ABSTRACT 
 
 
 
Title of Document:  CONTEXTUALIZATION OF THE E. COLI LSR 
SYSTEM: RELATIVE ORTHOLOGY, RELATIVE  
QS ACTIVITY, AND EMERGENT BEHAVIOR 
 
 
 
David Nathan Quan, Doctor of Philosophy, 2014 
 
 
Directed By:    Professor William E. Bentley 
Fischell Department of Bioengineering 
 
 
 
Within bacterial consortia there exist innumerable combinatorial circumstances, some of 
which may tip the scale toward pathogenicity, some of which may favor asymptomatic 
phenotypes. Indeed, the lines and intersections between commensal, pathogenic, and 
opportunistic bacteria are not always clean.  As a foothold to mediate pathogenicity 
arising from consortia, many have puzzled at communication between bacteria.  Primary 
among such considerations is quorum sensing (QS). Analogous to autocrine signaling in 
multicellular organisms, QS is a self-signaling process involving small molecules. 
Generally, QS activation is believed to have pleiotropic effects, and has been associated 
with numerous pathogenic phenotypes. The research herein focuses on autoinducer-2 
(AI-2) based QS signaling transduced through the Lsr system. Produced by over 80 
species of bacteria, AI-2 is believed to be an interspecies signaling molecule.  Outside of 
the marine bacteria genera Vibrio and Marinomonas, the only known AI-2 based QS 
transduction pathway is the Lsr system. We sought to deepen the characterization of the 
Lsr system in contexts outside of the batch cultures in which it was originally defined.  
 
 
 
 
First, we interrogated E. coli K-12 W3110 Lsr system orthologs relative to the same 
strain’s lac system.  Both systems are induced by the molecule which they import and 
catabolize.  We searched for homologs by focusing on the gene order along a genome, as 
gene arrangement can bear signaling consequences for autoregulatory circuits.  We found 
that the Lsr system signal was phylogenetically dispersed if not particularly deep, 
especially outside of Enterobacteriales and Pasteurellaceaes, indicating that the system 
has generally been conferred horizontally.  This contrasts with the lac system, whose 
signal is strong but limited to a select group of highly related enterobacteria.  We then 
modeled the Lsr system with ODEs, revealing bimodality in silico, bolstering preliminary 
experimental evidence.  This bifurcated expression was seen to depend upon nongenetic 
heterogeneity, which we modeled as a variation of a single compound parameter, basal, 
representing the basal rate of AI-2 flux into the cell through a low flux pathway. 
Moreover, in our finite difference-agent based models, bimodal expression could not 
arise from spatial stochasticity alone.  This lies in contrast with the canonical LuxIR QS 
system, which employs an intercellular positive feedback loop to activate the entire 
population.  We examined the consequences of this contrast, by modeling both systems 
under conditions of colony growth using finite difference-agent based methods.  We 
additionally investigated the confluence of Lsr signaling with chemotactic sensitivity to 
AI-2, which has been demonstrated in E. coli.  Finally, the consequences of bimodality in 
interspecies interactions were assessed by posing two populations containing different 
Lsr systems against each other.  While few natural consortia consist of only two 
interacting bacteria, these studies indicate that AI-2 based Lsr signaling may mediate a 
 
 
 
 
multitude of transitional intraspecies and interspecies bacterial dynamics, the specifics of 
which will vary with the context and the homologs involved.  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
CONTEXTUALIZATION OF THE E. COLI LSR SYSTEM: RELATIVE 
ORTHOLOGY, RELATIVE QS ACTIVITY, AND EMERGENT BEHAVIOR 
 
 
 
 
 
 
 
 
 
By 
 
David Nathan Quan 
 
 
 
 
 
 
Dissertation submitted to the Faculty of the Graduate School of the 
University of Maryland, College Park, in partial fulfillment 
of the requirements for the degree of 
Doctor of Philosophy 
2014 
 
 
 
 
 
 
 
 
 
Advisory Committee: 
Professor William E. Bentley, Chair 
Associate Professor Michael P. Cummings 
Associate Professor Nam Sun Wang 
Associate Professor Adam Hsieh 
Assistant Professor Ganesh Sriram 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
© Copyright by  
David Nathan Quan 
2015 
 
 
 
 
 
 
 
 
 
 
ii 
 
Dedication 
 
As a token of love and appreciation, to my parents, to my brother, and my broader family 
who have sacrificed so much, both small and great—to their unrelenting and 
unconditional support.  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
iii 
 
Acknowledgements 
      I wish to acknowledge all the members of my dissertation committee for their time, 
advice, and commitment to myself, their own graduate students, and the university. I 
especially recognize my committee chair and advisor, Dr. William Bentley, under whose 
guidance I have been allowed to follow my nose and in whose lab I have enjoyed a 
stimulating research environment. I also wish to acknowledge my fellow lab members 
both present and past, whose work serves as fundamental bedrock for my own, and 
whose ideas I have come to recount as readily as my own.  In particular, I wish to 
acknowledge Dr. Chen-Yu Tsao whose development of pCT5 and pCT6 has been utterly 
important for our lab and has resulted in compelling basic science.  I also acknowledge 
the work of those in the quorum sensing field as whole—it is only by standing upon their 
shoulders that I can view any forest at all.  I also wish to gratefully acknowledge my 
mentors at the Center for Cell Dynamics, the members of the Alliance for Cellular 
Signaling, and Dr. Gary Jarvis for giving me a wide berth when trying to do science 
during my formative years. In particular, I would like to recognize a former mentor at the 
Center for Cell Dynamics, Dr. Jonathon Alberts, whose Sim2D platform was modified 
and heavily relied upon here.  Finally, I acknowledge the reader: I hope you find what 
you need or perhaps even something interesting or worthwhile.   
 
 
 
 
 
 
 
 
 
 
iv 
 
Table of Contents 
Dedication ......................................................................................................................... ii 
Acknowledgements............................................................................................................iii 
Table of Contents................................................................................................................iv 
List of Tables .....................................................................................................................vi 
List of Figures .................................................................................................................. vii 
Chapter 1: Introduction ..................................................................................................... 1 
1.1  Background.......................................................................................................... 1 
1.1.1  QS in Vibrio Harveyi….………………………………….……………... 5 
1.1.2  LuxIR QS…….….………………………………………………………. 7 
1.1.3  AI-2 QS………….……………………………………………………..  10 
1.1.4  Lsr Regulon.……………………………………………………………. 15   
1.1.5  Known Lsr Associated Phenotypes.……………………………………. 16 
1.1.6  Broader AI-2 Associated Phenotypes…………...….………………….. 17 
1.1.7  QS in an Ecological Context….……………………………………..…. 20 
1.2  Research Motivation……………………………………………………………22 
1.3  Global Objective, Global Hypothesis, and Specific Aim……………………... 26 
1.4  Dissertation Outline…………………………………………………………… 27 
Chapter 2: Comparison of Homolog Identification for the Sugar Importing lac System and 
the QS Lsr System from E. coli K-12 W3110…………………………………….......... 28 
        2.1  Abstract……………………………………………………………………….. 28 
        2.2  Introduction…………………………………………………………………… 30 
        2.3  Methods and Algorithm...................................................................................... 34 
     2.3.1  Input……………………………………………………………………. 37 
                2.3.2  Scoring Heuristic………………………………………………………. 37 
                2.3.3  Weak and Stringent Criteria…………………………………………… 38 
    2.3.4  Ancillary LMNAST Search Tools……………………………………... 38 
        2.4  Results……………….........................................................................................40 
     2.4.1  E. coli K-12 W3110 lac Operon Query………………………………… 40 
     2.4.2  E. coli K-12 W3110 Lsr System Query………………………………… 48 
                2.4.3  Analysis of Lsr System Search Results……………...……………….... 57 
  2.4.3.1  Putative Lsr System in Rhizobiales…………………………….. 59 
        2.5  Concluding Remarks...………………………………………………………… 60 
        2.6  Supplemental Material………………………………………………………… 64 
Chapter 3: Quorum desynchronization leads to bimodality and patterned behaviors in 
microbial consortia …...................................................................................................... 74 
        3.1  Abstract.............................................................................................................. 74 
        3.2  Introduction........................................................................................................ 75 
        3.3  Methods.............................................................................................................. 79 
    3.3.1 Modeled Cell Behaviors……..……………………………………..........79 
  3.5.1.1 Chemotactic Swimming……………………………………….....79 
  3.5.1.2 Colony Growth…………………………………………………...79 
  3.5.1.3 LuxIR/AHL QS………………………………………………..…80 
3.5.1.4 Lsr/AI-QS……………………………………………………...…80  
3.3.2 Simulation Variants: Gene Deletions and Mixed Populations…………..80 
 
 
 
v 
 
        3.4  Results................................................................................................................ 82 
   3.4.1 Lsr Autoinduction in Pure Cultures……………………………….……..82   
3.4.2 Pattern formation in QS systems: LuxIR vs Lsr……….………………...84 
3.4.3 Cell Motility – Lsr QS based pattern emergence………………...………85 
3.4.4 Mixed Population Simulations…………………………………………...87 
3.5  Discussion………………………………………………………………………91 
3.6 Concluding Remarks……………………………………………………………96    
3.7 Supplemental Information………………………………………………………98 
3.7.1 Text and Discussion ……………………………….……………………98 
     3.7.2 Methods……………………...…………………………………………..99 
3.7.2.1 ODE Model…...……….…………………………………………99 
      3.7.2.1.1 Generalities and Scope.….…….…………………………...99 
   3.7.2.1.2 mRNA Expression……………………………..………….100 
  3.7.2.1.3 Protein Synthesis……………………………….…………101 
  3.7.2.1.4 mRNA and Protein Degredation...……………………..…101 
   3.7.2.1.5 Cell Growth………………………………………….....…101 
3.7.2.1.6 AI-2 Transport......…………………………………..…….102 
  3.7.2.1.7 AI-2 Degredation and Synthesis…………..………………103 
  3.7.2.1.8 Equations…………………………………..……………...103 
  3.7.2.1.9 Parameter Values.....………………………………..……..105 
3.7.2.1.10 Initial Values………………………………………..……107      
3.7.2.1.11 Numerical Solution…………………………………..…..108 
               3.7.2.2 Finite Difference-Agent Based Model…………………………...108 
  3.7.2.2.1 Modeled Environment………..…………………………...108  
  3.7.2.2.2 Adaptation of Equations and Solutions……………..…….108 
   3.7.2.2.3 Cell ODE Numerical Solution Method…...………..……. 109 
3.7.2.2.4 Cell Division……………………………………...……….109 
  3.7.2.2.5 Diffusion……………………………………………..……110 
  3.7.2.2.6 Time Interval Order………..………………………....….. 110 
3.7.3 Results…….…………………………..………………………………111 
3.7.3.1  Numerical solutions to ODE.………………………………….111 
3.7.3.2 System Sensitivity to basal rate of AI-2 uptake………………..111 
3.7.3.3 Two sets of Lsr with different rates of basal AI-2 uptake……..115  
3.7.3.4 Full population of cells with Lsr in a finite difference environment 
…………………………………………………………………………117 
3.7.3.5  Minimal role of spatial heterogeneity of AI-2……………...…119 
                          3.7.3.6 Agreement between numerical ODE and finite difference agent 
 based solutions………………………………………………………...121 
                       3.7.3.7 Heterogeneity of local Lsr and LuxIR QS activation in growing  
colonies………………………………………….………….…………121 
                           3.7.3.8  Evaluation of clustering when Lsr QS is coupled to AI-2  
chemoattraction…………………….……………………………..…..125 
                          3.7.3.9 Motility mode feedback onto population activation as a function  
of cell-cell distance…………………………………………………...126 
 
 
 
 
vi 
 
         
Chapter 4: Conclusions................................................................................................... 130 
References......................................................................................................................  133 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
vii 
 
List of Tables 
Table 2-S1. Accompaniment for Figure 2-S1, Trackback plots for Lsr system LMNAST 
extended window stringent search hits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 
Table 2-S2.  Results from three LMNAST searches for Lsr system homologs. . . . . 69 
Table 3-S1. Estimated parameter values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
viii 
 
List of Figures 
Figure 1-1. Synthesis of AHL and DPD. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  3 
Figure 1-2. QS molecule activation of QS signaling pathway. . . . . . . . . . . . . . . . . . . 4 
Figure 1-3. LuxIR activity at single cell and multicellular hierarchies. . . . . . . . . . . .  8    
Figure 1-4. The activated methyl cycle and the derivation of DPD. . . . . . . . . . . . . .  11 
Figure 1-5. The Lsr system found in most E. coli. . . . . . . . . . . . . . . . . . . . . . . . . . . . 14   
Figure 1-6. Lsr system effects biofilm development in multiple species. . . . . . . . . . .18 
Figure 1-7. Possible bimodality arising from Lsr activity. . . . . . . . . . . . . . . . . . . . . . 24 
Figure 2-1. Test Queries: lac Operon and Lsr System. . . . . . . . . . . . . . . . . . . . . . . . . 32 
Figure 2-2. LMNAST heuristic. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  35 
Figure 2-3. Lac operon LMNAST hits overlaid onto phylogenetic distributions of 
different scopes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41   
Figure 2-4. Coincidence heat map for lac operon LMNAST stringent search hits. . .  43   
Figure 2-5. 2D similarity plot of lac operon LMNAST stringent search hits overlaid with 
attributed annotation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  46 
Figure 2-6.  Lsr system LMNAST hits overlaid onto phylogenetic distributions of 
different scopes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  49 
Figure 2-7. Annotated 2D similarity plot for Lsr system LMNAST weak search hits. . . . . 
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  51 
Figure 2-8. Coincidence matrix for E. coli Lsr system LMNAST stringent search hits. . . . 
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  54  
Figure 2-9.  Coincidence matrix for E. coli Lsr system LMNAST extended window 
stringent search hits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  55  
 
 
ix 
 
Figure 2-10.  Phylogenetic distribution of Lsr at different phylogenetic scales using 
reconciled LMNAST results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  58 
Figure 2-S1. Trackback plots for Lsr system LMNAST extended window stringent search 
hits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  64 
Figure 2-S2.  GC content demonstrates consistent spiking dip at intergenic region. .   68 
Figure 3-1.  LuxIR and Lsr QS activity intracellularly and intercellularly. . . . . . . . .   76 
Figure 3-2. Lsr autoinduction of pure cultures leads to bimodal phenotype. . . . . . . . . 83 
Figure 3-3. QS dynamics coupled with gliding during colony growth. . . . . . . . . . . . .  86 
Figure 3-4. Cluster-disperse pattern from combination of Lsr and chemotaxis. . . . . .   88 
Figure 3-5. Mixed culture simulations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  89 
Figure 3-S1 Comparison of solution for population with a single basal value versus a 
population with a unimodal distributed value of basal . . . . . . . . . . . . . . . . . . . . . . . . . 112 
Figure 3-S2. Numerical solution for selected Lsr components and AI2-P. . . . . . . . . .  113 
Figure 3-S3. Parameter sensitivity to the parameter, . . . . . . . . . . . . . . . . . . . . . . 114  
Figure 3-S4.  Dual ODE system simulation where second population has varied 
parameter values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  116 
Figure 3-S5. Fraction of cell population QS activated decreases as the variation of the 
parameters Ksynth and VydgG increases. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 
Figure 3-S6. Comparison of results from single versus multiple finite difference elements 
to define environment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 
Figure 3-S7. Congruence of solution from finite difference based modeling versus 
implicit solution of pure ODEs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 
Figure 3-S8.  Measures of the difference between LuxIR and Lsr activation in the context 
of colony growth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 
Figure 3-S9. Clustering of cells with lsr activity and AI-2 chemoattraction as measured 
by cell-cell distance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 
 
 
x 
 
Figure 3-S10.  Measures of the difference between different modes of motility when 
coupled with Lsr/AI-2 dynamics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  129 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1 
 
Chapter 1: Introduction 1 
1.1 Background 2 
Originally termed autoinduction 1,2, “quorum sensing” (QS) is an admitted 3 
misnomer. The first description of a bacterial “quorum” clearly fell within the context of 4 
cell concentration 3.  Nonetheless, for the last two decades the popular equivalency of 5 
“quorum” to a “sufficiently large population” has been rigorously reevaluated 3.  With the 6 
aid of new microfluidic schemes, researchers have shown that QS can be activated for 7 
one or only a few isolated bacteria 4–6.  With the benefit of careful consideration, others 8 
have indicated that under some circumstances, QS might more appropriately be called 9 
“diffusion sensing” or “efficiency sensing” 7–9.  Yet others have argued for paring back 10 
the usage of the term “quorum sensing” from an evolutionary point of view 10.   11 
Indeed, QS studies have become home to a broad spectrum of bacterial 12 
intraspecies and interspecies autoinduction or autoinduction-like signaling.  This ready 13 
adoption is probably attributable not only to the fact that QS is an apparently widespread 14 
phenomenon, but could also be ascribed to the allure of the underlying paradigm: 15 
unicellular organisms using self-secreted molecules to drive multicellular behavior.  As 16 
an additional point of interest, many QS-driven behaviors, such as toxin production and 17 
biofilm formation, are tied to pathogenic phenotypes.  The current count of molecules 18 
generally considered autoinducers includes autoinducing peptide (AIP) from certain 19 
Gram positive bacterial species, autoinducer-1 (AI-1 or acyl homoserine lactones 20 
(AHLs)) from myriad beta- and alpha-proteobacteria, autoinducer-2 (AI-2 or (S)-4,5-21 
dihydroxy-2,3-pentandione (DPD) and its interconvertible stereoisomers) from a great 22 
 
 
2 
 
array of bacteria, autoinducer-3 (AI-3, of unidentified composition) from E. coli, Cholera 23 
autoinducer-1 (CA-1) and its analogs found in other Vibrio species, and Pseudomonas 24 
Quinolone Signal (PQS).  Notably, S-adenosyl methionine (SAM) lies directly upstream 25 
of both AHLs and AI-2 synthesis, as noted in Figure 1-1.  Serving as a methyl donor in 26 
the activated methyl cycle, marked cellular abundance, and high reactivity may make it a 27 
favored target for such repackaging. 28 
Each QS signal is matched to one or more effector pathways that fall under one of 29 
two broad categories, as depicted in Figure 1-2:  two component regulatory systems 30 
(TCRS) or direct transcription factor mediation.  TCRS are typically composed of a 31 
transmembrane receptor with histidine kinase functionality that activates a cytosolic 32 
regulatory partner 11.  For these systems, autoinducers remain extracellular when 33 
generating a response.  A separate category of QS activity is affected through the direct 34 
mediation of transcription factor activity by autoinducer or modified autoinducer.  35 
Occurring in the cytosol, this requires autoinducer internalization through either passive 36 
or active means.  Such cases include the activity of LuxIR found in numerous beta- and 37 
alpha-proteobacteria 12 and Lsr systems found dispersed mainly among gamma-38 
proteobacteria 13,14, respectively.  39 
 40 
 41 
 42 
 
 
3 
 
 43 
Figure 1-1. Synthesis of AHL and DPD (from 15). AHL and DPD share a common 44 
upstream reactant, SAM, from the activated methyl cycle.    45 
 46 
 47 
 48 
 49 
 50 
 51 
 52 
 53 
 54 
 55 
   56 
 
 
4 
 
 57 
Figure 1-2. QS molecule activation of QS signaling pathway.  A.  Signaling molecules 58 
activate the cell through receptor mediation at the cell surface, usually through a two 59 
component regulatory system.  B. Signaling molecules traverse the membrane and 60 
interact with transcription factors directly or after slight modification. 61 
  62 
 63 
 64 
 65 
 66 
 67 
 68 
 69 
 
 
5 
 
Perhaps reflecting the complexity of their environment, many bacteria appear to 70 
operate under the influence of multiple QS signal-effector units.  Sometimes this appears 71 
as topological redundancy or at least functional complementarity.  In other cases, QS 72 
signals appear to compete for influence, although this too could be a type of 73 
complementarity, when discordant phenotypes are separately timed.16  74 
1.1.1 QS in Vibrio harveyi 75 
For instance, Vibrio harveyi QS is known to rely on at least three autoinducers: 76 
3OHC4-HSL (an AHL), DPD, and a V. harveyi version of CA-1.  In this instance, all 77 
three autoinducers interact with separate cognate TCRS pairs, the signals of which are all 78 
funneled into and filtered through the same overlapping negative feedback loops, 79 
preventing premature activation under low autoinducer conditions and limiting 80 
expression at high autoinducer concentrations, ultimately phosphorylating and activating 81 
a single regulatory protein, LuxO.17,18  Despite this canalization, evidence suggests that 82 
different autoinducers accumulate at different rates and help propel separate genetic 83 
programs at distinct phases of growth 16, possibly triggering activity in combination with 84 
distinct sets of transcription factors.   85 
Although signal crosstalk between species can occur when AHLs are structurally 86 
similar19 and is likely where the same molecule has been synthesized by different species 87 
20, whether this necessarily represents QS qua QS is open to interpretation 10.  As a 88 
simplifying generalization, the activity of AHLs are commonly described as limited to 89 
the species from which they are synthesized 21,22.  For example, as far as is known, V. 90 
harveyi’s AHL, 3OHC4-HSL 23, conforms to this paradigm.  Furthermore, orthologs for 91 
 
 
6 
 
V. harveyi’s 3OHC4-HSL synthase and receptor, luxM and luxN respectively, are 92 
phylogenetically confined to a small number of Vibrios 24–26, and produce distinct AHLs. 93 
The limited scope of AHL signaling contrasts sharply with the more universally 94 
produced AI-2—which is to say that DPD synthase (LuxS) homologs are found in myriad 95 
species 27.  While DPD exists as a specific chemical species, it spontaneously 96 
interconverts between several distinct products, each possibly producing a different level 97 
of QS activity 28.  In V. harveyi, AI-2 interacts with the TCRS, LuxPQ 29, which 98 
according to iterative protein BLAST searches is highly abundant among Vibrio and 99 
Marinomonas species. 100 
 A particular derivative formed in marine environments where borate is abundant 101 
is borated DPD 29.  DPD is produced by practically all Vibrio species, and borated DPD 102 
appears to be the cognate AI-2 molecule for those same species 30.   103 
A third V. harveyi QS system involves cqsA and cqsS, a receptor-synthase pair 104 
that is conserved widely but exclusively among Vibrio species.  As with AHLs, the 105 
resulting autoinducers (CA-1-like) are not known to have wide signaling efficacies 106 
outside of the species by which they are produced 31.   107 
Mediation of QS processes at the membrane is a distinctive characteristic of these 108 
QS systems in comparison to other known QS machinery, and makes V. harveyi and the 109 
Vibrio and Marinomonas species (all marine bacteria) that share these architectures 110 
outliers.   111 
Along the separate continuum of QS processes that feedback onto autoinducer 112 
concentration, V. harveyi represents an intermediate in the QS landscape.  On one end of 113 
 
 
7 
 
this spectrum exists QS systems like the canonical luxIR pair, where AHL synthase 114 
expression is driven by autoinducer-activated transcription, producing a strong 115 
intercellular positive feedback loop.  On the other end is QS activation that is not only 116 
dissociated from autoinducer synthesis but also drives uptake, producing a strong 117 
negative intercellular feedback loop.  118 
1.1.2 LuxIR QS 119 
 Widely utilized in synthetic biology, the LuxIR system is composed of an AHL 120 
synthase, LuxI, and an AHL sensitive transcription factor, LuxR.  Its operations are 121 
depicted in Figure 1-3. Freely diffusing across the membrane, AHL extracellular 122 
concentration is reflected intracellularly and vice versa.  Given sufficient AHL with 123 
which to complex, LuxR becomes less prone to degradation, accumulates, and promotes 124 
the expression of both itself and LuxI, increasing AHL synthesis, thereby completing an 125 
intracellular positive feedback loop 21. Furthermore, this sharp increase in AHL synthesis 126 
leads to a corresponding increase in local extracellular concentration, coordinating the QS 127 
activation of neighboring cells which in turn influence their own neighbors, generating 128 
positive feedback at a multicellular scale. This system core and variations thereof 129 
represent the dominant pathway by which AHLs influence downstream phenotypes.   130 
 
 
8 
 
 131 
Figure 1-3. LuxIR activity at single cell and multicellular hierarchies.  A. AHL freely 132 
traverses the membrane and stabilizes the transcription factor, LuxR.  This activates a 133 
positive feedback loop, increasing LuxR expression as well as that of LuxI, the AHL 134 
synthase.  B. This positive feedback operates at the multicellular level as well.  135 
 136 
 137 
 138 
 139 
 140 
 141 
 142 
 143 
 144 
 
 
9 
 
LuxIR homologs are widespread throughout proteobacteria. As of 2008, 26% of 145 
sequenced bacterial genomes contained a complete set of homologs 12.  Among them are 146 
the pathogens Yersinia pestis, Agrobacteria tumefaciens, and Pseudomonas aeruginosa.  147 
Many species contain multiple and distinct luxIR pairs 12, which were likely acquired 148 
independent of each other even if their signaling converges downstream 32.   149 
Effective at organizing population expression, luxIR systems have often been 150 
repurposed toward applied ends.  Initial efforts usually involved adding exogenous genes 151 
under the luxR promoter, such as an enzymatic degrader of AHL to create an oscillating 152 
signal 33 or chloramphenicol to limit population density despite sufficient nutrient in the 153 
environment 34.  In certain cases, luxIR QS was used to ensure complete induction within 154 
a population 35. More advanced designs included separating luxIR sensing and 155 
autoinducer production capacities into multiple bacterial carriers whether for synthetic 156 
biology or therapeutic ends 36,37.   157 
In addition to complete luxIR pairs, most species contain more luxR homologs 158 
than luxI homologs.12  Possibly, at least some of these unmatched luxR homologs are not 159 
redundant, but serve as a means of testing for different species’ AHLs.  This is the case 160 
for Rhizobium leguminosarum bv. viciae which depends upon exogenous N-(3-hydroxy-161 
7-cistetradecenoyl)-L-homoserine lactone (3-OH-C14:1-HSL) to induce conjugation 38.  162 
The luxR homolog, sdiA, found in E. coli, Salmonella, and related enterobacterial species 163 
39, also appears to support such a role.  While these species do not synthesize any AHL, 164 
they nevertheless respond to several AHLs through SdiA 40–42.  As might be expected of a 165 
quorum sensing regulator, sdiA is at least in some species tied to biofilm activity 43, 166 
although the extent of its reported signaling has varied depending on whether the study 167 
 
 
10 
 
involved a knockout 44, plasmid based overexpression 42, or reporter fusion into the 168 
chromosome 45.       169 
1.1.3 AI-2 QS 170 
Perhaps even more broadly distributed than AHL based associated QS is AI-2 171 
driven QS.  However, this is not without caveats. For whereas many bacteria have AHL 172 
receptors without producing the specific AHL, many bacteria appear to produce AI-2 173 
without an apparent ability to perceive the signal.  This inversion has previously led some 174 
to suggest that AI-2 may represent a metabolic by-product rather than a true signaling 175 
molecule.  Indeed, LuxS in combination with Pfs are an integral part of regenerating 176 
homocysteine after methyl donation as part of the activated methyl cycle, as seen in 177 
Figure 1-4 46.  This pathway serves as an alternative to S-adenosyl homocysteine (SAH) 178 
hydrolase (SahH).  While eukaryotes use the SahH pathway exclusively, the proportion 179 
of bacteria expressing sahH is roughly split with that utilizing the luxs/pfs pathway 46.  180 
With only the rarest exception are luxS and sahH ever coincident to the same genome 181 
27,47.  Importantly, bacteria containing sahH are unable to produce AI-2 except by 182 
conversion from ribulose-5-phosphate 48,49.  Whether the flux from this secondary 183 
synthesis pathway is sufficient to effect even limited QS operations remains unknown 49.  184 
Still, while AI-2 production may not be universal, luxS is nonetheless well represented 185 
among bacterial genomes, and like AHL, AI-2 signaling is not restricted to the bacteria 186 
from which it is synthesized.  Moreover, whether various combinations of exogenous in 187 
vitro synthesized 27 or purely synthetic 50 AI-2 and exogenous sahH expression can 188 
rescue luxS mutant phenotypes provides an interesting point of study. 189 
 
 
11 
 
 190 
Figure 1-4. The activated methyl cycle and the derivation of DPD (from 51).  AI-2 is an 191 
umbrella term for a number of spontaneously cyclizing DPD derivatives.  DPD itself is a 192 
byproduct of the activated methyl cycle during the regeneration of homocysteine, at least 193 
in some bacteria.  Most other cells with an activated methyl cycle utilize the alternative, 194 
SahH, which bypasses the production of DPD. 195 
 196 
 197 
 
 
12 
 
Less phylogenetically dispersed than AI-2 synthesis, receptors for the molecule 198 
are nonetheless found in numerous bacteria. Predominantly counted among bacteria with 199 
receptors are gamma-proteobacteria, for which the luxS transgenomic signal appears to be 200 
the most monophyletic 46,52.  This is partly attributable to LuxP receptors found among 201 
numerous Vibrio and Marinomonas species, as previously discussed.  Among non-Vibrio 202 
and non-Marinomonas species, the only other known AI-2 QS receptor is the periplasmic 203 
receptor protein, LsrB, the binding component of an ABC type transporter.  Originally 204 
identified in Salmonella along with the rest of the LuxS regulated (Lsr) system 53, LsrB 205 
has been shown through reciprocal BLAST searching to be most prevalent, but not 206 
universal in, nor restricted to Pasteurellaceaes and Enterobacteriales 13.  Just as Vibrios 207 
and Marinomonas species exist in a shared environment, it could be suggested that LsrB 208 
has been transferred horizontally among bacteria sharing the same environment 14.   209 
In its sum, the Lsr system acts as a sugar-like importer system, composed of 210 
divergently arranged operons that share an intergenic region, lsrRK and lsrACDBFGE in 211 
Salmonella 53,54. lsrE is absent in E. coli, which otherwise has a nearly exact duplicate 55–212 
57.  LsrR belongs to the deoR family of transcriptional repressors, interacting with the 213 
intergenic region through a helix-turn-helix motif, keeping system expression to a 214 
minimum when sufficient substrate is unavailable.  LsrA is the nucleotide binding 215 
component of the ABC type transporter, binding and using ATP to drive AI-2 import.  216 
LsrC and LsrD are transmembrane proteins comprising a pore for AI-2 between the 217 
cytosol and the periplasm.  As described previously, homology indicates that LsrB binds 218 
AI-2 in the periplasm, where the complex joins to the transmembrane components.  Once 219 
inside the cytoplasm, AI-2 is phosphorylated by LsrK.  Notably, lsrK appears to have its 220 
 
 
13 
 
own constitutive promoter, albeit one that does not drive significant expression 56.  221 
Phosphorylated AI-2 (AI2-P) de-represses system expression by destabilizing LsrR’s 222 
interaction with DNA.  AI2-P is also broken down into multiple products by LsrF and 223 
LsrG.  The exact function of lsrE is unknown, although it is commonly alternatively 224 
annotated as ribulose-3-phosphate epimerase.  This, and its operon position suggests it 225 
may operate on AI2-P 51.  Unlike lsrF and lsrG, however, it fails to effect apparent AI2-P 226 
levels 54. 227 
The Lsr system is also affected by multiple factors outside its apparent regulon. 228 
For example, cAMP is required for system expression 55. Additionally, AI-2 export 229 
requires the transmembrane protein YdgG, without which AI-2 remains sequestered 230 
within the cell 58.  YdgG appears to belong to a well conserved superfamily of exporters 231 
that have no other ascribed function 59. Finally, it is believed that an additional low flux 232 
importer pathway is required for import when the Lsr system is inactive 55,60.    233 
 A few additional proteins have also been shown to mediate AI-2 activity at the 234 
cell membrane.   These include RbsB from the ribose importer system in Aggregatibacter 235 
actinomycetecomitans 61, TlpB which allows Helicobacter pylori to be chemotactically 236 
repelled by AI-2 62, and the PTS system in E. coli 63.  While TlpB is a broad array 237 
chemoreceptor, both RbsB and PTS are involved in the import of sugars across the 238 
membrane.  RbsB was shown to compete with LsrB for AI-2 and that it was required for 239 
Lsr activation 61.  In a certain sense, however this is unsurprising insofar as many high 240 
affinity transporters for one monosaccharide act as low affinity transporters for related 241 
monosaccharides.  For example, GalP imports glucose and galactose at high rates, and 242 
 
 
14 
 
 243 
Figure 1-5. The Lsr system found in many E. coli.  The ABC-type transporter and LsrK 244 
serve as positive feedback elements, working serially to produce AI-2P. LsrR and 245 
LsrF/G, on the other hand, serve as negative feedback elements, encouraging system 246 
repression or discouraging system de-repression respectively.  Outside of these elements 247 
exist the transporters YdgG (an exporter) and a low affinity alternative importer pathway 248 
(consisting, at least, of the PTS system). 249 
 250 
 251 
 252 
 253 
 254 
 
 
15 
 
lactose and fructose at lower rates when both of its primary targets are absent 64,65.  This 255 
also obtains for ALBP, which primarily transports allose across the membrane but also 256 
doubles as a low flux pathway for ribose 66.  Both allose and ribose contain aldehyde 257 
functional groups.  That both ribose and AI-2 are both 5 carbon molecules lends credence 258 
to the possibility of such a secondary role for RbsB.  While RbsB is at least indirectly 259 
involved in Aggregatibacter actinomycetemcomitans’s AI-2 response 67, the regulatory 260 
pathway is unknown.  Possibly, the rbs system comprises the alternative low flux 261 
importer affecting Lsr response, but this remains a subject of further research.  262 
Interestingly, a system annotated as rbs in Haemophilus influenzae also appears to 263 
regulate AI-2 influx.  Additionally, this rbs’s expression is also controlled by AI-2.  In 264 
the reported studies, none of the pentose sugars tested, including ribose, induced 265 
competitive inhibition, however, indicating that this may simply be a case of improper 266 
annotation 68, serving as a reminder of possible annotation biases associated with 267 
precedence.  Many of the same ideas regarding AI-2 cross-reactivity with rbs apply to the 268 
PTS system which serves as a flux pathway for a variety of sugars, and upon which Lsr 269 
signaling is dependent in E. coli 63.   270 
1.1.4 Lsr regulon  271 
The exact pathway by which signal is transduced by the Lsr system appears to 272 
vary from species to species.  Two separate studies have assessed the nature and number 273 
of potential LsrR binding sites.  According to ChIP-CHIP analysis, the LsrR regulon is 274 
limited to the Lsr system itself in Salmonella Typhimurium 69.  In E. coli, the LsrR 275 
regulon is ostensibly more expansive.  While the footprinting differential between lsrR 276 
knockouts and wildtype cells is strongest near the lsr intergenic region, additional 277 
 
 
16 
 
binding sites include regions near yegE, mppA, and yihF 70.  Both yegE and mppA assist 278 
in the transition from exponential growth toward a sessile lifestyle.  A probable 279 
diguanylate cyclase, yegE appears to help coordinate the concomitant downregulation of 280 
flagellar genes and upregulation of curli 71.  Separately, mppA is involved in recycling 281 
membrane associate peptides 72.   282 
Additional means by which Lsr may influence downstream phenotypes exist in 283 
the form of LsrF’s and LsrG’s enzymatic outputs.  While LsrF is believed to be an 284 
aldolase, its specific products remain unknown 73.  LsrG is thought to act as an isomerase 285 
under anaerobic conditions to produce 3,3,4-trihydroxy-2-pentanone-5-phosphate 74, 286 
whereas it cleaves AI2-P under aerobic conditions to produce 2-phosphoglycolic acid 287 
(PG) and an unidentified 3 carbon molecule 75.  PG acted upon by the phosphatase gph 288 
reenters metabolism as glycolic acid 76.  Whether any of these products affects 289 
downstream phenotypes remains unknown. 290 
1.1.5 Known Lsr Associated Phenotypes 291 
The reported effects of LsrR mutation and other Lsr system modifications mirrors 292 
the ChIP-CHIP results.  LsrR did not have an apparent impact on the measured 293 
phenotypes, invF and flagella gene expression, in Salmonella Typhimurium except when 294 
luxS was also deleted or when LsrR was overexpressed 77.  This suggests that LsrR 295 
binding to DNA may be easily destabilized by the presence of AI2-P in Salmonella, 296 
perhaps pointing to a reason why LsrR’s direct regulon may be limited in this species.  297 
On the other hand, in E. coli, deletion of either lsrR or lsrK led to marked inhibition of 298 
biofilm development 60 (Figure 1-6a).  Other experiments involving luxS knockouts with 299 
 
 
17 
 
homocysteine and SAM controls indicated that exogenous DPD restored biofilm height 300 
while not fully complementing biomass accumulation 78.  Interestingly, biofilm 301 
phenotype was also inhibited by lsrR deletion in Aggregatibacter 302 
actinomycetemcomitans, a bacteria common to the oral cavity 79.  In this case, although 303 
lsrRK knockout led to more dramatic changes than lsrR deletion alone, statistically 304 
significant changes arising from lsrK deletion remained unidentified, leading to a 305 
suspicion that an additional kinase may be working on cytosolic AI-2 79 (Figure 1-6b).  306 
This small cohort of studies indicates that the Lsr system can influence population scale 307 
phenotypes in a manner expected of QS systems.    308 
1.1.6 Broader AI-2 Associated Phenotypes 309 
In general, QS drives behaviors that work most efficiently when orchestrated at 310 
multicellular scales 80.  Along with swarming and sporulation, among the more strikingly 311 
cooperative of QS guided behaviors is biofilm formation.  In addition to the apparent 312 
cooperativity involved, biofilms are also known for their protective role, which lends to 313 
bacteria’s robustness against environmental insults (e.g. oxidation and antibiotics) and 314 
also plays a role in refractory infections.  This is a common theme in QS signaling, which 315 
appears to influence numerous aspects of virulence including motility switching, bacterial 316 
adhesion, invasion, and toxin production.  More broadly, AI-2 signaling might be 317 
expected to control numerous aspects of bacterial interactions in natural consortia.  318 
In addition to the species previously mentioned, AI-2 in particular has been 319 
shown to influence biofilm formation and other cooperative behaviors in many bacteria, 320 
often in a manner similar to that in E. coli, where, as discussed previously, exogenous  321 
 
 
18 
 
 322 
Figure 1-6. Lsr system effects biofilm development in multiple species. A. As adapted 323 
from 60, E. coli  biofilm development appears stalled upon deletion of either lsrR or lsrK.  324 
B.  The same appears to be true for A. aggregatibacter, except the effect from lsrK 325 
mutation was not statistically significant.  Dual deletion resulted in a sparser biofilm, and 326 
complementation on a plasmid restored that defect. 327 
 328 
 329 
 330 
 331 
 
 
19 
 
DPD partially complemented luxS deletion.  This inexact restoration may be attributable 332 
either to the dual signaling and metabolic role played by LuxS or to the transport 333 
limitations of exogenous DPD supplementation at both macroscales and microscales (i.e. 334 
around biofilm architecture or across the cell membrane). 335 
A couple studies have focused on bacterial species synthesizing AI-2 and 336 
containing Lsr homologs, without explicitly studying Lsr based expression.  DPD 337 
complemented luxS mutation in Photorhabdus luminescens for some differentially 338 
expressed genes to an unspecified degree while others were unaffected.  Fully or partially 339 
restored expression was found in virulence genes such as hemolysin and pyocin, 340 
oxidative response genes such as msrB (Peptide methionine sulfoxide reductase B) and 341 
uvrC (Exonuclease ABC subunit C), and motility related genes such as flhC (Flagellum 342 
biosynthesis transcription activator).  In this report researchers were additionally unable 343 
to duplicate DPD induced changes with homocysteine supplementation.81  In Bacillus 344 
cereus, exogenously added AI-2 affected biofilm growth in a closely titrated manner, 345 
initially increasing biofilm accumulation but also inhibiting biofilm growth when 346 
exposed to a higher concentration within the same order of magnitude 82.  347 
A separate group of bacteria producing AI-2 but lacking a known AI-2 receptor 348 
also appear to have AI-2 sensitive phenotypes.  Exogenously added DPD increased 349 
adhesion in a dose dependent fashion in both luxS mutant and wildtype Actinobacillus 350 
pleuropneumoniae populations 83. In Streptococcus pneumonia, researchers were able to 351 
demonstrate partial reestablishment of biofilm biomass to wildtype levels in a luxS 352 
mutant treated with exogenous DPD 84. In Streptococcus epidermidis, exogenous AI-2 353 
complementation restored PSM (pro-inflammatory phenol soluble modulin peptides) 354 
 
 
20 
 
expression in addition to that of many metabolic genes  that were differentially expressed 355 
in a luxS knockout compared to a WT strain 85. In Streptococcus intermedius, AI2 356 
exogenously added to luxS mutants again produced a distinct biofilm phenotype 357 
compared to WT and luxS mutants in both the presence and absence of antibiotics. 358 
Moreover, AI-2 conferred some ability to form a basal level of biofilm in the face of 359 
several classes of antibiotic.86  Among these reports, one involving Borrelia 360 
burgdorferi serves as a particularly illuminating example.  For this species and in many 361 
of its relatives, the activated methyl cycle is incomplete.  In particular, B. burgdorferi 362 
cannot generate methionine from homocysteine.  Nonetheless, AI-2 is produced through 363 
the LuxS/Pfs pathway, and moreover, DPD can complement luxS deletion, restoring 364 
expression of vlsE and erp, both of which translate into proteins that mediate the host-365 
pathogen interaction.87 In addition to its ability to chemorepel from AI-2, H. pylori also 366 
appears to regulate multiple flagellar genes with AI-2, as AI-2 complements luxS deletion 367 
and also enhances these flagellar gene expression levels when added to wildtype cells. 88  368 
A number of other bacteria that do not synthesize AI-2 also respond to 369 
exogenously supplied DPD.  Ranking among these is Sinorhizobium meliloti, which 370 
contains an Lsr homolog 14 and can deplete AI-2 from the extracellular environment 371 
without apparent consequence to colonization or growth phenotypes 89.  Pseudomonas 372 
aeruginosa expression of virulence genes is influenced by exogenous AI-2 90. Finally, 373 
DPD was shown to influence biofilm production in a dose responsive fashion via the 374 
induction of an oxidative stress response pathway in Mycobacterium avium 91.   Unlike S. 375 
meliloti neither P. aeruginosa nor M. avium has an identified AI-2 response pathway. 376 
 377 
 
 
21 
 
1.1.7 QS in an Ecological Context 378 
 From an evolutionary point of view, QS and AI-2 based-QS in particular 379 
represent a peculiarity due to perceived, or at least potential, cooperation between 380 
nonrelated bacterial species.  By way of illustration, pure culture QS signaling that 381 
encourages biofilm formation is an entirely suitable method by which to organize 382 
population-wide expression of public goods in an isolated setting.  In such instances, QS 383 
based cooperation can be attributed to kinship.92  When considering QS related mutations 384 
and the likely mixture of genetic backgrounds and species within a bacterial consortium, 385 
however, the exact manner in which evolutionary pressures inform and maintain QS 386 
signaling becomes less straightforward.   In even moderately more complex 387 
circumstances, questions of which bacteria produce signal, which bacteria respond to 388 
signal, and which bacteria benefit from any resulting product can quickly overwhelm the 389 
boundaries of current understanding. 390 
 391 
 392 
 393 
 394 
 395 
 396 
 397 
 
 
22 
 
1.2 Research Motivation 398 
 Lsr based QS in E. coli represents a practical model for studies herein.  Among 399 
the better characterized of the Lsr systems 49,55–58,60,63 and effecting virulence phenotypes 400 
60,78, the study of such a system is of natural interest for a diverse generalist bacteria like 401 
E. coli, certain strains of which are a common source of food supply contamination.  402 
Arising from such concerns, the identification of strains containing Lsr system homologs 403 
and how Lsr based QS might play a role in infection processes among these strains is of 404 
interest.  Yet even more broadly, it is important to identify other species in which the Lsr 405 
system may play a signaling role.  While much research has been conducted to identify 406 
AI-2 responsive bacteria, sometimes through the Lsr system, questions of 407 
generalizability, while somewhat epistemological in nature, are also useful when answers 408 
can be had.  By the identification of useful homology, and determining how homologs are 409 
constituted, one may begin a line of inquiry into questions of generalizability.  410 
The phylogenetic dispersion of AI-2 receptors has been assessed as a means of 411 
determining which bacteria might be responsive to AI-2 as an external signaling stimulus 412 
13.  The effectiveness of such searching is largely dependent upon the increasingly rapid 413 
pace of whole genome sequencing, and while many such receptors have been identified 414 
there remain numerous AI-2 responding bacteria with unknown receptors.  While it is 415 
possible additional sugar transporters moonlight as high affinity AI-2 transporters, sugar 416 
transporters acting as a lone pathway for AI-2 import has yet to be demonstrated in any 417 
species.  In general, the perception of AI-2 is of natural interest, as some cells may 418 
recognize it at lower concentrations than others, etc; and here, perception is meant to not 419 
only indicate reception, but also encompasses subsequent gene expression changes.  Here 420 
 
 
23 
 
we wish to only consider possible Lsr gene expression changes, while remaining aware 421 
that the Lsr system may effect a broader regulon.  Toward this end the identification of 422 
extant Lsr genes, operon organization 93, and the phylogenetic distribution of variants 423 
among Lsr homologs is of interest.      424 
A second cohort of studies described herein were initially motivated by 425 
experiments indicating that stable bimodal Lsr system expression may arise in pure 426 
cultures 94 (Figure 1-7a).  While it has been suggested that a transient bimodality may 427 
arise as a consequence of intracellular signaling topology 95, a more permanent 428 
population bifurcation may also develop from hyperlocal competition for AI-2 between 429 
cells, as depicted in Figure 1-7b.  Bimodal expression within a pure culture would 430 
represent a built-in population of QS cheats, and depending on the expected benefit could 431 
speak to issues about free riders and other evolutionary conundrums.      432 
The picture is further complicated for bacteria that are chemoattracted to AI-2, 433 
such as E. coli  96,97.  Although QS generally informs population scale behaviors, 434 
activation of Lsr and the resulting net recompartmentalization occur on time and spatial 435 
scales that intersect with chemotaxis. As a means of isolating this confluence, a mixed 436 
finite difference-agent based approach can be used to model emergent behaviors 98.  437 
Using such a platform, contextualized QS expression between two competing populations 438 
with different genotypically mixed signal/reception competency can be interrogated.  439 
Specifically, as a possible consequence of its universality, the literature suggests that AI-440 
2 signaling within a consortia is potentially quite complex.  How QS operates in an 441 
environment with multiple systems operating concurrently remains an open question.  442 
Thus, in addition to investigating QS by pure cultures, this framework can also be used to  443 
 
 
24 
 
 444 
Figure 1-7. Possible bimodality arising from Lsr activity.  A. FACS results adapted from 445 
94 suggest that a bimodality develops over time.  B. A diagram of a possible mechanism 446 
whereby bimodality could develop, as extracellular negative feedback operates 447 
hyperlocally. 448 
 449 
 450 
 451 
 452 
 453 
 454 
 455 
 456 
 457 
 458 
 
 
25 
 
investigate basic QS systems acting concurrently in both time and space.  Specifically, 459 
how different Lsr system variants might compete with one another can be investigated. 460 
Essentially, in addition to determining which species and strains might be 461 
responsive to AI-2, there exists additional opportunity to ask how they might be 462 
responsive to AI-2.  Research presented herein takes a closer look at the phylogenetic 463 
distribution of Lsr system homologs and the composing members of its homologs.  We 464 
then consider more closely how the Lsr system possibly produces bimodal expression 465 
through the mathematical modeling of two competing populations.  We further consider 466 
what the ramifications of system behavior might be in the context of motility, and how 467 
multiple bacteria with different Lsr system variants might compete for AI-2 with one 468 
another.      469 
 470 
 471 
 472 
 473 
 474 
 475 
 476 
 477 
 478 
 
 
26 
 
1.3 Global Objective, Global Hypothesis, and Specific Aims 479 
The global objective of this dissertation is to further investigate the nature of the Lsr 480 
system, using E. coli K-12 W3110’s Lsr system as a starting point, both interrogating 481 
how W3110’s Lsr homologs compare to each other and how they compare to other QS 482 
systems.  483 
Global Hypothesis: E. coli’s Lsr system is one among a spectrum of Lsr systems, the 484 
diversity of which bears signaling consequences both in pure cultures and mixed 485 
consortia.  486 
Specific Aim 1: Find E. coli Lsr system homologs. 487 
Specific Aim 2: Determine if the known components of the Lsr system, along with 488 
auxillary proteins can lead to stable, bimodal Lsr activation. 489 
Specific Aim 3: Assess how Lsr’s recompartmentalization dynamics may differentiate it 490 
from other QS systems within the context of population activation, emergent 491 
phenomenon, and competition between bacteria containing different Lsr systems.    492 
 493 
 494 
 495 
 496 
 497 
 
 
27 
 
1.4 Dissertation Outline 498 
Chapter 2 describes a novel algorithm to find homologs of modular networks and an 499 
analysis of the results when using the E. coli K-12 Lsr system.  This is contrasted to the 500 
results for the lac system from the same strain.  501 
Chapter 3 describes the development of a set of ODE’s that describes the development of 502 
bimodal system expression given parameter variation. 503 
Chapter 4 describes the development and results of a finite difference-agent based model 504 
contrasting LuxIR activation to Lsr activation within the context of emergent behaviors 505 
arising from the confluence of QS and varying motility modes. 506 
Chapter 5 summarizes the previous chapter’s work and indicates challenges to future 507 
work, denoting the larger implications and significance of the work. 508 
 509 
 510 
 511 
 512 
 513 
 514 
 
 
28 
 
Chapter 2: Comparison of Homolog Identification for the sugar 515 
importing Lac System and the QS Lsr System from E. coli K-12 W3110 516 
 517 
2.1 Abstract 518 
Bacterial cell-cell communication is mediated by small signaling molecules known as 519 
autoinducers. Importantly, autoinducer-2 (AI-2) is synthesized via enzyme LuxS in over 520 
80 species, some of which mediate their pathogenicity by recognizing and transducing 521 
this signal in a cell density dependent manner. AI-2 mediated phenotypes are not well 522 
understood however, as the means for signal transduction appears varied among species, 523 
while the AI-2 synthesis process appears conserved. Approaches to reveal the recognition 524 
pathways of AI-2 will shed light on pathogenicity as we believe recognition of the signal 525 
is likely as important, if not more, than the signal synthesis.  LMNAST (Local Modular 526 
Network Alignment Similarity Tool) uses a local similarity search heuristic to study gene 527 
order, generating homology hits for the genomic arrangement of a query gene sequence.  528 
We develop and apply this tool for the E. coli lac and LuxS regulated (lsr) systems.  Lsr 529 
is of great interest as it mediates AI-2 uptake and processing. Both test searches 530 
generated results that were subsequently analyzed through a number of different lenses, 531 
each with its own level of granularity, from a binary phylogenetic representation down to 532 
trackback plots that preserve genomic organizational information.  Through a survey of 533 
these results, we demonstrate the identification of orthologs, paralogs, hitchhiking genes, 534 
gene loss, gene rearrangement within an operon context, and also horizontal gene transfer 535 
(HGT).   We also found a variety of operon structures that are consistent with our 536 
 
 
29 
 
hypothesis that the signal can be perceived and transduced by homologous protein 537 
complexes, while their regulation may be key to defining subsequent phenotypic 538 
behavior.  539 
 540 
 541 
 542 
 543 
 544 
 545 
 546 
 547 
 548 
 549 
 550 
 551 
 552 
 553 
 554 
 
 
30 
 
2.2 Introduction 555 
Comparing prokaryotic whole genome sequences to identify operons is a mature 556 
area of research 99–102.  Orthologous operon identification can imply a secondary degree 557 
of relation between components, reaffirming Clusters of Orthologous Groups (COG) and 558 
other assignments of function as well as suggesting essentiality 103.  This conservation of 559 
components also speaks to the conservation of signaling capacity in orthologous modular 560 
signaling operon-based units.  That is, we are interested in ascertaining the genetic 561 
modularity of signal transduction processing, in particular those that operate within 562 
known, putative regulons.  Drawing partly on previous work investigating microsynteny 563 
and gene neighborhoods 101,104,105, we developed a general similarity search approach, we 564 
call a Local Modular Network Alignment Similarity Tool (LMNAST).  LMNAST applies 565 
a BLAST-like heuristic to gene order and arrangement.  Resultant search hits help 566 
capture the conservation and phylogenetic dispersion of a given query modular network.   567 
Using, as queries, contiguously abutting genes of prokaryotic modular signaling 568 
networks, LMNAST identifies and scores hits based on the minimum number of frank 569 
mutations in gene organization needed to arrive at a given putative system homolog 570 
starting from the query.  Here, homology refers to similarity in relative gene order and 571 
relative transcriptional direction, after nucleotide level threshold filtering of gene 572 
elements based on BLAST 106 E-value.   573 
For the purpose of evaluation, two small modular systems were used as test 574 
inputs: one was the E. coli lac system and the other was the LuxS regulated (Lsr) system. 575 
In some ways, the two systems are quite similar (Figure 2-1).   Both import and 576 
 
 
31 
 
catabolize the small molecules that induce system expression.  For the lac system, this 577 
small molecule is, of course, lactose.  For the Lsr system, the small molecule is 578 
autoinducer-2 (AI-2).  AI-2 is a signaling molecule common among at least eighty 579 
bacterial species 27.  As mediated either through the Lsr system or LuxPQ, bacteria are 580 
believed to use AI-2 to guide population based phenotypes, a phenomenon termed 581 
quorum sensing 54.  LuxPQ is a histidine kinase two component system, the regulon of 582 
which is distinct from Lsr and is not considered further.  Lsr is an interesting query 583 
because its distribution should help elucidate its putative, modular quorum sensing 584 
function 27 and because the known homologs differ in gene organization.54,55,82  585 
 586 
 587 
 588 
 589 
 590 
 591 
 
 
32 
 
 592 
Figure 2-1. Test queries: lac operon and Lsr system. A  The lac operon is composed of 593 
beta-galactosidase (LacZ), the lactose importer (LacY), and a beta-galactoside 594 
transcetylase (LacA).  Upstream of the operon, the operon repressor (LacI) is expressed 595 
in a co-directional orientation.  The primary function of the lac system is as a regulated 596 
importer/processing unit.  Lactose brought in through LacY is converted into allolactose 597 
or hydrolyzed into glucose and b-galactose.  Both reactions are catalyzed by LacZ.  598 
Allolactose then acts to release the repression of the system by LacI. B  The Lsr system is 599 
composed of two divergent operons. One operon consists of an AI-2 kinase, and a system 600 
repressor.  The other operon consists of an AI-2 transporter and phospho-AI2 (AI2-P) 601 
processing genes.  Contextual system behavior is partly governed by separately regulated 602 
parts including an alternative importer 61, an exporter 58, and the AI-2 synthase gene.  603 
Relative to the canonical lac system, the Lsr system is complicated by the fact that the 604 
cell synthesizes, exports, and imports AI-2, and by the negative regulation associated 605 
with the divergently arranged structure.  AI-2 exported by a mechanism involving YdgG 606 
traverses the outer membrane through a porin and enters the periplasmic space.  Through 607 
the ABC-type importer, LsrACDB, AI-2 is then transported back into the cytosol.  Once 608 
 
 
33 
 
there, AI-2 is phosphorylated by LsrK.  This phosphorylated form (AI2-P) de-represses 609 
the lsr system and is catabolized by LsrF and LsrG into separate downstream products.      610 
 611 
 612 
 613 
 614 
 615 
 616 
 617 
 618 
 619 
 620 
 621 
 622 
 623 
 624 
 625 
 626 
 
 
34 
 
2.3 Methods and Algorithm 627 
As previously indicated, LMNAST evaluates nucleotide records for similarity to a 628 
query network using a BLAST-like heuristic.  Specifically, it captures the gene order of 629 
query networks as a string of characters.  Queries are therefore not necessarily restricted 630 
to defined networks insofar as any gene ordering may be a query. A standard heuristic of 631 
penalties for various rearrangements of orthologous system is employed.  For searches 632 
described herein, a loose threshold was used to generate an exhaustive set of hits.  The 633 
overall scheme is depicted in Figure 2-2.  The program itself is available at 634 
http://www.bentley.umd.edu. 635 
 636 
 
 
35 
 
 637 
Figure 2-2. LMNAST heuristic.  LMNAST operates in a BLAST-like manner, using the 638 
results of BLAST searches themselves as a curated database.  639 
  640 
1. For each member of the query, in any nucleotide record, a homolog’s membership to a 641 
character type is assigned by scoring below a specified BLAST E-value threshold. Genes 642 
assigned to characters are highlighted blue.  Genes without sufficient homology to any 643 
character are represented by dashed boxes. 644 
2. Sufficiently long stretches of adjacent characters are identified as seeds (red). 645 
 
 
36 
 
3. Sufficiently proximal characters are connected to seeds or seeds are connected to each 646 
other when at a base pair distance < d.   647 
4. Rearrangements, losses, and deletions are scored according to a standard similarity 648 
heuristic.  Noncontinuous elements are dropped iteratively until a maximum score is 649 
achieved, arriving at…  650 
5. An LMNAST hit.   651 
 652 
 653 
 654 
 655 
 656 
 657 
 658 
 659 
 660 
 661 
 662 
 663 
 
 
37 
 
2.3.1 Input 664 
Input consisted of an ordered list of gene elements (for example, lacIZYA).  For 665 
each gene element a BLAST result file was generated using tblastn to search the nr/nt 666 
database for hits with E-values less than 0.1, narrowing the search space.  Each BLAST 667 
hit was assigned a character corresponding to the gene element queried.  BioPerl 107 was 668 
used to query Genbank databases and process data from retrieved files.  Nucleotide 669 
records with sufficiently proximal characters were investigated further. 670 
2.3.2 Scoring Heuristic 671 
The degree of similarity between a putative hit and a query was evaluated 672 
according to the number of deletions, insertions, and rearrangements required to generate 673 
the putative hit using the query as a starting point.  Intra-hit gene duplications were 674 
disallowed as a simplification.  Consequently, deletion could be noted by character type 675 
inclusion.  Insertions of uncharactered elements between gene homologs were scored 676 
according to an affine gap rule whereby a portion of the deduction was scaled to the 677 
insertion length.  Rearrangements refer to altered relative order and relative gene 678 
direction.  Changed relative direction was only considered when relative order was 679 
maintained.  When this criterion was satisfied, relative order was evaluated in terms of 680 
adjacent homolog distance, disregarding insertions and deletions.  For each such 681 
structural dissimilarity there was a standard deduction in score.  Noncontiguous elements 682 
were dropped iteratively until a maximum score was reached for each putative hit.  When 683 
putative hit versions elicited equal scores in the same round, the version of the hit with 684 
 
 
38 
 
the most characters was retained.  Putative hits with scores greater than zero were 685 
retained.  686 
2.3.3 Weak and Stringent Criteria 687 
For evaluation purposes and to find a suitable balance between false positives and 688 
coverage completeness, each test query was run under both weak and stringent 689 
conditions.  Stringent criteria searches assumed accurate annotation. Contrarily, weak 690 
criteria did not require genes to lie within the annotated coding sequence.  Moreover, 691 
characters annotated as “pseudo” or bounded outside “gene” annotation were accepted as 692 
homologous characters.  Weak criteria searches also allowed multiple genes to co-exist 693 
within the same annotation.  Additionally, as a concession to the possibility of longer 694 
range interactions between genes, reduced gap penalties were used in weak criteria 695 
searches.  Results described herein were derived using a gap penalty of 1 and 2 with an 696 
extension penalty of 0.3 and 1, for weak and stringent criteria searches respectively. 697 
2.3.4 Ancillary LMNAST Search Tools 698 
Mean element homology (meH) is a normalized, ancillary measure of string 699 
similarity as evaluated by BLAST.  Useful for contrasting BLAST results to LMNAST 700 
hits, meH was calculated by normalizing each gene homolog’s bit score to the maximum 701 
bit score for the entire corresponding BLAST result with a background subtraction of the 702 
minimum bit score.  These normalized bit scores were then averaged for all gene 703 
elements within an LMNAST hit.  A score of one indicates exact likeness whereas zero 704 
indicates the least degree of similarity. 705 
 
 
39 
 
Also, widening the query beyond the system of interest to include a nominal 706 
number of flanking genes, here termed “extended window searching,” afforded additional 707 
contextualization of LMNAST hit results. 708 
Finally, in evaluating certain low homology hits, nonscoring synonyms were 709 
used.  Nonscoring synonyms are elements with equivalent gene annotation but 710 
insufficient homology according to the initial E-value filter.  This is somewhat analogous 711 
to replacement in blastp.   712 
 713 
 714 
 715 
 716 
 717 
 718 
 719 
 720 
 721 
 722 
 723 
 724 
 
 
40 
 
2.4 Results  725 
2.4.1 E. coli K-12 W3110 lac Operon Query 726 
 We began evaluation of LMNAST by searching for the well characterized E. coli 727 
lac operon. Specifically, the E. coli lac genes lacI (BAE76127), lacZ (BAE76126), lacY 728 
(BAE76125), and lacA (BAE76124) (spanning bp 360473 to 366734 of the Genbank 729 
nucleotide record AP009048) were used as a query.  The stringent criteria search yielded 730 
fewer hits than the corresponding weak criteria search (189 vs. 236).  Of the hits derived 731 
from the stringent criteria search, complete and perfectly arranged lac systems were 732 
found in 26 unique E. coli strains and S. enterica arizonae serovar 62:z4,z23 (meH 0.8), 733 
the only Salmonella enterica serovar represented among all lac system hits, in keeping 734 
with its significant divergence from other serovars 74.  A representation of E. coli hits in a 735 
phylogenetic context is available in Figure 2-3a.  The average meH (0 < meH < 1) for 736 
these complete systems was 0.98.  An extended window query with five additional genes 737 
on either side of the original search frame, revealed eight complete systems with a 738 
hitchhiking, proximal cytosine deaminase after losing all other proximal genes.  Only one 739 
system with all four characters was entirely removed from the original query’s proximal 740 
gene set, suggestive of negligible stability for the canonical system outside of a highly 741 
limited phylogenetic domain. 742 
An additional 28 hits were bereft one lac system character (average meH 0.74).  743 
In all but three of these cases that missing gene was lacA.  Of the hits without lacA, ten 744 
had an additional frank structural change to a divergent expression pattern originating 745 
 
 
41 
 
 746 
Figure 2-3. lac operon LMNAST hits overlaid onto phylogenetic distributions of 747 
different scopes.  The larger, bolded leaves represent species that contain lac operon 748 
homologs, whereas the grayed italicized leaves were completely bereft. A  An E. coli 749 
specific phylogenetic tree as adapted from 102, wherein all genes from the core E. coli 750 
were used to construct a consensus tree.  Among the species/strains represented here, lac 751 
system homologs were absent from certain Shigella.  Additionally, BW2952, SE11, IAI1, 752 
HS, E243227A, CFT073, K-12 DH10B, and 11 of 18 O:H serotyped strains contained 753 
truncated systems.  Uniquely, CFT073 retains lacA, while missing lacI in an otherwise 754 
preserved lac system structure. B  The phylogenetic dispersion of the lac system is 755 
mostly limited to Escherichia and proximal species, as seen in the 16s based tree adapted 756 
from 108.  Bolded leaves indicate the presence of the lac system in at least one strain.  757 
 758 
 
 
42 
 
between lacI and lacZ characters (e.g. in E. cloacae), likely increasing system expression 759 
sensitivity to lacI repression in these cases 93.  Surprisingly, in other instances, extended 760 
window searching revealed the only proximal structural change to be a missing lacA 761 
gene.  This lacA degeneracy may be indicative of its relative functional unimportance 762 
compared to the other lac system members 109.   763 
Some of the patterns described above can be inferred from coincidence heat charts 764 
(Figure 2-4).  These matrices represent LMNAST results by the frequency of coincidence 765 
between gene characters within hits.  The shade of an index represents the frequency of 766 
hits where the row gene coincides with the column gene, normalized against the total 767 
number of hits containing the row gene, which itself is denoted by (#).  For example, in 768 
Figure 2-4, the left-most matrix is a representation of a theoretical set of homolog 769 
fragments (AB, BC, CD, ABC, BCD, and ABCD).  This simple set was constructed to 770 
only reflect unbiased homologous recombination presumably resulting only in 771 
chromosomal rearrangements.  In this set, B and C are extant in five inputs, while A and 772 
D are extant in three inputs.  All three inputs containing A also contain B, two also 773 
contain C, and one also contains D.  This is reflected in the shades of the grids in the top 774 
row. 775 
The middle matrix represents the coincidence distribution amongst LMNAST E. 776 
coli lac hits.  As an additional example, matrix element (2,1) is a rust color representing 777 
the 139 hits with a lacI character of the 155 also containing lacZ. Finally, the right-most 778 
matrix is the difference between the left and middle matrices.  This particular analysis 779 
 
 
43 
 
 780 
Figure 2-4. Coincidence heat map for lac operon LMNAST stringent search hits.  Each 781 
shaded index represents the normalized frequency of hits containing the row gene that 782 
also contain the column gene out of the total number of hits containing the row gene, as 783 
denoted by (#).  The matrix on the left is a representation of an unbiased set of evenly 784 
distributed homologs (AB, BC, CD, ABC, BCD, and ABCD).  The middle matrix is the 785 
actual coincidence data.  The matrix on the right is a heat map of the difference between 786 
the two.  LacI is heavily over-represented according to the difference matrix. LacY 787 
occurred less than would be expected according to a random distribution. 788 
 789 
 790 
 791 
 792 
 793 
 794 
 
 
44 
 
suggests, for example, that lacI is relatively over-represented across all hits, and that 795 
nearly all other coincidences are under-represented; surprisingly, this includes 796 
coincidences involving the permease, lacY.  Unlike lacA, lacY is believed necessary for 797 
lactose catabolism, possibly pointing to the use of a lower affinity transporter in such 798 
cases.  Obversely, the over representation of lacI indicates a preference for the regulation 799 
of lactose catabolism.  800 
Of the strong criteria search results, 138 hits contained only two lac gene 801 
homologs (average meH 0.28).  Two gene homologs represent the natural minimum of 802 
individual characters that a homologous system may contain. Such hits represented 803 
truncated systems, repurposed individual members, homoplasic convergence, or outright 804 
false positives.  The majority of these hits fell within clusters of shared Genbank 805 
annotation in 2D similarity plots, which compare meHs (averaged BLAST homologies) 806 
against LMNAST homologies, or, put differently, average amino acid identities against 807 
the system’s broader organizational identity.   Generically then, purely vertical 808 
displacements imply perfect conservation across species through either vertical or, more 809 
likely, recent horizontal gene transfer accompanied by amelioration, while purely 810 
horizontal displacements indicate recent gene loss and/or rearrangement.  For purposes of 811 
downstream analysis, it is interesting to speculate that the kinetics of the remaining genes 812 
are unaffected in cases of purely horizontal displacement.  For systems subject to HGT, 813 
such liberties must necessarily be taken with less confidence. 814 
In the case of the stringent lac search, similarity plots revealed a great deal of 815 
structural variability in the lac operon homologs of E. coli and near E. coli species 816 
(Figure 2-5).  Nonetheless, the canonical lac operon (26) and the paralogous evolved 817 
 
 
45 
 
beta-galactosidase system (43) 110 are clearly the most dominant lac operon-homologs, 818 
perhaps partially reflecting the relative preponderance of fully sequenced E. coli strains.  819 
Addressing the full breadth of two character homologs, 87 contained lacZ and 820 
lacY character types, all of which were adjacent, five of which were misdirected relative 821 
to one another. Numerous truncated systems had high meH but imperfect organizational 822 
similarity.  This cohort was restricted to strains of E. coli and closely related Shigella, 823 
Citrobacter, and Enterobacter species, reflecting a generally confined phylogenetic 824 
breadth among LMNAST lac hits (Figure 2-3b), and reinforcing the idea of limited lac 825 
horizontal gene transfer (HGT) 111.  The remainder of the hits consisted of adjacent 826 
repurposed characters with functional valence around sugar metabolism.   827 
This survey showed that LMNAST E. coli lac operon searches identified 828 
numerous ortholog and paralog instances.  Relative disparities in gene preservation, gene 829 
loss, and structural rearrangements bearing signaling implications were delineated.  830 
While there was a significant degree of conformity to the standard genomic arrangement, 831 
the amount of diversity indicates that attention paid to related, non-canonical signaling 832 
units may be worthwhile. 833 
 834 
 
 
46 
 
 835 
Figure 2-5. 2D similarity plot of lac operon LMNAST stringent search hits overlaid 836 
with attributed annotation.  Each gray dot represents the homology coordinate of a hit.  837 
The size of the dot scales directly with the number of hits at the same coordinate.  The 838 
dashed line is a 1:1 line along which hits have the same degree of homology by both 839 
BLAST and LMNAST measures.  Seemingly vertical displacements may imply 840 
horizontal gene transfer, while horizontal displacements may imply gene loss or 841 
arrangement within the same or proximal species.  Ovals indicate a clustering of similarly 842 
annotated hits. Dashed ovals denote cases where only the majority of hits therein share 843 
 
 
47 
 
the labeled Genbank annotation.  Here, the dominant features are the original structure 844 
and the evolved beta-galactosidase system.  Very little HGT is apparent while gene loss 845 
and rearrangement are ostensibly more common. 846 
 847 
 848 
 849 
 850 
 851 
 852 
 853 
 854 
 855 
 856 
 857 
 858 
 859 
 860 
   861 
 
 
48 
 
2.4.2 E. coli K-12 W3110 Lsr System Query 862 
            Further testing of LMNAST was conducted with weak, stringent, and expanded 863 
window searches of the E. coli Lsr system.  The query Lsr system consists of a kinase 864 
(LsrK: BAA15191), a repressor (LsrR: BAA15192), ABC transporter genes (LsrA: 865 
BAA15200, LsrC: BAA15201, LsrD: BAA15202, and LsrB: BAE76456), and AI-2-P 866 
processing genes (LsrF: BAE76457, LsrG: BAE76458).  Along with AI-2, the Lsr system 867 
consists of multiple overlapping positive and negative feedback loops.  Multimeric LsrR 868 
represses system expression emanating from the intergenic region.  AI-2-P, itself 869 
catabolized by LsrF and LsrG, allosterically relieves that repression.  Thus, both 870 
expression troughs and peaks are tightly regulated.112  For the LMNAST search we used 871 
the Lsr genes spanning bp 1600331 to 1609003 of E. coli K12 substrain W3110 872 
(Genbank nucleotide record AP009048).  The number of hits returned using stringent 873 
criteria totaled 419. 874 
Much like the lac operon, the Lsr system appeared subject to imperfect 875 
conservation.  Certainly, many fully sequenced E. coli bore exact Lsr homologs (meH > 876 
0.95).  Exceptions were the truncated systems found in strains BL21 113, REL606 113, and 877 
E24377A, and the specific excision of Lsr systems from an otherwise preserved gene 878 
order in B2 type E. coli (Figures 2-6 and 2-S1) as revealed through expanded window 879 
searching.  880 
 881 
 882 
 883 
 
 
49 
 
  884 
Figure 2-6.  Lsr system LMNAST hits overlaid onto phylogenetic distributions of 885 
different scopes. A   E. coli Lsr system LMNAST hits overlaid onto an E. coli specific 886 
phylogenetic distribution as developed in and adapted from 102.  The larger, bolded leaves 887 
contain Lsr system homologs, whereas the grayed italicized leaves do not.  Lsr system 888 
loss is evident in B2 strains. B  E. coli Lsr system LMNAST hits overlaid onto an 889 
Enterobacteriales and Pasteurellaceaes phylogenetic distribution adapted from 108.  The 890 
larger, bolded leaves contain Lsr system homologs, whereas the grayed italicized leaves 891 
do not.  Loss and gain events are denoted by – and + respectively based on parsimony.  892 
Compared to the distribution of the lac operon, Lsr is more phylogenetically dispersed, 893 
but also a bit shallower.  894 
 895 
 896 
 
 
50 
 
Unlike the lac operon, numerous Lsr system homologs had perfect LMNAST 897 
homology but markedly reduced meH (Figure 2-8a). This is suggestive of amelioration 898 
following recent HGT events (which may itself be a reflection of a carefully tuned signal 899 
requiring the full complement and correct arrangement of Lsr elements).  Indeed, Lsr 900 
system GC content varied in accordance with the background GC content, ranging from 901 
0.35 to 0.71. Finer scale GC analysis revealed a single consistent and curious feature 902 
across all hits with meH greater than 0.3: a sharply spiking dip in fractional GC content 903 
near the intergenic region (Figure 2-S2).  This dip is suggestive of a conserved DNA 904 
binding domain essential to the signal transduction process, which would also, however, 905 
be a regulatory feature outside the scope of LMNAST searches.   906 
Imperfect LMNAST hits with meH greater than 0.3, deviated from the theoretical 907 
distribution according to a bias toward the conservation of lsrB, F, and G, relative to the 908 
lsrA, C, and D importer genes (Figure 2-7).  This may be attributable to the fact that lsrB, 909 
F, and G likely pass cell signaling information downstream 73,74,97 , whereas loss of Lsr 910 
importer function might be partially redundant to a low affinity rbs pathway 61, the likely 911 
alternate AI-2 import pathway 49,55,112.  912 
In contrast to high meH systems, many systems with low meH (less than 0.3) 913 
were involved in the metabolism of 5 carbon sugars, mainly ribitol and xylose, according 914 
to Genbank annotation (Figure 2-8b).  Since AI-2 itself is mainly comprised of a 5 carbon 915 
ring, such homology is simultaneously intriguing and unsurprising.  More generally 916 
among these low similarity hits, lsrK characters were commonly coincident with Lsr 917 
 
 
51 
 
 918 
Figure 2-7. Annotated 2D similarity plot for Lsr system LMNAST weak search hits. A. 919 
HGT of homologous systems is evident among hits with perfect organizational homology 920 
but diminished mean element homology.  A great number of hits have low similarity 921 
 
 
52 
 
along both axes.  As seen in B., these hits are mostly involved in the metabolism of 5 922 
carbon carbohydrates according to their annotation.  This is likely reflective of the fact 923 
that AI-2 is of similar structure to 5 carbon sugars.  924 
 925 
 926 
 927 
 928 
 929 
 930 
 931 
 932 
 933 
 934 
 935 
 936 
 937 
 938 
 
 
53 
 
importer characters (lsrA, C, D, and B), indicative of the functional link between these 939 
characters.  These various features were laid more strongly in relief when measured 940 
against the proximal genetic background in an extended window search.     941 
While a representation of hit variability preserving structural information can be 942 
had from trackback plots (Figure 2-S1), additional salient results from stringent Lsr 943 
extended window searching could be evinced from the more summary coincidence heat 944 
maps (Figure 2-9).  The matrices indicate that lsrK and lsrA genes were strongly 945 
preserved among extended window hits.  Also, if either lsrF or lsrG were present, the 946 
remaining Lsr genes were likely present.  The complete system rescission mentioned 947 
before was hinted at, especially in rows 3 and 4, corresponding to the toxin/antitoxin 948 
hipAB system.  Intra-species variation of structural homology increased greatly when 949 
using stringent rather than weak criteria (data not shown), mainly as a result of gene loss 950 
to pseudo gene conversion, mostly among transporter genes—a bias most easily 951 
explained as a matter of pure probability since there are more transporter genes than any 952 
other type, and a fact whose functional significance is blunted by an alternative AI-2 953 
import pathway.    954 
These initial E. coli searches motivated other orthologous Lsr system queries. Full 955 
results for E. coli, S. Typhimurium, and B. cereus searches are available in Table 2-S2. 956 
These additional searches identified other possible Lsr system homologs, HGT partners, 957 
and non-canonical system-associated gene candidates.  In Table 2-S2, we delineate 958 
operon directionality and gene homology.  It is interesting to note that system variants 959 
exist among noted human pathogens: Yersinia pestis, Bacillus anthracis, and 960 
 
 
54 
 
 961 
 962 
Figure 2-8. Coincidence matrix for E. coli Lsr system LMNAST stringent search hits. 963 
This coincidence matrix depicts the subset of hits with a mean element homology > 0.3 964 
and also containing 4-6 gene characters.  This subset was chosen for its intermediate 965 
degree of homology to the query Lsr system.  LsrF and lsrG characters were found to be 966 
overrepresented among these hits coincident to hits also containing lsrR and lsrK 967 
characters. 968 
 969 
 970 
 971 
 972 
 973 
 974 
 
 
55 
 
 975 
Figure 2-9.  Coincidence matrix for E. coli Lsr system LMNAST extended window 976 
stringent search hits. Letters represent the respective lsr genes.  1-5 represent the five 977 
genes preceding the Lsr system: ydeT, yneL, hipA, hipB, and ydeK-lipoprotein.  14-17 978 
represent the four genes after the Lsr system: tam (transaconitate methyltransferase), 979 
yneE, uxaB, and a predicted diguanylate cyclase.  The figure suggests a) strong 980 
conservation of the association between Lsr system genes relative to its neighbors, b) hits 981 
in which the Lsr system has been excised entirely from its gene neighbors, and c) a weak 982 
coincidence between yneH-glutaminase characters and the Lsr system.  The matrix also 983 
indicates that the overall prevalence of lsrF and lsrG characters is lower than other 984 
canonical Lsr characters, although the presence of lsrF and lsrG characters is a good 985 
predictor of the presence of other Lsr genes. 986 
 987 
 988 
 
 
56 
 
Haemophilus influenzae.  In some instances, lsrRK are either absent (e.g. E. coli BL21) 989 
or are associated with altered intergenic regions suggesting altered regulatory control 990 
(e.g. Yersinia pestis Antiqua). In other cases transporter genes are distributed with altered 991 
bias due to position in the bidirectional operons (e.g. Yersinia pseudotuberculosis 992 
PB1/+).  In some cases there is no LsrFG component (e.g. Shigella flexerni 2002017). 993 
LsrF and LsrG are AI-2-P processing enzymes that lower the cytoplasmic AI-2-P level, 994 
thereby contributing to the repression of AI-2 induced genes.  995 
Given even only this modest degree of dispersion, it is nonetheless reasonable to 996 
suggest that the Lsr autoinduction system is, in fact, extant among scores of bacterial 997 
species and that because the organization of genes within the regulatory architecture is 998 
varied, the downstream phenotypic behaviors aligned with the AI-2 regulated QS genes is 999 
likewise variable. Thus, our results are in line with a general hypothesis that the AI-2 1000 
quorum sensing system is broadly distributed and that the specific needs of the bacteria in 1001 
a given niche are met by disparate operon arrangements. 1002 
     The overall phylogenetic distribution of the Lsr system mirrors that as 1003 
developed by Pereira, et al. in the cluster they denote as Group I.13  Here, however, 1004 
details were fleshed out with different emphases.  The Lsr LMNAST search captured the 1005 
diversity of pseudo gene conversion, structural rearrangement, and additional hitchhiking 1006 
genes associated with the Lsr system that exist in the present nr/nt database.  Moreover, 1007 
inferences regarding regulatory Lsr system signals could be made that might also map to 1008 
phylogenies or possibly, with much more effort, related ecological niches.  1009 
 1010 
 
 
57 
 
2.4.3 Analysis of Lsr System Search Results 1011 
Results from the various LMNAST searches were reconciled by taking the 1012 
highest scoring hit among overlaps within each nucleotide record.  Lsr system homologs 1013 
clustered mainly in gammaproteobacteria with the greatest density being among E. coli 1014 
strains.  Diffusely manifesting in more distantly related bacterial species, the Lsr system 1015 
appears to have been subject to several HGT events.  That is, the Lsr system is absent in 1016 
numerous Enterobacteriaceae species, while HGT gain events happened at the root of the 1017 
Bacillus cereus group, to R. sphaeroides and R. capsulatus separately, to Sinorhizobium 1018 
meliloti, and to Spirochaeta smaragdinae (Figure 2-10).  Curiously, while these bacteria 1019 
occupy distinct ecological niches, they are all common to soil or water environments.   1020 
Multiple extended window searches indicated that S. enterica was the most 1021 
proximal cluster for every Lsr system HGT candidate.  The sharing of a novel Lsr 1022 
system-associated “mannose-6-phosphate isomerase” (NP_460428) between Bacillus 1023 
cereus group members, S. smaragdine, and S. enterica, further strengthened the 1024 
suggestion of HGT partnership.  The gene annotated as “mannose-6-phosphate 1025 
isomerase” or “sugar phosphate isomerase,” has recently been shown to be part of the 1026 
LsrR regulon in Salmonella.114   Although not part of the E. coli regulon, it was also 1027 
associated with S. smaragdinae and B. cereus group orthologs.  In keeping with a 1028 
possible AI-2-P processing role, it was consistently adjacent to lsrK. 1029 
Among gammaproteobacteria, parsimony suggests that two gain events of the Lsr 1030 
system occurred: one deeply rooted in enterobacteriales and one in a pasteurellaceaes 1031 
ancestor.  In the enterobacteriales branch, besides Escherichia, Shigella, and Salmonella, 1032 
 1033 
 
 
58 
 
 1034 
Figure 2-10.  Phylogenetic distribution of Lsr at different phylogenetic scales using 1035 
reconciled LMNAST results. E. coli Lsr system LMNAST hits overlaid onto a bacterial 1036 
phylogenetic distribution as developed in and adapted from 52.  Each leaf bears a 1037 
representative member from a larger unseen collapsed branch.  The larger, bolded leaves 1038 
contain Lsr system homologs within the collapsed branch, whereas the grayed italicized 1039 
leaves do not.  Parenthesized numbers indicate the number of strains and species with Lsr 1040 
system LMNAST hits contained within the collapsed branch.   1041 
 1042 
 1043 
 1044 
 1045 
 
 
59 
 
Lsr organizational homologs were found in Enterobacter, Photorhabdus, and 1046 
Xenorhadbus species, although most of these instances lacked importer genes (lsrACDB).   1047 
While it is thought that regulatory proteins conserved across such long 1048 
phylogenetic distances often regulate different targets 115, the regulation of community-1049 
related functions by different manifestations of the Lsr system (such as biofilm 1050 
maturation checkpoints in E. coli 60 and possible biofilm dispersion in B. cereus 82) 1051 
suggests a convergent tendency to leverage a quorum/environment sensing capacity 1052 
inherent to the Lsr system.  Indirect influence over a broader regulon may be abetted by 1053 
the involvement of AI-2, the Lsr system substrate, in metabolic pathways 51.    1054 
2.4.3.1 Putative Lsr system in Rhizobiales 1055 
A grouping of Rhizobiales common to plant symbioses had a conserved set of 1056 
adjacent, low homology elements and nonscoring elements that consisted of 1057 
unidirectionally expressed genes annotated as: ABC-type transporters (ribose, putative, 1058 
putative), a DeoR family repressor (LsrR synonym), aldo-keto reductase (non-scoring), 1059 
glycerol-3-phosphate dehydrogenase (nonscoring LsrF synonym), fructose-biphosphate 1060 
aldolase (LsrF synonym), and xylulose kinase (LsrK synonym).  Somewhat inescapably, 1061 
synonymy is at least partly a function of precedence resulting in attribution bias.  The 1062 
indeterminacy of signaling similarity between these homologs and better characterized 1063 
Lsr systems suggests a need for further research.  1064 
 1065 
 1066 
 
 
60 
 
2.5 Concluding Remarks 1067 
LMNAST is a program that evaluates similarity or homology on the level of gene 1068 
organization, conducting a search for patterns and prevalence constrained by a BLAST E-1069 
value filter.  Program results overlaid onto phylogenetic data allow visual inspection of 1070 
phylogenetic density and dispersion.  2D homology plots display system variability 1071 
among LMNAST orthologs, and when overlaid with genera/species clustering, reveal the 1072 
degree of system conservation within and across genera/species when organizational 1073 
homology decreases and element homology is constant.  Clustering also enables the 1074 
identification of conserved system homologs.  Organizational information is lost when 1075 
using coincidence heat charts, but suggestions of the underlying structural variability 1076 
remain nonetheless.  This is particularly true for coincidence representations of extended 1077 
window searches.  For these searches, contextual associations with non-canonical genes 1078 
also may emerge.  Trackback plots illustrate both variety and structural information, 1079 
albeit in a less dense format.  These representations are especially useful in combination.  1080 
It should be noted that the results are nearly entirely comprised of excerpts from fully 1081 
sequenced genomes.  Results are also biased by BLAST input, as characters with more 1082 
element homologs (e.g. lsrA) appear more frequently in hits. 1083 
Generically, LMNAST identified query homologs with a variety of deletions, 1084 
insertions, misordering, and misdirections.  While nearly any source of mutagenesis may 1085 
result in a frank mutation, affecting a system’s organizational homology, homologous 1086 
recombination, insertion sequences, transposable elements, and combinations thereof are 1087 
likely to be of particular consequence for LMNAST searches.  Deletions may be a result 1088 
of pseudo gene conversion, of chromosomal rearrangements, or part and parcel of an 1089 
 
 
61 
 
insertion event—if the insertion results in a gap sufficiently large as to disconnect hit 1090 
elements from one another.  In the case of such insertions, sufficiently weak criteria may 1091 
be of use, with the caveat that decreased stringency increases the number of false 1092 
positives.  From a signaling perspective, depending on the impacted elements and the 1093 
nature of the inserted sequence, gap presence could result in system discoordination; and 1094 
the longer the gap the more probable and severe the discoordination, most likely to the 1095 
detriment of system function. 1096 
 As for the specific test queries examined herein, while the lac operon is well 1097 
characterized in its canonical form, there nonetheless exists a great deal of frank variation 1098 
from the textbook case.  Of particular interest are homologous instances where structural 1099 
rearrangement could influence self-regulation of component expression.  Also of note are 1100 
its multiple signaling component deletions.  Such abbreviated modules are frequently 1101 
repurposed in a related context.  Complete lac operons were found among nearly all E. 1102 
coli strains.  Incomplete lac operons were found to be distributed only among closely 1103 
related Enterobacteriaceae species comprised almost entirely of Escherichia, 1104 
Citrobacters, Enterobacters, and Serratias as expected based on limited lac operon HGT 1105 
111.  This difference between the rates of decay for the two homology signals over 1106 
phylogenetic space may be suggestive of distinct selection pressures guiding the two 1107 
systems.  Also identified through LMNAST were conserved, E. coli-specific evolved 1108 
beta-galactosidase systems 110, demonstrating a capacity to find closely related systems.   1109 
On par, Lsr system hit structural similarity was less well correlated with meH than 1110 
lac operon results, a phenomenon presumably associated with apparent Lsr system HGT.  1111 
The Lsr system was phylogenetically dispersed more widely than the lac operon, even 1112 
 
 
62 
 
while its distribution remained densest among gammaproteobacteria.  Much like the lac 1113 
operon, Lsr system structure was subject to significant variability.  lsrK and lsrR 1114 
characters were common to many hits.  lsrF and lsrG were the least common; the 1115 
inclusion of both elements nearly always coincided with the presence of all other Lsr 1116 
characters as well.  Lsr-contextually associated genes and novel putative Lsr systems 1117 
were also elucidated.   1118 
 The dispersion of Lsr to bacteria as far afield as the S. smaragdinae, the first 1119 
Spirochaeta to be fully sequenced 116, is intriguing.  It suggests that while the depth of 1120 
Lsr dispersion may not be significant, that its exposed breadth will expand incrementally 1121 
at a rate proportional to microbial genome sequencing.  While the direct regulon of such 1122 
HGT systems is expected to be limited 69,115, the proximity of the substrate to key 1123 
metabolic pathways may allow the Lsr system to confer contextual phenotypic 1124 
advantages by impacting downstream pathways with its capacity to recompartmentalize a 1125 
metabolic intermediate.  Moreover, the known regulatory requirements for functional 1126 
integration of the Lsr system are minimal, consisting entirely of interaction with cAMP-1127 
CRP complex, which is deeply rooted in eubacteria.  Gene organization differences 1128 
between dispersed Lsr homologs, may indicate distinct signaling outcomes, in turn 1129 
suggesting the appropriation of the Lsr system’s inherent quorum capacity to drive 1130 
distinct phenotypes suited to a given bacteria’s needs within its particular niche.          1131 
 Unlike the results for the lac operon, Lsr system results returned a large number 1132 
of other-annotated, low homology systems.  This speaks to both the inherent difficulty of 1133 
extrapolating based on homology and the utility of the additional, complementary 1134 
homology measure yielded by LMNAST searching. Overall, given the complexity of the 1135 
 
 
63 
 
results, numerous aspects may be of interest.  Some graphical tools of a complementary 1136 
nature (e.g. 2D similarity plots and coincidence heat maps) have been used here for 1137 
distillation and closer inspection.  The extant variation of the queried modular systems, as 1138 
captured by frank changes in gene organization, was revealed. 1139 
 1140 
 1141 
 1142 
 1143 
 1144 
 1145 
 1146 
 1147 
 1148 
 1149 
 1150 
 1151 
 1152 
 1153 
 
 
64 
 
2.6 Supplemental Materials 1154 
 1155 
Figure 2-S1. Trackback plots for Lsr system LMNAST extended window stringent 1156 
search hits. These diagrams describe the variety of LMNAST hits in greater detail.  A 1157 
straight diagonal line indicates complete agreement with the query.  Rearrangements are 1158 
represented by discontinuities.  Relative redirection is indicated by a flipping of the 1159 
diagonal orientation.  Deletion is indicated by horizontal dashed gaps.  Insertion is 1160 
indicated by vertical gaps.  The legend in the upper right hand corner indicates which 1161 
numbers correspond to which genes.  Trackback plots are organized into categories: A, B, 1162 
C, D, and E according to the following categories: I. Prototype Lsr systems, II. Modified 1163 
Lsr systems with pre and post-Lsr adjacent characters, III. Modified Lsr systems with 1164 
post-Lsr adjacent characters, IV. Modified systems without Lsr adjacent characters, and 1165 
V. Highly modified Lsr systems.   For exact subgroup membership see Table S1. 1166 
 1167 
 1168 
 
 
65 
 
 1169 
Table 2-S1  1170 
Group Subgr
oup 
Species-Strain 
I (pre-Lsr-post) A Escherichia coli ATCC 8739 
  Escherichia coli DH1 
 B Escherichia coli O103:H2 str. 12009 
  Escherichia coli O157:H7 str. EC4115 
  Escherichia coli SE11 
 C Escherichia coli O111:H- str. 11128 
  Escherichia coli O157:H7 str. TW14359 
  Escherichia coli O26:H11 str. 11368 
  Escherichia coli 55989 chromosome 
 D Escherichia coli IAI1 
 E Escherichia coli BW2952 
  Escherichia coli str. K12 substr. DH10B 
 F Escherichia coli O157:H7 str. Sakai 
  Escherichia coli O157:H7 EDL933 
 G Escherichia coli SMS-3-5 
II (pre-modified Lsr-
post) 
A Escherichia coli E24377A 
 B Escherichia coli B str. REL606 
  Escherichia coli BL21(DE3) 
 C Escherichia coli IAI39 
 D Shigella dysenteriae Sd197 
 E Escherichia coli 536 
  Escherichia coli APEC O1 
  Escherichia coli CFT073 
  Escherichia coli IHE3034 
  Escherichia coli S88 chromosome 
  Escherichia coli SE15  
  Escherichia coli UTI89 
 F Escherichia coli ED1a chromosome 
 G Escherichia coli 042 
  Shigella boydii CDC 3083-94 
III (modified Lsr-post) A Escherichia fergusonii ATCC 35469 
 B Shigella flexneri 2a str. 2457T 
  Shigella flexneri 2a str. 301 
 C Sinorhizobium meliloti 1021 plasmid pSymB 
  Rhodobacter sphaeroides  KD131 chromosome 2 
IV (modified Lsr) A Aggregatibacter aphrophilus NJ8700, complete genome 
  Enterobacter sp. 638 
  Haemophilus influenzae PittEE 
  Haemophilus somnus 129PT 
  Haemophilus somnus 2336 
 
 
66 
 
  Klebsiella pneumoniae 342 
  Klebsiella pneumoniae subsp. pneumoniae MGH 78578 
  Pasteurella multocida subsp. multocida str. Pm70 
  Salmonella enterica subsp. enterica serovar Agona str. SL483 
  Salmonella enterica subsp. enterica serovar Choleraesuis str. SC-B67 
  Salmonella enterica subsp. enterica serovar Enteritidis str. P125109 
  Salmonella enterica subsp. enterica serovar Gallinarum str. 287/91  
  Salmonella enterica subsp. enterica serovar Heidelberg str. SL476 
  Salmonella enterica subsp. enterica serovar Paratyphi A str. AKU_12601 
  Salmonella enterica subsp. enterica serovar Paratyphi A str. ATCC 9150 
  Salmonella enterica subsp. enterica serovar Paratyphi B str. SPB7 
  Salmonella enterica subsp. enterica serovar Paratyphi C str. RKS 4594 
  Salmonella enterica subsp. enterica serovar Schwarzengrund str. CVM19633 
  Salmonella enterica subsp. enterica serovar Typhi Ty2 
  Salmonella enterica subsp. enterica serovar Typhimurium str. D23580 
  Salmonella enterica subsp. enterica serovar Typhimurium str. LT2 
  Yersinia enterocolitica subsp. enterocolitica 8081  
  Yersinia pestis Angola 
  Yersinia pestis CO92  
  Yersinia pseudotuberculosis IP 31758 
  Yersinia pseudotuberculosis IP32953  
  Yersinia pseudotuberculosis PB1/+ 
  Yersinia pseudotuberculosis YPIII 
 B Klebsiella pneumoniae NTUH-K2044 DNA 
  Salmonella enterica subsp. enterica serovar Dublin str. CT_02021853  
  Salmonella enterica subsp. enterica serovar Newport str. SL254 
  Salmonella enterica subsp. enterica serovar Typhimurium str. 14082S 
  Yersinia pestis biovar Microtus str. 91001 
  Yersinia pestis Pestoides F 
 C Yersinia pestis Antiqua 
  Yersinia pestis Nepal516 
 D Yersinia pestis KIM 
 E Rhodobacter sphaeroides 2.4.1 chromosome 2 
  Rhodobacter sphaeroides ATCC 17029 chromosome 2 
 F Bacillus anthracis str. ‘Ames Ancestor’ 
  Bacillus anthracis str. Ames 
 G Bacillus anthracis str. A0248 
 H Bacillus anthracis str. CDC 684 
  Bacillus anthracis str. Sterne 
  Bacillus cereus ATCC 10987 
  Bacillus cereus ATCC 14579 
  Bacillus cereus B4264 
  Bacillus cereus E33L 
  Bacillus cereus G9842 
 
 
67 
 
  Bacillus cereus Q1 
  Bacillus thuringiensis serovar konkukian str. 97-27 
  Bacillus thuringiensis str. Al Hakam 
  Bacillus weihenstephanensis KBAB4 
  Shigella sonnei Ss046 
 I Bacillus cereus AH187 
  Bacillus cereus AH820 
  Bacillus cereus 03BB102 
V (Non-continous) A Escherichia coli ED1a chromosome 
  Serratia proteamaculans 568 
  Shewanella halifaxensis HAW-EB4 
  Yersinia pestis Angola 
  Yersinia pestis Antiqua 
  Yersinia pestis biovar Microtus str. 91001 
  Yersinia pestis CO92 
  Yersinia pestis KIM 
  Yersinia pestis Nepal516 
  Yersinia pestis Pestoides F 
  Yersinia pseudotuberculosis IP31758 
  Yersinia pseudotuberculosis IP32953 
  Yersinia pseudotuberculosis PB1/+ 
  Yersinia pseudotuberculosis YPIII 
 B Agrobacterium radiobacter K84 chromosome 2 
  Ochrobactrum anthropi ATCC 49188 chromosome 2 
  Rhizobium leguminosarum bv. trifolii WSM1325 plasmid pR132501 
  Rhizobium leguminosarum bv. viciae plasmid pRL12 
  Rhizobium sp. NGR234 plasmid pNGR234b 
  Silicibacter sp. TM1040 
  Sinorhizobium medicae WSM419 plasmid pSMED01 
  Sinorhizobium meliloti 1021 plasmid pSymB 
 C Burkholderia cenocepacia AU 1054 chromosome 1 
  Burkholderia cenocepacia HI2424 chromosome 1 
  Burkholderia cenocepacia J2315 chromosome 1 
  Burkholderia cenocepacia MC0-3 chromosome 1 
  Burkholderia multivorans ATCC 17616 DNA 
  Burkholderia sp. 383 chromosome 1 
  Burkholderia xenovorans LB400 chromosome 1 
 1171 
 1172 
 1173 
 
 
68 
 
 1174 
Figure 2-S2.  GC content demonstrates consistent spiking dip at intergenic region. GC 1175 
content graphs for Bacillus cereus ATCC10967, Yersinia Pestis Nepal516, Rhodobacter 1176 
sphaeroides ATCC 17029, and Yersinia pestis CO92.  Graphs are labeled with lsr system 1177 
beginning (end of lsrK), lsrR gene intersection with the intergenic region (labeled lsrR 1178 
end), and lsr system ending (end of lsrG).   Arrow direction indicates the direction of 1179 
lsrACDBFG expression.  Proximal to the intergenic region is a conserved dip in GC 1180 
content. 1181 
 1182 
 1183 
 1184 
 
 
69 
 
1185 
 
 
70 
 
1186 
 
 
71 
 
1187 
 
 
72 
 
 1188 
 
 
73 
 
Table 2-S2.  Results from three separate LMNAST searches for Lsr system homologs.  1189 
The E. coli K-12 W3110 search shown was completed using weak criteria (a lower gap 1190 
extension penalty and less rigid adherence to annotation), whereas the B. Cereus ATCC 1191 
10987 and S. enterica Typhimurium LT2 searches used stringent criteria (higher penalties 1192 
for deviation from the query pattern and adherence to supplied annotation).  “Join” 1193 
indicates a difficulty in handling irregular annotation where the first gene annotated in a 1194 
record spans the end and the   beginning of the record.  Arrow direction indicates the 1195 
direction of transcription along the genome.  Color is a stand-in for character type, and 1196 
the degree of shading indicates degree of element homology, with the darkest shade 1197 
representing 100% element homology. 1198 
 1199 
 1200 
 1201 
 1202 
 1203 
 1204 
 1205 
 1206 
 1207 
 
 
74 
 
Chapter 3: Quorum desynchronization leads to bimodality and 1208 
patterned behaviors in microbial consortia  1209 
3.1 Abstract 1210 
Quorum sensing (QS) is a type of signaling used to coordinate behavior in bacterial 1211 
consortia via secreted autoinducers.  Here we focus on Lsr (LuxS regulated) based QS 1212 
signaling that activates only a fraction of a cell population through the ‘universal’ QS 1213 
signal autoinducer-2 (AI-2).  Our modeling indicates that bimodality arises from 1214 
desynchronized, QS-activated importation of AI-2, and is sensitive to cell-cell distance 1215 
and cell density.  According to our agent-based models, through this mechanism, we 1216 
found Lsr QS drives spatial organization of cell signaling.  That is, Lsr induced AI-2 1217 
internalization results in emergent “cluster-then-disperse” behavior when acting in 1218 
concert with AI-2 chemoattraction, and speckled activation in surface attached bacterial 1219 
colonies.  This is contrasted against rapid and population-wide expansion of QS activity 1220 
emerging from LuxIR based QS in a growing bacterial colony.  Further, Lsr signaling 1221 
also results in differential QS activation between Lsr/LuxS types in in silico mixed 1222 
cultures.  One particular finding was that signal-negative cells were also found to be 1223 
effectively signal-blind in Lsr QS.  More broadly, interstrain communication modulated 1224 
QS activation patterns were elucidated, helping to frame the complex Lsr QS dynamics 1225 
arising in sociomicrobiological settings.   1226 
 1227 
  1228 
 
 
75 
 
3.2 Introduction 1229 
Quorum sensing (QS) is a bacterial response to self-secreted signaling molecules 1230 
generically known as “autoinducers”. While QS has been observed among individual 1231 
bacteria in experimentally manipulated settings 4–6, QS is often described as informing 1232 
the coordination of processes (such as virulence factor production and biofilm formation) 1233 
that are metabolically burdensome and ineffectual for individual cells, yet beneficial at 1234 
multicellular or population scales 80.  Population scale coordination typically arises from 1235 
the decompartmentalization of the autoinducer that acts as a shared pool of extracellular 1236 
signal available to coordinate individual cells that are sufficiently proximal.  This 1237 
regulatory strategy can reduce “noisy” inputs and other heterogeneity by focusing 1238 
phenotypic outcomes and organizing population activity 117,118.  Coordination varies in 1239 
degree and fashion, and depends upon the mechanistic underpinnings of the QS system 1240 
involved 94,119,120.  Importantly, a given “quorum’s” collective response to heterogeneity 1241 
is likely to reflect the topology of the cognate QS signaling “module” 121,122. Among the 1242 
better studied QS systems are LuxIR (background Figure 3-1), which is widely utilized 1243 
in synthetic biology because of its genetic simplicity, and Lsr (foreground Figure 3-1), 1244 
which is widely distributed among gammproteobacteria 14.  LuxIR is induced by 1245 
autoinducer-1 (AI-1) (an umbrella term encompassing a variety of species specific acyl-1246 
homoserine lactones (AHLs)) whereas Lsr is induced by autoinducer-2 (AI-2) (which is 1247 
produced by luxS in nearly half of eubacteria 46).  LuxIR signaling is comprised of an 1248 
intracellular positive feedback loop 21 that is tied to positive intercellular feedback at the 1249 
population scale.  This is contrasted against Lsr signaling the induction of which features  1250 
 
 
76 
 
 1251 
Figure 3-1.  LuxIR and Lsr QS activity intracellularly and intercellularly.  The LuxIR 1252 
system (background) involves a simple intercellular feedforward loop that entangles AHL 1253 
synthase production with LuxR regulatory production, which itself promotes the 1254 
expression of both LuxI and LuxR when AHL concentration is sufficiently high.  This 1255 
loop is connected to intercellular feedforward activity due to passive diffusion of AHL 1256 
across the cell membrane.  This is in contrast to the Lsr system (foreground), where the 1257 
feedforward loop of LsrACDB increasing AI-2 intracellularly, LsrK phosphorylating AI-1258 
2, and AI2-P derepressing LsrR which in turn stimulates LsrACDB and LsrK expression 1259 
does not drive extracellular AI-2 concentrations higher, but instead depletes them, 1260 
creating negative intercellular feedback. 1261 
 1262 
 1263 
 1264 
 1265 
 1266 
 1267 
 1268 
 
 
77 
 
the rapid depletion of extracellular AI-2 55, intertwining intracellular positive feedback 1269 
with extracellular negative feedback.  Lsr activity is subject to further negative feedback 1270 
by the enzymatic processing of phosphorylated AI-2 through LsrF and LsrG 73–75. Among 1271 
the small molecules produced by LsrF and LsrG under aerobic conditions are glycerol-3-1272 
phosphate 123 and phosphoglycolic acid 75, respectively, which serve as metabolic 1273 
intermediates for AI-2 reassimilation into primary metabolism.    1274 
As described herein, we explored and contrasted QS operation in a variety of 1275 
contexts through the use of finite element-agent based models.  In part, our modeling 1276 
efforts were motivated by indications that bimodal Lsr expression may arise in pure 1277 
cultures, where cell populations developed into QS-activated and non-QS activated 1278 
fractions 94.  While it has been suggested that a transient bimodality may arise as a 1279 
consequence of intracellular signaling topology 95, we hypothesized that a more 1280 
permanent population bifurcation might develop from hyperlocal competition for AI-2 1281 
between cells.  This possibility was probed here by employing a population of ODEs 1282 
wherein nongenetic heterogeneity was explicitly incorporated.  The ramifications of such 1283 
a hyperlocal competition were then considered in the contexts of cell motility and mixed 1284 
consortia.   1285 
LuxIR and Lsr QS were contrasted and found to generate different spatial patterns 1286 
of cell signaling in silico.  We recapitulated an earlier experimental study wherein the 1287 
sudden expansion of LuxIR QS activation from a cell colony center (or “Supernova”) 1288 
was observed 33. This stood in clear contrast to speckled QS activation found in Lsr-1289 
mediated cell colonies.  Further, the confluence of Lsr QS and AI-2 chemoattraction in E. 1290 
coli 97 was examined, the combination of which produced a “cluster-then-disperse” 1291 
 
 
78 
 
behavior that is consistent with concepts of sociomicrobiology 124 wherein “travel” to 1292 
better locales may be one of the currencies of public goods 125. 1293 
Additionally, the consequences of Lsr QS circuitry were examined in silico in 1294 
mixed cultures of different Lsr/LuxS cell types.  luxS and lsrFG mutants were chosen 1295 
for their prevalence and because they represent slower and faster QS activating 1296 
populations, respectively.  Coupling these mutants with their wildtype counterparts 1297 
illustrated how Lsr dynamics manifest in the preferential activation of one Lsr/LuxS cell 1298 
population over another.  Just as activation patterns arising from Lsr signaling were distinct 1299 
from those associated with LuxIR activity, interpopulation transactions between 1300 
Lsr/LuxS subgroups revealed different incentives for social cheating than have previously 1301 
been associated with generic QS signaling. 1302 
 1303 
 1304 
 1305 
 1306 
 1307 
 1308 
 1309 
 1310 
 1311 
 
 
79 
 
Methods 3.3 1312 
3.3.1 Modeled Cell Behaviors 1313 
During a given time step, based on cell growth characteristics and local 1314 
concentration of substrate, cells divided, moved, and responded to autoinducer 1315 
concentrations in their direct extracellular environment through an inflection of their QS 1316 
dynamics as modeled by ODE trajectories.  AI-2 was also exchanged between grids in an 1317 
approximation of diffusion.  Full model details and in particular those common to all 1318 
simulations are provided in SI Methods. 1319 
3.3.1.1 Chemotactic Swimming.  To interrogate QS activation patterning under different 1320 
regimes of motion, the mode of cell motility was varied.  To begin either swimming or 1321 
colony growth simulations, cells or colony centers were placed randomly in the simulated 1322 
environment. In simulations of chemotactic behavior, cells adjusted direction every 0.1 1323 
seconds.  If the average concentration of AI-2 experienced by the cell over the previous 1324 
0.1 seconds was less than a minimum sensitivity or less than the average concentration of 1325 
AI-2 from the previous 15 second period (reflecting the time required for accommodation 1326 
126), the cell moved in a random direction and at a speed produced from a distribution 1327 
with an average of 20 m/sec, in accordance with results from cell tracking studies 1328 
[unpublished].  If the short term average concentration was higher than the long term 1329 
average, the cell continued moving in the same direction as it had the previous time step, 1330 
with slight error.   1331 
3.3.1.2 Colony Growth.  Expanding colony growth was loosely mimicked by cells 1332 
moving according to space filling considerations.  In the interior of the colony, daughter 1333 
 
 
80 
 
cells pushed neighbors outward based on the shortest distance to the colony boundary 1334 
available.  At colony boundaries, cells moved into randomly chosen free space.  All 1335 
motions were constrained by simulation boundaries. 1336 
3.3.1.3 LuxIR QS.  To simulate LuxIR behavior, cells produced AHL at high, baseline, or 1337 
intermediate rates (10 M/min to 1 M/min) corresponding to active, inactive, or linearly 1338 
activating QS, respectively.  The rate of AHL production in the QS inactive state 1339 
(baseline AHL production) was varied across the population with the same coefficient of 1340 
variation that was applied to the parameter basal in Lsr simulations ( = 0.0225, log-1341 
normal distributions).  Diffusion across the membrane was modeled at a conductivity of 1342 
0.6 between the intracellular and extracellular spaces driven by concentration difference 1343 
127.  LuxIR expression was activated over a thirty minute period in a linear progression 1344 
once an intracellular AHL threshold of 2.9 M was exceeded, coarsely capturing inherent 1345 
system cooperativity and the time lag associated with transcription and translation.   1346 
3.3.1.4 Lsr QS.  The ODEs governing Lsr activity, associated parameters, and the 1347 
rationale behind the design choices are described in SI.  Lsr related ODEs relied heavily 1348 
upon Michaelis-Menten kinetics and cooperative Hill terms.  Parameters were chosen to 1349 
be biologically realistic and to conform to the behavior of the system as gleaned from 1350 
previously conducted experiments (see SI) 55,56,94,112.   1351 
3.3.2 Simulation Variants: Gene Deletions and Mixed Populations 1352 
 As metabolic burdens 128 were not considered, gene deletions were modeled by 1353 
setting the appropriate parameter to zero, reflecting ablated activity.  The gene deletions 1354 
examined consisted of luxS and lsrFG. These were modeled by setting parameter 1355 
 
 
81 
 
values to zero such that Ksynth = 0 and kCat = 0, respectively.  These variants differed in 1356 
their sensitivity to external AI-2.  Simulations placing variant populations and wildtype 1357 
cells in the same environment were run.  The two populations were equivalently 1358 
populated to initiate the simulation and divided according to the same heuristics.  1359 
Chemotaxing motility was stripped from these simulations in order to focus on the effect 1360 
from deletion of Lsr and auxiliary proteins.   1361 
In each of these cases, regardless of the cell type simulated, all bacteria and 1362 
growing colonies were accounted for at every time step and in every grid point.  Owing 1363 
to the complex nature of the system, a finite difference method was used.  Care was taken 1364 
to minimize error propagation (studies of grid size, time step, growth parameters, etc.). 1365 
 1366 
 1367 
 1368 
 1369 
 1370 
 1371 
 1372 
 1373 
 1374 
 
 
82 
 
3.4 Results 1375 
3.4.1 Lsr Autoinduction in Pure Cultures.  We first modeled desynchronized Lsr QS 1376 
activation. Specifically, the value of the model parameter basal (the rate of AI-2 transport 1377 
through the low-flux AI-2 importer), was distributed among modeled cells in a log 1378 
normal fashion.  All other parameters were held constant so that the relative background 1379 
AI-2 uptake rate and corresponding speed to Lsr QS activation were the only 1380 
distinguishing features among the entire cell population.  We observed that only a 1381 
fraction of the total cell population became QS activated, generating a bimodal pattern.   1382 
The exact on/off balance was influenced by the degree of variance in the basal 1383 
distribution (Figure 3-2A).  As discussed extensively in the SI, this bimodality ostensibly 1384 
resulted from QS active cells depriving inactive cells of AI-2 with which to induce.  1385 
Consider an ith cell with basali << basalMean that would undergo QS activation if basali = 1386 
basalj for all i,j, instead remaining inactive. When the coefficient of variance was zero, 1387 
the entire population became activated almost simultaneously.  Increasing the variance of 1388 
the value for basal increased the population fraction remaining inactive, attributable to 1389 
cells with higher values of basal from the leading edge of the distribution activating 1390 
earlier and earlier as the variance of the distribution increased.  The earlier the activation 1391 
of these cells relative to the mean, the faster extracellular AI-2 was drawn down, thus 1392 
preventing the activation of cells with lower basal values, which themselves required 1393 
more and more time to accumulate sufficient AI-2P.  This conceptual model serves as a 1394 
basis to explain previously reported flow cytometry data, where only about half of the 1395 
total population took on the QS-activated phenotype 94.   1396 
 1397 
 
 
83 
 
Figure 3-2. Lsr autoinduction of 1398 
pure cultures leads to bimodal 1399 
phenotype.  A The fraction of the 1400 
population QS activated over time 1401 
was influenced by the standard 1402 
deviation of the natural logarithm, 1403 
, of the log-normal distribution 1404 
for the parameter basal, here for  1405 
values ranging from 0 to 0.025, run 1406 
in triplicate.  Dark lines represent 1407 
the average values, whereas lighter 1408 
surrounding lines represent the 1409 
standard deviation.  Activation was 1410 
the slowest but most abrupt for the 1411 
cell population with a variance of 1412 
zero, quickly achieving complete 1413 
activation.  Inset is the distribution 1414 
of the value of basal for three 1415 
different values of .  B Inset is the 1416 
plateaued value of the population 1417 
fraction that was Lsr active for 1418 
populations of different median 1419 
basal values. In the main figure, 1420 
the overall extracellular 1421 
concentration of AI-2 per Lsr 1422 
induced cell is represented on the 1423 
primary ordinate axis.  Each curve 1424 
represents an average from three 1425 
different simulations of non-taxis 1426 
swimming cells.  The median 1427 
minimal distance between cells 1428 
averaged for 20 simulations of non-1429 
taxis swimming cells is represented 1430 
on the secondary ordinate axis.  At 1431 
the right top is a pictorial example 1432 
of the minimum distance between 1433 
cells (where dmin = min (d1, d2, d3,... 1434 
dn, n=6).   1435 
 1436 
 1437 
 1438 
 
 
84 
 
We also note that the distribution of the bifurcation was possibly influenced by 1439 
the related but competing factors of intercellular distance and cellular concentration. This 1440 
was suggested by a general correlation between decreasing basal and decreasing QS 1441 
activation as seen in the inset of Figure 3-2B. As indicated by the timing of spikes in 1442 
Figure 3-2B, reducing basal resulted in delayed activation.  This delayed activation 1443 
corresponded to diminished intercellular distance due to growth (light blue line) 1444 
calculated as the minimum distance between neighboring cells (a cartoon of which is 1445 
depicted on the top right corner of Figure 3-2B for an ith cell). Increased cell proximity 1446 
may have amplified the negative feedback associated with QS activation, as the local 1447 
depletion of AI-2 by QS active cells was able to delay or prevent the activation of more 1448 
cells.  Further evidence of this cell-cell distance effect is provided in the SI.  The putative 1449 
influence of intercellular distance appeared to be constrained, however. That is, when cell 1450 
concentration was too high, more QS activated cells were required to draw down local 1451 
AI-2 concentrations as indicated by the relatively broad peak of the last curve depicting 1452 
AI2 concentration divided by the number of QS activated cells.  Subsequently, depletion 1453 
of AI-2 from the environment was delayed, allowing more cells to become QS activated 1454 
than would have otherwise.  This competing effect was ostensibly dominant for 1455 
populations with the lowest median value of basal in Figure 3-2B, as the QS activated 1456 
fraction increased instead of continuing to decrease. 1457 
3.4.2 Pattern formation in QS systems: LuxIR vs Lsr.  In Figure 3-3A, we show that a 1458 
“supernova-like” LuxIR QS activation pattern emerged from near the colony center, 1459 
loosely recapitulating experimental results seen with LuxIR engineered cells 33,129. For 1460 
this simulation, cell motility was limited to gliding along the surface in a space filling 1461 
 
 
85 
 
capacity purely as a function of cell outgrowth (cell division resulting in occupation of 1462 
otherwise empty neighboring site by daughter cell) without directional bias.  Once 1463 
initiated, LuxIR activation (red color) spread quickly outward from its point of origin 1464 
often near the center of colony growth (due to positive feedback processes of cell growth 1465 
and autoinduced AHL production), engulfing inactive neighbors in a building wave of 1466 
high AHL concentration, coordinating population expression in what appears as a 1467 
traveling wave front of activated cells.   1468 
In Figure 3-3B, we depict simulated behavior for the Lsr system with identical 1469 
gliding motility rules as for the LuxIR simulations.  Instead of a blossoming wave of QS 1470 
activation (Figure 3-3A), scattered patches of activation appeared, presumably 1471 
influenced by the same Lsr driven AI-2 recompartmentalization dynamics associated with 1472 
bimodal activation.  Undergirding the distinction between these activation patterns were 1473 
differences in the interplay between non-genetic heterogeneity and intercellular QS 1474 
feedback, as discussed in SI, where activation heterogeneity was also quantified.  Thus, 1475 
in the case of Lsr, QS activation was not apparently localized, nor was it uniform. 1476 
Instead, fractional QS activation was observed among the population whole. 1477 
3.3.3 Cell Motility – Lsr QS based pattern emergence.  Swimming was modeled either 1478 
as an unbiased process or was governed by heuristics approximating AI-2 1479 
chemoattraction, described in the methods.  Early in the simulations, prior to QS 1480 
activation, chemotaxing cells assembled into clusters 130, attracted to each other by AI-2 1481 
molecules. Largely, cell populations gradually coalesced along simulation boundaries as 1482 
our boundary conditions preclude diffusion through these boundaries.  Since clusters  1483 
 
 
86 
 
Figure 3-3.  QS dynamics coupled 1484 
with gliding during colony growth.  1485 
A Images from a representative 1486 
simulation for LuxIR/AHL 1487 
dynamics coupled with colony 1488 
growth.  QS active cells are in red, 1489 
whereas inactive cells are in yellow.  1490 
B Images from a representative 1491 
simulation of Lsr/AI-2 dynamics 1492 
coupled with colony growth.  QS 1493 
active cells are in red, whereas 1494 
uninduced cells are yellow. 1495 
 1496 
 1497 
 1498 
 1499 
 1500 
 1501 
 1502 
 1503 
 1504 
 1505 
 1506 
 1507 
 
 
87 
 
were formations of relatively high local cell density and correspondingly high AI-2 1508 
concentration, cells within clusters activated more quickly than unclustered cells.  Once 1509 
Lsr activity was sufficiently pervasive, clusters were transformed into sites of the most 1510 
rapid AI-2 depletion, dissipating the AI-2 gradient and the impetus for clustering, the 1511 
ultimate result being cluster dispersal. This emergent behavior, depicted here for the first 1512 
time in a model, generates an interesting overall “collect-disperse” pattern.  This pattern 1513 
is exemplified in Figure 3-4. At zero minutes a randomized distribution of an initial cell 1514 
population was observed.  At 100 minutes, cell clusters began to coalesce, as seen at the 1515 
bottom and left edge and near the top left corner of the environment.  At 200 minutes, 1516 
clusters were more pronounced, and at 244 minutes, QS activated cells began to emerge 1517 
(red).  At 266 minutes more cells were QS active and the clusters became more dispersed.  1518 
By 300 minutes clusters had become entirely dispersed.  Through many repeated 1519 
simulations, the exact placement and number of clusters were found to be inconsistent.  1520 
Quantification of this clustering (by cell-cell distance) is available in SI.  Interestingly, as 1521 
different motility modes implied different cell-cell proximity, motility appeared to 1522 
feedback onto fractional activation, which is also discussed in SI.           1523 
3.4.4 Mixed Population Simulations.  Additional consequences of desynchronized Lsr 1524 
based AI-2 recompartmentalization were identified through the study of competition for 1525 
AI-2 among in silico mixed cultures of a wildtype population with varied derivative 1526 
mutant populations.  Here, initial conditions were the same as in previous simulations, 1527 
except that half the modeled cell population had modulated Lsr/LuxS activity, reflecting 1528 
different genotypes.  Figure 3-5A represents the results from a wildtype pure culture as a  1529 
 
 
88 
 
 1530 
 1531 
 1532 
 1533 
 1534 
 1535 
 1536 
Figure 3-4.  Cluster-disperse pattern from combination of Lsr and chemotaxis.  Images 1537 
from a representative simulation of Lsr/AI-2 dynamics coupled with chemoattraction to 1538 
AI-2.  Autoinduced cells appear in red, whereas uninduced cells appear in yellow.  1539 
 1540 
 1541 
 1542 
 1543 
 1544 
 1545 
 1546 
 1547 
 1548 
 1549 
 1550 
 
 
89 
 
 1551 
Figure 3-5.  Mixed culture simulations.  Each graph may be considered as being 1552 
composed of a series of histograms over time.  The vertical axis represents frequency, the 1553 
right axis (towards the reader) represents increasing log transformed concentration of 1554 
transporter protein, and the left hand axis (away from the reader) represents increasing 1555 
time in minutes.  A wildtype cells alone.  BluxS populations mixed with wildtype cells.  1556 
C lsrFG cells mixed with wildtype cells. 1557 
 1558 
 1559 
 1560 
 1561 
 1562 
 1563 
 1564 
 
 
90 
 
reference (lower panel).  Initially, all cells were QS inactive (represented here as 1565 
normalized to 0.0).  Around 200 minutes, subpopulations began to branch off into high 1566 
expressers, representing QS activation in these cells.  In Figure 3-5B where luxS 1567 
(above) and wildtype (below) cells were cultured together in silico, wildtype cells 1568 
became dominantly activated over the course of the simulation, with luxS cells 1569 
remaining largely Lsr inactive.  In fact, even by the end of the simulation, luxS 1570 
population Lsr expression levels were only nominally above their baseline.   1571 
As in Figure 3-5C, mixing of wildtype cells (below) with lsrFG cells (above) 1572 
also resulted in uneven activation between populations, but for ostensibly different 1573 
reasons.  Here, lsrFG cells were dominantly activated and became so early on, due to 1574 
the early activation of lsrFG cells, which accumulate AI2-P more readily than wildtype 1575 
cells 131.  This then deprived wildtype cells of AI-2, resulting in a smaller activated 1576 
fraction than when wildtype cells were cultured by themselves.   1577 
 1578 
 1579 
 1580 
 1581 
 1582 
 1583 
 
 
91 
 
3.5 Discussion 1584 
Bimodal expression arising from clonal origins is a common phenomenon, 1585 
frequently associated with pattern formation and differentiation in multicellular organisms 1586 
132,133.  As is commonly the case and here also, population bifurcation is expected to arise 1587 
as a consequence of nonlinear responses to nongenetic heterogeneity manifested in bimodal 1588 
protein expression 134,135.  Specifically, our results suggest that bimodal expression arises 1589 
when the intercellular negative feedback associated with Lsr activated AI-2 1590 
recompartmentalization is desynchronized across a population (here represented by varied 1591 
rates of AI-2 uptake and subsequent Lsr mediated gene expression).  This bimodal 1592 
expression may represent role diversification. Bacterial diversification is frequently framed 1593 
within the context of bet hedging 136 or as a graded response to environmental conditions, 1594 
an example being the different conditions that at the margins of biofilms compared to those 1595 
within the bulk 137.  Although Lsr signaling does influence biofilm development in E. coli 1596 
60, whether population diversification through Lsr QS represents bet-hedging or some other 1597 
transient specialization within the context of a population-wide transition to a sessile 1598 
lifestyle remains a point of significant interest. Regardless, in silico experiments here 1599 
clearly frame Lsr QS activity within a context of phenotype diversification.  We further 1600 
explored how this diversification operates spatially and between strains with different 1601 
Lsr/LuxS activity. Spatial self-arrangement of Lsr activity was evaluated and contrasted 1602 
against that of LuxIR.  LuxIR dynamics produced a sudden rapidly expanding activation 1603 
from the colony center.  With an autoinducer that requires active transport between cell 1604 
compartments, the Lsr system provided a stark contrast by producing speckled activation.  1605 
This distinction may be attributed to the coupling of positive intracellular and extracellular 1606 
 
 
92 
 
feedbacks to one another in LuxIR QS, whereas intracellular positive feedback drives an 1607 
extracellular negative feedback in Lsr QS (recall Figure 3-1).  1608 
Lsr QS populations subject to AI-2 chemoattraction displayed an interesting 1609 
collect-disperse behavior, which informed a complete cycle of nutritional resource 1610 
exploitation: seek, find, deplete, and seek again. To our best knowledge, this is the first 1611 
time that such behavior has been characterized by any mathematical model or otherwise 1612 
described in the context of the specific QS mechanisms investigated here.  Since 1613 
extracellular AI-2 could be viewed as a potential marker for the exponential growth of a 1614 
broad swath of bacterial species (its production is a direct consequence of primary carbon 1615 
metabolism) 27,46, motility toward such active growth is likely to be most fruitful for a 1616 
generalist like E. coli, potentially reinforcing chemotactic tendencies toward other 1617 
substrates.  As catabolite repression is epistatic to QS switching/Lsr activation 55, which 1618 
is itself correlated with late exponential/early stationary phase growth 54, the rapid 1619 
depletion of AI-2 from local extracellular space leading to cell redispersion could be seen 1620 
as part and parcel of a larger coordinated behavior of a population seeking out new 1621 
energy sources to exploit.  Alternatively, as Lsr signaling appears to mediate both the 1622 
influence of AI-2 on biofilm development 60 and its consumption as a secondary 1623 
metabolite 46, partial population activation in this context may represent a mechanism for 1624 
the alignment of secondary metabolism consumption with the metabolic burden of 1625 
producing public goods.  As Lsr activation loosely corresponds with growth phase 1626 
transitions, one specific instance of this concept is the possibility that chemotaxis to self-1627 
secreted AI-2 draws a population toward surfaces at which preparations for a sessile 1628 
 
 
93 
 
lifestyle are initiated by the QS activated population fraction while the remaining cells 1629 
continue to seek out other fertile grounds for colonization. 1630 
Aside from phenotype diversification within a population, the ability to move 1631 
toward a favorable growth environment and absorb signaling molecules from the 1632 
extracellular space could have multiple motivations.   For example, influx of AI-2 via Lsr 1633 
activation could be used to limit QS sensing processes in competitors — processes 1634 
involved in the transition to quiescence or sessility including: increased antibiotic 1635 
resistance in local bacteria 138; increased predation defense 139,140, by preventing other AI-2 1636 
consumers from competing for a similar niche; exploiting available secondary 1637 
metabolites 46, or some combination thereof.  Additionally, as an established population, 1638 
competition for a secondary metabolite might be an expensive proposition — especially 1639 
if the metabolite or its breakdown products moonlight as epistatic signals for the 1640 
transition to stationary phase phenotypes.  However, denying potential competitor 1641 
populations secondary metabolite and/or limiting the prevalence of potential chemotactic 1642 
cues could be a worthwhile tradeoff, particularly if only a select fraction of the 1643 
population is required for the job. 1644 
 Whichever if any of these possibilities obtain for any particular circumstance, 1645 
they all fall under the broad umbrella of hoarding amid the relative phylogenetic 1646 
abundance of AI-2 in the microbial world 27.  To more carefully consider how changes to 1647 
the Lsr system itself might inform hoarding dynamics in consortia, mixed population 1648 
simulations were performed. 1649 
 
 
94 
 
In mixed simulations, we found that luxS mutants remained largely inactive when 1650 
paired with wildtype cells.  This is attributable to a zero baseline of intracellular AI-2 in 1651 
such cells, thus requiring a much higher extracellular threshold of AI-2 prior to 1652 
activation.  These populations represent a type of QS cheat termed “signal-negative” 141. 1653 
Such cells are metabolically unburdened by the production of autoinducer even though 1654 
they participate in the production of social goods and enjoy their benefits.  In this 1655 
particular case, the luxS population was also unable to compete for AI-2 against signal-1656 
producing wildtype cells, and remained largely QS inactive.  Thus, in mixed cultures, 1657 
luxS negative cells were effectively signal blind as well.  As long as QS products are 1658 
public goods, luxS mutants are likely to be doubly non-cooperative.  If the benefit of 1659 
activation is direct or the resulting product is not a public good for the consortia, such a 1660 
defect would likely minimize signal-negative cheating.  This incentive for cells to avoid 1661 
signal-negative cheating is compounded by any metabolic burdens arising from an 1662 
incomplete methyl cycle 46, creating an incentive for cooperation even more direct than 1663 
those previously considered for generic QS processes 142.  Alternatively, luxS negative-1664 
Lsr positive cells could play a niche role within bacterial consortium, consuming AI-2 1665 
made by competing species 89.   1666 
LsrF and lsrG double mutants were chosen to study in part because they are the 1667 
most commonly missing elements among Lsr homologs 14.  They were also found to be 1668 
more sensitive to AI-2 in synthetic biology contexts 131, internalizing and accumulating 1669 
AI-2 more rapidly than wildtype cells. lsrFG cells thereby suppressed wildtype 1670 
activation.  In general, lsrFG mutants present a sort of paradox. Even though these 1671 
populations may be slightly advantaged by a faster activation and corresponding AI-2 1672 
 
 
95 
 
internalization, they are unable to utilize this increased pool metabolically, resulting 1673 
merely in a higher AI2-P intracellular steady state.  This may be warranted if the only 1674 
role of AI2-P is to effect multiple downstream phenotypes via pleiotropic LsrR 1675 
derepression.  However, the degree to which LsrR derepression upregulates non-Lsr 1676 
genes remains unknown 143.     1677 
Alternatively, as representatives of other strains or species, interactions of lower 1678 
AI-2 sensitivity strains (luxS) and higher AI-2 sensitivity lsrFG) strains with wildtype 1679 
cells suggest that consortial Lsr activation for cell types with different Lsr/AI-2 sensitivity 1680 
groups is likely to be highest for any particular subset when all constituent cells share a 1681 
similar or lower Lsr/AI-2 sensitivity.  This adds another layer of complexity to the 1682 
efficiency sensing view of QS, where bacteria use QS machineries to evaluate whether 1683 
dependent secreted public goods will serve their intended purpose, insofar as Lsr QS may 1684 
allow interrogation of the environment for other cells with varied Lsr/LuxS affinity 1685 
types.  This potential outcome is one of many stemming from the intercellular negative 1686 
feedback of the Lsr system operating even amid simple consortia as studied here.        1687 
 1688 
 1689 
 1690 
 1691 
 1692 
 
 
96 
 
3.6 Concluding Remarks 1693 
While the simulations contained herein attempt to model quorum sensing 1694 
contextually, the extent to which our in silico experiments correctly predict emergent 1695 
behaviors must be considered 125. 1696 
As a general matter, the reported results rely upon the validity of assumptions 1697 
regarding the sensitivity of QS and chemotactic behaviors to AI-2. As a rule we have 1698 
tried to make assumptions that are supported by the literature.  For example, the minimal 1699 
concentration at which AI-2 triggers chemotactic behaviors appears to be lower than that 1700 
at which QS can be triggered for wildtype Lsr systems 131.  Even if both behaviors were 1701 
triggered at similar concentrations, clustering phenomena still likely obtain due to time 1702 
lags associated with transcription, translation, and rate-limited transport through low flux 1703 
pathways. 1704 
Additionally, as previously mentioned, QS phenomenon apply only where 1705 
extracellular transport is dominated by diffusive processes, when convection is minimal; 1706 
otherwise, autoinducer dilution prevents activation 144.  Therefore, QS has often been 1707 
reframed as diffusion or efficiency sensing 145.  Regardless of these semantics and 1708 
connected interpretations, these phenomenon are likely to prevail within a protected 1709 
volume or cavity with restricted access to flow conditions. 1710 
Of further complication is the fact that the greater the diffusivity of the AI-2, the less 1711 
pronounced chemotactic phenotypes are likely to be, as gradients rapidly dissipate.  Also, 1712 
the greater the diffusivity of AHL, the slower the colony will be to activate. While 1713 
diffusivity coefficient values were specific to AI-2 and AHL, they were estimates based 1714 
 
 
97 
 
on Wilke-Chang correlation calculations146. However, while increasing diffusivity would 1715 
blunt chemotactic behaviors, bimodality would be exacerbated as the effective distance 1716 
between cells became smaller.  An additional concern regarding the extracellular 1717 
transport of chemical species is that bacterial movement might be reasonably expected to 1718 
result in superdiffusive conditions directly proximal to the cell membrane.  Due to the 1719 
low Reynold’s number at these scales, however, this superdiffusivity should not extend 1720 
into the surrounding bulk.   1721 
Moreover, when asserted noise is tuned lower, Lsr bifurcation is also likely to 1722 
dissipate.  While the heterogeneity chosen is probably conservative, studies of population 1723 
heterogeneity that track a significant number of single cell expression histories were 1724 
unable to be identified. 1725 
Finally, from a purely qualitative perspective, swimming populations generally 1726 
appear to slow upon entering late-exponential or early-stationary phase growth.  That is 1727 
to say that the average speed of propulsion becomes markedly distributed, in a manner 1728 
probably associated with growth phase heterogeneity.  Commonly, in such 1729 
circumstances, macroscopic self-propulsion comes to a halt altogether for some cells.  As 1730 
QS activity is often associated with growth phase transitions, it is beguiling to imagine a 1731 
scenario where chemotaxis draws a population toward surfaces at which sessile behaviors 1732 
are initiated by the less mobile population upon QS activation, while cells that still can 1733 
respond chemotactically are less likely to be QS activated and continue to seek out more 1734 
fertile grounds for colonization. 1735 
 1736 
 
 
98 
 
3.7 Supplemental Information 1737 
3.7.1 Text and Discussion 1738 
The focus of the main text, the Lsr system, is one of two known signaling transduction 1739 
systems that responds to the ‘universal’ quorum sensing signal, autoinducer-2 (AI-2).  In 1740 
E. coli 60, Salmonella 77, and A. acetinomycetemcomitans 67, Lsr based QS drives or at 1741 
least influences cooperative behaviors among multicellular consortia of bacteria.  1742 
Experiments from Tsao, et al. 94 used a two plasmid system to amplify Lsr based 1743 
expression by transcribing T7RNAP under pLsrR control on a single copy plasmid and 1744 
linking GFP expression to a T7 promoter on a mid-to-high-copy pET200 plasmid.  By 1745 
one interpretation, this created a reporter system for Lsr QS activation 144,147.  1746 
Interestingly, pure cultures of reporter transfected-wildtype bacteria developed a 1747 
fluorescence distribution similar to that of mixed cultures.  This suggested that one 1748 
fraction of the wildtype population became QS active, while the other remained off in a 1749 
bimodal fashion, belying the general association of QS with whole population 1750 
coordination.   1751 
Previous studies have modeled the prototypical LuxIR QS system as producing a 1752 
strong switching behavior given sufficiently high cell density 127, usually attributed to the 1753 
positive feedback inherent in the system topology 148.  The Lsr system too, has been 1754 
shown to generate behavior reflecting a highly sensitive switch 112.  Modeling of Lsr 1755 
activity in distinct cells within a population has not previously demonstrated lasting 1756 
bimodality of system expression 95, however.   1757 
 
 
99 
 
Unlike the LuxIR system, Lsr topology resembles that of sugar importer systems 1758 
like the lac operon, which has been shown to produce bimodal activity when exposed to a 1759 
nonmetabolizable inducer 149.  While thiomethyl galactoside and isopropyl--D-1760 
thiogalactopyranoside served as a constant stimulus in those experiments, sufficient 1761 
production of AI-2 throughout the time course of interest could serve as an analog in the 1762 
case of the Lsr system. 1763 
Using parallel ODEs to represent individual cells drawing from the same 1764 
extracellular AI-2 pool, our model suggested that one subpopulation could prevent 1765 
another from activating when paramet+ers were sufficiently perturbed.  This analysis was 1766 
extended to demonstrate that drawing parameter values from log normal distributions 1767 
with different variances resulted in a range of partially activated populations. 1768 
Additionally, our model suggested that noise from spatially associated stochasticity was 1769 
by itself insufficient to generate bimodal expression patterns.   1770 
3.7.2 Methods 1771 
3.7.2.1 Lsr ODE Model 1772 
3.7.2.1.1 Generalities and scope.  The model described herein was developed to simulate 1773 
Lsr system behavior for bacteria in a batch reactor between lag and early stationary 1774 
growth phases.  At the endpoint, AI-2 should be functionally depleted from the 1775 
supernatant 54,55,57, with extracellular concentrations peaking between 4-6 hours 150.  The 1776 
developed equations relied on Hill and Michaelis-Menten like expressions to encapsulate 1777 
reaction rate behaviors.  In order to investigate population bifurcation, separate sets of 1778 
 
 
100 
 
ODEs were run simultaneously, modeling multiple cells sharing the same extracellular 1779 
space.   1780 
Lsr system components were modeled as translated from polycistronic mRNA 1781 
species lsrRK and lsrACDBFG, the transcription of which was considered a function of 1782 
LsrR and intracellular AI-2 concentrations.  The proteins LsrK and LsrR were modeled as 1783 
distinct species, LsrF and LsrG were modeled as a single entity.  LsrA, LsrC, LsrD, and 1784 
LsrB were folded into a single type of entity, OP, as they form a complex ABC type 1785 
importer.  Additionally, “LsrFG”, for while their products are distinct they both use AI2-1786 
P as their substrate and neither’s products are known to feed back onto Lsr activity.  Thus 1787 
LsrF and LsrG were functionally equivalent for our specific purposes 54.  Of the proteins 1788 
outside the divergent Lsr operons that influence Lsr activity in E. coli (luxS, ydgG, crp, 1789 
and pts), none are themselves modulated by Lsr activity 55,58,63.  Moreover, the 1790 
concentration of these species have not been shown to markedly vary within the time 1791 
frame of interest.  These proteins were therefore considered constant and their activity 1792 
rates were simplified to a Michaelis-Menten like dependence on the substrate alone.   1793 
3.7.2.1.2 mRNA expression.  Production of mRNA species was modeled with a modified 1794 
Hill equation derived from an expression for the fraction of free DNA to total DNA (free 1795 
DNA + repressed DNA) as a function of LsrR and AI2P concentrations.  Whereas active 1796 
LsrR was modeled as a tetramer 151, AI2P cooperativity was asserted.  Cooperativity for 1797 
AI2P derepression was assumed to be on the order of cooperativity for protein/DNA 1798 
binding.  A lower bound for such cooperativity might be 1.38-2.72, which was measured 1799 
for a nonspecific interaction between DNA and a multidomain protein in Mycobacterium 1800 
smegmatis 152. An upper bound for such interactions is believed to be around 10 153.  As 1801 
 
 
101 
 
each LsrR has a distinct AI2P binding domain, we chose 4 as a baseline level of 1802 
cooperativity.  The trajectory of the system using this cooperativity was similar to that for 1803 
degrees of cooperativity greater than 4.  At tested cooperativities less than 4, the drop in 1804 
activation sharpness was sufficient to noticeably dull the rate of Ai-2 1805 
recompartmentalization from the extracellular space.  The transcription rates for the two 1806 
different polycistronic mRNA species were set as equivalent, as the data on the relative 1807 
strength of expression toward LsrA and LsrR is inconsistent 56,70.   1808 
3.7.2.1.3  Protein synthesis.  Lsr proteins expressed from the same polycistronic mRNA 1809 
species were modeled as having the same translation rate.  While different ribosome 1810 
binding sites along the polycistrons are likely to have different affinities for ribosomes, in 1811 
the absence of data, the co-regulation implied by operon structure and protein function 1812 
was given overriding consideration.  1813 
3.7.2.1.4  mRNA and protein degradation.  As an additional simplification, both 1814 
polycistronic mRNA species were modeled with the same rate of degradation, with a 1815 
half-life on the order of 10 minutes, which was an order of magnitude faster than that 1816 
used for proteins.  While Lsr proteins are expected to have different vulnerabilities to 1817 
proteolytic degradation, the result is either higher or lower quasi-steady state levels.  The 1818 
effects of such differences could be accounted for by either lowering or increasing 1819 
corresponding unconstrained enzymatic rates.    1820 
3.7.2.1.5  Cell growth.  Cell growth was modeled as Monod growth. The maximum rate 1821 
and Monod saturation rate constants of 0.032/min and 75 M, respectively, were had 1822 
from a fit to OD600 measurements from a batch growth at 37oC.  The growth rate also 1823 
 
 
102 
 
informed the dilution of cellular components.  As a generality, the rate of cell growth 1824 
strongly influenced simulation results, as increasing cell density accelerated extracellular 1825 
AI-2 accumulation.  That is, faster growth decreased the time to autoinduction. 1826 
3.7.2.1.6  AI-2 transport.  Instantaneous concentration and dilution associated with 1827 
transport between extracellular and intracellular spaces was treated by including a 1828 
dilution/concentration term of 1012 as seen in equation (10).  This dilution factor 1829 
accounted for the difference between the femtoliter intracellular environment and the 1830 
milliliter term associated with cell concentration. 1831 
Influx into the periplasm through porins was modeled as a diffusion process with 1832 
a Michaelis-Menten form, rate limited at high concentrations, but otherwise proportional 1833 
to the concentration difference between extracellular and periplasmic compartments.  1834 
Influx through Lsr ABC type complexes was modeled as a function of importer complex 1835 
concentration and periplasmic AI-2.  The transporter’s component proteins were treated 1836 
as a single species.  Transporter complex was assumed to form with a cooperativity of 4, 1837 
as a reflection of the fact that four independent components are involved in complex 1838 
formation, namely LsrA, LsrB, LsrC, and LsrD.  Although the nucleotide binding 1839 
component, LsrA, operates as a dimer, any additional degree of freedom was considered 1840 
to be eliminated by the fact that the LsrA dimer pair is fused together.  Further, this 1841 
transporter complex was asserted to act upon periplasmic AI-2 according to a Michaelis-1842 
Menten like dynamic.  Michaelis-Menten like dynamics were also asserted for the 1843 
operation of the alternative importer as well as for AI-2 export through YdgG.  That is, 1844 
these processes were assumed to be constrained by a maximal velocity above a saturating 1845 
 
 
103 
 
AI-2 concentration, and that at lower AI-2 concentrations activity was approximately 1846 
linear with respect to AI-2 itself. 1847 
3.7.2.1.7  AI-2 degradation and synthesis.  The rate of extracellular AI-2 degradation was 1848 
assumed to be minimal based on a lack of attenuated AI-2 activity in bioassays after 1849 
incubation of in vitro synthesized samples overnight at 37oC.  The rate of cytoplasmic 1850 
AI-2 degradation was set significantly higher to account for experiments measuring 1851 
cytoplasmic AI-2 in E. coli without functional Lsr activity, wherein cytoplasmic AI-2 1852 
concentrations dropped significantly once the stationary phase had been well established 1853 
150.  While the time frame of this marked decrease fell outside the scope of in silico 1854 
experiments herein, it nonetheless suggests the existence of mechanisms that degrade AI-1855 
2 intracellularly, independent of Lsr system expression.  Periplasmic degradation of AI-2 1856 
was modeled as intermediate to the extracellular and cytosolic rates, and did not bear 1857 
strongly on the activity of the Lsr system, since the absolute moles in periplasmic pools 1858 
were limited compared to extracellular and intracellular species. 1859 
 The same experiments that provide evidence for this degradation also suggest that 1860 
the rate of synthesis is not constant 150, even if the expression of luxS and pfs are not 1861 
strongly varied across time 56.  Nonetheless, we model AI-2 synthesis as constant, which 1862 
over the course of the time scale of interest we assume to be operationally approximate.   1863 
3.7.2.1.8  Equations 1864 
The specific form of the equations used and the parameters values used herein 1865 
were as follows: 1866 
  1867 
 
 
104 
 
 1868 
(1)      
max
[ ] [ ] [ ]
[ ] [ ]
[ ]
( [ ] ) ( ( ))[ ]
[ ]
peri out peri
porin
porin out peri
pericoop
in AIperi peri
ABC peri
d AI AI AI
V
dt Km AI AI
AI
V OP basal t AI
Km AI
 

 
 
  

 , 1869 
(2) 
[ ][ ]
( [ ] )
[ ]
[ ] [ ]
[ ] ( ( ))[ ]
[ ] [ ]
pericoopin
synth in
ABC peri
in in
Phos ydgG AIin in
Phos in export in
AId AI
K V OP basal
dt Km AI
AI AI
k LsrK V t AI
Km AI K AI
 
   

  
 
, 1870 
(3) 
2
[ ][ 2 ]
[ ]
[ ]
[ 2 ]
[ ] ( ( ))[ 2 ]
[ 2 ]
in
Phos
Phos in
Cat AI P
Cat
AId AI P
k LsrK
dt Km AI
AI P
k LsrFG t AI P
Km AI P
 
 

 

, 1871 
(4) [ ] [ 2] ( ( ))[ ]P Td OP k mRNA t OPdt    
, 1872 
(5) [ ] [ 2] ( ( ))[ ]P FGd LsrFG k mRNA t LsrFGdt    
, 1873 
(6) [ ] [ 1] ( ( ))[ ]B Rd LsrR k mRNA t LsrRdt    
, 1874 
(7) [ ] [ 1] ( ( ))[ ]B Kd LsrK k mRNA t LsrKdt    
, 1875 
(8) 
1 12
2
3
2
[ 1] 1 ( ( ))[ 1]
[ ]1
[ 2 ]
mR MRcoop
coop
d mRNA V t mRNA
k LsrRdt
r AI P
   


, 1876 
(9) 
2 22
2
3
2
[ 2] 1 ( ( ))[ 2]
[ ]1
[ 2 ]
mR MRcoop
coop
d mRNA V t mRNA
k LsrRdt
r AI P
   


, 1877 
(10) 
[ ] [ ][ ]
[ ](
[ ] [ ]
[ ]
)
[ ]
peri outout
max porin
porin out peri
in
ydgG
export in
AI AId AI
dilutionF cells V
dt Km AI AI
AI
V
K AI

 
 

, 1878 
(11) [ ] [ ][ ] [ ]mid
d cells yield Substrcellsdt K Substr 
, 1879 
(12) [ ] [ ][ ] [ ]mid
d Substr yield Substrtake cellsdt K Substr  
 1880 
 
 
105 
 
, where [Substr] represents the concentration of substrate, [AIperi] represents the 1881 
concentration of periplasmic AI-2, [AIin] represents the concentration of cytoplasmic AI-1882 
2, [AIout] represents the concentration of extracellular AI-2, [AI2P] represents 1883 
phosphorylated AI-2, [LsrR] represents LsrR concentration, [LsrK] represents LsrK 1884 
concentration, [LsrFG] represents LsrF and LsrG concentration, [mRNA2] represents 1885 
lsrRK concentration, [mRNA1] represents lsrACDBFG concentration, and [OP] 1886 
represents transporter protein concentration.      1887 
3.7.2.1.9  Parameter Values. A table of parameter values can be found in Table 3-S1.  1888 
The maximal rate of transport through the alternative importer, as governed by basal, was 1889 
set lower than the rate of export through YdgG, allowing extracellular AI-2 accumulation 1890 
at baseline levels of Lsr activity.  The remaining parameters were set such that this initial 1891 
AI-2 flux out was overcome by Lsr activation. 1892 
Specifically, according to BB170 bioassays of culture supernatant, AI-2 appears 1893 
to peak between 4-5 hours in a batch culture.  FRET-LuxP assays and bioassays suggest 1894 
that the peak concentration is less than 80 M and greater than 40 M, respectively 55,150.  1895 
Once extracellular accumulation stalls, AI-2 appears to be drawn down to low 1896 
concentrations of AI-2 in less than an hour and then to concentrations below bioassay 1897 
sensitivity within the next hour.   1898 
Asserting that transcription is directly coupled to translation for LacZ, 1899 
transcription was fit to Miller assays of LacZ expression from pLsrA and pLsrR 1900 
promoters, indicating that transcription begins evolving prior to 4 hours after culture 1901 
initiation, reaching a several fold higher level of expression prior to 6 hours 56.   1902 
 
 
106 
 
Table S1. Estimated parameter values  1903 
Parameter Value Parameter Value Parameter Value 
 8000 M/min  0.15 /min  7x10
-5 M4 
                 1 M  50 /min  4 
  2x10
12 
/(min*M3) 
 1.4 M       6 
  4  18.8 /min  0.693 /min 
  500 M/min  0.01 /min   0.693 /min 
 0.5 M  0.01 /min   10
-12 
  0.015 /min  18.8 /min   0.3
 /min 
  220 M/min  0.01 /min   0.032 /min 
   80 /min  0.01 /min   2250 M 
  1.4M  2x10
-5 
M/min 
  4.5x10
-7 
[Substr]/[cell] 
  1250 M/min  2x10
-5 
M/min 
  
  0.5 M  3.5x10
8    
 1904 
 1905 
 1906 
 1907 
 1908 
 1909 
 1910 
 1911 
 
 
107 
 
While the selected parameter set in combination with the given ODE’s produced a 1912 
much faster depletion of extracellular AI-2 than was observed experimentally, the 1913 
fraction of QS induced cells evolved over an extended period of time 94, and it was this 1914 
evolving population’s averages to which the model was fit (Figure 3-S1).  In order to 1915 
generate an evolving fractional Lsr autoinduction, a tractable number of cells was 1916 
modeled within a finite difference agent based scheme.  Each cell’s processes were 1917 
modeled by its own set of ODEs and the cell population was slightly desynchronized in 1918 
order to generate the fractional induction seen by Tsao, et al 94. 1919 
A parameter search was carried out to identify parameter sets satisfying the above 1920 
criteria. Among other measures of fit, the time to Lsr autoinduction was sensitive to 1921 
changes in all parameters effecting transport due to their direct bearing upon the balance 1922 
of accumulating cytoplasmic AI-2 and corresponding AI2-P species.  The parameter 1923 
space satisfying the available data including extracellular AI-2 concentration, was 1924 
nonetheless broad with similar sensitivities and behaviors over a range of values.  1925 
3.7.2.1.10  Initial Values.  Initial values were set according to approximate steady state 1926 
values for a system without lacking AI-2 production, where the Lsr system was 1927 
uninduced: [cells] = 3e7, [Substr] = 1500 M, [AIperi] = 0.015 M, [AIin] = 0.2 M, 1928 
[AIout] = 0.02 M, [AI2P] = 0.03 M, [LsrR] = 0.0012 M, [LsrK] = 0.0012 M, [LsrFG] 1929 
= 0.0012 M, [OP] = 0.0012 M, and [mRNA2] = [mRNA1] = 0. 1930 
 1931 
 
 
108 
 
3.7.2.1.11  Numerical solutions.  The equations and parameters were solved using 1932 
NDSolve in Mathematica 8.0 utilizing “StiffnessSwitching” methods.  The solution 1933 
required the “StiffnessSwitching” option, without which, the solution was unstable.  1934 
3.7.2.2 Finite difference-agent based model 1935 
3.7.2.2.1  Modeled Environment.  The environment was defined as a 500 x 500 x 6 m 1936 
volume, and was divided into 2 x 2 x 6 m elements.  Cells either exported or imported 1937 
AI-2 from their intracellular space into or from the finite difference element in which 1938 
their cell centers were found.  AI-2 also diffused between finite difference elements as 1939 
modeled by a forward in time-central in space scheme with an assumed diffusion 1940 
coefficient of 5x10-7 cm2/s.  The boundaries of the simulation were modeled as 1941 
impermeable. 1942 
3.7.2.2.2  Adaptation of equations and solutions.  With few exceptions, the previously 1943 
described ODEs were repurposed without modification in the finite difference/agent 1944 
based model.  Growth was one exception, as division became a discontinuous stochastic 1945 
event.  Monod growth dynamics informed the median of a log normal distribution of 1946 
doubling time with a  of 0.05 where the median rate was updated every time step.  The 1947 
exchange of AI-2 between the environment and a cell was localized to the space in which 1948 
the cell center was found at the beginning of the time step.  This allowed a synchronized 1949 
update of the final grid concentration at the end of the time step.  Furthermore, the AI-2 1950 
dilution/concentration factor was adjusted from 1012 to 24 to account for the difference 1951 
between the milliliter volume associated with cell concentration and the implied volume 1952 
of grid elements.  1953 
 
 
109 
 
3.7.2.2.3  Cell ODE Numerical Solution Method.  The numerical method used in the 1954 
agent based modeling was a second order Runga-Kutta with an ad hoc allowance for 1955 
stiffness.  In order to achieve efficient calculation, we used an explicit method with the 1956 
exception of periplasmic AI-2 after transporter concentration exceeded 4 times its initial 1957 
concentration.  This threshold was chosen based upon NDSolve interpolation, as a point 1958 
at which the Lsr system including transporter expression and subsequent concentration 1959 
had already begun its transition to an active state.  After this threshold was surpassed, 1960 
porins were assumed to be the rate limiting element in the transport of AI-2 from the 1961 
extracellular space to the cytosol, and modeling was made to reflect this by substituting 1962 
terms describing ABC-type transporter activity with terms describing porin activity; 1963 
furthermore, periplasmic AI-2 was held at zero, reflecting the numerical solution from the 1964 
ODE system described in previously.  1965 
3.7.2.2.4  Cell Division.  Cell division was governed by individual counters that 1966 
incremented at each time step.  Once a cell’s counter exceeded its doubling time, division 1967 
occurred.  At the time zero, cell counters were set randomly between zero and the 1968 
maximum allowed value.  Doubling time was varied between cells according to a mean 1969 
growth rate based on Monod kinetics (with parameters found in Table S1) with a log 1970 
normal distribution (variance of 0.05) in order to desynchronize cell doubling.  Upon 1971 
division, both mother and daughter cells acquired new growth rates from a log normal 1972 
distribution, while bearing duplicate properties including initial position, and their age 1973 
counters were reset to zero.  Newly initialized cells were assigned a basal rate of AI-2 1974 
flux through alternative importer pathways (the parameter, basal) or a rate of AHL 1975 
synthesis, for Lsr or LuxIR simulations respectively.    1976 
 
 
110 
 
3.7.2.2.5  AI-2 Diffusion.  AI-2 diffusion was modeled using a central difference 1977 
approximation.  While both truncation and roundoff error arise from this process, the 1978 
overall behavior of the simulation was not expected to be dramatically impacted, as 1979 
clustering behavior also obtained when elements were four times the size.  The diffusion 1980 
coefficient used (~5x10-7 cm2/s) was approximated using the Wilke-Chang correlation 146. 1981 
3.7.2.2.6  Time interval ordering.  The order of calculations within a time step was as 1982 
follows.  First, the average growth rate was determined from a Monod growth dynamic 1983 
solved by a second order Runga-Kutta method as a function of substrate concentration 1984 
and E. coli density.  Second, each E. coli divided or did not divide, marginally 1985 
accommodated its AI-2 chemotactic threshold (if appropriate), and moved (according to 1986 
the particular scheme employed).  For the specific purposes of the simulations reported 1987 
here, cells moved randomly in space each time step at an average rate of 20 m/sec, 1988 
coming to an average distance of 1.3 m per time step.  Correspondingly, cells rarely 1989 
moved more than one grid element at a time for any given time step.  Finally, each grid 1990 
was subject to calculations approximating diffusion.  That is, AI-2 was allowed to move 1991 
according to molecular diffusivity and distance/time calculations.  After all diffusion 1992 
processes were calculated over the entire environment, the exchange of AI-2 between 1993 
bacteria and the grids in which the bacteria were present was calculated according to Lsr 1994 
dynamics governed by previously described ODEs.  The sum of these changes to AI-2 1995 
concentration were then applied to each grid and each cell.   Time step and grid size were 1996 
chosen such that moderate changes to these measures resulted in qualitatively indistinct 1997 
outcomes. 1998 
 
 
111 
 
All random assignments were drawn according to the Mersenne Twister algorithm 1999 
154.  Seeds for the algorithm were changed every simulation.   2000 
3.7.3  Results 2001 
3.7.3.1  Numerical solutions to ODEs.  Given the ODEs and parameter values used 2002 
herein, the system evolved from a state of low expression to a state of high expression 2003 
(QS activated), as can be seen in Figure 3-S1, where the trajectories of different state 2004 
variables are represented, normalized to their maximum values.  This corresponds to the 2005 
known pattern of Lsr activity development.  At a particular AI2-P-to-LsrR threshold, 2006 
system expression is de-repressed, leading to the increased expression of transporter 2007 
proteins, which itself leads to an influx of extracellular AI-2 and a sustained increase in 2008 
the expression of lsr mRNA species.  2009 
In the model, heightened activity appeared to persist even after the depletion of 2010 
extracellular AI-2, by virtue of an altered flux balance from net AI-2 exportation to one 2011 
of net importation of processing.  That Lsr autoinduction leads to a shifts in the 2012 
compartmentalization of AI-2 species from one dominated by extracellular species to one 2013 
dominated by cytosolic species is one interpretation of Figure 3-S2.   2014 
3.7.3.2  System Sensitivity to basal rate of AI-2 uptake.  As expected, system behavior 2015 
was sensitive to parameters that markedly affected the rate of AI2-P accumulation when 2016 
Lsr system expression was low.  Among these was the parameter basal (Figure 3-S3). 2017 
With increasing  value, the rate of AI2-P accretion increased, accelerating the time 2018 
to Lsr activation.    2019 
 
 
112 
 
 2020 
Figure 3-S1.  Comparison of solution for population with a single basal value versus 2021 
a population with a unimodally distributed value of basal.   Juxtaposition of the 2022 
solution for extracellular AI-2 for a simulation of cells with a single basal value versus 2023 
the average solution of extracellular AI-2 for a simulation of cells with a log normal 2024 
distribution of the parameter basal 2025 
 2026 
 2027 
 2028 
 2029 
 2030 
 2031 
 2032 
 
 
113 
 
 2033 
Figure 3-S2.  Numerical solution to developed ODE equations. A  Selected 2034 
interpolated trajectories of the solution to the ODEs from 3.3.1.7 with the parameters 2035 
from 3.3.1.8. B  Mole fraction of total AI-2 species located either extracellularly or 2036 
cytosolically.  The denominator was the mole sum of extracellular, periplasmic, and 2037 
cytosolic AI-2.   2038 
 
 
114 
 
 2039 
Figure 3-S3.  Parameter sensitivity to the parameter basal as seen by changes in the 2040 
time until system activation.  System trajectories given different values of the parameter 2041 
basal. 2042 
 2043 
 2044 
 2045 
 2046 
 2047 
 2048 
 
 
115 
 
Specifically, an ~8% increase in the rate of AI-2 influx through the alternative pathway 2049 
from the base value resulted in a ~21% reduction in the time to activation.  The values 2050 
shown in this figure represent a range of basal that corresponds roughly to that of ribose 2051 
uptake via its low affinity mechanism.  Or, correspondingly, that basal level denoted 530 2052 
corresponds to 10% of the maximum achieved when the uptake mechanism is switched 2053 
on.  All these values (490-530) represent metabolically realistic values of small molecule 2054 
uptake. 2055 
3.7.3.3  Two sets of Lsr with different rates of basal AI-2 uptake.  More importantly, 2056 
when two sets of ODEs drew from the same pool of extracellular AI-2, one system of 2057 
ODEs could prevent the other from activating given sufficient parameter perturbation, as 2058 
seen in Figure 3-S4.  As the perturbation dwindled, the two ODEs converged to the same 2059 
high expression state.  That is, two sets of ODEs, representing two different populations, 2060 
were arranged to use the same extracellular AI-2.  Solutions to ODE’s in Figure 3-S4A 2061 
shared the same parameter values except for that of basal, which reflects the rate of AI-2 2062 
flux into the cell from the periplasmic space through the alternative pathway.  The value 2063 
of basal was set to 510 for one population and was adjusted continually lower for a 2064 
second, gradually slower activating, population.  When the difference between the two 2065 
values of basal was sufficiently large, the faster activating population depleted 2066 
extracellular AI-2 prior to the slower population achieving a sufficient intracellular 2067 
threshold to trigger Lsr induction, thus preventing QS activation.   2068 
 
 
116 
 
  2069 
Figure 3-S4. Solution to dual ODE system where second population had varied 2070 
values of parameters related to the transport of AI-2.  Population 1’s parameter values 2071 
were held constant. A  Over separate simulations, population 2’s parameter, basal was 2072 
varied at comparatively lower values.  The parameter values are in parentheses.  The 2073 
trajectory of AI2-P accumulation for each of these cases is represented here. B  Over 2074 
separate simulations, population 2’s parameter, Vin was varied at comparatively lower 2075 
values.  The trajectory of AI2-P accumulation for each of these cases is represented here.  2076 
The values in parentheses reflect Vin/1012. 2077 
 2078 
 
 
117 
 
While basal represented a specific activity, the phenomenon of a faster Lsr 2079 
activating group preventing a slower group from inducing proved a more general 2080 
property of the model, holding when other parameters to which Lsr induction was 2081 
sensitive were varied instead.  This can be seen in Figure 3-S4B, where the variation in 2082 
the assigned value of the parameter Vin, representing the rate of AI-2 transport through 2083 
the Lsr ABC-type transporter, was substituted for variation in the value of basal.  A 2084 
decreasing series of values for Vin , all smaller than that of population 1, were assigned to 2085 
population 2.  Again, once the difference in Vin between populations became sufficient, 2086 
population 1 was able to prevent Lsr activation in population 2.   2087 
3.7.3.4  Full population of cells with Lsr in a finite difference environment.  While 2088 
coupling two systems of ODEs to the same pool of extracellular AI-2 allowed us to 2089 
interrogate possible Lsr signal bifurcation of two different populations, we extended this 2090 
analysis to include a full population of cells assigned basal values from a log-normal 2091 
distribution instead of an effective binomial one.  This analysis was conducted using 2092 
mixed finite difference-agent based simulations, as described in the methods.   2093 
As discussed in the main text, Lsr induction was bimodal, with the exact balance 2094 
dependent on the variance of the basal distribution.  This likely arose due to cells with 2095 
higher values of basal from the leading edge of the distribution activating earlier and 2096 
earlier as the variance of the distribution increased.  The earlier the activation of these 2097 
cells relative to the mean, the faster extracellular AI-2 species were drawn down, thus 2098 
preventing the activation of cells with lower basal values, which themselves on average 2099 
were liable to activate later and later.   2100 
 
 
118 
 
 2101 
Fig. S5 - Fraction of cell population QS activated decreases as the variation of the 2102 
parameters Ksynth (A) and VydgG (B) increases. The influence of standard deviation of 2103 
the natural logarithm, , of the log-normal distributions on the fraction of the population 2104 
that was QS activated is represented here for standard deviation values ranging from 0 to 2105 
0.025, each run in triplicate.  Standard deviation of the natural logarithm for the log-2106 
normal distribution was used because the coefficient of variance is solely dependent on 2107 
this measure.  Dark lines represent the average values, whereas lighter surrounding lines 2108 
represent the standard deviation.   2109 
 
 
119 
 
Apart from basal, separately varying Ksynth and VydgG , the rate of AI-2 synthesis and AI-2 2110 
export, respectively, achieved similar changes to the bimodal nature of Lsr activation 2111 
(Figure 3-S5).  Although activation levels plateaued at different fractions of the 2112 
population when the same standard deviations were applied to the distribution of these 2113 
parameters, increased variation in both Ksynth and VydgG consistently decreased the fraction 2114 
of the population that was ultimately activated.  This serves to further indicate that 2115 
population level bimodal expression of the Lsr system may be a function of any 2116 
heterogeneity that desynchronizes AI-2 recompartmentalization. 2117 
3.7.3.5  Minimal role of spatial heterogeneity of AI-2.  In addition to evaluating the 2118 
role of heterogeneous expression at the population scale, whether spatially associated 2119 
stochasticity might influence bimodal expression was also inferred, mainly from 2120 
comparison of simulations using a standard finite difference scheme against simulations 2121 
where the entire environment was defined by a single element.  Treating the environment 2122 
homogenous made all AI-2 simultaneously available to all cells, whereas cells only 2123 
interacted with AI-2 in their own element using a standard finite difference approach.  As 2124 
shown in Figure 3-S6, the approaches yielded highly similar AI2-P trajectories when cell 2125 
motility was undirected. This was the case for all state variables modeled.  Furthermore, 2126 
in standard finite element environments, when governed by a single parameter set, cell 2127 
populations became wholly activated over a very small window (Figure 3-S5,  = 0). If 2128 
heterogeneity arising from spatial stochasticity influenced the bimodal phenotype, 2129 
population activation would be expected to be incomplete.  The absence of such an effect 2130 
implied that spatial stochasticity did not play a marked role in shaping bimodal response.      2131 
 
 
120 
 
 2132 
Figure 3-S6.  Comparison of results from single versus multiple finite difference 2133 
elements to define environment.  The average trajectory of AI2-P for cells with the 2134 
same parameter sets in simulations where the environment was defined as either a single 2135 
finite difference element or by the standard array of elements as defined in the methods.  2136 
Modeling with a single finite difference element eliminates spatial noise as a source of 2137 
difference between cells.  The addition of noise through the full implementation of finite 2138 
difference elements, adds spatially associated noise to the simulation.  This did not result 2139 
in a significant change in the average trajectory of AI2-P. 2140 
 2141 
 2142 
 2143 
 2144 
        2145 
 
 
121 
 
3.7.3.6  Agreement between numerical ODE and finite difference agent based 2146 
solutions.  As a general comment, the agreement between numerical ODE solutions and 2147 
the finite difference-agent based approach was inexact.  In particular, the time to 2148 
activation was offset between the two solutions as seen in Figure 3-S7A.  Nonetheless, 2149 
the solution trajectories were similar and an evaluation of the time to activation as a 2150 
function of basal indicated that parameter sensitivities between the solution approaches 2151 
were congruous, as seen in Figure 3-S7B.   2152 
3.7.3.7  Heterogeneity of local Lsr and LuxIR QS activation in growing colonies.  2153 
Smoothing of heterogenous input by LuxIR QS was not shared by Lsr QS.  This is 2154 
implied in Figure 3-S8A, which shows the distribution of key parameters from each 2155 
simulation type.  For LuxIR simulations (left), the uninduced rate of AHL synthesis was 2156 
varied among cells according to the depicted distribution. The higher the rate of synthesis 2157 
for uninduced cells, the faster the QS activation.  For Lsr simulations (right), the basal 2158 
rate of AI-2 influx was varied.  Here too, the higher the rate of basal AI-2 influx, the 2159 
faster the QS activation.  The red line in both distributions represents the average 2160 
parameter value for the first cell to activate from twenty simulations.  For both Lsr and 2161 
LuxIR simulations, the first cells to activate were on average all from the higher end of 2162 
the imposed heterogeneity.  For LuxIR QS (left), however, the first cells to activate were 2163 
not always those with the highest rate of basal AHL synthesis, as indicated by the wide 2164 
standard deviation.  This is clearer when compared to Lsr QS (right), where the first 2165 
activators were exclusively found at the tip of distribution for basal values (reflected in 2166 
the limited variance).  Essentially, while LuxIR/AHL activation smoothed out 2167 
heterogeneity associated with the rate of AHL production, Lsr/AI-2 dynamics were  2168 
 
 
122 
 
  2169 
Figure 3-S7.  Congruence of solution from finite difference-agent based modeling 2170 
versus implicit solution of pure ODE’s. A  AI2-P trajectory from implicit numerical 2171 
methods and the average AI2-P concentration from the finite difference-agent based 2172 
approach.  Here, cells from the finite-difference-agent based solution all held the same 2173 
parameter values as that from the pure ODE solution.  In the pure ODE approach, cells 2174 
were modeled as a dependent variable.  Ideally, the two solutions would bear identical 2175 
traces. B  The rate to activation was assessed by fitting the function, f(t), from 12-152 2176 
minutes to a first order linear regression, g(t).  The first time point at which f(t)-g(t)>2g(t) 2177 
was considered the point of activation.  The time to activation for each value of basal was 2178 
calculated and the bearing on the solution by the modeling and numerical method used 2179 
was evaluated by direct comparison along the primary axis and according to the ratio of 2180 
activation times for the finite difference-agent based solution to the pure ODE solution on 2181 
the secondary axis. 2182 
 
 
123 
 
 2183 
Figure 3-S8.  Measures of the difference between LuxIR and Lsr activation in the 2184 
context of colony growth.  A Histograms of randomly generated parameter values, each 2185 
with an event count of 10,000.  The average of the parameter values associated with the 2186 
first cell to QS activate in each simulation and its standard deviation from among 20 2187 
simulations are overlaid on the histogram in red. In LuxIR simulations, while the 2188 
selection is biased, a significant number of cells with less than the median parameter 2189 
value indicating that LuxIR has a smoothing effect on heterogeneity. B  The dark lines 2190 
represent the average local heterogeneity of 20 simulations, while the lighter, surrounding 2191 
shades represent the standard deviation of those values.  Inset is local heterogeneity of 2192 
QS activation, where active cells were placed randomly within the colony as a function of 2193 
percent of cells that were QS active for that colony (n=100).   2194 
 2195 
 2196 
 2197 
 2198 
 2199 
 
 
124 
 
unable to smooth similar non-genetic heterogeneity, instead producing population 2200 
desynchronization.  This difference in response to non-genetic heterogeneity was 2201 
presumably a reflection of topological differences between QS signaling modules.     2202 
This distinction between LuxIR and Lsr QS responses was also reflected in the 2203 
spatial heterogeneity of activation.  Local heterogeneity of QS activity was measured by 2204 
using a heuristic that scored highest when every cells’ neighbors, up or down, left or 2205 
right, were of the opposite QS state.  That is, a value of one indicated a perfect 2206 
checkerboard pattern of alternating activation, whereas a score of zero indicated that all 2207 
cells were of the same QS state.  That score was then averaged over the entire population 2208 
to arrive at the score reported in Figure 3-S8B. For perspective, inset is a graph of the 2209 
same measure for colonies whose QS active cells were randomly distributed, with the 2210 
ordinate axis reflecting the likelihood that any one cells was QS activated.  As the 2211 
measure was only concerned with changes between QS state, scoring was symmetric 2212 
about 0.5.   2213 
At time zero, all simulations began with a score of zero (QS unactivated).  Lsr/AI-2214 
2 simulations were run with a median basal of 487.8 and a coefficient of variance of 2215 
0.052 ( = 0.0225).  For such an Lsr/AI-2 population, the first non-zero local QS 2216 
heterogeneity values emerged near 120 minutes.  The average of twenty simulations 2217 
(dark blue line) and the standard deviation (surrounding light blue band) are depicted.  As 2218 
more cells became activated, local QS heterogeneity increased as both outlying and inner 2219 
cells were induced.  Local QS heterogeneity reached a plateau near 0.34 as QS activation 2220 
began to abate.  Given that the local QS heterogeneity for a cell colony wherein the same 2221 
fraction of QS active cells were completely randomly distributed was 0.4 (inset), the local 2222 
 
 
125 
 
QS heterogeneity for Lsr activation was high, while retaining some degree of non-2223 
randomness.  This stood in contrast to the LuxIR case.  LuxIR activation began near the 2224 
colony center at 225 minutes, and increased rapidly, reflecting the rapid expansion of 2225 
activation.  However, spatial heterogeneity scoring of LuxIR QS activity peaked near 0.1 2226 
despite a ~50% QS activation rate at that point.  Cell colonies where 50% of the cells 2227 
were QS active but where those cells were randomly placed within the colony averaged a 2228 
heterogeneity score near 0.5.  Low spatial heterogeneity associated with LuxIR/AHL 2229 
activation is attributed to two primary factors: QS activity originates from a single 2230 
centralized location and QS “On”/”Off” distinctions are quickly obliterated as QS active 2231 
cells turn on their QS inactive neighbors. 2232 
3.7.3.8  Evaluation of clustering when Lsr QS is coupled to AI-2 chemoattraction. 2233 
Cluster and dispersal patterns were observable from inspection of Figure 3-S9, where the 2234 
distance between cells across simulated time is shown for different swimming modes.  2235 
Here, fully functioning cells (green, Lsr + chemotaxis) were compared to non-2236 
chemotaxing populations (blue, randomly moving) or populations lacking the ability to 2237 
recompartmentalize AI-2 (purple, lsr operon negative and non-specific uptake minus; Vin 2238 
= 0, basal = 0).  These alternative populations represented groups of cells that did not 2239 
cluster or clustered but did not disperse, respectively.  Initially, the median minimal 2240 
distances for all population types were identical.  The median minimal distance between 2241 
cells decreased for all populations as a function of growth, as expected.  However, the 2242 
median minimal distance between cells for the two AI-2 chemotaxing populations 2243 
decreased more rapidly than their non-taxis counterpart due to clustering.  Cells unable to 2244 
recompartmentalize (or uptake) AI-2 had a higher net flux out (they synthesize by do not  2245 
 
 
126 
 
 2246 
Figure 3-S9. Clustering of cells with lsr activity and AI-2 chemoattraction as 2247 
measured by cell-cell distance.   The median minimal distance between cells for (i) a 2248 
population with both Lsr activity and AI-2 chemoattraction, (ii) a population lacking Lsr 2249 
recompartmentalization ability, and (iii) another population lacking AI-2 chemoattraction 2250 
but with recompartmentalization. Each condition was simulated twenty times, the average 2251 
median minimal distance between cells is represented by the darker line, while the 2252 
standard deviation of those values is represented by the surrounding lighter regions.  The 2253 
darker green box indicates the time over which the switch in median minimum distance 2254 
between cells from that of an AI-2 chemoattracted population to one of a population not 2255 
chemoattracted to AI-2 occurred for the wildtype population, while the lighter green 2256 
region indicates the time over which Lsr induction began for the wildtype population. 2257 
The bump around 250 minutes is indicative of this change.  Inset is a pictorial example of 2258 
the minimum distance between cells.   2259 
 2260 
 2261 
 2262 
 
 
127 
 
take in AI-2) than wildtype cells, achieving a higher extracellular concentration of AI-2 2263 
over a longer distance.  This may account for the slower or looser clustering of such cells 2264 
compared to the wildtype phenotype.  For wildtype AI-2 chemotaxing cells (Lsr+), the 2265 
median minimal distance between cells began increasing around 225 minutes a short 2266 
while after the population began to QS activate around 180 minutes.  The delay was 2267 
likely a function of the time required to begin drawing down the AI-2 concentration in 2268 
these clusters, while the increasing distance between cells reflected the dispersion 2269 
phenomena, which is apparent by 300 minutes. We note that the dispersion phenomena 2270 
here is underrepresented locally because the calculated value of the intercell distance is 2271 
averaged over the entire population, including all clusters and dispersed cells. In the end, 2272 
the distance between wildtype cells matched the distance between cells lacking AI-2 2273 
chemoattraction, indicating that the clusters had fully dispersed. While this is an exciting 2274 
outcome that could have broad ramifications, we know of no directly aligned 2275 
observations. 2276 
3.7.3.9  Motility mode feedback onto population activation as a function of cell-cell 2277 
distance.  In general, we found that among the populations simulated here, non-taxis 2278 
swimming populations were the slowest to QS activate. We note, however, that these 2279 
same cells ultimately achieved the largest proportion of stably QS activated cells. This 2280 
inverse correlation persisted across motility types (Figure 3-S10A).  Cells in simulations 2281 
of growing colonies activated the fastest but also experienced the smallest fraction of 2282 
stably activated cells, whereas chemotaxing populations experienced intermediate levels 2283 
of both speed to activation and of the final proportion activated.  In other words, as a 2284 
generalization, the higher the cell density the earlier the activation. However, higher 2285 
 
 
128 
 
density also appeared to produce stronger negative extracellular feedback from QS 2286 
activation as a smaller fraction of the population was ultimately activated. (Figure 3-2287 
S10B)  2288 
 2289 
 2290 
 2291 
 2292 
 2293 
 2294 
 2295 
 2296 
 2297 
 2298 
 2299 
 2300 
 2301 
 
 
129 
 
 2302 
Figure 3-S10.  Measures of the difference between different modes of motility when 2303 
coupled with Lsr/AI-2 dynamics.  A  The fraction of the population that was QS 2304 
activated over time, in simulations of different motility, with average values (n = 20) set 2305 
in a darker thinner line and a lighter surrounding shade representing the standard 2306 
deviation.  B  The median minimum distance between cells for populations influenced by 2307 
different combinations of motility and AI-2 uptake.  Dark lines are an average value (n = 2308 
20), while the surrounding lighter shades reflect the corresponding standard deviation.  2309 
For example, cells undergoing colony growth had a predefined, regular distance between 2310 
them, thus a single value prevailed across the entire time course and variability was zero.     2311 
 2312 
 2313 
 2314 
 2315 
 2316 
 2317 
 2318 
 2319 
 2320 
 
 
130 
 
Chapter 4: Conclusions 2321 
Living within consortia is likely the primary mode of existence for bacteria.  2322 
Therein, they must frequently negotiate numerous interactions with other species.  While 2323 
AI-2 is likely to be only one molecule within a vast sea by which such interactions are 2324 
mediated, due to the prevalence of LuxS 46 and YdgG 59 homologs, it is one that is likely 2325 
to be found in a plethora of ecological contexts.  This prevalence alone makes it a likely 2326 
candidate for QS and QS-like operations, the actual downstream functions of which are 2327 
expected to be manifold.  In an ecological context, QS and QS-like operations are likely 2328 
to mediate the most types of cell-cell interactions, consisting of cooperation among and 2329 
between species, coercion of one species’s action by another, and as cues from one 2330 
bacteria to another without the intent of cooperation.  Representing coercion, signal blind 2331 
mutants within an otherwise cooperating community cause signal responding bacteria to 2332 
QS.  When the QS response involves the production of public secreted goods, signal 2333 
blind mutants benefit without the metabolic expenditure and are coercing their signal 2334 
responding counterparts into, at the very least, premature QS.  Alternatively, purely as an 2335 
indicator of cell density, autoinducers could potentially serve as a cue.   2336 
The widespread phylogenetic signal of the Lsr system as described in the second 2337 
chapter suggests that the Lsr system plays a functional role in numerous bacteria.  The 2338 
exact nature of that role is possibly varied from species to species, especially considering 2339 
that it is unlikely that the affinity of LsrR for the intergenic region coevolved in an exact 2340 
manner.  As the Lsr system is known to affect biofilm development, it is believed to 2341 
influence cooperative behaviors at least to this extent 60.  In the third chapter, 2342 
mathematically modeling the form native to E. coli, we lend further credence to the idea 2343 
 
 
131 
 
that the Lsr system leads bacteria to serve a dual role, as both a cooperator and coercer as 2344 
a result of bimodal expression arising from population heterogeneity.   2345 
We also demonstrated through modeling, that bimodality results in multiple 2346 
emergent phenomenon depending on the mode of motility it is paired with, which itself 2347 
fed back onto the bimodal phenotype in silico.  As multiple homologs of the Lsr system 2348 
exist, it is highly likely that three categories of Lsr homolog exist relative to E. coli’s 2349 
system: more sensitive, equally sensitive, and less sensitive to AI-2.  In terms of speed to 2350 
activation these might roughly translate to: faster, equal, and slower to activate 2351 
respectively.  Here, we interrogated how bacteria containing such homologous systems 2352 
interact with a population possessing a base E. coli Lsr system, and suggest lsrFG 2353 
mutants, double lsrFG luxS mutants, and luxS mutants as representatives of each 2354 
respective category.  We showed how these competing populations could either largely 2355 
inhibit, act in concert with, or cause wholesale activation of the wildtype population, 2356 
respectively. For a luxS mutant population, the cost of coercion is the loss of the activated 2357 
methyl cycle, indicating that at least for this case there is a built in disadvantage to free 2358 
ridership.          2359 
Placing the Lsr system within the larger context of other QS architectures, our 2360 
modeling strongly indicates that consequent to bimodal activation, the patterns of 2361 
expression arising from the Lsr system are in stark contrast to those associated with 2362 
LuxIR QS.  Moreover, if placed in the same environment as the AI-2 activated TCRS, 2363 
LuxPQ, our studies indicate that Lsr activation would likely curtail LuxPQ based 2364 
signaling coordination, at least up to a point.  Indeed, in isolation the Lsr system operates 2365 
more closely to sugar importation systems than other QS systems.  Clearly, the 2366 
 
 
132 
 
distinction between sugar systems and the Lsr system is the context of self-production 2367 
and self-secretion.  Based on previous examination of QS 117,118, we believe that this 2368 
secretion is likely to result in greater coordination compared to a population with the 2369 
same average rate of AI2-P accretion but without YdgG or other means of AI-2 export.  2370 
Aside from this consideration, our homology search indicates that the Lsr system 2371 
phylogenetic signal is much less monophyletic than that for the lac system, even if it is 2372 
more widespread. 2373 
 2374 
 2375 
 2376 
 2377 
 2378 
 2379 
 2380 
 2381 
 2382 
 2383 
 2384 
 2385 
 
 
133 
 
References 2386 
 2387 
1. Greenberg, E., Hastings, J. & Ulitzur, S. Induction of luciferase synthesis in Beneckea 2388 
harveyi by other marine bacteria. Arch. Microbiol. 120, 87–91 (1979). 2389 
2. Eberhard, A. Inhibition and activation of bacterial luciferase synthesis. J. Bacteriol. 109, 2390 
1101–1105 (1972). 2391 
3. Fuqua, W. C., Winans, S. C. & Greenberg, E. P. Quorum sensing in bacteria: the LuxR-2392 
LuxI family of cell density-responsive transcriptional regulators. J. Bacteriol. 176, 269 2393 
(1994). 2394 
4. Hagen, S. J., Son, M., Weiss, J. T. & Young, J. H. Bacterium in a box: sensing of quorum 2395 
and environment by the LuxI/LuxR gene regulatory circuit. J. Biol. Phys. 36, 317–327 2396 
(2010). 2397 
5. Carnes, E. C. et al. Confinement-induced quorum sensing of individual Staphylococcus 2398 
aureus bacteria. Nat. Chem. Biol. 6, 41–45 (2009). 2399 
6. Boedicker, J. Q., Vincent, M. E. & Ismagilov, R. F. Microfluidic Confinement of Single 2400 
Cells of Bacteria in Small Volumes Initiates High‐Density Behavior of Quorum Sensing 2401 
and Growth and Reveals Its Variability. Angew. Chem. Int. Ed. 48, 5908–5911 (2009). 2402 
7. Alberghini, S. et al. Consequences of relative cellular positioning on quorum sensing and 2403 
bacterial cell‐to‐cell communication. FEMS Microbiol. Lett. 292, 149–161 (2009). 2404 
8. Hense, B. A. et al. Does efficiency sensing unify diffusion and quorum sensing? Nat. Rev. 2405 
Microbiol. 5, 230–239 (2007). 2406 
9. Redfield, R. J. Is quorum sensing a side effect of diffusion sensing? Trends Microbiol. 10, 2407 
365–370 (2002). 2408 
10. Diggle, S. P., Gardner, A., West, S. A. & Griffin, A. S. Evolutionary theory of bacterial 2409 
quorum sensing: when is a signal not a signal? Philos. Trans. R. Soc. B Biol. Sci. 362, 2410 
1241–1249 (2007). 2411 
11. Mascher, T., Helmann, J. D. & Unden, G. Stimulus perception in bacterial signal-2412 
transducing histidine kinases. Microbiol. Mol. Biol. Rev. 70, 910–938 (2006). 2413 
12. Case, R. J., Labbate, M. & Kjelleberg, S. AHL-driven quorum-sensing circuits: their 2414 
frequency and function among the Proteobacteria. ISME J. 2, 345 (2008). 2415 
13. Pereira, C. S., de Regt, A. K., Brito, P. H., Miller, S. T. & Xavier, K. B. Identification of 2416 
functional LsrB-like autoinducer-2 receptors. J. Bacteriol. 191, 6975–6987 (2009). 2417 
14. Quan, D. N. & Bentley, W. E. Gene network homology in prokaryotes using a similarity 2418 
search approach: Queries of quorum sensing signal transduction. PLoS Comput. Biol. 8, 2419 
e1002637 (2012). 2420 
15. Bassler, B. L. Small talk: cell-to-cell communication in bacteria. Cell 109, 421–424 (2002). 2421 
16. Anetzberger, C. et al. Autoinducers act as biological timers in Vibrio harveyi. PloS One 7, 2422 
e48310 (2012). 2423 
17. Tu, K. C., Long, T., Svenningsen, S. L., Wingreen, N. S. & Bassler, B. L. Negative 2424 
Feedback Loops Involving Small Regulatory RNAs Precisely Control the Vibrio harveyi 2425 
Quorum-Sensing Response. Mol. Cell 37, 567–579 (2010). 2426 
18. Tu, K. C., Waters, C. M., Svenningsen, S. L. & Bassler, B. L. A small‐RNA‐mediated 2427 
negative feedback loop controls quorum‐sensing dynamics in Vibrio harveyi. Mol. 2428 
Microbiol. 70, 896–907 (2008). 2429 
19. Winson, M. K. et al. Construction and analysis of luxCDABE‐based plasmid sensors for 2430 
investigating N‐acyl homoserine lactone‐mediated quorum sensing. FEMS Microbiol. Lett. 2431 
163, 185–192 (1998). 2432 
20. Swift, S. et al. Quorum sensing in Aeromonas hydrophila and Aeromonas salmonicida: 2433 
identification of the LuxRI homologs AhyRI and AsaRI and their cognate N-2434 
acylhomoserine lactone signal molecules. J. Bacteriol. 179, 5271–5281 (1997). 2435 
 
 
134 
 
21. Ng, W.-L. & Bassler, B. L. Bacterial quorum-sensing network architectures. Annu. Rev. 2436 
Genet. 43, 197–222 (2009). 2437 
22. Waters, C. M. & Bassler, B. L. Quorum sensing: cell-to-cell communication in bacteria. 2438 
Annu Rev Cell Dev Biol 21, 319–346 (2005). 2439 
23. Cao, J.-G. & Meighen, E. Purification and structural identification of an autoinducer for the 2440 
luminescence system of Vibrio harveyi. J. Biol. Chem. 264, 21670–21676 (1989). 2441 
24. Milton, D. L. et al. The LuxM Homologue VanM from Vibrio anguillarum directs the 2442 
Synthesis of N-(3-Hydroxyhexanoyl) homoserine Lactone and N-Hexanoylhomoserine 2443 
Lactone. J. Bacteriol. 183, 3537–3547 (2001). 2444 
25. Hanzelka, B. L. et al. Acylhomoserine lactone synthase activity of the Vibrio fischeri AinS 2445 
protein. J. Bacteriol. 181, 5766–5770 (1999). 2446 
26. Makino, K. et al. Genome sequence of Vibrio parahaemolyticus: a pathogenic mechanism 2447 
distinct from that of V cholerae. The Lancet 361, 743–749 (2003). 2448 
27. Sun, J., Daniel, R., Wagner-Döbler, I. & Zeng, A.-P. Is autoinducer-2 a universal signal for 2449 
interspecies communication: a comparative genomic and phylogenetic analysis of the 2450 
synthesis and signal transduction pathways. BMC Evol. Biol. 4, 36 (2004). 2451 
28. Schauder, S., Shokat, K., Surette, M. G. & Bassler, B. L. The LuxS family of bacterial 2452 
autoinducers: biosynthesis of a novel quorum‐sensing signal molecule. Mol. Microbiol. 41, 2453 
463–476 (2001). 2454 
29. Chen, X. et al. Structural identification of a bacterial quorum-sensing signal containing 2455 
boron. Nature 415, 545–549 (2002). 2456 
30. Tait, K., Hutchison, Z., Thompson, F. L. & Munn, C. B. Quorum sensing signal production 2457 
and inhibition by coral‐associated Vibrios. Environ. Microbiol. Rep. 2, 145–150 (2010). 2458 
31. Ng, W. et al. Signal production and detection specificity in Vibrio CqsA/CqsS quorum‐2459 
sensing systems. Mol. Microbiol. 79, 1407–1417 (2011). 2460 
32. Gray, K. M. & Garey, J. R. The evolution of bacterial LuxI and LuxR quorum sensing 2461 
regulators. Microbiology 147, 2379–2387 (2001). 2462 
33. Danino, T., Mondragón-Palomino, O., Tsimring, L. & Hasty, J. A synchronized quorum of 2463 
genetic clocks. Nature 463, 326–330 (2010). 2464 
34. Bulter, T. et al. Design of artificial cell–cell communication using gene and metabolic 2465 
networks. Proc. Natl. Acad. Sci. U. S. A. 101, 2299–2304 (2004). 2466 
35. Dai, Y., Toley, B. J., Swofford, C. A. & Forbes, N. S. Construction of an inducible cell‐2467 
communication system that amplifies Salmonella gene expression in tumor tissue. 2468 
Biotechnol. Bioeng. 110, 1769–1781 (2013). 2469 
36. Tamsir, A., Tabor, J. J. & Voigt, C. A. Robust multicellular computing using genetically 2470 
encoded NOR gates and chemical/wires/’. Nature 469, 212–215 (2011). 2471 
37. Saeidi, N. et al. Engineering microbes to sense and eradicate Pseudomonas aeruginosa, a 2472 
human pathogen. Mol. Syst. Biol. 7, (2011). 2473 
38. Danino, V. E., Wilkinson, A., Edwards, A. & Downie, J. A. Recipient‐induced transfer of 2474 
the symbiotic plasmid pRL1JI in Rhizobium leguminosarum bv. viciae is regulated by a 2475 
quorum‐sensing relay. Mol. Microbiol. 50, 511–525 (2003). 2476 
39. Ahmer, B. M., van Reeuwijk, J., Timmers, C. D., Valentine, P. J. & Heffron, F. Salmonella 2477 
typhimurium encodes an SdiA homolog, a putative quorum sensor of the LuxR family, that 2478 
regulates genes on the virulence plasmid. J. Bacteriol. 180, 1185–1193 (1998). 2479 
40. Yao, Y. et al. Structure of the Escherichia coli Quorum Sensing Protein SdiA: Activation of 2480 
the Folding Switch by Acyl Homoserine Lactones. J. Mol. Biol. 355, 262–273 (2006). 2481 
41. Houdt, R., Aertsen, A., Moons, P., Vanoirbeek, K. & Michiels, C. W. N‐acyl‐l‐homoserine 2482 
lactone signal interception by Escherichia coli. FEMS Microbiol. Lett. 256, 83–89 (2006). 2483 
42. Michael, B., Smith, J. N., Swift, S., Heffron, F. & Ahmer, B. M. SdiA of Salmonella 2484 
enterica is a LuxR homolog that detects mixed microbial communities. J. Bacteriol. 183, 2485 
5733–5742 (2001). 2486 
 
 
135 
 
43. Smith, J. N. & Ahmer, B. M. Detection of other microbial species by Salmonella: 2487 
expression of the SdiA regulon. J. Bacteriol. 185, 1357–1366 (2003). 2488 
44. Sharma, V. K., Bearson, S. M. & Bearson, B. L. Evaluation of the effects of sdiA, a luxR 2489 
homologue, on adherence and motility of Escherichia coli O157: H7. Microbiology 156, 2490 
1303–1312 (2010). 2491 
45. Dyszel, J. L. et al. E. coli K-12 and EHEC genes regulated by SdiA. PLoS One 5, e8946 2492 
(2010). 2493 
46. Winzer, K. et al. LuxS: its role in central metabolism and the in vitro synthesis of 4-2494 
hydroxy-5-methyl-3 (2H)-furanone. Microbiology 148, 909–922 (2002). 2495 
47. Bodor, A., Elxnat, B., Thiel, V., Schulz, S. & Wagner-Döbler, I. Potential for luxS related 2496 
signalling in marine bacteria and production of autoinducer-2 in the genus Shewanella. 2497 
BMC Microbiol. 8, 13 (2008). 2498 
48. Tavender, T. J., Halliday, N. M., Hardie, K. R. & Winzer, K. LuxS-independent formation 2499 
of AI-2 from ribulose-5-phosphate. BMC Microbiol. 8, 98 (2008). 2500 
49. Li, J. et al. A stochastic model of Escherichia coli AI‐2 quorum signal circuit reveals 2501 
alternative synthesis pathways. Mol. Syst. Biol. 2, (2006). 2502 
50. Meijler, M. M. et al. Synthesis and biological validation of a ubiquitous quorum‐sensing 2503 
molecule. Angew. Chem. Int. Ed. 43, 2106–2108 (2004). 2504 
51. Vendeville, A., Winzer, K., Heurlier, K., Tang, C. M. & Hardie, K. R. Making’sense’of 2505 
metabolism: autoinducer-2, LuxS and pathogenic bacteria. Nat. Rev. Microbiol. 3, 383–396 2506 
(2005). 2507 
52. Daubin, V., Gouy, M. & Perriere, G. A phylogenomic approach to bacterial phylogeny: 2508 
evidence of a core of genes sharing a common history. Genome Res. 12, 1080–1090 (2002). 2509 
53. Taga, M. E., Semmelhack, J. L. & Bassler, B. L. The LuxS‐dependent autoinducer AI‐2 2510 
controls the expression of an ABC transporter that functions in AI‐2 uptake in Salmonella 2511 
typhimurium. Mol. Microbiol. 42, 777–793 (2001). 2512 
54. Taga, M. E., Miller, S. T. & Bassler, B. L. Lsr‐mediated transport and processing of AI‐2 in 2513 
Salmonella typhimurium. Mol. Microbiol. 50, 1411–1427 (2003). 2514 
55. Wang, L., Hashimoto, Y., Tsao, C.-Y., Valdes, J. J. & Bentley, W. E. Cyclic AMP (cAMP) 2515 
and cAMP receptor protein influence both synthesis and uptake of extracellular autoinducer 2516 
2 in Escherichia coli. J. Bacteriol. 187, 2066–2076 (2005). 2517 
56. Wang, L., Li, J., March, J. C., Valdes, J. J. & Bentley, W. E. luxS-dependent gene 2518 
regulation in Escherichia coli K-12 revealed by genomic expression profiling. J. Bacteriol. 2519 
187, 8350–8360 (2005). 2520 
57. Xavier, K. B. & Bassler, B. L. Regulation of uptake and processing of the quorum-sensing 2521 
autoinducer AI-2 in Escherichia coli. J. Bacteriol. 187, 238–248 (2005). 2522 
58. Herzberg, M., Kaye, I. K., Peti, W. & Wood, T. K. YdgG (TqsA) controls biofilm 2523 
formation in Escherichia coli K-12 through autoinducer 2 transport. J. Bacteriol. 188, 587–2524 
598 (2006). 2525 
59. Rettner, R. E. & Saier Jr, M. H. The autoinducer-2 exporter superfamily. J. Mol. Microbiol. 2526 
Biotechnol. 18, 195–205 (2010). 2527 
60. Li, J. et al. Quorum sensing in Escherichia coli is signaled by AI-2/LsrR: effects on small 2528 
RNA and biofilm architecture. J. Bacteriol. 189, 6011–6020 (2007). 2529 
61. Shao, H., James, D., Lamont, R. J. & Demuth, D. R. Differential interaction of 2530 
Aggregatibacter (Actinobacillus) actinomycetemcomitans LsrB and RbsB proteins with 2531 
autoinducer 2. J. Bacteriol. 189, 5559–5565 (2007). 2532 
62. Rader, B. A. et al. Helicobacter pylori perceives the quorum-sensing molecule AI-2 as a 2533 
chemorepellent via the chemoreceptor TlpB. Microbiology 157, 2445–2455 (2011). 2534 
63. Pereira, C. S. et al. Phosphoenolpyruvate phosphotransferase system regulates detection and 2535 
processing of the quorum sensing signal autoinducer‐2. Mol. Microbiol. 84, 93–104 (2012). 2536 
 
 
136 
 
64. Thompson, J. Galactose transport systems in Streptococcus lactis. J. Bacteriol. 144, 683–2537 
691 (1980). 2538 
65. Henderson, P. J. Proton-linked sugar transport systems in bacteria. J. Bioenerg. Biomembr. 2539 
22, 525–569 (1990). 2540 
66. Chaudhuri, B. N., Ko, J., Park, C., Jones, T. A. & Mowbray, S. L. Structure of d-allose 2541 
binding protein from Escherichia coli bound to d-allose at 1.8 Å resolution. J. Mol. Biol. 2542 
286, 1519–1531 (1999). 2543 
67. Shao, H., Lamont, R. J. & Demuth, D. R. Autoinducer 2 is required for biofilm growth of 2544 
Aggregatibacter (Actinobacillus) actinomycetemcomitans. Infect. Immun. 75, 4211–4218 2545 
(2007). 2546 
68. Armbruster, C. E. et al. RbsB (NTHI_0632) mediates quorum signal uptake in nontypeable 2547 
Haemophilus influenzae strain 86‐028NP. Mol. Microbiol. 82, 836–850 (2011). 2548 
69. Thijs, I. M. et al. The AI-2-dependent regulator LsrR has a limited regulon in Salmonella 2549 
Typhimurium. Cell Res. 20, 966–969 (2010). 2550 
70. Byrd, C. M. Local and global gene regulation analysis of the autoinducer-2 mediated 2551 
quorum sensing mechanism in Escherichia coli. (2011). 2552 
71. Pesavento, C. et al. Inverse regulatory coordination of motility and curli-mediated adhesion 2553 
in Escherichia coli. Genes Dev. 22, 2434–2446 (2008). 2554 
72. Park, J. T., Raychaudhuri, D., Li, H., Normark, S. & Mengin-Lecreulx, D. MppA, a 2555 
Periplasmic Binding Protein Essential for Import of the Bacterial Cell Wall Peptidel-2556 
Alanyl-γ-d-Glutamyl-meso-Diaminopimelate. J. Bacteriol. 180, 1215–1223 (1998). 2557 
73. Diaz, Z., Xavier, K. B. & Miller, S. T. The crystal structure of the Escherichia coli 2558 
autoinducer-2 processing protein LsrF. PloS One 4, e6820 (2009). 2559 
74. Marques, J. C. et al. Processing the Interspecies Quorum-sensing Signal Autoinducer-2 (AI-2560 
2) Characterization phospho-(S)-4, 5-dihydroxy-2, 3-pentanedione isomerization by LsrG 2561 
protein. J. Biol. Chem. 286, 18331–18343 (2011). 2562 
75. Xavier, K. B. et al. Phosphorylation and processing of the quorum-sensing molecule 2563 
autoinducer-2 in enteric bacteria. ACS Chem. Biol. 2, 128–136 (2007). 2564 
76. Pellicer, M. T., Nuñez, M. F., Aguilar, J., Badia, J. & Baldoma, L. Role of 2-2565 
phosphoglycolate phosphatase of Escherichia coli in metabolism of the 2-phosphoglycolate 2566 
formed in DNA repair. J. Bacteriol. 185, 5815–5821 (2003). 2567 
77. Choi, J. et al. LsrR-mediated quorum sensing controls invasiveness of Salmonella 2568 
typhimurium by regulating SPI-1 and flagella genes. PloS One 7, e37059 (2012). 2569 
78. Niu, C. et al. LuxS influences Escherichia coli biofilm formation through autoinducer‐2‐2570 
dependent and autoinducer‐2‐independent modalities. FEMS Microbiol. Ecol. 83, 778–791 2571 
(2013). 2572 
79. Torres-Escobar, A., Juárez-Rodríguez, M. D., Lamont, R. J. & Demuth, D. R. 2573 
Transcriptional regulation of Aggregatibacter actinomycetemcomitans lsrACDBFG and 2574 
lsrRK operons and their role in biofilm formation. J. Bacteriol. 195, 56–65 (2013). 2575 
80. Darch, S. E., West, S. A., Winzer, K. & Diggle, S. P. Density-dependent fitness benefits in 2576 
quorum-sensing bacterial populations. Proc. Natl. Acad. Sci. 109, 8259–8263 (2012). 2577 
81. Krin, E. et al. Pleiotropic role of quorum-sensing autoinducer 2 in Photorhabdus 2578 
luminescens. Appl. Environ. Microbiol. 72, 6439–6451 (2006). 2579 
82. Auger, S., Krin, E., Aymerich, S. & Gohar, M. Autoinducer 2 affects biofilm formation by 2580 
Bacillus cereus. Appl. Environ. Microbiol. 72, 937–941 (2006). 2581 
83. Li, L. et al. Analysis on Actinobacillus pleuropneumoniae LuxS regulated genes reveals 2582 
pleiotropic roles of LuxS/AI-2 on biofilm formation, adhesion ability and iron metabolism. 2583 
Microb. Pathog. 50, 293–302 (2011). 2584 
84. Vidal, J. E., Ludewick, H. P., Kunkel, R. M., Zähner, D. & Klugman, K. P. The LuxS-2585 
dependent quorum-sensing system regulates early biofilm formation by Streptococcus 2586 
pneumoniae strain D39. Infect. Immun. 79, 4050–4060 (2011). 2587 
 
 
137 
 
85. Li, M., Villaruz, A. E., Vadyvaloo, V., Sturdevant, D. E. & Otto, M. AI-2-dependent gene 2588 
regulation in Staphylococcus epidermidis. BMC Microbiol. 8, 4 (2008). 2589 
86. Ahmed, N. A., Petersen, F. C. & Scheie, A. A. AI-2/LuxS is involved in increased biofilm 2590 
formation by Streptococcus intermedius in the presence of antibiotics. Antimicrob. Agents 2591 
Chemother. 53, 4258–4263 (2009). 2592 
87. Von Lackum, K. et al. Functionality of Borrelia burgdorferi LuxS: The Lyme disease 2593 
spirochete produces and responds to the pheromone autoinducer-2 and lacks a complete 2594 
activated-methyl cycle. Int. J. Med. Microbiol. 296, 92–102 (2006). 2595 
88. Rader, B. A., Campagna, S. R., Semmelhack, M. F., Bassler, B. L. & Guillemin, K. The 2596 
quorum-sensing molecule autoinducer 2 regulates motility and flagellar morphogenesis in 2597 
Helicobacter pylori. J. Bacteriol. 189, 6109–6117 (2007). 2598 
89. Pereira, C. S., McAuley, J. R., Taga, M. E., Xavier, K. B. & Miller, S. T. Sinorhizobium 2599 
meliloti, a bacterium lacking the autoinducer‐2 (AI‐2) synthase, responds to AI‐2 supplied 2600 
by other bacteria. Mol. Microbiol. 70, 1223–1235 (2008). 2601 
90. Duan, K., Dammel, C., Stein, J., Rabin, H. & Surette, M. G. Modulation of Pseudomonas 2602 
aeruginosa gene expression by host microflora through interspecies communication. Mol. 2603 
Microbiol. 50, 1477–1491 (2003). 2604 
91. Geier, H., Mostowy, S., Cangelosi, G. A., Behr, M. A. & Ford, T. E. Autoinducer-2 triggers 2605 
the oxidative stress response in Mycobacterium avium, leading to biofilm formation. Appl. 2606 
Environ. Microbiol. 74, 1798–1804 (2008). 2607 
92. Strassmann, J. E., Gilbert, O. M. & Queller, D. C. Kin discrimination and cooperation in 2608 
microbes. Annu. Rev. Microbiol. 65, 349–367 (2011). 2609 
93. Wu, K. & Rao, C. V. The role of configuration and coupling in autoregulatory gene circuits. 2610 
Mol. Microbiol. 75, 513–527 (2010). 2611 
94. Tsao, C.-Y., Hooshangi, S., Wu, H.-C., Valdes, J. J. & Bentley, W. E. Autonomous 2612 
induction of recombinant proteins by minimally rewiring native quorum sensing regulon of 2613 
E. coli. Metab. Eng. 12, 291–297 (2010). 2614 
95. Gonzalez Barrios, A. F. & Achenie, L. E. Escherichia coli autoinducer-2 uptake network 2615 
does not display hysteretic behavior but AI-2 synthesis rate controls transient bifurcation. 2616 
Biosystems 99, 17–26 (2010). 2617 
96. Englert, D. L., Manson, M. D. & Jayaraman, A. Flow-based microfluidic device for 2618 
quantifying bacterial chemotaxis in stable, competing gradients. Appl. Environ. Microbiol. 2619 
75, 4557–4564 (2009). 2620 
97. Hegde, M. et al. Chemotaxis to the quorum-sensing signal AI-2 requires the Tsr 2621 
chemoreceptor and the periplasmic LsrB AI-2-binding protein. J. Bacteriol. 193, 768–773 2622 
(2011). 2623 
98. Zahedmanesh, H. & Lally, C. A multiscale mechanobiological modelling framework using 2624 
agent-based models and finite element analysis: application to vascular tissue engineering. 2625 
Biomech. Model. Mechanobiol. 11, 363–377 (2012). 2626 
99. Suyama, M. & Bork, P. Evolution of prokaryotic gene order: genome rearrangements in 2627 
closely related species. Trends Genet. 17, 10–13 (2001). 2628 
100. Moreno-Hagelsieb, G. & Collado-Vides, J. A powerful non-homology method for the 2629 
prediction of operons in prokaryotes. Bioinformatics 18, S329–S336 (2002). 2630 
101. Overbeek, R., Fonstein, M., D’souza, M., Pusch, G. D. & Maltsev, N. The use of gene 2631 
clusters to infer functional coupling. Proc. Natl. Acad. Sci. 96, 2896–2901 (1999). 2632 
102. Touchon, M. et al. Organised genome dynamics in the Escherichia coli species results in 2633 
highly diverse adaptive paths. PLoS Genet. 5, e1000344 (2009). 2634 
103. Fang, G., Rocha, E. P. & Danchin, A. Persistence drives gene clustering in bacterial 2635 
genomes. BMC Genomics 9, 4 (2008). 2636 
104. Harrington, E. D., Jensen, L. J. & Bork, P. Predicting biological networks from genomic 2637 
data. FEBS Lett. 582, 1251–1258 (2008). 2638 
 
 
138 
 
105. Rogozin, I. B., Makarova, K. S., Wolf, Y. I. & Koonin, E. V. Computational approaches for 2639 
the analysis of gene neighbourhoods in prokaryotic genomes. Brief. Bioinform. 5, 131–149 2640 
(2004). 2641 
106. Altschul, S. F., Gish, W., Miller, W., Myers, E. W. & Lipman, D. J. Basic local alignment 2642 
search tool. J. Mol. Biol. 215, 403–410 (1990). 2643 
107. Stajich, J. E. et al. The Bioperl toolkit: Perl modules for the life sciences. Genome Res. 12, 2644 
1611–1618 (2002). 2645 
108. Williams, K. P. et al. Phylogeny of gammaproteobacteria. J. Bacteriol. 192, 2305–2314 2646 
(2010). 2647 
109. Roderick, S. L. The lac operon galactoside acetyltransferase. C. R. Biol. 328, 568–575 2648 
(2005). 2649 
110. Arraj, J. A. & Campbell, J. H. Isolation and characterization of the newly evolved ebg beta-2650 
galactosidase of Escherichia coli K-12. J. Bacteriol. 124, 849–856 (1975). 2651 
111. Stoebel, D. M. Lack of evidence for horizontal transfer of the lac operon into Escherichia 2652 
coli. Mol. Biol. Evol. 22, 683–690 (2005). 2653 
112. Hooshangi, S. & Bentley, W. E. LsrR quorum sensing ‘switch’ is revealed by a bottom-up 2654 
approach. PLoS Comput. Biol. 7, e1002172 (2011). 2655 
113. Studier, F. W., Daegelen, P., Lenski, R. E., Maslov, S. & Kim, J. F. Understanding the 2656 
Differences between Genome Sequences of Escherichia coli B Strains REL606 and BL21 2657 
(DE3) and Comparison of the E. coli B and K-12 Genomes. J. Mol. Biol. 394, 653–680 2658 
(2009). 2659 
114. Xue, T., Zhao, L., Sun, H., Zhou, X. & Sun, B. LsrR-binding site recognition and regulatory 2660 
characteristics in Escherichia coli AI-2 quorum sensing. Cell Res. 19, 1258–1268 (2009). 2661 
115. Price, M. N., Dehal, P. S. & Arkin, A. P. Horizontal gene transfer and the evolution of 2662 
transcriptional regulation in Escherichia coli. Genome Biol 9, R4 (2008). 2663 
116. Mavromatis, K. et al. Complete genome sequence of Spirochaeta smaragdinae type strain 2664 
(SEBR 4228T). Stand. Genomic Sci. 3, 136 (2010). 2665 
117. Tanouchi, Y., Tu, D., Kim, J. & You, L. Noise reduction by diffusional dissipation in a 2666 
minimal quorum sensing motif. PLoS Comput. Biol. 4, e1000167 (2008). 2667 
118. Tabareau, N., Slotine, J.-J. & Pham, Q.-C. How synchronization protects from noise. PLoS 2668 
Comput. Biol. 6, e1000637 (2010). 2669 
119. Anetzberger, C., Pirch, T. & Jung, K. Heterogeneity in quorum sensing‐regulated 2670 
bioluminescence of Vibrio harveyi. Mol. Microbiol. 73, 267–277 (2009). 2671 
120. Sayut, D. J., Kambam, P. K. R. & Sun, L. Noise and kinetics of LuxR positive feedback 2672 
loops. Biochem. Biophys. Res. Commun. 363, 667–673 (2007). 2673 
121. Kittisopikul, M. & Süel, G. M. Biological role of noise encoded in a genetic network motif. 2674 
Proc. Natl. Acad. Sci. 107, 13300–13305 (2010). 2675 
122. Fenley, A. T., Banik, S. K. & Kulkarni, R. V. Computational modeling of differences in the 2676 
quorum sensing induced luminescence phenotypes of Vibrio harveyi and Vibrio cholerae. J. 2677 
Theor. Biol. 274, 145–153 (2011). 2678 
123. Marques, J. C. et al. LsrF, a coenzyme A-dependent thiolase, catalyzes the terminal step in 2679 
processing the quorum sensing signal autoinducer-2. Proc. Natl. Acad. Sci. 111, 14235–2680 
14240 (2014). 2681 
124. Parsek, M. R. & Greenberg, E. Sociomicrobiology: the connections between quorum 2682 
sensing and biofilms. Trends Microbiol. 13, 27–33 (2005). 2683 
125. Defoirdt, T. Can bacteria actively search to join groups? ISME J.-Int. Soc. Microb. Ecol. 5, 2684 
569 (2011). 2685 
126. Sourjik, V. & Armitage, J. P. Spatial organization in bacterial chemotaxis. EMBO J. 29, 2686 
2724–2733 (2010). 2687 
127. Dockery, J. D. & Keener, J. P. A mathematical model for quorum sensing in Pseudomonas 2688 
aeruginosa. Bull. Math. Biol. 63, 95–116 (2001). 2689 
 
 
139 
 
128. Bentley, W. E., Mirjalili, N., Andersen, D. C., Davis, R. H. & Kompala, D. S. Plasmid‐2690 
encoded protein: the principal factor in the ‘metabolic burden’ associated with recombinant 2691 
bacteria. Biotechnol. Bioeng. 35, 668–681 (1990). 2692 
129. Melke, P., Sahlin, P., Levchenko, A. & Jönsson, H. A cell-based model for quorum sensing 2693 
in heterogeneous bacterial colonies. PLoS Comput. Biol. 6, e1000819 (2010). 2694 
130. Mittal, N., Budrene, E. O., Brenner, M. P. & van Oudenaarden, A. Motility of Escherichia 2695 
coli cells in clusters formed by chemotactic aggregation. Proc. Natl. Acad. Sci. 100, 13259–2696 
13263 (2003). 2697 
131. Wu, H. et al. Autonomous bacterial localization and gene expression based on nearby cell 2698 
receptor density. Mol. Syst. Biol. 9, (2013). 2699 
132. Albert, R. & Othmer, H. G. The topology of the regulatory interactions predicts the 2700 
expression pattern of the segment polarity genes in Drosophila melanogaster. J. Theor. 2701 
Biol. 223, 1–18 (2003). 2702 
133. Lawrence, P. A., Casal, J. & Struhl, G. hedgehog and engrailed: pattern formation and 2703 
polarity in the Drosophila abdomen. Development 126, 2431–2439 (1999). 2704 
134. Birtwistle, M. R. et al. Emergence of bimodal cell population responses from the interplay 2705 
between analog single-cell signaling and protein expression noise. BMC Syst. Biol. 6, 109 2706 
(2012). 2707 
135. Shalek, A. K. et al. Single-cell transcriptomics reveals bimodality in expression and 2708 
splicing in immune cells. Nature (2013). 2709 
136. Veening, J.-W., Smits, W. K. & Kuipers, O. P. Bistability, epigenetics, and bet-hedging in 2710 
bacteria. Annu Rev Microbiol 62, 193–210 (2008). 2711 
137. De Beer, D. & Stoodley, P. Relation between the structure of an aerobic biofilm and 2712 
transport phenomena. Water Sci. Technol. 32, 11–18 (1995). 2713 
138. Coulthurst, S. J., Kurz, C. L. & Salmond, G. P. luxS mutants of Serratia defective in 2714 
autoinducer-2-dependent ‘quorum sensing’show strain-dependent impacts on virulence and 2715 
production of carbapenem and prodigiosin. Microbiology 150, 1901–1910 (2004). 2716 
139. Høyland-Kroghsbo, N. M., Mærkedahl, R. B. & Svenningsen, S. L. A quorum-sensing-2717 
induced bacteriophage defense mechanism. MBio 4, e00362–12 (2013). 2718 
140. Sun, S., Kjelleberg, S. & McDougald, D. Relative contributions of Vibrio polysaccharide 2719 
and quorum sensing to the resistance of Vibrio cholerae to predation by heterotrophic 2720 
protists. PloS One 8, e56338 (2013). 2721 
141. Diggle, S. P., Griffin, A. S., Campbell, G. S. & West, S. A. Cooperation and conflict in 2722 
quorum-sensing bacterial populations. Nature 450, 411–414 (2007). 2723 
142. Dandekar, A. A., Chugani, S. & Greenberg, E. P. Bacterial quorum sensing and metabolic 2724 
incentives to cooperate. Science 338, 264–266 (2012). 2725 
143. Byrd, C. M. & Bentley, W. E. Quieting cross talk - the quorum sensing regulator LsrR as a 2726 
possible target for fighting bacterial infections. Cell Res 19, 1229–1230 2727 
144. Luo, X. et al. Biofabrication of stratified biofilm mimics for observation and control of 2728 
bacterial signaling. Biomaterials 33, 5136–5143 (2012). 2729 
145. Goryachev, A. B. Understanding Bacterial Cell− Cell Communication with Computational 2730 
Modeling. Chem Rev 111, 238–250 (2011). 2731 
146. Wilke, C. & Chang, P. Correlation of diffusion coefficients in dilute solutions. AIChE J. 1, 2732 
264–270 (1955). 2733 
147. Terrell, J. L. et al. Integrated biofabrication for electro‐addressed in‐film bioprocessing. 2734 
Biotechnol. J. 7, 428–439 (2012). 2735 
148. Becskei, A., Séraphin, B. & Serrano, L. Positive feedback in eukaryotic gene networks: cell 2736 
differentiation by graded to binary response conversion. EMBO J. 20, 2528–2535 (2001). 2737 
149. Ozbudak, E. M., Thattai, M., Lim, H. N., Shraiman, B. I. & Van Oudenaarden, A. 2738 
Multistability in the lactose utilization network of Escherichia coli. Nature 427, 737–740 2739 
(2004). 2740 
 
 
140 
 
150. Zhu, J. & Pei, D. A LuxP-based fluorescent sensor for bacterial autoinducer II. ACS Chem. 2741 
Biol. 3, 110–119 (2008). 2742 
151. Wu, M., Tao, Y., Liu, X. & Zang, J. Structural basis for phosphorylated autoinducer-2 2743 
modulation of the oligomerization state of the global transcription regulator LsrR from 2744 
Escherichia coli. J. Biol. Chem. 288, 15878–15887 (2013). 2745 
152. Ganguly, A., Rajdev, P., Williams, S. M. & Chatterji, D. Nonspecific Interaction between 2746 
DNA and Protein allows for Cooperativity: A Case Study with Mycobacterium DNA 2747 
Binding Protein. J. Phys. Chem. B 116, 621–632 (2011). 2748 
153. Von Dassow, G., Meir, E., Munro, E. M. & Odell, G. M. The segment polarity network is a 2749 
robust developmental module. Nature 406, 188–192 (2000). 2750 
154. Matsumoto, M. & Nishimura, T. Mersenne twister: a 623-dimensionally equidistributed 2751 
uniform pseudo-random number generator. ACM Trans. Model. Comput. Simul. TOMACS 2752 
8, 3–30 (1998). 2753 
 2754