure 4.3, ɤ was calculated using 6 largest degrader densities less than the critical density and β and ν were calculated using 7 points on either side of the critical density. XIII. SENSITIVITY OF THE CRITICAL EXPONENTS TO THE THRESHOLD. Figure S4.11: Sensitivity of the critical exponents to the threshold. We changed the threshold by adding or subtracting 10%, 20% and 30% to the threshold and calculated the changes to the critical exponents. The critical exponents of a. the average cluster size, b. order parameter, c. correlation length. 106 Figure S4.12: Windowing the experimental images (2x2) and measuring the scaling properties. (a) To window the image we divide the full microscopy image to four equal sized segments. (b) Average cluster size. (c) Order parameter. (d) Correlation length. The blue dots represent the values extracted from the full image while the red dots represent the values from the windowed images. The error bars represent the standard error from the mean of all the images. Table S4.2: The values of the critical exponents from full and windowed experimental images. The error represented here is the standard error (SE) of the slope. Critical exponents 𝜸 𝐏 β 𝐏 Ν 2D percolation universality class 2.39 0.14 1.33 Full image 2.28 ± 0.36 0.17± 0.06 1.36± 0.12 2x2 windowed image 2.31± 0.45 0.17± 0.05 1.48 ± 0.48 107 Figure S4.13: Calculating the critical exponents of the transition from reaction-diffusion simulations for the average cluster size. Data from Figure 4.4C showing the relationship between <s> and the (𝒏 𝑨𝒊𝒊𝑨 − 𝒏 𝑨𝒊𝒊𝑨 𝒄 ) are plotted on a log-log plot. All the data points on either side of the 𝒏 𝑨𝒊𝒊𝑨 𝒄 are fitted to a linear plot using least-square fit method. The error bars represent the log propagation of the errors from Figure 4.4C. The gradient value of the best fit slope with the standard error, (see caption of Table 4.1) is 2.42± 0.16. 108 Figure S4.14: Calculating the critical exponents of the transition from reaction-diffusion simulations for the order parameter. Data from Figure 4.4C showing the relationship between P∞ and the (𝒏 𝑨𝒊𝒊𝑨 − 𝒏 𝑨𝒊𝒊𝑨 𝒄 ) are plotted on a log-log plot. All the data points on either side of the 𝒏 𝑨𝒊𝒊𝑨 𝒄 are fitted to linear plot using least-square fit method. The error bars represent the log propagation of the errors from Figure 4.4C. The gradient value of the best fit slope with the standard error is 0.16± 0.03. 109 Figure S4.15: Calculating the critical exponents of the transition from reaction-diffusion simulations for the correlation length. Data from Figure 4.4C showing the relationship between ξ and the (𝒏 𝑨𝒊𝒊𝑨 − 𝒏 𝑨𝒊𝒊𝑨 𝒄 ) are plotted on a log-log plot. All the data points on either side of the 𝒏 𝑨𝒊𝒊𝑨 𝒄 are fitted to linear plot using least-square fit method. The error bars represent the log propagation of the errors from Figure 4.4C. The gradient value of the best fit slope with the standard error is 1.35± 0.06. 110 Figure S4.16: Finite-size scaling of the simulation results. Simulations, as shown in Figure 4.4, were rerun for systems smaller than the standard size of L=150. From these simulations, the (a) average cluster size, (b) order parameter, and (c) correlation length were calculated for variable amounts of degrader cells. The error bars represent the standard error from 100 simulations. L represents the system size. We observe similar scaling behaviors for different system sizes. 111 Table S4.3: The values of the critical exponents from the finite scaled simulation results. Here L represents the system size. The error represented here is the standard error (SE) of the slope. Figure S4.17: Critical exponents diverges from the classical values for simulations of smaller systems. From left to right, we have listed critical exponents of the average cluster size, order parameter and the correlation length for different system sizes. We have added the classical value to the bar plots for comparison. The error represented here is the standard error of the critical exponents. Critical exponents 𝜸 𝐏 β 𝐏 ν 2D percolation universality class 2.39 0.14 1.33 L=150 2.42± 0.16 0.16± 0.03 1.35 ± 0.06 L=125 2.35± 0.35 0.18± 0.07 1.47± 0.35 L=100 2.65± 0.39 0.23± 0.12 1.53± 0.45 L=75 1.95± 0.53 0.24± 0.14 1.14± 0.62 L=50 1.85± 0.50 0.29± 0.10 1.09± 0.53 112 Figure S4.18: A strain with a lower production rate of the degradative enzyme. (a) The plasmid map of the degrader strain with higher production rate of the degradative enzyme. (b) The plasmid map of the degrader strain with lower production rate of the degradative enzyme. The production rate is regulated by isopropyl β-D-1-thiogalactopyranoside (IPTG) (c) We mix 50 µl of the senders with 50 µl of the degraders in the plate reader. Data shows that when we increase the IPTG concentration, the fluorescence per cell for the sender strain is reduced, indicating control over the production rate of the degradative enzyme. The green line shows the fluorescence per cell of the senders when mixed with the higher production rate degraders, containing the plasmid is (a). For the experiments with the plate assay, an IPTG concentration of 1 mM was used. Error bars are the standard error from three replicates. Critical exponents 𝜸 𝐏 β 𝐏 Ν 2D percolation universality class 2.39 0.14 1.33 From experiments 2.53± 0.22 0.19± 0.05 1.51± 0.21 From reaction- diffusion simulations 2.23± 0.20 0.18± 0.05 1.44 ± 0.14 Table S4.4: The values of the critical exponents for the case of the degrader with lower production rate. The error represented here is the standard error (SE) of the slope. 113 Figure S4.19: Well mixed simulations of the changes in the AHL concentration with respect to the amount of degraders. In simulations, we mix 10 9 senders with different amount of degraders (0,10 8 , 2x10 8 , 3x10 8 , 4x10 8 , 5x10 8 , 7x10 8 , 10 9 ) and observe that at least 5x10 8 degraders are needed to bring the AHL concentration below the threshold (𝜽 𝟏 = 𝟕𝟎 𝒏𝑴 , see Table S4.1). 114 Chapter 5: A neural network model predicts community-level signaling states in a diverse microbial community This work is currently submitted to PLOS Computational Biology. 5.1: Abstract Signal crosstalk within biological communication networks is common, and such crosstalk can have unexpected consequences for decision making in heterogeneous communities of cells. Here we examined crosstalk within a bacterial community composed of five strains of Bacillus subtilis, with each strain producing a variant of the quorum sensing peptide ComX. In isolation, each strain produced one variant of the ComX signal to induce expression of genes associated with bacterial competence. When strains were combined, a mixture of ComX variants was produced resulting in variable levels of gene expression in each strain. To examine gene regulation in mixed communities, we implemented a neural network model. Experimental quantification of asymmetric crosstalk between pairs of strains parametrized the model, enabling the accurate prediction of activity within the full five-strain network. Unlike the single strain system in which quorum sensing activated upon exceeding a threshold concentration of the signal, crosstalk within the five-strain community resulted in multiple community-level quorum sensing states, each with a unique combination of quorum sensing activation among the five strains. Quorum sensing activity of the strains within the community was influenced by the combination and ratio of strains as well as community dynamics. The community-level signaling state was altered through an external signal perturbation, and the output state depended on the timing of the perturbation. Given the ubiquity of signal crosstalk in diverse microbial communities, the application of such neural network models will increase accuracy of predicting activity within microbial consortia and enable new strategies for control and design of bacterial signaling networks. 5.2: Introduction In microbiology, quorum sensing (QS) is a process where bacteria produce and secrete small chemical molecules known as auto-inducers. Many bacteria regulate gene expression in 115 response to the external concentration of auto-inducer, including regulation of processes related to biofilm formation, virulence, and horizontal gene transfer [52,95,104,160]. Although QS is historically viewed as a process of a single species regulating its own gene expression, numerous reports have shown signal exchange between species contributed to regulation of QS phenotypes [11–15]. Such crosstalk between cells is usually the result of two bacterial strains producing chemical variants of a QS signal. QS signals have many naturally occurring chemical variants, including 56 distinct variations of acyl homoserine lactones and 231 variants of auto- inducing peptides (AIP) [98,99]. Chemically similar variants of a signal interact with QS receptors, leading to excitation or inhibition of QS activation to a variable degree [3,12,15,24]. Multiple signal inputs to a given receptor protein lead to variable levels of gene expression, making it difficult to predict community-level behaviors in the presence of two or more signaling molecules. Signal crosstalk was first recognized when Vibrio cholera and Vibrio parahaemolyticus produced a QS response in Vibrio harveyi [19]. Riedel et al. [161] observed QS crosstalk between Burkholderia cepacia and Pseudomonas aeruginosa where P. aeruginosa activated QS in B. cepacia. Mclean et al. [15] showed that a Chromobacterium violeceum biosensor produced different levels of QS activation when introduced with a variety of distinct Acyl-Homoserine Lactones (AHLs) separately. Geisinger et al. [162] detected QS activity in pairwise combinations of the Agr-I-IV QS AIP system to uncover the contribution of divergent QS alleles to variant expression of virulence determinants within Staphylococcus aureus. Although these studies identified the potential for crosstalk in the presence of pairwise combinations of cells, in a given environment QS crosstalk can be more complex when several QS bacteria coexist in the same community. For instance, in the human microbial gut 300-500 bacterial species are present [30] and among these populations at least ten QS species have been identified and more than eight QS signal variants have been recognized [163]. Thompson et al. [31] showed that the ratio between Firmicutes and Bacteroidetes can be influenced by introducing an external source of QS signals, demonstrating that changes in QS signals can have community-level influences on activity. Furthermore, in rhizosphere soil out of the 350-550 bacterial species, at least 30 species were capable of producing multiple QS signals [164], and another study found 8% of genomes in the soil contained the genes needed to activate an AHL reporter strain [165]. 116 Interest in engineering microbial communities to utilize multiple QS signals led to a broader characterization of gene regulation in contexts with multiple signals. Several early examples of cellular networks with multiple signals utilized the LasRI and RhlRI networks from Pseudomonas aeruginosa, which produce nearly orthogonal signals C4-HSL and 3-oxo-C12- HSL, with essentially zero crosstalk. Later work combined signaling networks with measurable crosstalk. Scott et al. [16] conducted an extensive study on the pairwise effects of the AHLs 3- oxo-C6 HSL, 3-oxo-C8-HSL and 3-oxo-C12-HSL on the LuxR, LasR, RpaR and TraR QS systems to understand how to construct higher-level genetic circuitry for the use in microbial consortia. Wu et al. [14] also characterized the QS pairwise interactions of the AHLs 3-oxo-C6- HSL and 3-oxo-C12-HSL on the LuxI/R and LasI/R QS systems to better understand new directions in engineering gene networks. Our previous study investigated the robustness of signaling networks to interference by quantifying crosstalk between the LuxI/R QS system and the AHLs 3-oxo-C6-HSL and C4-HSL [12].Although these studies identified the pairwise interactions of QS in the presence of one or two auto-inducers, there is limited knowledge on how species composition, signal diversity and external perturbations would affect QS activation when more than two QS species are present, see Fig. 5.1a. In such heterogeneous environments, with several QS signals, activation of QS in each species is interdependent, making it a challenge to predict community-wide QS activity. Here we introduce a neural network model to examine the consequences of QS crosstalk within bacterial communities producing mixtures of QS signals, see Fig. 5.1b. Neural network models have been commonly used to understand the network-level consequences of interactions in many complex systems, including both biological and non-biological contexts. Neural networks have been implemented in advanced analytical techniques such as deep learning, pattern recognition and image compression [166–169]. In finance and economics, neural networks are trained with historical market data to discover trends and to make successful current market predictions [170]. In the pharmaceutical industry, based on the bio activity of a large set of chemicals, neural networks are used to identify new types of drugs that can be used to treat diseases [171]. In a neural network, components have a variable state, in simplest cases active or inactive. Interactions between network components influence state dynamics are represented as a weight, with the magnitude of the weight indicating the strength of the interaction and the sign of the weight indicating whether the interaction promotes or inhibits 117 activation. We have previously implemented a neural network to theoretically analyze the information capacity within a QS networks composed to multiple Staphylococcus aureus strains [146]. Here we extend these ideas, combining both experimental and theoretical results, to test whether neural network models can be used to predict and control the activation of QS in communities of bacteria producing multiple signals. Figure 5.1: A neural network simplifies the complexity of QS crosstalk in diverse communities. a. A graphical representation of QS activation in both simple and complex communities. In isolation, individual strains activate QS. Signal crosstalk potentially changes QS activation in pairs of strains. In diverse communities, a mixture of signals determines QS activity in each strain. b. In multi-signal contexts QS activation acts as a neural network, with each strain integrating the contribution of each signal to regulate QS activity. c. Five variants of Bacillus subtilis produce five variants of the ComX signal [11]. These strains, labeled A-E, test the ability of the neural network model to predict QS activation in a diverse community. The producers and testers are described in the text. 118 5.3: Results 5.2.1: Quantifying pairwise crosstalk between strains of Bacillus subtilis. To test our ability to predict QS activation within a community producing multiple variants of a signal, we used the a five strains of Bacillus subtilis previously reported by Ansaldi et al. [11]. Each of the five strains produces a unique variant of the ComX QS signal [11,172], see Fig. 5.1c. Each strain also had variation in the sequence of the ComP receptor protein [10,172]. In previous work [11], crosstalk between pairs of these five strains was reported, revealing a mix of both excitatory and inhibitory crosstalk between strains of variable strength. These measurements were not sufficient to construct a neural network representation of QS activity within the community, as the ratio of signals was not varied, so pairwise crosstalk between each signal was measured. We measured QS activity within mixtures of two ComX signals by combining ratios of supernatant from stationary phase culture of two “producer” strains and measuring QS activation using a “tester” strain containing a lacZ QS reporter (as used in the Ansaldi et al. study), see Fig. 5.2a,b. Producer strains produce the ComX signals and self- activate QS while the tester strains do not produce the ComX signal, but can activate QS if there is an external supply of ComX. QS activity was measured with a fluorogenic LacZ assay in a 96 well plate reader, see Fig. S5.1 and methods for tester responses with cognate signal. The fold change in LacZ expression is the ratio of LacZ expression in the presence and absence of ComX signal, see methods for further details. In individual strains, QS activity followed a sigmoidal activation behavior, Fig. 5.2c and Fig. S5.2. The yellow circles shown in Fig. 5.2c are a representation used throughout the manuscript to symbolize the addition of either a producer cells (P) or a supernatant from producer cells (S) to a tester (T). The tester is indicated in the middle of the circle while the producer or supernatant strains are indicated on the circumference. 119 Figure 5.2: Experimental setup oto measure QS activity. a. In producer cells, ComX QS signal is processed by ComQ and transported out of the cell. Binding of ComX to ComP receptor in cell membrane activates the response regulator ComA, leading to changes in the expression of QS-regulated genes. b. Tester cells, with a deletion of ComX and ComQ, do not produce QS signal, but increase expression of a lacZ, driven by QS regulated srfA promoter, in response to the addition of exogenous signal. The fluorogenic dye, Fluorescein di-β-D-galactopyranoside (FDG), enables fluorescent readout of QS activity c. The fold change in lacZ expression in tester strain C in response to addition of supernatant from strain C (in blue) and the supernatant from both strain C and 10 µl of strain D (in red). With the introduction of signal from strain D, the Fold change in LacZ decreased resulting in no activation of QS. The yellow circle indicates the tester strain (T) used and the addition of either producer cells (P) or the supernatant of producer cells (S). The activation curve of individual strains defines the threshold for QS activation, with QS activated if the fold change exceeds the threshold fold change. Next we examined QS activity upon exposure to mixtures of signal. As shown in Fig. 5.2c, in the presence of second signal D, the QS activity in strain C is inhibited. To quantify the interaction weight between strains, we constructed a mathematical model. The model is based on a set of differential equations and accounts for signal crosstalk by introducing a crosstalk weights for each pair of receptor and signal, similar to models used previously [12,146]. Specifically, the expression of the QS-regulated gene lacZ follows: 𝜕𝐿 𝜕𝑡 = 𝜌 𝐿 𝑛 𝑖 (𝑓 𝑖 ( 𝐶 𝑒𝑓𝑓 ,𝑖 𝑚 𝐶 𝑒𝑓𝑓 ,𝑖 𝑚 +𝜃 𝑖 𝑚 ) + 1) − 𝛾 𝐿 𝐿 , (5.1) 120 where the effective concentration of signal as the result of crosstalk is, 𝐶 𝑒𝑓𝑓 ,𝑖 = ∑ 𝑤 𝑖 ,𝑗 𝑐 𝑗 𝑗 . (5.2) In a population of 𝑛 𝑖 cells, lacZ expression occurs at a basal rate of 𝜌 𝐿 and upon QS activation the production rate is increased by a fold change 𝑓 𝑖 . A Hill’s function is used to represent the scaling of QS-activity with signal concentration, with a Hill coefficient of m and the concentration of half maximum of 𝜃 𝑖 . LacZ will degrade at a rate of 𝛾 𝐿 . An effective concentration, 𝐶 𝑒𝑓𝑓 ,𝑖 is used to account for the excitatory or inhibitory influence of each signal on QS activation in the i th strain. The interaction weight 𝑤 𝑖 ,𝑗 accounts for the magnitude and sign of the interaction between a ComX signal from B. subtilis strain j on QS activation in B. subtilis strain i. As in a neural network mode, the sum of these weighted interactions predicts the activity of each node (strain) for mixtures of inputs (signals). The self-weight (𝑤 𝑖 ,𝑖 ) is one for all strains, and if 𝐶 𝑒𝑓𝑓 < 0, we assume that 𝐶 𝑒𝑓𝑓 = 0. Similar to previous studies [12,25,111,114], the i th tester grow at a rate µ in a volume v, through a logistic growth equation. 𝜕 𝑛 𝑖 𝜕𝑡 = 𝜇 𝑖 𝑛 𝑖 (1 − 𝑛 𝑡𝑜𝑡𝑎𝑙 𝑠 𝑣 ) (5.3) Here 𝑛 𝑖 is the amount of the i th tester cell, 𝜇 𝑖 is the growth rate of the cells (given in Fig. S5.3), ntotal is the total number of cells in the system, and s is the maximum density of cells reached by the culture. In simulations, tester strains in the well of a plate grew from a cell density of 10 8 to 10 9 cell per mL. Over time, QS-regulated lacZ was produced following Equations 5.1, 5.2 and 5.3. LacZ concentrations in the culture were simulated for tester cells exposed to no signal as well as for a specified mixture of signal to calculate the fold-change in lacZ expression, see methods section and Table S5.1 for model parameters [114,173,174]. Simulations were used to fit experimental data for 𝑓 𝑖 and 𝜃 𝑖 for each strain, further details given in the methods section, see Fig. S5.4, Fig. S5.5 and Table 5.1. The best fit 𝑓 𝑖 and 𝜃 𝑖 values were used to generate the simulation curve, and the fold change in LacZ of this simulated curve at 𝜃 𝑖 was defined as the threshold value of LacZ fold change needed for QS activation, see Fig. 5.2c and Fig. S5.5. In the methods section we describe how to convert from supernatant volume to signal concentration. The five strains had five distinct threshold values needed for QS activation, see Fig. S5.5. 121 Strain 𝑓 𝑖 𝜃 𝑖 (nM) A 5.955 ± 0.002 1.393 ± 0.002 B 4.083 ± 0.001 1.633 ± 0.004 C 24.839 ± 0.004 1.364 ± 0.002 D 8.345 ± 0.005 1.394 ± 0.001 E 6.951 ± 0.002 1.815 ± 0.003 Table 5.1: The best fit values for 𝒇 𝒊 and 𝜽 𝒊 . These values were extracted by minimizing the root-mean squared error between the experimental and simulated data points of the LacZ fold changes vs. amount of supernatant. Combining the model with experimental measurements of QS activity enabled the calculation of the interaction weight between all 5 strains of Bacillus subtilis. The model predicted the pattern of QS activation that would be expected for different combinations of two signals, as depicted in Fig. 5.3a. The interaction weight of the second signal determined which combinations of signals led to QS activation. Fig. 5.3b shows how the pattern of activity, or the QS activation landscape, depended on the interaction weight, with each weight mapping to a specific activation pattern. The QS activation landscape represents the QS response of the testers with yellow and blue boxes indicating combinations of signals for which QS did or did not activate, respectively. Experimental measurements of the expression of a QS-regulated reporter gene at different ratios of signals were used to calculate the interaction weight for each pair of signal and receptor. As shown in Fig. 5.3c, for each strain, between 0 and 25 L of supernatant from each producer strain was mixed with 0 to 25 L of supernatant from a second strain. 25 L was chosen as the maximum volume of supernatant as individual strains required only 10 L to activate QS, giving sufficient dynamic range to measure even strong inhibition. Fig. 5.3c shows the activation landscapes for strain C, see Fig. S5.6-S5.10 for data for the other strains. The weights calculated for each strain revealed a rich network of signaling interactions within the 5 strain community, Fig. 5.3d and Fig. S5.11. For example QS activation in strain C was activated by signal from strains A or C, strongly inhibited by signal from strains B or D, and only weakly responded to signal from strain E. Strain E on the other hand only responded to its own signal and was not influenced by the signal from any other strain tested. 122 Figure 5.3: Pairwise measurements. a. QS activation in the presence of two signals is measured by combing supernatant containing the cognate signal (Sc) with supernatant containing the interacting strain (Sint) with a tester strain. b. Simulation results show the activation landscape depended on the crosstalk weight of the interacting signal. Results are shown for strain C, where f = 24.839 and 𝜃 = 1.364 nM. Yellow and blue boxes indicate combinations of signals for which QS did or did not activate, respectively. QS activation is defined as the fold change in lacZ expression associated with a cognate signal concentration of 𝜃 . c. Experimentally measured activation landscapes for strain C. d. Experimental measurements were used to extract the 123 crosstalk weight for each pair of signal and receptor. The error bars represent the range of weight values which gives the same activation landscape, see Fig. S5.6-S5.10. 5.2.2: Pairwise weights predicts quorums sensing activation patterns in the 5-strain community. The extraction of the pairwise interaction weights in the previous section enabled us to apply the neural network model to predict QS activation patterns in groups of 3 or more strains. The response of each strain to three input signals was simulated, as shown in Fig. 5.4a. The model predicted QS activity for a tester strain exposed to three input signals. The QS activity of strain C changed as the concentrations of a signal increased. As shown in Fig. 5.4b and 5.4c, model predictions were verified in experimental measurements. In experiments, supernatants from three different strains were mixed at a specific ratio with a tester strain, and the expression of QS genes was measured using the fluorogenic LacZ indicator. Experimental measurements reproduce the predicted pattern of QS activation as ratios of supernatant were varied. Analysis of QS activation in the presence of three signals revealed the concept of a community signaling state. In a microbial community producing multiple signal variants, crosstalk between these signaling systems potentially leads to the activation of QS in subsets of the community. The exact combination and ratio of signals, as well as the structure of the crosstalk network, will determine which strains activate QS. The neural network model predicted the community-level signaling state and its sensitivity to changes in signal concentrations, as shown in Fig. 5.4d and Fig. S5.12. The signaling state of the 5-strain community is shown as the amount of signal from strain E was varied. The community-level QS state can be represented as a binary string, with a 1 or 0 in each position of the string indicating whether QS will activate or not activate, respectively, for each strain. For example, at 0 L of supernatant from strain E, the predicted string was (1,0,1,0,0), which indicated that only strains A and C would activate QS under these conditions. In communities in which the amount of strain E supernatant was increased, the community-level state changed twice, first to (1,0,0,0,1) and then to (1,1,0,0,1). These predictions from the model 124 were borne out in experimental measurements, as the two transitions of the community-level signaling state were observed as the volume of supernatant from strain E increased. Figure 5.4: Experimental validations of the model predictions when more than two signals are present. a. Simulations using the neural network model predicted the activity of a tester strain in the presence of three ComX signals. The response of tester Tc was simulated in the presence of SC, SE and SD. The cube of colored boxes shows how the tester strain C will respond to different combinations of signal, blue boxes represent QS off and yellow boxes represent QS on. The dashed lines show slices of the cube that were tested experimentally. Experimental validation of theoretical predictions of QS activation in the presence of 3 signals. One signal concentration was varied as the other two ComX supernatants were kept at a constant value. For b. SC = 25 µl and SE = 10 µl and SD was changed from 0-25 µl. For c. SE = 10 µl and SD = 5 µl and SC was changed from 0-25 µl. d. In the presence of all five ComX signal supernatants, we observed a range of activation patterns for the community. The community-level signaling state, 125 represented by a binary string of numbers indicating the QS activity of each strain, changes depending on the ratio of signals. SA, SB, SC and SD were held constant at 10 µl, 4 µl, 15 µl and 1 µl respectively while SE was varied from 0 to 25 µl, in 5 µl increments. Experimental measurements of QS activity in the tester strains matched theoretical predictions. In the yellow circles, the red text indicates the signal concentration that was varied. 5.2.3: The network state depends on the inoculation ratio of strains. In the previous sections, we determined that a neural network model could predict the consequences of QS crosstalk within a 5-strain community of Bacillus subtilis. These predictions were tested in experiments in which specified ratios of supernatant from stationary phase cultures were combined with tester strains to measure QS activity. Next we wanted to verify that the model could also predict QS activity for mixtures of strains growing from low density culture, see Fig. 5.5a. These experiments will reveal how inoculation ratios influenced QS activation in a multi-strain community. As cells grow, release signals, and potentially activate QS, the signal concentrations will change over time until reaching a steady state [12,25,114]. To simulate signal production, cell growth, and the expression of QS-regulated lacZ, Equations 5.1-5.4 were used. Equation 5.4 describes the change in signal concentration (𝑐 𝑖 ) over time for the i th strain as the result of signal production and degradation. The signal production of the i th strain will be influenced by all the other strains that are present, as captured by ceff,i defined in Equation 5.2. 𝜕 𝑐 𝑖 𝜕𝑡 = 𝜌 𝑐 𝑖 𝑛 𝑖 (𝑓 𝑖 ( 𝑐 𝑒𝑓𝑓 ,𝑖 𝑚 𝑐 𝑒𝑓 𝑓 ,𝑖 𝑚 +𝜃 𝑖 𝑚 ) + 1) − 𝛾 𝑐 𝑖 𝑐 𝑖 . (5.4) In a population of 𝑛 𝑖 cells of the i th strain, signal production occurs at a basal rate of 𝜌 𝑐 𝑖 and at QS activation, this rate is increased by a fold change 𝑓 𝑖 . A Hill’s function is used to represent the scaling of signal production with effective signal concentration 𝑐 𝑒𝑓𝑓 ,𝑖 , with a Hill coefficient of m and the concentration of half maximum of 𝜃 𝑖 . 𝑐 𝑒𝑓𝑓 ,𝑖 incorporates the QS crosstalk as mentioned previously. The signals will degrade at a rate of 𝛾 𝑐 𝑖 . For this case, in 126 Equation 5.3, 𝑛 𝑡𝑜𝑡𝑎𝑙 = ∑ 𝑛 𝑗 𝑘 𝑗 =1 , where k is the total amount of producer strains mixed together. Parameter values are listed in Table S5.1. In simulations, P A and PB were mixed at different ratios and the concentration profile of the ComX signals for both were plotted with respect to time, see Fig. S5.13 and methods for further details. In Fig. 5.5b, we represent the change in the simulated fold change in LacZ normalized with the threshold needed for QS activation with respect to the inoculation ratio of PA and PB. Here the fold change in LacZ is determined by applying the supernatant of the mixture PA and PB after 10 hours growth, to the testers TA and TB separately. We observed that above a 7:1 inoculation ratio of PA to PB , strain B does not produce a sufficient concentration for QS activation, whereas strain A activated QS at all strain ratios. In experiments, we mixed P A and PB at ratios of 1:1, 2:1, 10:1, 100:1, 1000:1, and grew them for 10 hours, see Fig. 5.5c. Supernatants of these mixtures activated QS in TA at all ratios, whereas TB was only activated at 1:1 and 2:1 ratios. These results demonstrate that the inoculation ratio of strains dictated the final QS state of the community. Figure 5.5: Initial loading ratio of the producers dictates the final QS activation state. a. Producers A and B were mixed together at a specified ratio and grown to stationary phase. Supernatant of the coculture was applied to tester strains to determine the QS activity of each strain. b. Simulations predict the fold change in LacZ normalized with the threshold level of LacZ expression defined as QS activation. Strain A activated QS at all ratios of cells, whereas QS activation of strain B occurred only at specific inoculation ratios of the two strains. c. Experimental validation of the simulation results. Supernatant from the coculture of producers 127 grown 2 hours past exponential phase was used to determine QS activity. The arrow map shows the nature of crosstalk between A and B, with A inhibiting B and B activating A. 5.2.4: Using ComX perturbations to switch community-level QS activation states. We have shown that the QS activation of species depends on species composition, ratio of the number of species, signal composition and signal concentration. Next we tested if the steady-state QS activity of a multi-strain community could be influenced by a perturbation of added signal, and whether the timing of signal addition impacted the response to the perturbation. As depicted in Fig. 5.6a, in both simulation and experiments we mixed together producers A and B at a ratio of 1:1, and measure the QS activity of tester strains A and B. In the control experiment, the coculture was not perturbed, and we expected results as in Fig. 5.5, strains A and B both activate QS. Duplicate cocultures were perturbed by the addition of 20 µl of supernatant from strain C at various points of time after inoculation. As shown in Fig. 5.6B, both simulations and experimental confirm that the addition of signal C had the potential to alter the activation of QS in the coculture, see Fig. S5.14. Signal C inhibits QS activation in strain B and promotes QS activation in strain A. However, whether the community-level signaling state changed from (1,0) to (1,1) depended on the time of the perturbation. Addition of supernatant from strain C prior to 4 hours resulted in strain B not activating QS, whereas a perturbation at 4 hours or later did not alter QS activation of either strain. As above, the QS activity for these cases was tested with supernatants extracted from cultures that were grown 2 hours past exponential phase. 128 Figure 5.6: Perturbing the system to change the QS activation state. a. When PA: PB are mixed at a ratio of 1:1, both A and B will activate QS. A perturbation of signal from strain C was introduced to the system after tp hours. b. The map in a shows the nature of crosstalk between strains. Here, we represent the fold change in LacZ normalized to the threshold fold change needed for QS activation for both TA and TB vs. the time at which the perturbation was introduced. The lines represent the simulation results and the data points are from experiments. The error bars represent the standard error from 3 sets of replicates. We observe that the change in the activation state of TB depends on the time at which the perturbation was introduced. 5.3: Discussion In this study, we analyzed QS activation within a community of five different B. subtilis strains that produce distinct ComX signals. The pairwise interactions that we measured within the 5-strain community were consistent with what was reported previously in Ansaldi et al. [11], Fig. S5.15. Although crosstalk between pairs of QS systems have been observed previously [11– 13,101,114,172], here for the first time we predict the consequences of signal crosstalk on QS activity in communities utilizing multiple signal variants. In natural environments, QS crosstalk is prominent and often regulates genes associated with virulence and biofilm formation [3,32,40,175–177]. Many natural communities produce diverse sets of signaling 129 molecules, resulting in entangled and interdependent gene expression within the community. Neural networks, inspired by the decision making within densely interconnected neurons, enable predictions of activity within complex systems with many interacting components. Here we demonstrate the consequences of signal crosstalk on community-level gene expression states can be accurately predicted using a neural network model with pairwise interactions. Many previous studies on QS crosstalk have implied that QS activation state is solely dependent on the type, or in our case sign,of the crosstalk [14,40]. In the presence of all the five ComX signals, we observed that the QS activation state of one strain will be different depending on the exact mixture of ComX concentrations. Therefore simply knowing the community composition will not give an accurate of the potential for QS activation. Communities with identical membership can be driven towards multiple community-level signaling states; ratios of strains and the structure of the interaction network determine which strains or species activate QS. The history of the community also dictates activation patterns, as the temporal dynamics of signal accumulation, as influenced by strain inoculation ratios and perturbations, dictate the QS activation states. The neural network model may enable the design of perturbations to redirect QS activation within bacterial communities. As QS activation is known to influence the real- world problems such as infections and biofouling. This predictive model of community-level activity should be relevant for industrial and medical applications. Although the utility of a neural network model of QS was demonstrated for only one class of QS signals, cyclic autoinducing peptides used by several species of Gram-positive bacteria, the model should be easily extended to communities using the acyl-homoserine lactone signals common in Gram-negative QS. Crosstalk as the result of two cells producing chemically related signaling molecules is not unique to cyclic autoinducing peptides [14,178]. Crosstalk within AHL networks or other signaling networks may have a completely different distribution of weights, which would impact the number and sensitivity of community-level signaling states. In addition, although the 5-strain B. subtilis community was densely connected, i.e. each strain was connected to every other strain in the network, the modelling framework could be applied to networks in which only subsets of strains participate in quorum sensing crosstalk or any other microbial interaction [179]. This would model situations in which multiple signal types are present in the same system, such as strains which utilize AI-2 and AHLs or V. cholera which 130 information from four chemical distinct signals [180]. In networks with such orthogonal signaling exchange could be accounted for as zero interactions weights except for in nodes (strains) which integrate the two signals through regulatory feedback. Mapping these situations onto the neural network model should give deeper insights into QS regulation in diverse communities. 5.4: Materials and methods. 5.4.1: Bacterial strains and growth media and conditions. All the strains used in this study was obtained from the study Ansaldi et al. [11]. The producer cells constitutively produced the ComX pro-protein which gets modified and processed by ComQ, in turn, releasing the ComX pheromone to the extracellular environment. Released ComX pheromones bind to the ComP membrane protein to phosphorylate ComA and activate QS. The testers cannot produce the ComX signal due to the disruption of the comQ and comX genes but can activate QS when the signal is exogenously added to produce LacZ under the QS- regulated srfA promoter. All the testers and producers were isogenic apart from producing distinct pheromones and receptors. To construct the strains labeled from B-E, the comQXP genes of the Bacillus subtilis 168 were replaced with the foreign comQXP genes from four distinct Bacillus species. The Bacillus subtilis strains were grown in the Bacillus competence media as described previously [181]. This media contained (w/v) 1.00% sodium Lactate, 0.25% yeast extract, 0.20% ammonium sulfate, 1.40% dipotassium phosphate, 0.60%; mono potassium phosphate, 0.10% sodium citrate-2H20, 0.02% magnesium sulfate*7H20 and 0.40% of glucose. All cultures were grown at 37oC and 200 RPM in competence media. Tester strains grown overnight, diluted 1/1000 in fresh media, and grown for an additional 10 hours before measurements. To extract supernatant, producers were grown for 10 hours after 1/1000 dilution of overnight culture. Cultures were centrifuged for 4 min at 4000 RPM to collect supernatant. The supernatant was passed through a 0.2 μm VWR syringe filter. Filtered supernatants were stored at 20 0 C. A single batch of supernatant from each strain was used for all experiments. 131 When mixing the producers strains together producers were grown in separate cultures for 10 hours, diluted 1/1000 in fresh competence media and grown for an additional 3 hours. Growth for 3 hours at low density was to ensure QS was not active prior to mixing strains together. After 3 hours of growth, cultures reached a final OD of 0.2-0.3. Cultures of producer strains were mixed together at different ratios. For example, in the 1:1 case, we mixed 10 μl of PA with 10 μl of PB together into 3 mL of competence media. Mixed cultures were grown for 10 hours. Since all producers and testers were based on the B. subtilis 168 and they were near isogenic, no growth interactions occurred between strains when mixed together. To measure activity, supernatant was extracted from producer cocultures as described above. In perturbation experiments, supernatant from strain C was added to the culture 0 and 7 hours after mixing producer strains together. 5.4.2: β-galactosidase assay. For the indicator, we used Fluorescein di-β-D-galactopyranoside (FDG) [127,172], and the fluorescence was detected by using a Tecan plate reader with a 96 well plate. In the 96 well plate, we mixed 25 µl of the testers, 25 µl of the FDG at 0.04 mg ml -1 (FDG was mixed in the competence media), 75 µl of the competence media and the rest with the different combinations of filtered supernatants and spent media such that the total volume in a well is 200 µl. For consistency, in all experiments, we made sure that all the wells had a total of 75 µl of spent media from the supernatants. The testers were loaded after 10 hours growth. The fluorescence was detected with an excitation wavelength of 480 nm and an emission wavelength of 514 nm. For the experiments with the mixed producers, the supernatants were extracted after 10 hours growth and 25 µl of this supernatant was mixed with the testers. The absorbance was detected at 600 nm. We calculated the fluorescence per cell as described previously [114]. Similar to previous studies [11,172], the LacZ expression level was calculated by obtaining the gradient of the fluorescence per cell vs. time plot in the linear region between 0-5 hrs, see Fig. S5.1. The fold change in LacZ was calculated by taking the ratio of a given LacZ expression level with the LacZ expression level when no signal is present. 132 5.4.3: Growth measurements. To obtain the growth rates, the cells were grown in the Bacillus competence media, as mentioned above. We then diluted the cells 1000 fold in 3 mL of competence media and grew them at 37 o C and 200 RPM. Dilutions of the culture were plated on competence media agar plates at every 1 hour interval during growth for 7 hrs, and these plates were incubated for 16 hrs at 37 0 C. For each time point 2 sets of replicates were considered, and after 10 hours growth, the number of colonies were counted to get the colony forming units (C.F.U). The growth rates were obtained by considering a fit to the linear region in the growth curve. 5.4.4: Mathematical modelling and simulations. To perform the simulations described in the manuscript we used MATLAB 2016b. We used the finite difference method to simulate the change in LacZ expression with respect to time. We calculated the LacZ expression level by simulating the increase in LacZ using Equation 5.1, with respect to time and by calculating the gradient of the LacZ expression level within the linear region, which was within the 0-5 hours. The fold change in LacZ for the simulations was calculated the same way as described in the experimental section. For each strain i, as shown in Table 5.1, 𝑓 𝑖 and 𝜃 𝑖 were distinct. We obtained 𝑓 𝑖 and 𝜃 𝑖 by minimizing the root-mean squared error and determining the best fit curve between the experimental data points and simulation curve, see Fig. S5.4, Fig. S5.5 and Table 5.1. The thresholds to determine QS activation for each strain was different based on the unique values for 𝑓 𝑖 and 𝜃 𝑖 , see Fig. S5.5. To plot the QS activation patterns of each tester for the pair-wise case, using unique 𝑓 𝑖 and 𝜃 𝑖 , values we simulated the fold change in LacZ for each combination of signals and tested if the fold change in LacZ was above or below the threshold of the tester. If it was above, we assigned the yellow color and if it was below, we assigned a blue color. The activation pattern of the simulations, as seen in Fig. 5.3a, was compared with the experimental patterns to extract the weights. We observe the same activation pattern for a range of weights, see Fig. S5.6-S5.10. 133 To convert the loading volume of the supernatant to the signal concentration we considered the following method. Previous studies have reported the final concentration of ComX in media after 2 hours growth from late exponential phase is approximately 30 nM [182,183]. Since a volume of x µl of the supernatant at 30 nM gets diluted in a total of 200 µl in the 96 well plates, see methods for further details, we can calculate the signal concentration = 30 nM (x/200). For the simulations with the P A and PB grown together, we considered an initial density of 6x10 4 cells ml -1 . This was based on the dilutions done in the experiments. We mixed P A and PB at different ratios and calculated the ComX concentration of each strain at 10 hours. We then applied 25 µl of the mixed supernatant with both testers A and B to obtain the LacZ fold change separately. In the simulations with the perturbation, we followed a similar procedure. Here, the perturbations were introduced at times, tp=0 to 7 hrs, every 1 hour, and the supernatant of the mixture was tested with the testers separately. Finally, the fold change in the testers were measured. 134 5.5: Supplementary information. Figure S5.1: Fluorescence per cell vs time for the testers when mixed with their cognate supernatants (indicated by the yellow circles). For each case, the multiple lines represent the amount of supernatant (S) used. Errorbars represent the standard error from two sets of replicates. 135 Figure S5.2: The fold changes in the LacZ expression of each tester A, B, C, D and E respectively, when mixed with its cognate producer supernatant. Error bars represent the standard error from 10 sets of replicates. 136 Figure S5.3: Doubling times for the producers and testers, see methods for further details. Growth was measured by counting the colony forming units with respect to time in agar plates with the competence media. A linear fit was considered for the data points between 2hrs-6hrs, using the least squared fitting method and the gradient was extracted to obtain the growth rates. The estimated growth rates are, µ 𝑷 𝑨 = 𝟎 . 𝟗𝟐𝟎 , µ 𝑷 𝑩 = 𝟏 . 𝟎𝟎𝟗 , µ 𝑷 𝑪 = 𝟎 . 𝟗𝟐𝟒 , µ 𝑷 𝑫 = 𝟎 . 𝟗𝟏𝟖 , µ 𝑷 𝑬 = 𝟎 . 𝟗𝟑𝟕 , µ 𝑻 𝑨 = 𝟎 . 𝟗𝟏𝟔 , µ 𝑻 𝑩 = 𝟎 . 𝟗𝟗 𝟎 , µ 𝑻 𝑪 = 𝟎 . 𝟗𝟒𝟑 , µ 𝑻 𝑫 = 𝟎 . 𝟗𝟐𝟔 , µ 𝑻 𝑬 = 𝟎 . 𝟗𝟏𝟔 . Parameter Value Obtained from 𝜇 𝑖 0.916-1.009 hrs -1 Calculated in this study, see Fig. S5.3 𝑠 10 9 cells ml -1 Silva et al. (2017) [114] 𝜌 𝐿 3.180 x 10 -7 nM cell −1 hrs −1 Haustenne et al. (2015) [174] 𝛾 𝐿 0.600 hrs -1 Santilla et al. (2008) [173] 𝜌 𝑐 𝑖 3.180 x 10 -7 nM cell −1 hrs −1 [174] Haustenne et al. (2015) [174] 𝛾 𝑐 𝑖 0.420 hrs -1 [173] Santilla et al. (2008) [173] 𝑚 3 [173] Santilla et al. (2008) [173] Table S5.1: Parameter values used in the study for the modeling. 137 Figure S5.4: The root mean squared error (indicated with the color intensity) was minimized for changing values of θ and f. The θ and f values that gave the minimum root mean squared error was averaged to obtain the values used in the simulations. 138 Figure S5.5: The best fit curves for f and θ. These curves, obtained from simulations, were fit by minimizing the root mean square error for changing values of both f and θ. Once the θ was extracted, the fold change LacZ for that θ from the fitted curve, and obtain the threshold needed for QS activation. Figure S5.6: The quorum sensing activation landscape for the tester A when mixed with the supernatant of producer A and the supernatant of producer A, B, C, D and E respectively 139 (from left to right). A blue square represents quorum sensing activation off and a yellow square represents quorum sensing activation on. The weight values listed on top represent the range of values we observe the same activation landscape. Figure S5.7: The quorum sensing activation landscape for the tester B when mixed with the supernatant of producer B and the supernatant of producer A, B, C, D and E respectively (from left to right). A blue square represents quorum sensing activation off and a yellow square represents quorum sensing activation on. The weight values listed on top represent the range of values we observe the same activation landscape. Figure S5.8: The quorum sensing activation landscape for the tester C when mixed with the supernatant of producer C and the supernatant of producer A, B, C, D and E respectively (from left to right). A blue square represents quorum sensing activation off and a yellow square represents quorum sensing activation on. The weight values listed on top represent the range of values we observe the same activation landscape. 140 Figure S5.9: The quorum sensing activation landscape for the tester D when mixed with the supernatant of producer D and the supernatant of producer A, B, C, D and E respectively (from left to right). A blue square represents quorum sensing activation off and a yellow square represents quorum sensing activation on. The weight values listed on top represent the range of values we observe the same activation landscape. Figure S5.10: The quorum sensing activation landscape for the tester E when mixed with the supernatant of producer E and the supernatant of producer A, B, C, D and E respectively (from left to right). A blue square represents quorum sensing activation off and a yellow square represents quorum sensing activation on. The weight values listed on top represent the range of values we observe the same activation landscape. 141 Figure S5.11: The overall pairwise quorum sensing interaction map. In the left, the green arrow represents activation, the red flat head represents inhibition. On the right, we have elaborated the strength of each pairwise interaction. Each pentagram represents the strength of the pairwise interaction between a tester and all the supernatants. For instance, the smallest pentagram represents the response of TA. The colors represent the weight values. Green represents positive weights while red represents negative weigths. 142 Figure S5.12 : LacZ fold change comparisons of the 5 supernatant case from simulations and experiments. Here the values indicated are the fold change in LacZ when increasing SE from simulations (left) and experiments (right). Here QS on is represented by the color yellow and QS off is represented by the color blue. The threshold fold changes for QS activation of the strains A, B, C, D, E are 3.43, 2.75, 10.56, 4.17 and 4.68 respectively. Figure S5.13: Simulation results of the ComX concentration normalized by Ɵ over time for PA (green, left y-axis) and PB (brown, right y-axis) when they are mixed (PA: PB) at ratios of (1:1), (2:1), (4:1), (7:1). We observe that at a ratio above 7:1, PB will not activate QS. 143 Figure S5.14: Simulation results of the ComX concentration normalized by Ɵ over time for PA (green, left y-axis) and PB (brown, right y-axis) when they are mixed (PA: PB) at a ratio (1:1). As observed in Fig. S5.12, when the ratio is set to 1:1, both PA and PB will activate QS. When SC is added externally at a volume of 20 µl, the activation of PB depends on the time at which SC was introduced (tp = time of perturbation) to the mixture. Since C will activate A, P A activates QS for all tp. 144 Figure S5.15: The comparisons of the pairwise crosstalk of our results with Ansaldi et al. [11]. Figure S15: Figure S15: The comparisons of the pairwise crosstalk of our results with Ansaldi et al. [11]. In this previous study, the LacZ activity of each tester was measured and crosstalk was tested by mixing the cognate ComX signal and a non-cognate ComX signal at a ratio of 1:1 obtained from purifying supernatants of E. coli producers. They tested if the LacZ expression was above or below the LacZ level when there was only the cognate ComX present. Here we compare these results with the LacZ measurements of the testers when the ComX signals were introduced at a ratio of 1:1. The normalized LacZ activity was the ratio between the LacZ activity of the pairwise combinations and the LacZ activity when only the cognate ComX signal is present. The ComX signal of our experiments were obtained from the B. subtilis producers as described from Tortosa et al. [172] We obtained the results of the previous study using the software plot digitizer. The blue bars represent the results from our study and the red bars represent the results from Ansaldi et al. 145 Chapter 6: Concluding remarks In this final chapter, I will describe the impact of my work and discuss possible directions for future research. 6:1: Impact of my work 6:1:1 Robust method of quantifying quorum sensing interactions In Chapters 2, 3 and 5 I have described how we can use either plate based experimental setup or well mixed plate reader experimental setup to measure and quantify quorum sensing crosstalk and quorum quenching between strains. As discussed extensively throughout the thesis, there is a high demand to quantify quorum sensing crosstalk and quenching and my studies have shown robust methods to do such quantification by comparing experimental results with mathematical modeling results. In Chapter 2, we quantified how a quorum quenching strain can impact the propagation of activation of a quorum sensing sender strain. In Chapters 3 and 5, we used a novel approach to quantify quorum sensing crosstalk by introducing interaction weights in the mathematical model. We fit our simulation curves to our experimental curves to obtain the weights, and using these weights, we made accurate theoretical predictions that agreed with our experimental observations. I believe that calculating interaction weights to predict the level of gene expression is a novel and powerful tool when it comes to building bio synthetic circuits, where every input has to be accurately characterized to determine the correct level of genetic output. In the thesis, I also discuss how the pairwise weights can be used to predict patterns of quorum sensing activation within communities containing several quorum sensing species. This is a useful approach for scientists who study the role of quorum sensing in biofilms as these environments consists of a multitude of quorum sensing species. 146 6:1:2 The plate based approach enabled us to understand how quorum sensing is influenced by quorum sensing crosstalk or quenching within diverse communities of cells In Chapter 2 and 3, we used a classic LuxI/R sender-receiver type plate based assay to observe the propagation of a quorum sensing activation gradient. For the case of crosstalk, from theoretical simulations we concluded that the shift in the activation time can be explained by the competition of autoinducers binding to the receptors. We extracted weights based on the comparison of experimental data and theoretical simulations. Although there was a shift in the activation time, the quorum sensing activation of a spatially distributed LuxI/R system was largely robust to interacting signals produced by LasI and RhlI. This robustness was not present when we added the AiiA quencher, where the activation zone of the LuxI/R system was strongly dependent on the amount of interference that was used. Therefore, one can use such an assay to distinguish between quorum sensing crosstalk and interference in natural isolates. I believe that with the plate based approach one can examine quorum sensing interactions in more complex and realistic context, such as in the presence of more than two strains, when cells are heterogeneously distributed in space, or even when transport dynamics are spatially dependent. Since the experiments in these contexts would be challenging, our assay and model provide a straightforward path towards predicting signaling dynamics in complex conditions. 6:1:3 Neural network approach to make accurate predictions and controlling quorum sensing activation in well-mixed bacterial populations In Chapter 5, we describe how a neural network approach can be used to simplify our task of understanding quorum sensing crosstalk in well-mixed systems where a large number of autoinducer types are present. I showed how to calculate pairwise interaction weights and using a neural network I was able to predict the quorum sensing activation states when there is more than two types of autoinducer molecules present in a mixture. This approach enabled us to understand how quorum sensing activation not only depended on the type of crosstalk, but also species number, autoinducer concentration and time sensitive external perturbations. We learned that the 147 activation state of certain species can be controlled and guided towards a different state and in a microbial environment with a large number of quorum sensing species, the activation states are interdependent and non-trivial to predict. A neural network is a novel model to understand complex interactions within microbiological systems and can be applied to predict and understand signaling exchange within diverse microbial communities. 6:1:4 Criticality observed in disrupting large-scaled communication through quenching In Chapter 2, I showed how the spatial spread of quorum sensing activation can be tuned by changing the amount of quorum quenchers we introduce. Using this idea, in Chapter 4 I described how the long-range communication of a quorum sensing strain can be disrupted by a critical amount of degrader strain which causes quorum sensing interference. As quorum sensing is known to regulate virulence, biofilm formation and bacterial infections, disrupting the communication can lead to a strategy of preventing the robust functionality of spatially spread out microbial communities. The transition from long-range to small-range communication showed a critical behavior and by calculating the critical exponents I was able confirm that this transition was a percolation transition. From a physics prospective, it was interesting to observe such a collective phenomenon in the context of microbial communication as it is an emergent behavior of quorum sensing interactions. Perhaps there is a deeper level of coordination that contributes to governing emergent behaviors of biological systems. In this study, I also emphasized the value of using physical models to study biological systems, and by doing so we were able gain insights into the complex behavior of the emergent properties of communicating bacteria. This can potentially reveal new strategies for robust control of system function. 6:2: Future directions In the first two chapters, work was done to understand on how to quantify quorum sensing interactions and the impact on the quorum sensing activation in the presence of crosstalk or quorum quenching. Our studies can be expanded to understand the quorum sensing interactions in a spatially distributed heterogeneous environment of bacteria such as biofilms. 148 Today there is a large interest in understanding the role of quorum sensing interactions in biofilms and in the future, one can use our experimental and theoretical procedure as a stepping stone towards this goal. One of the limitations of our setup was that the activation gradient was sensitive to non-quorum sensing interactions as well. As microbes have many interactions with each other such as metabolic, chemotactic, and growth influences, it can be challenging to isolate the quorum sensing effects. Therefore more work needs to be done on understanding how quorum sensing is coupled with non-quorum sensing interactions. We have shown this to some extent using our theoretical model to incorporate growth influences, but more work needs to be done. In Chapter 2 we discuss how quorum sensing crosstalk can be represented by a single parameter known as the weight. Although the definition of the weight was given to be the impact a certain autoinducer will have on the regulation of genetic expression, work needs to be done to identify what physical processes dictate a weight. For instance, signal-receptor dimerization, transcription rates, interactions between the receptor, DNA and RNA polymerases will be accounted for in the weight. Although this is no easy task, by doing so will answer some of the fundamental questions on how signal crosstalk effects the regulatory pathways of species. The percolation transition observed in Chapter 4 gives us a basic understanding of criticality that is observed in emergent behaviors of bacterial populations. With our experimental system we understood that a percolation transition can occur when the large scale communication of quorum sensing bacteria is disrupted. Work needs to be done to understand if a percolation transition can be observed just by increasing the number of cells that load on to a plate. For instance if the quorum sensing bacteria were seeded onto an area starting from a low density to a high density, will the transition from small scale to large scale communication follow a percolation transition? Such a study can shed light on to whether randomness in the distribution of cells follow complex biophysical rules. In Chapter 5 a neural network model was used to disentangle the complexities of quorum sensing interactions within a well-mixed heterogeneous population of quorum sensing bacteria. We identified how species composition, signal diversity and time sensitive signal perturbations are important when it comes to quorum sensing activation within well mixed culture. In the future, work needs to be done to understand the advantages of using a neural network approach 149 to study quorum sensing activation in spatially distributed populations of microbes. Although we have addressed this to some extent in Chapter 3, quantifying crosstalk will be more complex since we need to know the signal-receptor binding energies. In the neural network approach, fitting parameters is far direct and straightforward. I also propose that neural network can be used to study why certain bacteria contain more than one quorum sensing system. Since these quorum sensing systems are often coupled, the interaction strengths can be quantified using weights and once the weights are known one can gain valuable insights on the advantages of having multiple quorum sensing systems. 150 References [1] G. Rampioni, L. Leoni, and P. Williams, Bioorg. Chem. 55, 60 (2014). [2] K. Zengler and B. O. Palsson, Nat. Rev. Microbiol. 10, 366 (2012). [3] P. Williams, K. Winzer, W. C. Chan, and M. Cámara, Philos. Trans. R. Soc. London B Biol. Sci. 362, 1119 (2007). [4] M. J. Federle and B. L. Bassler, J. Clin. Invest. 112, 1291 (2003). [5] W. C. Fuqua, S. C. Winans, and E. P. Greenberg, J. Bacteriol. 176, 269 (1994). [6] K. H. Nealson, T. Platt, and J. W. Hastings, J. Bacteriol. 104, 313 (1970). [7] M. B. Miller and B. L. Bassler, Annu. Rev. Microbiol. 55, 165 (2001). [8] A. Eberhard, A. L. Burlingame, C. Eberhard, G. L. Kenyon, K. H. Nealson, and N. J. Oppenheimer, Biochemistry 20, 2444 (1981). [9] B. L. Bassler, M. Wright, R. E. Showalter, and M. R. Silverman, Mol. Microbiol. 9, 773 (1993). [10] L. S. P. Tran, T. Nagai, and Y. Itoh, Mol. Microbiol. 37, 1159 (2000). [11] M. Ansaldi, D. Marolt, T. Stebe, I. Mandic-Mulec, and D. Dubnau, Mol. Microbiol. 44, 1561 (2002). [12] K. P. T. Silva, P. Chellamuthu, and J. Q. Boedicker, PLoS Comput. Biol. 13, 1 (2017). [13] J. C. March and W. E. Bentley, Curr. Opin. Biotechnol. 15, 495 (2004). [14] F. Wu, D. J. Menn, and X. Wang, Chem. Biol. 21, 1629 (2014). [15] K. H. McClean, M. K. Winson, L. Fish, A. Taylor, S. R. Chhabra, M. Camara, M. Daykin, J. H. Lamb, S. Swift, B. W. Bycroft, G. S. Stewart, and P. Williams, Microbiology 143, 3703 (1997). [16] S. R. Scott and J. Hasty, ACS Synth. Biol. 5, 969 (2016). [17] K. H. Mcclean, M. K. Winson, L. Fish, A. Taylor, S. R. Chhabra, M. Camara, M. Daykin, H. John, S. Swift, B. W. Bycroft, G. S. A. B. Stewart, and P. Williams, Microbiology 143, 151 3703 (1997). [18] L. Molina, F. Constantinescu, L. Michel, C. Reimmann, B. Duffy, and G. Défago, FEMS Microbiol. Ecol. 45, 71 (2003). [19] B. L. Bassler, E. P. Greenberg, and A. M. Stevens, J. Bacteriol. 179, 4043 (1997). [20] Y. Dong, J. Xu, X. Li, and L. Zhang, Proc. Natl. Acad. Sci. 97, 3526 (2000). [21] C. Grandclement, T. Melanie., S. Mor, Y. Dessaux, and D. D. Faure, FEMS Microbiol. Rev. 40, 1 (2015). [22] C. Tmin and C. Koh, Sensors 12, 4846 (2012). [23] Y. Dong, L. Wang, and L.-H. Zhang, Philos. Trans. R. Soc. Lond. B. Biol. Sci. 362, 1201 (2007). [24] P. D. Pérez, J. T. Weiss, and S. J. Hagen, BMC Syst. Biol. 5, 153 (2011). [25] G. E. Dilanji, J. Langebrake, P. De Leenheer, and J. Stephen, J. Am. Chem. Soc. 134, 1 (2012). [26] S. Fetzner, J. Biotechnol. 201, 2 (2014). [27] R. C. Kelly, M. E. Bolitho, D. A. Higgins, W. Lu, W. Ng, P. D. Jeffrey, J. D. Rabinowitz, M. F. Semmelhack, F. M. Hughson, and B. L. Bassler, Nat. Chem. Biol. 5, 1 (2009). [28] M. Fujiya, M. W. Musch, Y. Nakagawa, S. Hu, J. Alverdy, Y. Kohgo, O. Schneewind, B. Jabri, and E. B. Chang, 2, 299 (2007). [29] A. Culotti and A. I. Packman, PLoS One 9, 1 (2014). [30] E. M. M. Quigley, Gastron. Hepatol. 9, 560 (2013). [31] J. A. Thompson, R. A. Oliveira, and C. Ubeda, CellReports 10, 1861 (2015). [32] E. J. G. Pollitt, S. A. West, S. A. Crusz, M. N. Burton-chellew, and P. Diggle, Infect. Immun. 82, 1045 (2014). [33] Y. Wang, Y. Dai, Y. Zhang, Y. Hu, B. Yang, and S. Chen, Sci. China. Ser. C, Life Sci. 50, 385 (2007). 152 [34] J. Q. Boedicker, L. Li, T. R. Kline, and R. F. Ismagilov, Lab Chip 8, 1265 (2008). [35] A. Kumar and N. Chordia, Appl. Microbiol. Open Access 03, 2 (2017). [36] J. W. Costerton, P. S. Stewart, and E. P. Greenberg, Science (80-. ). 284, 1318 (1999). [37] S. Basu, Y. Gerchman, C. H. Collins, F. H. Arnold, and R. Weiss, Nature 434, 1130 (2005). [38] L. Tsimring, J. Hasty, T. Danino, O. Mondrago, O. Mondragón-Palomino, L. Tsimring, and J. Hasty, Nature 463, 326 (2010). [39] R. M. Davis, R. Y. Muller, K. A. Haynes, R. M. Davis, and I. A. Fulton, Front. Bioeng. Biotechnol. 3, 1 (2015). [40] G. Rampioni, L. Leoni, and P. Williams, Bioorg. Chem. 55, 60 (2014). [41] R. Kenna and B. Berche, J. Phys. A 50, 1 (2017). [42] Y. Zhu and X. Chen, Cond-Mat.Stat-Mech 818, 1 (2017). [43] L. L.-S. YU Mei-Ling, XU Ming-Mei, LIU Zheng-You, Chinese Phys. C 33(7), 552 (2009). [44] J. W. Larkin, X. Zhai, K. Kikuchi, S. E. Redford, A. Prindle, J. Liu, S. Greenfield, A. M. Walczak, J. Garcia-Ojalvo, A. Mugler, and G. M. Süel, Cell Syst. 7, 137 (2018). [45] K. D. Micheva, B. Busse, N. C. Weiler, N. O’Rourke, and S. J. Smith, Neuron 68, 639 (2010). [46] S. Herculano-Houzel, Front. Hum. Neurosci. 3, 1 (2009). [47] W. Maass, Neural Networks 10, 1659 (1997). [48] T. R. de Kievit and B. H. Iglewski, Infect. Immun. 68, 4839 (2000). [49] Y.-H. Li and X. Tian, Sensors 12, 2519 (2012). [50] Q. Yang and T. Defoirdt, 17, 960 (2015). [51] J. Boedicker and K. Nealson, IEEE Trans. Mol. Biol. Multi-Scale Commun. 1, 310 (2016). 153 [52] K. H. Nealson and J. W. Hastings, Appl. Environ. Microbiol. 72, 2295 (2006). [53] D. M. Cornforth, R. Popat, L. McNally, J. Gurney, T. C. Scott-Phillips, A. Ivens, S. P. Diggle, and S. P. Brown, Proc. Natl. Acad. Sci. 111, 4280 (2014). [54] C. D. Sifri, Clin. Infect. Dis. 47, 1070 (2008). [55] S. Heilmann, S. Krishna, and B. Kerr, 6, 1 (2015). [56] J. S. Kavanaugh, M. Thoendel, and A. R. Horswill, Mol. Microbiol. 65, 780 (2007). [57] D. Haas and K. Heurlier, 296, 93 (2006). [58] S. P. Darch, S.E., West, S.A., Winzer, K. , Diggle, Proc. Natl. Acad. Sci. 109, 8259 (2012). [59] C. Grandclément, M. Tannières, S. Moréra, Y. Dessaux, and D. Faure, FEMS Microbiol. Rev. 40, 86 (2016). [60] Y. Cao, M. D. Ryser, S. Payne, B. Li, C. V. Rao, and L. You, Cell 165, 620 (2016). [61] C. A. Masiello, Y. Chen, X. Gao, S. Liu, H.-Y. Cheng, M. R. Bennett, J. A. Rudgers, D. S. Wagner, K. Zygourakis, and J. J. Silberg, Environ. Sci. Technol. 47, 11496 (2013). [62] T. Sohka, R. A. Heins, R. M. Phelan, J. M. Greisler, C. A. Townsend, and M. Ostermeier, Proc. Natl. Acad. Sci. 106, 10135 (2009). [63] T. L. Wentao Kong, Andrew E. Blanchard, Chen Liao, Nucleic Acids Res. 1 (2016). [64] C. Liu, X. Fu, L. Liu, X. Ren, C. K. L. Chau, S. Li, L. Xiang, H. Zeng, G. Chen, L.-H. Tang, P. Lenz, X. Cui, W. Huang, T. Hwa, and J.-D. Huang, Science (80-. ). 334, 238 (2011). [65] B. Borek, J. Hasty, and L. Tsimring, PLoS One 11, e0153679 (2016). [66] T. Danino, O. Mondragón-Palomino, L. Tsimring, and J. Hasty, Nature 463, 326 (2010). [67] J. B. Langebrake, G. E. Dilanji, S. J. Hagen, and P. De Leenheer, J. Theor. Biol. 363, 53 (2014). [68] A. Trovato, F. Seno, M. Zanardo, S. Alberghini, A. Tondello, and A. Squartini, FEMS 154 Microbiol. Lett. 352, 198 (2014). [69] W. Kong, V. Celik, C. Liao, Q. Hua, and T. Lu, Bioresour. Bioprocess. 1, 24 (2014). [70] M. D. Servinsky, J. L. Terrell, C.-Y. Tsao, H.-C. Wu, D. N. Quan, A. Zargar, P. C. Allen, C. M. Byrd, C. J. Sund, and W. E. Bentley, ISME J. 10, 1 (2015). [71] M. Weitz, A. Mückl, K. Kapsner, R. Berg, A. Meyer, and F. C. Simmel, J. Am. Chem. Soc. 136, 72 (2013). [72] W. S. Choi, D. Ha, S. Park, and T. Kim, Biomaterials 32, 2500 (2011). [73] N. Marchand and C. H. Collins, Biotechnol. Bioeng. 110, 3003 (2013). [74] Y. Chen, J. K. Kim, A. J. Hirning, K. Josić, and M. R. Bennett, Science 349, 986 (2015). [75] F. Zhang, A. Kwan, A. Xu, and G. M. Süel, PLoS One 10, 15 (2015). [76] A. Fekete, C. Kuttler, M. Rothballer, B. A. Hense, D. Fischer, M. Lucio, M. Johannes, and P. Schmitt-kopplin, FEMS Microbiol. Ecol. 72, 22 (2010). [77] P. S. Stewart, J. Bacteriol. 185, 1485 (2003). [78] T. Ramalho, A. Meyer, A. Mückl, K. Kapsner, U. Gerland, and F. C. Simmel, PLoS One 11, e0145829 (2016). [79] M. Englmann, A. Fekete, C. Kuttler, M. Frommberger, X. Li, I. Gebefügi, J. Fekete, and P. Schmitt-Kopplin, J. Chromatogr. A 1160, 184 (2007). [80] S. Payne, B. Li, Y. Cao, D. Schaeffer, M. D. Ryser, and L. You, Mol. Syst. Biol. 9, 697 (2013). [81] A. Tamsir, J. J. Tabor, and C. A. Voigt, Nature 469, 212 (2011). [82] K. Brenner, L. You, and F. H. Arnold, Trends Biotechnol. 26, 483 (2008). [83] V. S. Coker, N. D. Telling, G. van der Laan, R. A. D. Pattrick, C. I. Pearce, E. Arenholz, F. Tuna, R. E. P. Winpenny, and J. R. Lloyd, ACS Nano 3, 1922 (2009). [84] A. Y. Chen, Z. Deng, A. N. Billings, U. O. S. Seker, M. Y. Lu, R. J. Citorik, B. Zakeri, and T. K. Lu, Nat. Mater. 13, 515 (2014). 155 [85] J. Shong, M. R. Jimenez Diaz, and C. H. Collins, Curr. Opin. Biotechnol. 23, 798 (2012). [86] N. I. Johns, T. Blazejewski, A. L. Gomes, H. H. Wang, M. Vasquez, and S. Hallam, Curr. Opin. Microbiol. 31, 146 (2016). [87] C. Ratzke and J. Gore, Nat. Microbiol. 1, 1 (2016). [88] C. J. Kastrup, J. Q. Boedicker, A. P. Pomerantsev, M. Moayeri, Y. Bian, R. R. Pompano, T. R. Kline, P. Sylvestre, F. Shen, S. H. Leppla, W.-J. Tang, and R. F. Ismagilov, Nat. Chem. Biol. 4, 742 (2008). [89] P. D. Straight, J. M. Willey, and R. Kolter, J. Bacteriol. 188, 4918 (2006). [90] J. L. Connell, J. Kim, J. B. Shear, A. J. Bard, and M. Whiteley, Proc. Natl. Acad. Sci. 111, 18255 (2014). [91] H. J. Kim, J. Q. Boedicker, J. W. Choi, and R. F. Ismagilov, Proc. Natl. Acad. Sci. 105, 18188 (2008). [92] A. Prindle, P. Samayoa, I. Razinkov, T. Danino, L. S. Tsimring, and J. Hasty, Nature 481, 39 (2011). [93] R. Milo, Bioessays 35, 1050 (2013). [94] I. M. Bomze, 447453, (1995). [95] M. B. Miller and B. L. Bassler, Annu. Rev. Microbiol. 55, 165 (2001). [96] S. T. Rutherford and B. L. Bassler, Cold Spring Harb Perspect Med 2, 1 (2012). [97] R. Teresa and B. H. Iglewski, Infect. Immun. 68, 4839 (2000). [98] A. Rajput, K. Kaur, and M. Kumar, Nucleic Acids Res. 44, D634 (2016). [99] E. Wynendaele, A. Bronselaer, J. Nielandt, M. D’Hondt, S. Stalmans, N. Bracke, F. Verbeke, C. Van De Wiele, G. De Tré, and B. De Spiegeleer, Nucleic Acids Res. 41, 655 (2013). [100] D. T. Hughes and V. Sperandio, Nat. Rev. Microbiol. 6, 111 (2008). [101] M. T. G. Holden, S. R. Chhabra, R. De, P. Stead, N. J. Bainton, P. J. Hill, M. Mane, N. 156 Kumar, M. Labatte, D. England, S. Rice, M. Givskov, P. C. Salmond, G. S. A. B. Stewart, W. Barrie, S. Kjelleberg, and P. Williams, Mol. Microbiol. 33, 1254 (1999). [102] P. K. Grant, N. Dalchau, J. R. Brown, F. Federici, T. J. Rudge, B. Yordanov, O. Patange, A. Phillips, J. Haseloff, and P. K. Grant, Mol. Syst. Biol. 12: 849, 1 (2016). [103] Y. Gao, J. Song, B. Hu, L. Zhang, Q. Liu, and F. Liu, Curr. Microbiol. 58, (2009). [104] B. Lasarre and M. J. Federle, Microbiol. Mol. Biol. Rev. 77, 73 (2013). [105] S.-W. Teng, J. N. Schaffer, K. C. Tu, P. Mehta, W. Lu, N. P. Ong, B. L. Bassler, and N. S. Wingreen, Mol. Syst. Biol. 7, 491 (2011). [106] B. Canton, A. Labno, and D. Endy, Nat. Biotechnol. 26, 787 (2008). [107] A. Prindle, P. Samayoa, I. Razinkov, T. Danino, L. S. Tsimring, and J. Hasty, Nature 481, 39 (2012). [108] S. Basu, Y. Gerchman, and C. H. Collins, Nature 434, 1130 (2005). [109] T. Ramalho, A. Meyer, A. Mückl, and K. Kapsner, PLoS One 11, 1 (2016). [110] H. Song, J. Ozaki, C. H. Collins, M. Barnet, F. H. Arnold, and S. R. Quake, Mol. Syst. Biol. 1 (2008). [111] A. Fekete, C. Kuttler, M. Rothballer, B. A. Hense, D. Fischer, K. Buddrus-Schiemann, M. Lucio, J. Müller, P. Schmitt-Kopplin, and A. Hartmann, FEMS Microbiol. Ecol. 72, (2010). [112] A. Pai, Y. Tanouchi, and L. You, Proc. Natl. Acad. Sci. 109, 19810 (2012). [113] D. Van Valen, M. Haataja, and R. Phillips, Biophys. J. 96, 1275 (2009). [114] K. P. Silva, P. Chellamuthu, and J. Q. Boedicker, Biophys. J. 112, 1037 (2017). [115] S. Gnanendra, S. Anusuya, and J. Natarajan, J. Mol. Model. 18, 4709 (2012). [116] C. Garde, T. Bjarnsholt, M. Givskov, T. H. Jakobsen, M. Hentzer, A. Claussen, K. Sneppen, J. Ferkinghoff-borg, and T. Sams, J. Mol. Biol. 396, 849 (2010). [117] S. Swift, A. V Karlyshev, L. Fish, E. L. Durant, M. K. Winson, S. R. A. M. Chhabra, P. 157 Williams, S. Macintyre, and G. S. A. B. Stewart, J. Bacteriol. 179, 5271 (1997). [118] K. Jangid, P. S. Parameswaran, Y. S. Shouche, and M. C. Collection, Sensors 12, 3814 (2012). [119] X. Chan, K. How, W. Yin, and K. Chan, Front. Cell. Infect. Microbiol. 6, 1 (2016). [120] T. Hosni, C. Moretti, G. Devescovi, Z. R. Suarez-Moreno, M. B. Fatmi, C. Guarnaccia, S. Pongor, A. Onofri, R. Buonaurio, and V. Venturi, ISME J. 5, 1857 (2011). [121] L. Chalupowicz, S. Manulis-Sasson, M. Itkin, A. Sacher, G. Sessa, and I. Barash, Mol. Plant. Microbe. Interact. 21, 1094 (2008). [122] G. Pizarro, D. Griffeath, and D. Noguera, J. Environ. Eng. 127, (2001). [123] S. Payne, B. Li, Y. Cao, D. Schaeffer, M. D. Ryser, and L. You, Mol. Syst. Biol. 9, 1 (2013). [124] S. A. Chugani, M. Whiteley, K. M. Lee, D. D’Argenio, C. Manoil, and E. P. Greenberg, Proc Natl Acad Sci U S A 98, 2752 (2001). [125] J. Friedman and J. Gore, Curr. Opin. Syst. Biol. 1, 114 (2017). [126] M. E. Hibbing, C. Fuqua, M. R. Parsek, and S. B. Peterson, Natl. Rev. Microbiol. 8, 15 (2010). [127] N. N. Kreamer, R. Phillips, D. K. Newman, and J. Q. Boedicker, Sci. Rep. 5, 18238 (2015). [128] X. Guo and J. Q. Boedicker, PLoS Comput. Biol. 12, 1 (2016). [129] B. Kirkup, U. States, M. Woods, and H. Oceanographic, Science (80-. ). 337, (2012). [130] M. Marı, N. Fuster-valls, M. Herna, and J. Rodrı, Food Control 19, 308 (2008). [131] T. I. Yusufaly and J. Q. Boedicker, Phys. Rev. E 94, 62410 (2016). [132] M. R. Frederick, C. Kuttler, B. A. Hense, J. Müller, and H. J. Eberl, Can. Appl. Math. Q. 18, 267 (2011). [133] Y. B. Tatek and G. W. Slater, Physica A 362, 2 (2006). 158 [134] J. S. Dickschat, Nat. Prod. Rep. 27, 343 (2010). [135] A. M. L. Barnard, S. D. Bowden, T. Burr, S. J. Coulthurst, R. E. Monson, and G. P. C. Salmond, Philos. Trans. R. Soc. London B Biol. Sci. 362, 1165 (2007). [136] K. Anguige, J. R. King, and J. P. Ward, J. Math. Biol. 51, 557 (2005). [137] F. Constantinescu, L. Michel, and C. Reimmann, FEMS Microbiol. Ecol. 45, 71 (2003). [138] R. M. Ziff, Phys. Rev. Lett. 69, 2670 (1992). [139] S. N. Dorogovtsev, A. V Goltsev, and J. F. F. Mendes, Rev. Mod. Phys. 80, 1 (2008). [140] J. H. Stathis, J. Appl. Phys. 86, 5757 (1999). [141] K. L. and J. F. W. Richard Chandler, Joel Koplik, J. Fluid Mech. 119, 249 (1982). [142] R. Parshani, S. Carmi, and S. Havlin, Phys. Rev. Lett. 258701, 1 (2010). [143] A.-L. Barabási, G. Musella, N. Samay, R. Sinatra, S. Morrison, A. Husseini, and P. Hoevel, CC BY-NC-SA 2.0 (2014). [144] E. Ben-jacob, M. Lu, D. Schultz, and J. N. Onuchic, Front. Cell. Infect. Microbiol. 4, 1 (2014). [145] P. Freemont, K. M. Polizzi, and L. Goers, J. R. Soc. Interface 11, 1 (2014). [146] T. I. Yusufaly and J. Q. Boedicker, Phys. Biol. 14, 1 (2017). [147] H. Wioland, F. G. Woodhouse, J. Dunkel, and R. E. Goldstein, Nat. Phys. 12, 341 (2016). [148] J. Hidalgo, J. Grilli, S. Suweis, M. A. Muñoz, J. R. Banavar, and A. Maritan, Proc. Natl. Acad. Sci. 111, 10095 (2014). [149] X. Chen, X. Dong, A. Be, H. L. Swinney, and H. P. Zhang, Phys. Rev. Lett. 148101, 1 (2012). [150] T. J. Rudge, P. J. Steiner, A. Kan, and J. Haselo, ACS Synth. Biol. 2, 705 (2013). [151] J. D. Shrout, D. L. Chopp, C. L. Just, M. Hentzer, M. Givskov, and M. R. Parsek, Mol. Microbiol. 62, 1264 (2006). 159 [152] T. Mora and W. Bialek, J Stat Phys 144, 1 (2011). [153] W. Bialek, A. Cavagna, I. Giardina, T. Mora, and E. Silvestri, Proc Natl Acad Sci U S A 109, 4786 (2012). [154] D. Krotov, J. O. Dubuis, T. Gregor, and W. Bialek, Proc. Natl. Acad. Sci. 111, 3683 (2014). [155] C. Furusawa, Phys. Rev. Lett. 208103, 1 (2012). [156] M. Nykter, N. D. Price, M. Aldana, S. A. Ramsey, S. A. Kauffman, L. E. Hood, O. Yli- harja, and I. Shmulevich, Proc. Natl. Acad. Sci. 105, 1897 (2007). [157] J. M. Beggs and D. Plenz, J. Neurosci. 23, 11167 (2003). [158] W. Kong, D. R. Meldgin, J. J. Collins, and T. Lu, Nat. Chem. Biol. 14, 821 (2018). [159] H. G. Garcia, H. J. Lee, J. Q. Boedicker, and R. Phillips, Biophys. J. 101, 535 (2011). [160] C. Fuqua and E. P. Greenberg, Nat. Rev. Mol. Cell Biol. 3, (2002). [161] K. Riedel, M. Hentzer, O. Geisenberger, B. Huber, A. Steidle, H. Wu, N. Høiby, M. Givskov, S. Molin, and L. Eberl, Microbiology 147, 3249 (2001). [162] E. Geisinger, J. Chen, and R. P. Novick, J. Bacteriol. 194, 2854 (2012). [163] J. B. Kaper and V. Sperandio, Infect. Immun. 73, 3197 (2005). [164] K. M. Deangelis, S. E. Lindow, and M. K. Firestone, FEMS Microbiol. Ecol. 66, 197 (2008). [165] L. L. Williamson, B. R. Borlee, P. D. Schloss, C. Guan, H. K. Allen, and J. Handelsman, Appl. Environ. Microbiol. 71, 6335 (2005). [166] A. Rajavelu, M. T. Musavi, and M. V Shirvaikar, Neural Networks 2, 387 (1989). [167] C. Cortes and V. Vapnik, Mach. Learn. 20, 273 (1995). [168] Y. Lecun, Y. Bengio, and G. Hinton, Nature 521, 436 (2015). [169] J. G. Daugman, IEEE Trans. Acoust. 36, 1169 (1988). 160 [170] A. Chen, M. T. Leung, and H. Daouk, Comput. Oper. Res. 30, 901 (2003). [171] W. B. Langdon, S. J. Barrett, and B. F. Buxton, Lect. Notes Comput. Sci. 2278, 60 (2002). [172] P. Tortosa, L. Logsdon, B. Kraigher, Y. Itoh, I. Mandic-Mulec, and D. Dubnau, J. Bacteriol. 183, 451 (2001). [173] M. Santilla, Biophys. J. 94, 2065 (2008). [174] L. Haustenne, G. Bastin, P. Hols, and L. Fontaine, Front. Microbiol. 6, 1 (2015). [175] A. M. L. Barnard, S. D. Bowden, T. Burr, S. J. Coulthurst, R. E. Monson, and G. P. C. Salmond, Philos. Trans. R. Soc. London B Biol. Sci. 362, 1165 (2007). [176] E. P. Davies, D. G., Matthew R. P., James P. P., Barbara H. I., Costerton, J.W., Greenberg, Science (80-. ). 280, 295 (1998). [177] Y. Dong, L. Wang, and L. Zhang, Philos. Trans. R. Soc. London B Biol. Sci. 362, 1201 (2007). [178] N. Kylilis, Z. A. Tuza, G. B. Stan, and K. M. Polizzi, Nat. Commun. 9, (2018). [179] J. Friedman, L. M. Higgins, and J. Gore, Nat. Ecol. Evol. 1, 1 (2017). [180] M. J. Eickhoff and B. L. Bassler, Cell 174, 1328 (2018). [181] C. Anagnostopoulos and J. Spizizen, J. Bacteriol. 741 (1960). [182] M. Okada, I. Sato, J. Cho, and Y. Sakagami, Tetrahedron 62, 8907 (2006). [183] R. Magnuson, J. Solomon, and A. Grossman, Cell 77, 207 (1994).

Abstract (if available)

Abstract Quorum sensing is a process in which microbes exchange molecular signals to coordinate cellular activity within populations of cells. Although typically viewed as a mechanism for a single bacterial species to measure their own cell density, recent work has shown that quorum sensing enables cells to gain information about multiple facets of the local environment, including physical and chemical conditions as well as local community composition. Often in these environments, signaling activity is influenced by the activity of neighboring cells. Some cells produce signals capable of binding to receptors in other cell types, resulting in signal crosstalk between two cell types. Other neighboring cells produce enzymes that degrade signal in the local environment, a process known as quorum quenching. ❧ Although previous studies characterized several examples of quorum sensing crosstalk and quorum quenching, less work has been done to understand the role of such cell to cell interactions in modulating quorum sensing activation within communities of quorum sensing bacteria. Therefore, my major goals were to broaden our understanding of the importance of quorum sensing crosstalk and quenching on regulating quorum sensing activation within diverse communities of bacteria, to understand the biophysics of cell signaling with microbial communities. ❧ In this thesis, theoretical and experimental tools from both physics and quantitative biology, including thermodynamics, neural networks, synthetic gene circuits, and reaction-diffusion models, were used to find the rules that govern quorum sensing activation within diverse communities of cells. I observed that the spatial propagation of quorum sensing activation was robust to crosstalk but was sensitive to quorum quenching. The sensitivity to quorum quenching was utilized to observe that a critical amount of quorum quenching was required to shut down the large-scaled quorum sensing activation of a spatially dispersed bacterial colony. I confirmed that the breakdown of such large-scaled quorum sensing coordination followed a percolation transition. In a well-mixed group of quorum sensing bacteria I was able to quantify pairwise crosstalk to predict and control the quorum sensing outcomes within a five-strain community of Bacillus subtilis. With my work, we have been able to broaden our understanding of the role of quorum sensing and quorum quenching in communities of bacteria. Such knowledge can be utilized to devise better methods to prevent quorum sensing regulated bacterial infections and to improve the implementation of cell signaling networks in synthetic microbial communities.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Mathematical modeling in bacterial communication and optogenetic systems

PDF

Microbial interaction networks: from single cells to collective behavior

PDF

Electronic, electrochemical, and spintronic characterization of bacterial electron transport

PDF

Electrochemical studies of outward and inward extracellular electron transfer by microorganisms from diverse environments

PDF

Genetic diversity and bacterial death in the context of adaptive evolution

PDF

Electrochemical studies of subsurface microorganisms

PDF

Plant-microbial interactions in mangrove sediments under different nutrient conditions

PDF

Bacterial nanowires of Shewanella oneidensis MR-1: electron transport mechanism, composition, and role of multiheme cytochromes

PDF

From cables to biofilms: electronic and electrochemical characterization of electroactive microbial systems

PDF

Exploring bacteria-mineral interactions

PDF

Annual pattern and response of the bacterial and microbial eukaryotic communities in an aquatic ecosystem restructured by disturbance

PDF

Kinetic Monte Carlo simulations for electron transport in redox proteins: from single cytochromes to redox networks

PDF

Using sequencing techniques to explore the microbial communities associated with ferromanganese nodules and sediment from the South Pacific gyre

PDF

From single molecules to bacterial nanowires: functional and dynamic imaging of the extracellular electron transfer network in Shewanella oneidensis MR-1

PDF

Bacterial survival under low nutrient conditions in the laboratory and the deep biosphere

PDF

Amelogenin-ameloblastin protein interaction and function in dental enamel formation

PDF

Electrochemical investigations and imaging tools for understanding extracellular electron transfer in phylogenetically diverse bacteria

PDF

The evolution of gene regulatory networks

PDF

Extracellular electron transport: Investigating the diversity and mechanisms behind an understudied microbial process with global implications

PDF

Electromagnetic scattering models for satellite remote sensing of soil moisture using reflectometry from microwave signals of opportunity

Asset Metadata

Creator Silva, Kalinga Pavan Thushara (author)

Core Title Quorum sensing interaction networks in bacterial communities

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Physics

Publication Date 03/14/2019

Defense Date 01/24/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag bacteria,biophysics,Microbiology,OAI-PMH Harvest,Physics,quantifying interactions,quorum sensing,synthetic biology,Systems Biology

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Boedicker, James Quill (committee chair), Di Felice, Rosa (committee member), El-Naggar, Moh (committee member), Kresin, Vitaly (committee member), Nealson, Kenneth (committee member)

Creator Email kalingas@usc.edu,kptsilva@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-134670

Unique identifier UC11676671

Identifier etd-SilvaKalin-7164.pdf (filename),usctheses-c89-134670 (legacy record id)

Legacy Identifier etd-SilvaKalin-7164.pdf

Dmrecord 134670

Document Type Dissertation

Format application/pdf (imt)

Rights Silva, Kalinga Pavan Thushara

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA