Transmission of Vibrio cholerae Is Antagonized by LyticPhage and Entry into the Aquatic EnvironmentEric J. Nelson1, Ashrafuzzaman Chowdhury2, James Flynn3, Stefan Schild4, Lori Bourassa1, Yue Shao3,

Regina C. LaRocque5, Stephen B. Calderwood5, Firdausi Qadri6, Andrew Camilli1*

1 Howard Hughes Medical Institute and the Department of Molecular Biology and Microbiology, Tufts University School of Medicine, Boston, Massachusetts, United States

of America, 2 Microbiology Department, Jahangirnagar University, Savar, Dhaka, Bangladesh, 3 Tufts Expression Array Core (TEAC) Facility, Tufts University School of

Medicine, Boston, Massachusetts, United States of America, 4 Institut fuer Molekulare Biowissenschaften, Karl-Franzens-Universitaet Graz, Graz, Austria, 5 Division of

Infectious Diseases, Massachusetts General Hospital, Boston, Massachusetts, United States of America, and Harvard Medical School, Boston, Massachusetts, United States

of America, 6 International Centre for Diarrhoeal Disease Research, Dhaka, Bangladesh

Abstract

Cholera outbreaks are proposed to propagate in explosive cycles powered by hyperinfectious Vibrio cholerae and quenchedby lytic vibriophage. However, studies to elucidate how these factors affect transmission are lacking because the fieldexperiments are almost intractable. One reason for this is that V. cholerae loses the ability to culture upon transfer to pondwater. This phenotype is called the active but non-culturable state (ABNC; an alternative term is viable but non-culturable)because these cells maintain the capacity for metabolic activity. ABNC bacteria may serve as the environmental reservoir foroutbreaks but rigorous animal studies to test this hypothesis have not been conducted. In this project, we wanted todetermine the relevance of ABNC cells to transmission as well as the impact lytic phage have on V. cholerae as the bacteriaenter the ABNC state. Rice-water stool that naturally harbored lytic phage or in vitro derived V. cholerae were incubated in apond microcosm, and the culturability, infectious dose, and transcriptome were assayed over 24 h. The data show that themajor contributors to infection are culturable V. cholerae and not ABNC cells. Phage did not affect colonization immediatelyafter shedding from the patients because the phage titer was too low. However, V. cholerae failed to colonize the smallintestine after 24 h of incubation in pond water—the point when the phage and ABNC cell titers were highest. Thetranscriptional analysis traced the transformation into the non-infectious ABNC state and supports models for theadaptation to nutrient poor aquatic environments. Phage had an undetectable impact on this adaptation. Taken together,the rise of ABNC cells and lytic phage blocked transmission. Thus, there is a fitness advantage if V. cholerae can make a rapidtransfer to the next host before these negative selective pressures compound in the aquatic environment.

Citation: Nelson EJ, Chowdhury A, Flynn J, Schild S, Bourassa L, et al. (2008) Transmission of Vibrio cholerae Is Antagonized by Lytic Phage and Entry into theAquatic Environment. PLoS Pathog 4(10): e1000187. doi:10.1371/journal.ppat.1000187

Editor: Frederick M. Ausubel, Massachusetts General Hospital, United States of America

Received June 13, 2008; Accepted September 24, 2008; Published October 24, 2008

Copyright: � 2008 Nelson et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permitsunrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Funding: The research was supported by the National Institutes of Health R01 AI 055058 (A. C.), UO1 AI 058935 (S. B. C.), RO3 AI063079 (F. Q.), and InternationalResearch Scientist Development Award K01 TW007144 (R. C. L.). A. Camilli is a Howard Hughes Medical Institute investigator. E. J. Nelson is a recipient of theFogarty/Ellison Fellowship in Global Health awarded by the Fogarty International Center at the National Institutes of Health (D43 TW005572).

Competing Interests: The authors have declared that no competing interests exist.

* E-mail: andrew.camilli@tufts.edu

Introduction

Diarrheal disease is the second most common cause of death

among children under 5 years of age globally – it is the leading

cause of morbidity [1,2]. The Gram-negative bacterium Vibrio

cholerae is a facultative pathogen having both human and

environmental stages, and is the etiologic agent of the secretory

diarrheal disease cholera [3]. Today, the burden of cholera is

estimated to reach several million cases a year in both Asia and

Africa, with fewer cases in Latin America [4]. Aquatic reservoirs

harbor V. cholerae during extended periods between outbreaks [5],

but there is little known about how fast V. cholerae moves from one

patient to the next during an outbreak. Transmission between

patients may be quite rapid. For example, two devastating

outbreaks strike Dhaka, Bangladesh annually. The high burden

of disease [6], collapsed water infrastructure, poverty, and

crowding make Dhaka an ideal setting for the fast transmission

of a facultative pathogen such as V. cholerae. At the host population

level, first degree relatives in households are more likely to be

infected with V. cholerae [7]. At the pathogen level, the di-annual

cholera outbreaks may be clonal [8,9,10], and there are rapid

shifts in drug resistance patterns [11,12]. Despite these epidemi-

ological observations that support a model for rapid transmission

during an outbreak, little is known about the selective forces that

drive facultative pathogens – like V. cholerae – out of the

environment and into the next host.

Using the infant-mouse model of cholera, we recently

demonstrated that genes induced late in the infection provide a

fitness advantage for the transition to aquatic environments [13].

In this study, V. cholerae from cholera patients or in vitro culture

were transferred to an aquatic environment. We tested three

factors as potential selective forces for driving V. cholerae out of the

aquatic environment and into the next host. These factors are

shared among several facultative pathogens and are as follows: the

viable but non-culturable state, hyperinfectivity, and lytic phage.

Escherichia coli, Shigella sonnei, Listeria monocytogenes, Campylobacter jejuni,

and V. cholerae are examples of facultative pathogens that lose the

ability to culture on standard media upon transfer to aquatic

PLoS Pathogens | www.plospathogens.org 1 October 2008 | Volume 4 | Issue 10 | e1000187

environments [14,15]. This phenotype was traditionally called the

viable but non-culturable state (VBNC) because the cells maintain

the capacity for metabolic activities such as protein synthesis,

respiration, and have intact membranes despite their inability to

culture [16]. However, we prefer to use the active but non-

culturable (ABNC) term for reasons explained by Kell et al [17].

The critical debate over terminology is if it is possible for bacteria

with a known in vitro growth condition to be viable and (but)

nonculturable. Since the answer to this question seems unresolved,

the ABNC term is a more conservative definition. In the case of V.

cholerae, animals become infected when inoculated with high doses

of ABNC bacteria (.106 or .1000-fold above the typical ID50 in

animal models) suggesting that ABNC bacteria can be rescued for

vegetative growth in vivo [14]. The experimental designs in these

studies were unfortunately not overly relevant to conditions in the

field; the ABNC state was induced by prolonged incubation at

4uC. ABNC V. cholerae have been observed in rural and urban

water samples in Bangladesh between and during outbreaks [5].

ABNC V. cholerae are found as single cells or associated in

aggregates with phytoplankton and zooplankton [5,18,19,20]. For

these reasons, ABNC V. cholerae are proposed to be the

environmental reservoir that maintains V. cholerae between

outbreaks and seeds new outbreaks. However, the role this

reservoir plays during an outbreak is unclear because cholera

outbreaks accelerate faster than the stochastic contribution of V.

cholerae from an environmental reservoir [21].

The second factor we measured was hyperinfectivity. This

phenotype was discovered when V. cholerae from patients were

found to be more infectious in the infant-mouse model than in vitro

grown V. cholerae [22,23]. This phenotype has also been

documented in Citrobacter rodentium [24], and can be modeled with

mouse passaged bacteria [25]. The role hyperinfectivity plays in

transmission is largely unknown, but V. cholerae from patients

remain hyperinfectious for at least 5 h in pond water [23]. Models

suggest that outbreaks start when an index case consumes V.

cholerae from an environmental reservoir, but the acceleration of

the outbreak is driven by hyperinfectious V. cholerae. Unlike the

stochastic contribution of environmental V. cholerae, mathematical

models that incorporate hyperinfectivity produce the steep rise in

case numbers that are consistent with the actual rise in cases

observed in Dhaka, Bangladesh during an outbreak [21].

The third factor we examined was lytic phage; we note here that

this report concerns only lytic vibriophage and not cholera toxin

phage or other lysogenic phage. Lytic vibriophage in the

environment have been studied from almost the time that V.

cholerae was first discovered [26], but recent phage epidemiology

papers provide new insights into the role phage play in outbreaks.

The percentage of patients passing lytic phage rises as a cholera

outbreak progresses; at the same time, phage titers in the

environment increase [27,28]. Towards the end of an outbreak,

the vast majority of cholera patients (.90%) void lytic vibriophage

in addition to V. cholerae. Over a 5-year study of patients at the

International Centre for Diarrhoeal Disease, Bangladesh

(ICDDR,B), at least half of cholera patients harbored lytic

vibriophage [29]. The ubiquity of lytic phage at the end of an

outbreak suggests phage may play an important role in stopping

an outbreak. This hypothesis is also supported by mathematical

models [30], as well as epidemiological data that indicate

household contacts of an index case that does not harbor lytic

phage are at an increased risk of infection with V. cholerae [29].

In summary, a cholera outbreak is currently modeled as follows:

An outbreak begins with the consumption of ABNC V. cholerae

from the environment, is accelerated by hyperinfectious bacteria

shed from patients, and is terminated by a rise in lytic phage. This

model however does not provide a reason (selective pressure) for V.

cholerae to leave the aquatic environment and go to the next host.

Contrary to the current model regarding the importance of ABNC

V. cholerae for transmission, we show that the loss of culturability is

a negative selective pressure for transmission, and non-culturable

cells are not the major contributors to infection. Instead we show

here that culturable V. cholerae recently shed by patients are the

major contributors to infection, and upon prolonged incubation in

pond water, lytic phage and ABNC cells rise in the aquatic

environment to cooperatively block transmission. In addition,

transcriptional analysis suggests that bacteria quickly adjust to the

stresses of the aquatic environment, and lytic phage have an

undetectable influence on this adaptation. Despite this adaptation,

rice-water stool V. cholerae rapidly become ABNC. In the absence

of high-titer phage, our results support the model that recently

shed hyperinfectious V. cholerae drive cholera outbreaks.

Materials and Methods

Bacterial strains and growth conditionsThe strains used in this study are provided in Table S1. Strains

were grown on Luria-Bertani (LB) agar or in LB broth with

aeration at 37uC with streptomycin (SM) 100 mg/ml unless

otherwise specified. SM sensitive strains were cultured on LB or

a Vibrio spp. selective medium, TTGA [31]; the plating efficiency

on TTGA and LB was equivalent (data not shown). The in vitro

derived V. cholerae were prepared by growth for 4 h at 37uC with

gentle rocking in M9 minimal medium (pH 9.0) supplemented

with trace metals, vitamins (Gibco MEM Vitamins, Invitrogen),

and 0.5% glycerol [32]; this medium is referred to as ‘M9 pH 9’.

Pond microcosm systemWater was collected from a pond in central Dhaka each day of

experimentation using a mechanical pump and intake hose system

that collected water approximately 0.5 m below the water surface

to avoid fluctuations in osmolarity due to rain water stratified at

the top layer of the pond. This pond has historically cultured

positive for V. cholerae; however in this study, V. cholerae and phage

lytic for V. cholerae were below the limit of detection by standard

methods [29] on the days of experimentation. Eighty liters of

unfiltered pond water were transferred to a barrel lined with a

Author Summary

The biological factors that control the transmission ofwater-borne pathogens like Vibrio cholerae during out-breaks are ill defined. In this study, a molecular analysis ofthe active but non-culturable (ABNC) state of V. choleraeprovides insights into the physiology of environmentaladaptation. The ABNC state, lytic phage, and hyperinfec-tivity were concurrently followed as V. cholerae passagedfrom cholera patients to an aquatic reservoir. Therelevance to transmission of each factor was weighedagainst the others. As the bacteria transitioned from thepatient to pond water, there was a rapid decay into theABNC state and a rise of lytic phage that compounded toblock transmission in a mouse model. These two factorsgive reason for V. cholerae to make a quick transit throughthe environment and onto the next human host. Thus, inover-crowded locations with failed water infrastructure,the opportunity for fast transmission coupled with theincreased infectivity and culturability of recently shed V.cholerae creates a charged setting for explosive choleraoutbreaks.

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 2 October 2008 | Volume 4 | Issue 10 | e1000187

pond-water washed autoclave bag, an aquarium bubbler was

placed in the barrel to oxygenate the water as well as to avoid

stratification, the barrels were positioned in an open shed shielded

from direct sunlight but freely exposed to the outside air: water

temp. 26–28uC, dissolved oxygen <6%, conductivity 260–

300 mS/cm, total dissolved solutes 137.8 mg/l, salinity = 0.1 ppt,

and pH 6.6–6.9. One liter of the water was centrifuged at 2,744 g

at room temperature (RT) for 5 min, and filter sterilized through a

0.2 mm filter (FS pond water). This FS pond water was used to

resuspend in vitro and in vivo derived V. cholerae, as well as for

chemical analysis. After each experiment, bleach was added to

each barrel to 0.5% and held for 24 hrs to sterilize the contents.

Cultures on LB agar were taken to confirm complete sterilization

before the water and bag were disposed. Inorganic chemical

analysis on stool supernatant and pond water samples was

performed by Dr. R. Auxier at the Center for Applied Isotope

Studies (U. of Georgia, Athens, GA). Dr. A. Parastoo at the

Complex Carbohydrate Research Center (U. of Georgia, Athens,

GA) determined the major sugars in the FS pond water samples

using mass spectrometry.

Preparation of stool and in vitro derived V. cholerae forincubation in the pond microcosm

Stool samples were collected from adult patients (.15 yrs of

age) with acute watery diarrhea and no prior treatment with

antibiotics. The samples were examined by darkfield microcopy to

confirm the presence of V. cholerae [29], and were included in the

study if .95% of the cells were highly motile and vibrioid in

shape. All samples were screened and found to be negative for

ETEC, the ratio of V. cholerae to non-V. cholerae bacteria was

determined, and the presence of lytic phage was assayed as

previously described [29].

Stool samples meeting the inclusion criteria were clarified of

mucus and debris by centrifugation at 988 g for 3 minutes at RT,

and then V. cholerae were pelleted by 15 minutes of centrifugation

at 26,892 g. Bacterial pellets were resuspended in an equal volume

of FS pond water at a final concentration of approximately 16108

CFU/ml; alternatively, pellets were resuspended in RNAlater

(Ambion, INC), flash frozen, and stored at 280uC for subsequent

microarray analysis. Fifty ml aliquots of the resuspension were

transferred to dialysis tubes with a 12 KDa cutoff (Fisher Scientific

INC), and the tubes were immediately transferred to the pond

microcosm described above. The tubes were kept just below the

surface of the water, and bacteria and phage did not traverse the

dialysis tubing (data not shown). The time from stool collection in

the hospital to incubation in pond water was under 1 h. The

collection of the rice-water stool from human subjects was

reviewed and approved by both the Research Review Committee

and Ethical Review Committee at the International Centre for

Diarrhoeal Disease Research, Bangladesh, and by the Human

Research Committee at the Massachusetts General Hospital.

V. cholerae were isolated by single colony purification from stool

on either LB SM or TTGA media [31]. The in vitro derived V.

cholerae were prepared as described above in M9 pH 9 at a final

concentration of 16108 cfu/ml (approximately equivalent to the

density in stool). After incubation at 37uC with gentle rocking for

4 h, the cells were then pelleted by centrifugation at 26,892 g for

15 min at RT, resuspended in an equal volume of FS pond water,

transferred to dialysis tubes, and placed in the pond microcosm in

a manner similar to the stool derived V. cholerae described above.

Additionally, a portion of the pellets were stored in RNAlater as

above for subsequent microarray analysis. At 5 and 24 h, the

contents of the dialysis tubes (patient and in vitro derived) were

transferred to sterile centrifuge tubes. For microarray analysis, the

contents were centrifuged at 26,892 g for 15 min at RT, and the

pellets were stored in RNAlater as above. At the 0, 5, and 24 h

time points of collection for patient or in vitro derived cells for

microarray analysis, paired samples were simultaneously taken for

animal experiments described below.

Infection experiments in the infant mouse modelThe competitive index of V. cholerae pre-incubated in M9 pH 7,

M9 pH 9, or rice-water stool supernatant was determined using 5

to 6-day-old Swiss Webster mice as described previously [22]. In

brief, the O1 El Tor Inaba strain N16961 (LacZ2) of V. cholerae

was grown overnight on LB agar with SM, and colonies were

resuspended in LB broth. The cells were washed and incubated in

M9 pH 7, M9 pH 9, or phage negative stool supernatant. During

the incubation of the in vitro samples, stool samples from cholera

patients were screened for V. cholerae and processed as described

above. After the 1 h incubation of the in vitro grown strains, infant

mice were inoculated intragastrically with 105 CFU of a 1:1

mixture of the paired LacZ+ stool V. cholerae and in vitro grown

LacZ2 wild-type N16961 strain. At 24 h post inoculation, the

small intestine was harvested and the homogenized contents were

serially diluted and plated on LB SM, X-gal 40 mg/ml agar plates.

After overnight incubation at 37uC, blue and white colonies were

counted to determine the competitive index.

To study the infectivity of V. cholerae transferred to the pond

microcosm, the ID50 was determined for both stool and in vitro

derived V. cholerae after 0, 5, and 24 h of incubation in pond water.

At 0 h, stool derived V. cholerae were prepared as described above

and serially diluted in LB. The in vitro derived V. cholerae were

prepared as described above with the 4 h preincubation at 37uC in

M9 pH 9, and subsequent serial dilution in LB. Groups of 5–6

day-old Swiss Webster mice were then inoculated intragastrically

with doses that ranged from approximately 1 to 105 CFU per

mouse. Mice were euthanized at 24 h post inoculation, and the

small intestinal homogenates were plated as described above.

Values of $1,000 CFU/mouse (limit of detection = 100 CFU)

were recorded as positive for infection. A dose-response curve was

made by plotting the fraction of infected mice against the log10 of

the input V. cholerae cell count – either by CFU or direct counts.

The ID50 was estimated from this curve by a standard nonlinear

regression using the Hill Equation – the Hill slope was fixed at 1.0

when there were ,3 data points between the values of 0.1 and 0.9

on the Y axis. The 95% confidence intervals for the ID50 (CI) and

coefficient of determination (R2) are provided. The ID50 for the

stool and in vitro derived V. cholerae incubated in the pond

microcosm was determined at 5 h and 24 h in the same manner.

The in vivo dynamic between V. cholerae and lytic phage was

investigated by the co-infection of both the bacteria and lytic phage

in ID50 experiments as described above. Lytic phage isolates were

obtained in a pair-wise manner from the same patients that the V.

cholerae isolates were obtained. Phage were isolated from stool

supernatant by a standard plaque assay on a bacterial lawn made of

the V. cholerae isolate from the same stool sample [33]. Phage were

picked from 3 serial clear lytic plaques. V. cholerae were prepared for

the animal studies by overnight growth and a 4 h incubation at 37uCin M9 pH 9 as described above. The V. cholerae isolated from a given

patient and the paired lytic phage were combined for 8 min prior to

infection at a phage::bacterium multiplicity of infection (MOI) that

reflected what was observed in the rice-water stool and pond

microcosm in this project: 0.001 to 5 PFU/CFU. The inocula were

then serially diluted and groups of at least five infant mice were

inoculated with doses that ranged from approximately 1 to 105 CFU

per mouse. The ID50 was calculated for each experiment as

described above. At a given dose of bacteria and phage, the burden

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 3 October 2008 | Volume 4 | Issue 10 | e1000187

of infection was determined by calculating the median CFU/ml for

each group of at least five mice. This was repeated for a total of 3

strains at all doses of bacteria and paired phage. The three medians

were plotted individually, and the average of the three medians was

also plotted. A Student’s t-test was performed between the average

for the no-phage control and each phage dose.

Analysis of Cell Surface PolysaccharidesWe investigated if colonization of infant mice by V. cholerae in the

presence of lytic phage was because the bacteria had become

resistant to the phage. One way that bacteria can become resistant to

phage is by altering the phage receptor which is most commonly LPS

for vibriophage [34,35]. A basic test for putative LPS mutants is

agglutination in LB [36]; this test lacks absolute specificity as other

phenotypes can also cause agglutination such as expression of the

toxin co-regulated pilus (TCP), but TCP is not expressed in LB [37].

We validated the agglutination assay with LPS extraction and gel

electrophoresis of several putative LPS mutants identified by

agglutination (below). We chose strain EN159 for this study because

it is SM resistant. From each animal coinfected with EN159 (all

doses) and the paired EN159 phage (all doses), eight isolates were

colony purified (36) and frozen for further evaluation of phage

sensitivity. These isolates were grown in LB SM broth overnight and

agglutination of the cells was assessed if the media clarified after

20 min of static incubation. The fraction of the 8 isolates from a

given mouse that agglutinated was recorded as a fraction of isolates

from a given mouse that were putative LPS mutants [36]. All isolates

from mice infected with the highest dose of V. cholerae (1.56105

CFU/mouse) and all phage MOI’s (0, 0.005, 0.1, and 2.0) were

further tested for phage resistance by the standard plaque assay.

For validation of the agglutination assay, LPS was extracted

from a total of five isolates from 5 different mice infected with the

highest bacterial dose (1.56105 CFU/mouse) and at highest MOI

(2.0). As a control, LPS was extracted from a total of five isolates

from 5 different mice infected with the highest dose of bacteria

(1.56105 CFU/mouse) and no phage. The input strain also served

as an additional control. Cell surface polysaccharides from the

eleven strains were isolated and analyzed as described recently

[38,39]. Briefly, Proteinase K-digested whole cell extracts were

isolated according to Hitchcock and Brown [40] and analyzed by

electrophoresis on 16.5% SDS-polyacrylamide gels. The complete

synthesized LPS and the lipid A-core oligosaccharide precursor

were visualized by silver staining [41].

Microarray experimentsRNA was prepared from the samples collected at 0, 5, and 24 h

of dialysis in pond water (described above). The frozen suspensions

of bacteria in RNAlater (Qiagen) were thawed on ice, spun at

15,000 g for 20 min at 4uC, the supernatant was discarded, RNA

was extracted from the pellet using the Qiagen (Valencia, CA)

RNeasy Mini Kit, and DNA was removed using the Qiagen on-

column RNase-Free DNase set. For qRT-PCR validation,

complete DNA removal was achieved using the Ambion (Applied

Biosystems/Ambion, Austin, TX) DNA-free DNase Treatment kit.

Each RNA sample was spiked with an in vitro transcribed

Arabidopsis RNA which served as a reference for color balancing

during scanning; the control RNA was provided by the Pathogen

Functional Genomics Resource Center (PFGRC) at the J. Craig

Venter Institute (formerly TIGR). Labeling of cDNA was

performed as described previously [42] with the exception that

the reverse transcription reaction used Superscript III (Invitrogen,

Carlsbad, CA) at a reaction temperature of 52uC for 1 h and 8 mg

RNA. The cDNA from each reaction was split and labeled with

either Cy3 or Cy5 (dye swapped). Unless indicated otherwise, at

least 4 technical microarray replicates (2 dye swaps) were

performed per biological replicate. There were two biological

replicates for each condition: patient derived with phage, patient

derived with no phage, and in vitro derived.

Microarrays were provided by PFGRC and consisted of glass

slides with genes spotted in quadruplicate with 70 bp oligonucleo-

tides for each of 3810 V. cholerae ORFs. Hybridizations were

performed as described previously [42]. Microarrays were scanned

with a Perkin-Elmer Scanner, and the raw data were analyzed using

the Perkin-Elmer Scan Array Express, Imigene, and Spotfire

software packages. Cy3 and Cy5 data from each slide were split

into the relevant biological groupings as single channel data. All

items with a raw intensity of less than 50 were assigned a minimum

intensity value of 49.9 [43]. The complete data set was log2

transformed and normalized against all other scans by the 75th

percentile. The values for a given gene across all scans were then

normalized by the z-score for that specific gene. The normalized

data was then compared by ANOVA according to the relevant

biologic grouping [44,45,46]. For ANOVA analysis between 6

groups, a Bonferroni correction was applied to account for bias due

to multiple tests by dividing the desired level of significance (a= 0.01)

by the total number of comparisons performed (22,860 = 3,810

genes with 6 comparisons) [44,45,46]. Therefore, the corrected false-

positive rate was a= 4.461027 which was rounded to a= 1.061027;

P values that fell below 1.061027 were considered statistically

significant. Cluster analysis was performed by Spotfire with the

following metrics: clustered by Unweighted Pair-Group Method

with Arithmetic mean (UPGMA), correlated by Pearson Product

Momentum Correlation, and ordered by Input Rank. As an

independent measure of similarity between biological groupings,

Principal Component Analysis (PCA) was performed on all samples

using Spotfire. After the ANOVA, all replicates were ungrouped,

and the cluster analysis and PCA were performed in an unsupervised

fashion with respect to the technical replicates and biological

groupings. Fold-changes between two biological groupings were

calculated using distinction calculations performed by Spotfire, and

fold-changes with P values,6.661026 (Bonferroni corrected) were

considered significant. Microarray data are available in the

supplemental material (Tables S3, S4, S5, S6, S7, S8, S9 and S10).

Validation of microarray by quantitative qRT-PCRThere was sufficient sample to obtain cDNA template from

three phage positive patients (EN159, EN182, EN191), and three

phage negative patients (EN124, EN150, EN174). RNA was

isolated as described above, and qRT-PCR was performed as

previously described [13]. In brief, cDNA was synthesized from

1 mg of RNA using the SuperScript II First Strand Synthesis

System for qPCR (Invitrogen Inc.). The qRT-PCR experiments

were performed with iQ SYBR Green supermix (Biorad). Each

reaction contained 200 nM primers, approximately 10 ng of the

template, and the ROX reference dye. All primer pairs (Table S1)

amplified the target with efficiencies of 92% or greater (data not

shown). The mean cycle threshold for the test transcript was

normalized to the reference transcript sanA [47] and argS. The

reference argS was chosen because no expression changes were

detected in this microarray project as well as all publicly available

V. cholerae microarray databases. Values .1 indicate that the

transcript is in higher concentration than the reference.

Results

Sample collection from cholera patients at the ICDDR,BThis project focuses on rice-water stool samples collected from

three patients (EN159, EN182, and EN191) who harbored lytic

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 4 October 2008 | Volume 4 | Issue 10 | e1000187

vibriophage for V. cholerae, and the respective phage and V. cholerae

isolates from these three patients. In addition, rice-water stool was

collected from three patients who did not harbor lytic vibriophage

(EN124, EN150, and EN174), and V. cholerae was isolated from

each of these patients. Therefore, the biological replicates for each

arm of the study were three unless stated otherwise; sufficient

numbers of infant mice for ID50 testing were available only for the

three patient samples that harbored phage. At the time of

collection, all patients were severely dehydrated as defined by the

World Health Organization [48].

Chemical analysis of pond waterAs V. cholerae passes from the patient into pond there is dramatic

shift in osmolarity and in the concentrations of inorganic nutrients

and carbon sources. Some of these factors are depicted in Fig. 1.

NaCl and KCl are major contributors to osmolarity and both have

a decline from 2,600 to 22 ppm (120-fold) and 820 to 6 ppm (140-

fold) between the rice-water stool and pond supernatant,

respectively. The conductivity difference between the rice-water

stool (as well as LB broth) and pond water is approximately a 50-

fold decline. Phosphate and fixed nitrogen are typically limiting

inorganic nutrients in fresh water ponds. Phosphate and fixed

nitrogen (NH4+) decline from 160 to 0.1 ppm (1,600-fold) and 52

to 0.5 ppm (104-fold), respectively. V. cholerae was placed in filtered

pond water and then dialyzed in 12 KDa tubing with live pond

water. Therefore, carbon sources such as large polymers like

chitinous exoskeletons would not be present in the dialysis bags.

Carbon sources detected were rhamnose (29 Mol.%; 16 nM),

fucose (20% Mol.%; 11 nM), glucose (2.7 Mol.%; 1 nM), and

unidentified sugars (48.9 Mol.%). This chemistry collectively

framed many of the physiological events that occurred as V. cholerae

adapted to the aquatic system. This adaptation and pond

microcosm system is not necessarily specific to Bangladesh as the

chemical composition shown herein is comparable to pond water

used in transition studies with pond water obtained in Boston, MA

[13].

Exponential drop of culturability in a pond microcosmThe culturability of V. cholerae transferred to the pond

microcosm was monitored by culture and direct microscopy

counts. We define the non-culturable cells as ‘active but non-

culturable’ (ABNC) because there were clear transcriptional

changes between 5 and 24 h detected by both microarray and

qRT-PCR analysis (below). Thus, our measure of ‘active’ was

global transcriptional change. Culturability was rapidly lost upon

transfer to the pond microcosm at 5 and 24 h with declines of 63%

(SD+/216%) and 98% (SD+/21.0%), respectively (Fig. 2A). The

V. cholerae isolates from the respective patients were grown in vitro

(M9 pH 9) and transferred to the pond microcosm; the declines in

culturability in the pond microcosm were similar for the in vitro

derived samples compared to the patient derived samples (Fig. 2A).

Despite the drop in culturable cells, the total cell numbers

remained constant by direct counts (Fig. 2B) for all sample types;

the cell number was also constant for phage negative patient

samples and the paired in vitro grown strains (Fig. 2B). The culture

counts are not available for the phage negative patient samples

because two isolates were unexpectedly SM sensitive. The plating

efficiency of starting cultures neared 100%. For example, the

average concentration of V. cholerae from patients (EN159, EN182,

EN191) at 0 h by culture counts and direct counts was 1.06108

CFU/ml (+/21.16108 CFU/ml) and 1.656108 CFU/ml (+/

20.356108 CFU/ml), respectively.

Lytic phage from patients bloom in the pond microcosmThe PFU titer was monitored at 0, 5 and 24 h in the pond

microcosm (Fig. 2C). At 0 h, the average ratio of phage to V. cholerae

for all three patient stools was 2.261026 (SD+/23.561026 ). At 5 h,

this ratio increased by 4 orders of magnitude to 1.061022 (SD+/

21.261022) by culture counts, or 3 orders of magnitude to

1.561023 (SD+/21.361023) by direct counts. At 24 h, this ratio

increased an additional 2 orders of magnitude to 4.061021 (SD+/

23.961021) by culture counts, but remained steady at 3.861023

(+/23.261023) by direct counts. From 5 to 24 h, this ratio changed

because the culturable counts decreased 14-fold. These findings are

supported by micrographs that illustrate altered morphology of V.

cholerae only in the patient derived samples from phage positive

patients (Fig. 3A). Lytic and lysogenic vibriophage have been

previously characterized from patients [27,33,35,49,50,51,52]; our

phage isolates are consistent in terms of the tropism of those lytic

phage previously published [27] because our phage had specificity

for the Inaba or Ogawa serotype of the O1 El Tor V. cholerae biotype,

and the phage were unable to form plaques on O139 V. cholerae (data

not shown). These data indicate the phage receptor may be O1 LPS

as has been demonstrated previously [34,35]. Support for this

hypothesis is the generation of LPS mutants in the presence of lytic

phage (presented below).

Hyperinfectivity is maintained for at least 5 h of dialysisin pond water

The ID50 for V. cholerae freshly shed from the patients (113 CFU;

95% confidence interval [CI] = 65–196 CFU) was lower compared

to the in vitro grown reference (596 CFU; 95% CI = 193–1834

CFU; Fig. 4A). Hyperinfectivity was also observed after 5 h of

dialysis between the patient (51 CFU; 95% CI = 13–202 CFU) and

in vitro culture (680 CFU; 95% CI = 276–1673 CFU; Fig. 4B).

These findings are consistent with competition experiments

previously published that suggest V. cholerae maintains hyperinfec-

tivity for at least 5 h after exit from the patient [23]. We tested if

hyperinfectivity could be induced by the medium alone (stool-

supernatant), and we found that hyperinfectivity could not be

Figure 1. Change in concentration of biologically relevantelements upon passage from patients to the pond environ-ment. Data represent the average and standard deviation for threeindependent patient rice-water stool samples (EN159, EN182, EN191)and the pond water samples collected on the respective days ofexperimentation. ‘0 h Pond’ (dark grey) and ‘24 h Pond’ (light grey)represent filter sterilized pond water from the inside of dialysis bagsused to incubate V. cholerae, respectively. Zn and Mn were below thelevel of detection in the pond. In addition, Al, B, Ba, Cd, Co, Cu, Fe, Mo,Ni and Pb were below the level of detection; ppm = parts per million.doi:10.1371/journal.ppat.1000187.g001

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 5 October 2008 | Volume 4 | Issue 10 | e1000187

induced in vitro by incubation in stool supernatant (pH 9) or

minimal media (M9 pH 9) (Fig. S1). Unique to the present study

was that the single strain infection experiments revealed that the

fraction of mice infected with high doses of patient derived V.

cholerae was reduced at 5 h and 24 h compared to the in vitro

reference (Fig. 4B–E). Indeed, the ID50 was not able to be

calculated for the patient derived samples at 24 h because less than

50% of the animals were infected (Fig. 4D–E). The 24 h time

point corresponds with the point when the titer of PFU was highest

and the titer of culturable cells was lowest (Fig. 2); note again that

the no phage control for this experiment are in vitro derived cells.

We hypothesized, and show below, that the incomplete coloniza-

tion observed is due to the presence of lytic phage in the inocula.

Culturable V. cholerae are the major contributors toinfection

We wanted to investigate the relevance of the ABNC state to the

transmission of V. cholerae. To do this we tracked the ID50 over

time by both culturable counts and direct counts. We focus here

on the ID50 data from the in vitro derived V. cholerae because the

phage positive patient derived samples failed to fully colonize at 5

and 24 h. In the context of the pond system, the total cell counts

remained constant but the proportion of culturable cells decreased

over time. We tested three competing hypotheses: (i) If culturable

cells are equally infectious as non-culturable cells, then the ID50 by

total cell counts will be constant as the percent of culturable cells

decreases. (ii) If culturable cells are more infectious than non-

culturable cells, then the ID50 by total cell counts will increase as

the percent of culturable cells decreases. (iii) If culturable cells are

less infectious than non-culturable cells, then the ID50 by total cell

counts will decrease as the percent of culturable cells decreases. As

mentioned above, the culture cell counts fell from 100% to 27% to

3% at 0, 5 and 24 h during the experiment. The corresponding

ID50 by culturable counts remained constant as the culturable

counts decreased at 0, 5 and 24 h (Fig. 4A, B, D). However, the

ID50 by total cell counts rose from 596 (95% CI = 193–1834) to

1683 (95% CI = 683–4145) to 7383 (95% CI = 3970–13731) as the

culturable counts decreased at 0, 5 and 24 h (Inset table in Fig. 4).

Therefore, hypothesis (ii) appears to be correct that culturable cells

are more infectious than non-culturable cells, and thus, the major

contributors to infection are culturable V. cholerae.

Coinfection of lytic phage and V. cholerae alters theburden of infection

Because lytic phage are present in aquatic reservoirs and in at

least half of cholera stool samples, we wanted to determine the

relevance of lytic phage to the transmission of V. cholerae. We

hypothesized that the reduction in colonization at 5 and 24 h for

the patient derived samples (Fig. 4) was caused by the bloom of

lytic phage because no such reduction was observed for phage

minus in vitro derived V. cholerae. We tested this hypothesis by

coinfecting infant mice with V. cholerae isolated from the three

phage positive patients (EN159, EN182, EN191), and the paired

lytic phage isolate from each patient (described above). The

inocula were made by mixing bacteria and phage at various

MOI’s that were relevant to those observed in rice-water stools

and after incubation in the pond microcosm (Fig. 2C). A linear

dose response (R2 = 0.99; slope = 20.57; 95% CI = 20.83 to

20.32; Fig. 5A) was observed in the infant mice inoculated with a

constant bacterial dose (1–26104 CFU/mouse) and variable

phage dose (1.06102322.0 MOI). In contrast, at a high dose of

V. cholerae (1–26105 CFU/per mouse) there was a significant

reduction in colonization at all MOI tested (Fig. 5B).

The coinfection experiments were also performed as ID50

experiments with variable concentrations of V. cholerae and 4

constant doses of phage (MOI = 0, 0.005–0.1, 0.05–0.1, 0.5–2.0).

There was no difference in the ID50 between the no phage control

and animals coinfected with phage at an MOI of 0.05–0.1 (Fig. 5C)

and 0.005–0.01 (data not shown). However, the ID50 in mice co-

Figure 2. V. cholerae and phage counts during a 24 h incubationin pond water. A. The CFU decline between the patient derived (soliddiamonds, N = 3) and in vitro derived (clear diamonds, N = 3) V. choleraeis similar. All patient samples had phage (EN159, EN182, EN191; N = 3).B. Direct counts (DC) do not change over 24 h. DC for V. cholerae frompatient samples with phage (solid diamond, N = 3) and with no phage(shaded squares, N = 3). The in vitro grown strains isolated from phagepositive stools (N = 3) or from phage negative stools (N = 3) are cleardiamonds and clear squares, respectively. C. Phage bloom at 5 h (N = 2).The ratio of plaque forming units (PFU) to V. cholerae is plotted in twoways: PFU to CFU ratio (solid line); PFU to DC ratio (dashed line).doi:10.1371/journal.ppat.1000187.g002

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 6 October 2008 | Volume 4 | Issue 10 | e1000187

infected with phage at a MOI of 0.5–2.0 (18 CFU; 95% CI = 9–

36) was significantly lower than the no phage control (65 CFU;

95% CI = 37–111 CFU). These experiments support the hypoth-

esis that phage can limit infection at doses of V. cholerae greater

than 103 CFU. However at high phage MOI and low doses of in

vitro grown V. cholerae, the phage may have an unexpected positive

impact on the ID50 (Fig. 5D). EN159 isolates from experiments in

which the phage may have had a positive impact on infectivity

(MOI of 2.0; 100–200 CFU) were found to be phage sensitive and

not LPS mutants (data not shown). In competition in the infant

mouse model, these isolates competed 1:1 with the input strain

suggesting there was no gain of function from prior co-culture with

the phage (data not shown).

Although the phage reduced the burden of V. cholerae infection,

complete clearance of the bacteria was not observed, as had occurred

in the pond microcosm (Fig. 4E). To investigate a reason behind this

we closely examined isolates from the EN159 coinfection studies

because EN159 is SM resistant. 40–70% of isolates from mice

coinfected with EN159 V. cholerae at a dose of 1–26105 CFU/per

mouse agglutinated in LB independent of the phage dose – a

phenotype consistent with LPS mutants [36]. No isolates from mice

coinfected with V. cholerae at a dose of less than or equal to 1–26103

CFU/per mouse agglutinated. To confirm that agglutination was

indicative of LPS mutations in our system we analyzed LPS from

several isolates. The LPS from a total of five isolates from five

different mice infected with the EN159 V. cholerae and phage

(MOI = 1.0) was compared to the LPS from a total of five isolates

from different mice infected with EN159 V. cholerae and no phage. All

five mouse passaged isolates from coinfection experiments with

phage were resistant to the phage and agglutinated, and the five

mouse passaged isolates without phage were sensitive to the phage

and did not agglutinate. These phage sensitive colonies demonstrat-

ed a wild-type LPS with the typical two band pattern consisting of

the lipid A-core oligosaccharide precursor as the lower band and the

complete LPS with attached O antigen as the upper band (Fig. 5F).

In contrast, phage resistant colonies exhibited an O antigen deficient

phenotype (Fig. 5E). We were concerned about lysogeny among

phage resistant colonies. Phage sensitive V. cholerae were infected with

phage in vitro and subsequent treatment of phage resistant isolates

with and without mitomycin-C [53] yielded no phage; this

experiment was repeated for all strains and phage in this study.

These data suggest there was no lysogeny. However, experiments

with additional stresses (osmotic shock, UV, etc.) and phage isolates

genetically marked with an antibiotic resistance marker would be

required to definitively show the absence of lysogeny. Taken

together, these data indicate that at a high dose of V. cholerae,

spontaneous LPS mutants will dominate during in vivo colonization

in the presence of lytic phage.

Figure 3. Phage positive patient derived V. cholerae have irregular shape at 24 h (arrow). Representative images of a lytic phage positivepatient sample (EN182; A, B, C) and phage negative patient sample (EN124; D, E, F) incubated in pond water over 24 h. The isogenic strain EN182was isolated away from the phage, grown in vitro and incubated for 24 h in the pond water (G, H, I). V. cholerae are labeled with a FITC conjugatedmonoclonal Ab to O1 LPS (left) and nucleic acid is labeled with DAPI (middle) which indicates non-V. cholerae; merge is shown on right. Data arerepresentative of 3 independent experiments (data not shown). V. cholerae from the 0 h and 5 h time points for all sample types wereindistinguishable by morphology and represented ‘F’ and ‘I’. Scale bar = 10 mm.doi:10.1371/journal.ppat.1000187.g003

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 7 October 2008 | Volume 4 | Issue 10 | e1000187

Global transcriptional analysis of ABNC V. cholerae in theaquatic environment

Because ABNC V. cholerae have low infectivity, yet represent the

predominant state of the bacteria after 5 h of incubation in the

pond microcosm, we measured possible transcriptional changes

during this transition. The goal was to ascertain whether the

bacteria were adapting to the nutrient poor conditions in a

manner dependent or independent of their source of origin

(patient or in vitro). Samples for the microarray fell into six

biological groups: patient derived samples (EN159, EN182)

incubated in the pond microcosm for 0, 5 and 24 h (designated

T0P*, T5P* and T24P*, respectively; Fig. 6A) and the paired in

vitro derived isolates incubated in the pond for 0, 5 and 24 h

Figure 4. ID50 experiments of V. cholerae incubated in pond water. V. cholerae were incubated in pond water for 0 (A), 5 (B, C) and 24 h (D,E). Each symbol represents the average of the fraction of infant mice infected with V. cholerae from three phage positive patient samples (soliddiamond/solid line), or the same three isogenic strains from in vitro culture (clear diamond/dashed line). The data are plotted on the X axis by the logof the culturable inputs (CFU) on the left or direct count inputs (DC) on the right; the log (ID50) is depicted with the grey dashed vertical line. The insettable summarizes the fold changes in the ID50 over 24 h for the in vitro derived samples. CI = 95% confidence interval for the ID50 determined fromthe fitted curve (red) modeled with the Hill Equation. Plating efficiency at 0 h neared 100% (see text). Each symbol represents the average of the 3biological replicates (EN159, EN182, EN191) which pools data from at least 15 mice total.doi:10.1371/journal.ppat.1000187.g004

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 8 October 2008 | Volume 4 | Issue 10 | e1000187

(designated T0I, T5I and T24I, respectively; Fig. 6A). Both patient

(EN159, EN182) samples harbored phage, which is indicated with

an asterisk. Two additional patient samples that did not harbor

phage (EN124, EN150) were included as controls for transcrip-

tional changes induced by phage (designated T0P, T5P and T24P,

respectively; Fig. 7A). The patient samples EN174 and EN191 and

in vitro samples EN124 and EN150 were excluded because of

insufficient material for microarray analysis. The qRT-PCR

validation is provided in Table S2. Cluster analyses in Fig. 6

and Fig. 7 isolated key expression patterns; genes within these

Figure 5. Coinfection of infant mice with V. cholerae and lytic vibriophage. A. Average colonization (bar) of mice inoculated with in vitroderived strains (EN159 EN182, EN191) co-infected with the paired phage isolate (solid diamonds). The bacterial dose is constant (1–26104 CFU/mouse) and the phage dose is variable ranging from a MOI of 1023 to 1.0. NP = no phage control (clear diamonds). Each diamond represents themedian of 5 technical replicates (mice) for a given biological replicate (EN159, EN182, EN191). CI = 95% confidence interval for the slope of the linearline. B. Same as ‘A’ except a constant bacterial dose of 1–26105 CFU/mouse. *Significant difference compared to the no phage control (Student’s t-Test; P,0.05). C. ID50 plotted by the log(CFU/Mouse) against the fraction of mice infected. Inoculation with a constant phage dose (MOI = 0.05–0.1;solid diamond/line) and a variable dose of in vitro derived V. cholerae. No phage control = clear diamonds/dashed line; CI = 95% confidence intervalfor the ID50 calculated from the fitted curve (red) modeled with the Hill Equation. Each symbol represents the average of 3 biological replicates ($5mice per biological replicate). D. Same as ‘C’ except with a constant phage dose of MOI = 0.5–2.0. E. Analysis of LPS isolated from wild-type EN159(Wt), or mouse passaged EN159 (1–5) isolated from mice infected at 1.56105 CFU/mouse as in ‘B’ and an MOI of 2.0. Upper arrow indicates fullysynthesized LPS with attached O antigen, lower arrow indicates the lipid A-core oligosaccharide precursor. F. Same as ‘E’ except mouse passagedisolates (1–5) were in the absence of phage (NP).doi:10.1371/journal.ppat.1000187.g005

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 9 October 2008 | Volume 4 | Issue 10 | e1000187

Figure 6. Transcriptional profiles of patient derived and in vitro derived V. cholerae incubated in a pond microcosm. A. Heat map of435 genes that are differentially regulated (ANOVA P,161027; N = 2) in at least one of the following six conditions: V. cholerae from patients (T0P*,T5P*, T24P*) or in vitro culture (T0I, T5I, or T24I) incubated for 0, 5, or 24 h; the patient samples harbored phage*. Yellow and blue represent induced(max = 46-fold) and repressed (max = 20-fold) genes, respectively. The sample labels at the bottom are color-coded to match the right panel. The thinvertical dotted line breaks the genes into four major groups (nodes) provided in the supplement from top to bottom as Nodes 1-4 (Tables S3, S4, S5,S6 and S7). Node two is subdivided into 2A (upper; Table S4) and 2B (lower; Table S5). B. Principal Component Analysis (PCA) of the 435 geneexpression values in ‘A’. The arrows indicate the ‘transcriptional movement’ from 0 to 5 to 24 h for the patient and in vitro derived V. cholerae.doi:10.1371/journal.ppat.1000187.g006

Figure 7. Transcriptional profiles of patient derived V. cholerae with and without phage incubated in a pond microcosm. A. Shown isa heat map of the subset of transcripts that are differentially regulated (ANOVA P,161027; 182 genes) in at least one of the following six conditions:V. cholerae from phage positive patients (T0P*, T5P*, T24P*) or phage negative patients (T0P, T5P, or T24P) incubated for 0, 5, or 24 h. Yellow and bluerepresent induced (max = 46-fold) and repressed (max = 16-fold) genes, respectively. The sample labels at the bottom are color-coded to match theright panel; one T5P sample clustered with T5P* (as shown) and one T5P clustered with the T24 samples (not individually labeled). The thin verticaldotted line breaks the genes into two major groups (nodes) provided in the supplement as Node 1 (upper group; Table S8) and Node 2 (lower group;Table S9). B. Principal Component Analysis (PCA) of the 182 gene expression values in ‘A’. The arrows indicate the ‘transcriptional movement’ from 0to 5 to 24 h. Biological Replicates = 2. Yellow cubes are the T24I in vitro derived samples depicted in Fig. 6B as a reference.doi:10.1371/journal.ppat.1000187.g007

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 10 October 2008 | Volume 4 | Issue 10 | e1000187

groupings are described by biological function in Tables S3, S4,

S5, S6, S7, S8, S9 and S10. A complete list of all fold changes is

available in Table S10.

The transcriptional profile of patient and in vitro derived

V. cholerae did not converge during transition to the ABNC

state. The transcriptomes at each time point (0, 5 and 24 h)

were highly concordant within each of the various biological

groups (Fig. 6A and Fig. 7A). ANOVA analysis was used to select a

population of 435 genes (P,1.061027) that were differentially

regulated in at least one of the six groupings stated above. These

435 genes are represented by the heat map and cluster analysis in

Fig. 6A. Each technical replicate clustered into the appropriate

biological grouping. At no time point did the transcriptome of the

patient and in vitro derived samples converge on the global level.

These results were cross-validated by independent analysis using

principal component analysis (PCA) (Fig. 6B).

Transcriptional adjustment was rapid. The cluster

analysis and PCA in Fig. 6 group the 5 and 24 h in vitro derived

samples together. This suggests that the majority of the

transcriptional adaptation to the pond occurs by 5 h. We

explored this observation by performing cluster analysis between

the 0, 5 and 24 h time points for each sample individually.

ANOVA analysis was used to select a population of approximately

300 genes (P,1.061027) that were differentially regulated

between the 0, 5 and 24 time points. Five of the six cluster

analyses demonstrated rapid adjustment to the pond as indicated

by the 5 and 24 time points grouping together (data not shown)

which is consistent with Fig. 6A.

The absence of convergence was not caused by

phage. We tested if phage were responsible for the lack of

convergence between the patient and in vitro derived samples by

comparing phage positive (EN159, EN182) and negative (EN124,

EN150) patient samples. A new ANOVA analysis was used to

select a population of 182 genes (P,1.061027) that were

differentially regulated in at least one of the six groupings: T0P*,

T0P, T5P*, T5P, T24P* and T24P. These 182 genes are

represented by the cluster analysis and PCA in Fig. 7. The

cluster analysis grouped the 0 h samples together independent of

whether the samples harbored phage. The 24 h samples also

clustered together independent of whether they harbored phage.

One of the 5 h phage negative patient samples (T5P) clustered

with the 24 h samples, and the second 5 h sample (T5P) clustered

in a unique clade with both the 5 h phage positive patient samples

(T5P*). The similarity between the patient derived samples

suggests that phage had a limited influence on the

transcriptome. A second analysis was performed on only the 435

genes depicted in Fig. 6. The in vitro specific transcripts were

subtracted, and the remaining genes were analyzed by cluster

analysis and PCA for the patient derived samples with and without

phage. Again, there was no difference in profiles between the two

types of patient derived samples (data not shown). One question is

how the microarray data can be so similar between phage positive

and negative samples at 24 h despite gross differences in cellular

morphology (Fig. 3). One explanation is that the majority of the

transcriptional changes occur by 5 h which is supported by the

cluster analysis presented above. This explanation requires the

RNA to remain intact to some degree from 5 to 24 h.

Regulon specific analysis of ABNC V. cholerae in theaquatic environment

De-repression of virulence regulons in the pond. The

microarray findings are consistent with previously published

microarray studies that demonstrate at the point of exit from the

human host there is a profound repression of virulence genes, and

their relevant activators, that includes the toxin co-regulated pilus

(TCP), cholera toxin (CT), and chemotaxis regulons relative to the

in vitro derived reference [23,47]. For example, cholera toxin genes

(ctxA and ctxB) have divergent expression at 0 and 5 h but

convergent expression at 24 h relative to the in vitro derived

reference (Table S10). The repression of chemotaxis is

hypothesized to be a contributing factor to the hyperinfectivity

phenotype [22,54]. By microarray, all three chemotaxis operons

(#1 VC1394-1406, #2 VC2059-20-65, #3 VCA1088-1096) were

repressed between rice-water stool V. cholerae and the in vitro

derived reference. qRT-PCR analysis is consistent with these

findings showing repression of approximately 2-fold and 4-fold for

cheW-1 (operon #2) and cheY-4 (operon #3) with respect to the in

vitro reference, respectively (Table S2). Expression rises

approximately 4-fold and 8-fold for cheW-1 and cheY-4 by 5 h,

respectively by qRT-PCR (Table S2). The mechanism for the

repression of chemotaxis genes in rice-water stool and de-

repression in the pond remains unknown.

Adaptation to nutrient limitation. Transition from rice-

water stool into the ABNC state in the aquatic environment is a

process of adaptation to nutrient limitation (Fig. 1). There was

commonality of expression profiles for genes related to phosphate

limitation independent of where the sample was derived from

(patient or in vitro). For example, phoB is part of the pho regulon that

senses and responds to phosphate limitation [55,56]. By qRT-

PCR, phoB is induced approximately 5-fold and 39-fold in the first

5 h for patient and in vitro derived samples, respectively (Table S2).

These data are consistent with the array data, but the absolute

numbers are different because the microarray data are

compressed. Fixed nitrogen limitation is a second nutrient of

limited availability. Ammonia is incorporated to make glutamine

by the action of GlnA (Glutamine Synthetase) and GlnB [57]. V.

cholerae has two paralogs of glnB: glnB-1 and glnB-2. By qRT-PCR,

glnB-1 is induced in the patient derived samples by 5-fold in the

first 5 h; the in vitro derived samples are already induced in the M9

pH 9.0 and decline in expression levels by 3-fold over the 24 h

(Table S2). The qRT-PCR values are again consistent with the

microarray values (Table S10). These data support a model for

immediate adaptation to nutrient limitation coincident with entry

into the ABNC state in the aquatic environment.

Suppression of protein synthesis. A second level of

adaptation to the aquatic environment is at the level of global

protein synthesis. In Fig. 6A, clusters of genes fall into 4 major

groups (nodes) based on 4 expression profiles. Node 3 contains 174

genes that are strongly expressed in patients as well as in vitro and

rapidly turn off in the pond. 34% of these genes (60 total) are

involved in protein synthesis, and 14% (14 total) are involved in

energy metabolism (Table S6). Many of the protein synthesis genes

are contained within the VC2599-VC2570 ribosomal protein gene

locus. This cluster was identified previously as a locus induced in

patients [23,42] and repressed in stationary phase compared to log

phase [47]. The ribosomal gene rplC (VC2596) is near the

beginning of the operon and is induced 2-fold by qRT-PCR

(Table S2), compared to the in vitro reference at 0 h (Table S2). By

microarray, 29 out of 30 genes at this locus decrease expression

between 2-fold to 310-fold over 24 h (Table S10). This decline in

the expression of ribosomal proteins suggests a general decrease in

the capacity for protein synthesis as the cells enter the ABNC state.

RpoS as a candidate regulator of environmental

adaptation. One component of the regulation of the

ribosomal protein locus VC2599-VC2570 is the stationary phase

alternative sigma factor RpoS [47]. RpoS has been determined to

play a role in virulence in the infant mouse model [58] as well as

regulate the mucosal escape response in the rabbit ileal loop model

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 11 October 2008 | Volume 4 | Issue 10 | e1000187

[47]. A microarray of an in vitro grown rpoS mutant compared to

wild-type V. cholerae showed that the ribosomal protein locus

VC2599-VC2570 was repressed in the mutant strain. This

suggests that RpoS positively regulates VC2599-VC2570 [47].

However in our microarray, the VC2599-VC2570 locus is

repressed in the pond while rpoS is simultaneously induced at

least two-fold (Table S2). The regulation of VC2599-VC2570 is

therefore likely complicated by additional unknown repressive

factors that undermine the positive control by RpoS. Nevertheless,

at the global level, RpoS regulated genes represented 28% of the

genes depicted in Fig. 6 (123/435 genes) and Fig. 7 (51/182

genes). In comparison, RpoS regulates 11% of the transcriptome

in stationary phase in LB (418/3810 genes) [47]. The difference

between 28% and 11% is significantly different (P,0.01; Pearson’s

chi-square test), and the increased representation of RpoS

regulated genes in the pond is therefore not due to chance

alone. Future studies of rpoS mutants in pond water may aid in

determining the role rpoS plays in the entry of V. cholerae into the

ABNC state.

Discussion

This project was designed to concurrently test three critical

factors for their relevance to the transmission of cholera. The

patient to pond microcosm system allowed us to evaluate (i) the

infectivity of V. cholerae as the cells enter into (ii) the ABNC state in

(iii) the presence or absence of lytic vibriophage. The ID50 data

suggest that the major contributors to infection are culturable V.

cholerae. Phage did not affect colonization immediately after

passage from the patients because the PFU titer was likely too

low. However, V. cholerae failed to colonize the animals after 24 h

of adaptation to pond water – the point when the PFU titer and

ABNC cells were highest. Taken together, these data challenge the

concept that the aquatic environment is an amenable refuge for V.

cholerae during transit between human hosts.

The entry into the ABNC state has been challenging to

standardize experimentally because it is difficult to sufficiently

dilute the cells to the point that they do not self fertilize key

nutrients, and at the same time maintain a high enough cell

density for tractable experimentation. These problems were

overcome by using dialysis tubes containing V. cholerae in

suspension in a large volume (80 L) of live pond water. In this

system, culturability reproducibly fell by approximately 60% and

98% by 5 and 24 h, respectively, independent of the origin of the

bacteria. Defining the physiologic state of cells that do not culture

has been controversial. Herein we limit our work to two

populations: cells that culture and those that do not culture. We

do not differentiate within the population of non-culturable cells

that may contain a subpopulation of dead cells. That said, the non-

culturable cells are likely to be alive for several reasons: propidium

iodide staining for intact membranes indicated the majority of cells

in all arms of the study at 24 h had intact membranes (data not

shown). Secondly, the RNA yield was similar between 0 and 24 h

despite the 98% loss in culturable cells. Finally, the microarray

data at 24 h showed continued adaptation in the pond. For

example, genes involved with adaptation to low phosphate and

fixed nitrogen were induced [59,60]. Tests for metabolic activity

were not performed on all samples. Instead, we define ‘active’ in

the context of this project as the capacity for transcriptional

change despite a lack of culturability.

Having defined the proportion of non-culturable cells at 24 h,

we tested the relevance of ABNC cells to infection. In the context

of the pond microcosm, the total cell counts remained constant but

the proportion of culturable cells decreased over time. We tested

several hypotheses to determine the role of non-culturable cells in

transmission. One hypothesis stated that if culturable cells are

more infectious than non-culturable cells, then the ID50 calculated

by total cell counts will increase as the percent of culturable cells

decreases. The data revealed that the ID50 by total cell counts rose

as the culturable counts decreased at 0, 5 and 24 h. Therefore,

these data suggest that culturable V. cholerae were the major

contributors to infection. Previous studies have demonstrated

infection with ABNC V. cholerae is possible without an in vitro

pregrowth in rich media, but the doses used in these experiments

were often quite high [14]. We do not propose diminishing the

significance of the ABNC state as ABNC bacteria may still play a

vital role in maintaining environmental reservoirs of facultative

pathogens between outbreaks. However, our results indicate that

the relevance of ABNC V. cholerae during an outbreak may be

limited. These results are consistent with other systems that draw

into question the role of ABNC cells in infection without in vitro

pregrowth [61,62].

In Dhaka, Bangladesh, lytic vibriophage are common in human

patients and the environment [27,28,29]. The phage fluctuate in

number seasonally in delayed concordance with cholera outbreaks

[26,27,28], and household contacts of index cases that do not have

lytic phage are at an increased risk of being infected with V. cholerae

[29]. Despite this epidemiology, the dynamic role that phage play

in the environment has not been studied. Phage carried over from

the rice-water stool samples bloomed in the pond microcosm by

5 h. There was no significant rise in the phage titer between 5 and

24 h. However, the ratio of phage to CFU increased because of

the continued decline in culturable counts between 5 and 24 h.

Production of lytic phage is dependent on the growth of its host.

Since the bacteria had no net increase/decrease in cell number in

the pond system in the presence or absence of phage, it is likely

that there was not sufficient growth capacity to make more phage.

At the highest dose of V. cholerae, there was only partial

colonization of mice infected with patient derived V. cholerae at

5 h, and fewer mice infected at 24 h. These data suggest that

phage may reduce colonization, and at 24 h, the negative impact

of phage on infectivity is exacerbated by the decline of culturable

V. cholerae. Coinfection experiments with V. cholerae and lytic phage

confirmed that phage have a negative impact on colonization. At

low doses of bacteria and high doses of phage, the bacteria became

more infectious. The relevance of this phenotype to the natural

environment remains to be determined. The ready generation of

LPS mutants in the coinfection experiment provides one

mechanism by which bacteria may escape phage, but this is

detrimental for the bacteria as LPS mutants are attenuated

[63,64]. This attenuation may be one reason why LPS mutants

may not accumulate in the environment. Future studies to

elucidate the mechanisms by which lytic phage may influence

the infectivity of V. cholerae will add an additional dynamic to

consider when modeling cholera transmission. At the most basic

level, a better quantification of the infectious dose of V. cholerae in

the natural setting, and the seasonal titer of phage and V. cholerae in

the environment, will be a starting point for these future studies.

The microarray platform was similar to those previously used, but

the method of analyzing the data by ANOVA and PCA to compare

biological groups is novel for the V. cholerae field. ANOVA and PCA

have been used in this manner in other fields when comparing large

numbers of varied sample types [65,66,67]. qRT-PCR was used to

validate the microarray. The results between qRT-PCR and the

microarray were concordant; no false positives were observed by

microarray when crosschecked with qRT-PCR. However as

expected, the qRT-PCR was more sensitive than the microarray,

and the opportunity for false negatives in the microarray analysis is

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 12 October 2008 | Volume 4 | Issue 10 | e1000187

therefore increased. The decrease in sensitivity by microarray was

most pronounced with the patient derived samples after 24 h in

pond water – with and without phage. One explanation for this

decrease in sensitivity is that the RNA may be damaged at 24 h. If

true, this would provide some insight into the physiological status of

the bacteria at 24 h. qRT-PCR was normalized to an internal

reference gene (sanA and argS) whereas the microarray was

normalized to the global expression level. Therefore, qRT-PCR is

less affected by RNA damage in this case because the references will

theoretically also be equally affected by damaged RNA. If the RNA

from cells incubated for 24 h is indeed damaged, this suggests that

ABNC bacteria may be less capable of maintaining ribonucleic acid

integrity. Despite these concerns, the congruence between the 5 h

and 24 h samples at the global level demonstrates that the 24 h data

are still informative albeit with reduced sensitivity at the level of

individual genes.

The microarrays reveal immediate and striking transcriptional

adjustment to the pond within the first 5 h. As the bacteria enter

the ABNC state, they adapt to the nutrient limited nature of pond

water by upregulating genes required for low phosphate and fixed

nitrogen conditions. This adaptation is similar between patient

and in vitro derived samples. In addition, there is a general down

regulation of ribosomal proteins that indicates a general decline in

the ability to synthesize new proteins. These congruent expression

patterns are contrasted by divergent profiles of patient and in vitro

derived samples at the global level. At both the 5 and 24 h time

points, the patient derived samples do not converge with the in vitro

derived samples by cluster analysis or by PCA. This result was not

caused by the lytic phage in the patient derived samples because

patient derived samples without phage were also divergent from

the in vitro derived samples. These findings are supported by in vitro

studies with lytic phage and E. coli and P. aeruginosa that detected

changes in less than 4% of genes [68,69].

The lack of convergence between patient and in vitro derived

samples has relevance to vaccine development. One vaccine

strategy is to vaccinate with V. cholerae in a ABNC state expressing

‘environmental’ surface proteins. The method behind this strategy

was to transfer LB grown bacteria to pond water and allow the

bacteria to enter the ABNC state and express a new repertoire of

environmental antigens. This strategy is still valid, but we caution

that in vitro derived cells incubated in pond water may indeed be

ABNC, but they may express a different set of antigens than those

from patients incubated in pond water. Thus, it may prove

necessary to express surface proteins of ABNC bacteria derived

from patients by genetic modification using inducible expression

systems. Simple incubation of in vitro derived V. cholerae in pond

water may not be adequate.

In summary, the dynamic interaction between lytic phage and

bacteria in the pond environment suggests that the model of

cholera transmission be reconsidered with respect to the urgency

for transmission to the next host. Phage did not affect colonization

immediately after shedding from the patients because the phage

titer was too low. However, V. cholerae failed to colonize the

animals after 24 h of incubation in pond water – the point when

the phage and ABNC cell titers were highest. At 24 h, the rise of

ABNC cells and lytic phage blocked transmission. The dialysis

system was open with respect to small molecules but was closed

with respect to the phage. The impact of the phage in the natural

environment will be a function of the dilution of the bacteria away

from the phage in large bodies of water that are free-flowing open

systems such as rivers or closed systems such as ponds. Real-time

studies of phage and bacterial titers in such bodies of water will

provide a critical temporal factor to consider when gauging the

negative selective pressure imposed by lytic phage and the ABNC

state. Understanding this dynamic may ultimately demonstrate

that the unfavorable conditions in the environment provide the

critical selective pressure for toxigenic V. cholerae to maintain its

facultative pathogen life-history strategy.

Supporting Information

Figure S1 Hyperinfectivity is not induced in vitro.

Found at: doi:10.1371/journal.ppat.1000187.s001 (108 KB PDF)

Table S1 Bacterial strains and primers used in this study.

Found at: doi:10.1371/journal.ppat.1000187.s002 (13 KB PDF)

Table S2 Median qRT-PCR values per biological comparison

(N = 3).

Found at: doi:10.1371/journal.ppat.1000187.s003 (68 KB PDF)

Table S3 Genes with differential expression (P,161027) in at

least one of the six conditions described in Fig. 6A–Node 1.

Found at: doi:10.1371/journal.ppat.1000187.s004 (20 KB PDF)

Table S4 Genes with differential expression (P,161027) in at

least one of the six conditions described in Fig. 6A–Node 2A.

Found at: doi:10.1371/journal.ppat.1000187.s005 (27 KB PDF)

Table S5 Genes with differential expression (P,161027) in at

least one of the six conditions described in Fig. 6A–Node 2B.

Found at: doi:10.1371/journal.ppat.1000187.s006 (22 KB PDF)

Table S6 Genes with differential expression (P,161027) in at

least one of the six conditions described in Fig. 6A–Node 3.

Found at: doi:10.1371/journal.ppat.1000187.s007 (32 KB PDF)

Table S7 Genes with differential expression (P,161027) in at

least one of the six conditions described in Fig. 6A–Node 4.

Found at: doi:10.1371/journal.ppat.1000187.s008 (18 KB PDF)

Table S8 Genes with differential expression (P,161027) in at

least one of the six conditions described in Fig. 7A–Node 1.

Found at: doi:10.1371/journal.ppat.1000187.s009 (32 KB PDF)

Table S9 Genes with differential expression (P,161027) in at

least one of the six conditions described in Fig. 7A–Node 2.

Found at: doi:10.1371/journal.ppat.1000187.s010 (16 KB PDF)

Table S10 Microarray data for the patient derived V. cholerae

(phage+) vs. the in vitro derived V. cholerae comparison are provided

in an Excel spreadsheet.

Found at: doi:10.1371/journal.ppat.1000187.s011 (50 KB PDF)

Acknowledgments

We would like to thank D. Lazinski, K. Nair, R. Isberg, J. Coburn, and H.

Wortis for their critical evaluation and direction throughout this project.

This project would not have been possible without the diligence and

commitment from the animal facility at the ICDDR,B run by K.M.N.

Islam. Kind assistance with animal experiments was also provided by R.

Tamayo and A. Bishop. E. Snesrud and the J. Craig Venter Institute, F.

Wallace-Gadsden, D. S. Merrell, and G. Schoolnik provided guidance on

the microarray design and analysis. We thank R. Auxier and A. Parastoo at

the University of Georgia for their enthusiasm to chemically analyze rice-

water stool.

Author Contributions

Conceived and designed the experiments: E. Nelson, J. Flynn, S. Schild, L.

Bourassa, R. LaRocque, S. Calderwood, F. Qadri, A. Camilli. Performed the

experiments: E. Nelson, A. Chowdhury, S. Schild, L. Bourassa, Y. Shao.

Analyzed the data: E. Nelson, J. Flynn, A. Camilli. Contributed reagents/

materials/analysis tools: E. Nelson, J. Flynn, R. LaRocque, S. Calderwood, F.

Qadri, A. Camilli. Wrote the paper: E. Nelson, A. Camilli.

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 13 October 2008 | Volume 4 | Issue 10 | e1000187

References

1. Bryce J, Boschi-Pinto C, Shibuya K, Black RE (2005) WHO estimates of the

causes of death in children. Lancet 365: 1147–1152.

2. Kosek M, Bern C, Guerrant RL (2003) The global burden of diarrhoeal disease,as estimated from studies published between 1992–2000. Bulletin of the World

Health Organization 81: 197–204.

3. Koch R (1884) An address on cholera and its bacillus. Br Med J 2: 403–407.

4. Sack DA, Sack RB, Chaignat CL (2006) Getting serious about cholera.N Engl J Med 355: 649–651.

5. Alam M, Sultana M, Nair GB, Siddique AK, Hasan NA, et al. (2007) Viable butnonculturable Vibrio cholerae O1 in biofilms in the aquatic environment and their

role in cholera transmission. Proc Natl Acad Sci U S A 104: 17801–17806.

6. Ryan ET, Dhar U, Khan WA, Salam MA, Faruque AS, et al. (2000) Mortality,

morbidity, and microbiology of endemic cholera among hospitalized patients inDhaka, Bangladesh. Am J Trop Med Hyg 63: 12–20.

7. Harris JB, Larocque RC, Chowdhury F, Khan AI, Logvinenko T, et al. (2008)Susceptibility to Vibrio cholerae Infection in a Cohort of Household Contacts of

Patients with Cholera in Bangladesh. PLoS Negl Trop Dis 2: e221. doi:10.1371/

journal.pntd.0000221.

8. Alam M, Hasan NA, Sadique A, Bhuiyan NA, Ahmed KU, et al. (2006)Seasonal cholera caused by Vibrio cholerae serogroups O1 and O139 in the coastal

aquatic environment of Bangladesh. Appl Environ Microbiol 72: 4096–4104.

9. Islam MS, Talukder KA, Khan NH, Mahmud ZH, Rahman MZ, et al. (2004)

Variation of toxigenic Vibrio cholerae O1 in the aquatic environment of

Bangladesh and its correlation with the clinical strains. Microbiol Immunol48: 773–777.

10. Stine OC, Alam M, Tang L, Nair GB, Siddique AK, et al. (2008) Seasonalcholera from multiple small outbreaks, rural Bangladesh. Emerg Infect Dis 14:

831–833.

11. Glass RI, Huq I, Alim AR, Yunus M (1980) Emergence of multiply antibiotic-

resistant Vibrio cholerae in Bangladesh. J Infect Dis 142: 939–942.

12. Saha D, Karim MM, Khan WA, Ahmed S, Salam MA, et al. (2006) Single-dose

azithromycin for the treatment of cholera in adults. N Engl J Med 354:2452–2462.

13. Schild S, Tamayo R, Nelson EJ, Qadri F, Calderwood SB, et al. (2007) Genes

induced late in infection increase fitness of Vibrio cholerae after release into the

environment. Cell Host Microbe 2: 264–277.

14. Colwell RR, Brayton PR, Grimes DJ, Roszak DB, Huq SA, et al. (1985) Viable

but non-culturable Vibrio cholerae and related pathogens in the environment:Implications for release of genetically engineered microorganisms. Nature

Biotechnology 3: 817–820.

15. Besnard V, Federighi M, Declerq E, Jugiau F, Cappelier JM (2002)

Environmental and physico-chemical factors induce VBNC state in Listeria

monocytogenes. Vet Res 33: 359–370.

16. Oliver JD (2005) The viable but nonculturable state in bacteria. J Microbiol 43Spec No: 93–100.

17. Kell DB, Kaprelyants AS, Weichart DH, Harwood CR, Barer MR (1998)Viability and activity in readily culturable bacteria: a review and discussion of

the practical issues. Antonie Van Leeuwenhoek 73: 169–187.

18. Rawlings TK, Ruiz GM, Colwell RR (2007) Association of Vibrio cholerae O1 El

Tor and O139 Bengal with the Copepods Acartia tonsa and Eurytemora affinis. ApplEnviron Microbiol 73: 7926–7933.

19. Islam MS, Jahid MI, Rahman MM, Rahman MZ, Kabir MS, et al. (2007)Biofilm acts as a microenvironment for plankton-associated Vibrio cholerae in the

aquatic environment of Bangladesh. Microbiol Immunol 51: 369–379.

20. Colwell RR, Huq A, Islam MS, Aziz KM, Yunus M, et al. (2003) Reduction of

cholera in Bangladeshi villages by simple filtration. Proc Natl Acad Sci U S A100: 1051–1055.

21. Hartley DM, Morris JG Jr, Smith DL (2006) Hyperinfectivity: a critical elementin the ability of V. cholerae to cause epidemics? PLoS Med 3: e7. doi:10.1371/

journal.pmed.0030007.

22. Butler SM, Nelson EJ, Chowdhury N, Faruque SM, Calderwood SB, et al.

(2006) Cholera stool bacteria repress chemotaxis to increase infectivity. MolMicrobiol 60: 417–426.

23. Merrell DS, Butler SM, Qadri F, Dolganov NA, Alam A, et al. (2002) Host-induced epidemic spread of the cholera bacterium. Nature 417: 642–645.

24. Wiles S, Dougan G, Frankel G (2005) Emergence of a ‘hyperinfectious’ bacterialstate after passage of Citrobacter rodentium through the host gastrointestinal tract.

Cell Microbiol 7: 1163–1172.

25. Alam A, Larocque RC, Harris JB, Vanderspurt C, Ryan ET, et al. (2005)

Hyperinfectivity of human-passaged Vibrio cholerae can be modeled by growth inthe infant mouse. Infect Immun 73: 6674–6679.

26. Pasricha CL, de Monte AJH, Gupta IK (1932) Cholera and cholera likevibriophages. Indian med gaz 67.

27. Faruque SM, Naser IB, Islam MJ, Faruque AS, Ghosh AN, et al. (2005)Seasonal epidemics of cholera inversely correlate with the prevalence of

environmental cholera phages. Proc Natl Acad Sci U S A 102: 1702–1707.

28. Faruque SM, Islam MJ, Ahmad QS, Faruque AS, Sack DA, et al. (2005) Self-

limiting nature of seasonal cholera epidemics: Role of host-mediatedamplification of phage. Proc Natl Acad Sci U S A 102: 6119–6124.

29. Nelson EJ, Chowdhury A, Harris JB, Begum YA, Chowdhury F, et al. (2007)Complexity of rice-water stool from patients with Vibrio cholerae plays a role in the

transmission of infectious diarrhea. Proc Natl Acad Sci U S A 104:

19091–19096.

30. Jensen MA, Faruque SM, Mekalanos JJ, Levin BR (2006) Modeling the role ofbacteriophage in the control of cholera outbreaks. Proc Natl Acad Sci U S A

103: 4652–4657.

31. Qadri F, Wenneras C, Albert MJ, Hossain J, Mannoor K, et al. (1997)

Comparison of immune responses in patients infected with Vibrio cholerae O139

and O1. Infect Immun 65: 3571–3576.

32. Sambrook J, Fritsch EF, Maniatis T (1989) Molecular Cloning: a laboratory

manual. NY: Cold Spring Harbor.

33. Hava DL, Camilli A (2001) Isolation and characterization of a temperature-sensitive generalized transducing bacteriophage for Vibrio cholerae. J Microbiol

Methods 46: 217–225.

34. Guidolin A, Manning PA (1988) Molecular analysis of the packaging signal in

bacteriophage CP-T1 of Vibrio cholerae. Mol Gen Genet 212: 514–521.

35. Nesper J, Kapfhammer D, Klose KE, Merkert H, Reidl J (2000) Character-

ization of Vibrio cholerae O1 antigen as the bacteriophage K139 receptor and

identification of IS1004 insertions aborting O1 antigen biosynthesis. J Bacteriol182: 5097–5104.

36. Schild S (2008) Personal communication.

37. Skorupski K, Taylor RK (1997) Cyclic AMP and its receptor protein negativelyregulate the coordinate expression of cholera toxin and toxin-coregulated pilus

in Vibrio cholerae. Proc Natl Acad Sci U S A 94: 265–270.

38. Nesper J, Kraiss A, Schild S, Blass J, Klose KE, et al. (2002) Comparative and

genetic analyses of the putative Vibrio cholerae lipopolysaccharide core

oligosaccharide biosynthesis (wav) gene cluster. Infect Immun 70: 2419–2433.

39. Schild S, Lamprecht AK, Reidl J (2005) Molecular and functional character-ization of O antigen transfer in Vibrio cholerae. J Biol Chem 280: 25936–25947.

40. Hitchcock PJ, Brown TM (1983) Morphological heterogeneity amongSalmonella lipopolysaccharide chemotypes in silver-stained polyacrylamide gels.

J Bacteriol 154: 269–277.

41. Tsai CM, Frasch CE (1982) A sensitive silver stain for detecting lipopolysac-

charides in polyacrylamide gels. Anal Biochem 119: 115–119.

42. Larocque RC, Harris JB, Dziejman M, Li X, Khan AI, et al. (2005)

Transcriptional profiling of Vibrio cholerae recovered directly from patientspecimens during early and late stages of human infection. Infect Immun 73:

4488–4493.

43. Meibom KL, Li XB, Nielsen AT, Wu CY, Roseman S, et al. (2004) The Vibrio

cholerae chitin utilization program. Proc Natl Acad Sci U S A 101: 2524–2529.

44. Conners SB, Montero CI, Comfort DA, Shockley KR, Johnson MR, et al.

(2005) An expression-driven approach to the prediction of carbohydratetransport and utilization regulons in the hyperthermophilic bacterium Thermotoga

maritima. J Bacteriol 187: 7267–7282.

45. Pysz MA, Conners SB, Montero CI, Shockley KR, Johnson MR, et al. (2004)

Transcriptional analysis of biofilm formation processes in the anaerobic,

hyperthermophilic bacterium Thermotoga maritima. Appl Environ Microbiol 70:6098–6112.

46. Wolfinger RD, Gibson G, Wolfinger ED, Bennett L, Hamadeh H, et al. (2001)Assessing gene significance from cDNA microarray expression data via mixed

models. J Comput Biol 8: 625–637.

47. Nielsen AT, Dolganov NA, Otto G, Miller MC, Wu CY, et al. (2006) RpoS

controls the Vibrio cholerae mucosal escape response. PLoS Pathog 2: e109.doi:10.1371/journal.ppat.0020109.

48. (World Heath Organization. 4th rev. – 2005) The treatment of diarrhoea – Amanual for physicians and other senior health workers. Geneva, Switzerland.

49. Kapfhammer D, Blass J, Evers S, Reidl J (2002) Vibrio cholerae phage K139:complete genome sequence and comparative genomics of related phages.

J Bacteriol 184: 6592–6601.

50. Reidl J, Mekalanos JJ (1995) Characterization of Vibrio cholerae bacteriophage

K139 and use of a novel mini-transposon to identify a phage-encoded virulencefactor. Mol Microbiol 18: 685–701.

51. Shimodori S, Takeya K, Takade A (1984) Lysogenicity and prophage type of thestrains of Vibrio cholerae O-1 isolated mainly from the natural environment.

Am J Epidemiol 120: 759–768.

52. Takeya K, Shimodori S, Gomez CZ (1967) Kappa-type phage detection as a

method for the tracing of cholera El Tor carriers. Bull World Health Organ 37:806–810.

53. Faruque SM, Asadulghani, Alim AR, Albert MJ, Islam KM, et al. (1998)Induction of the lysogenic phage encoding cholera toxin in naturally occurring

strains of toxigenic Vibrio cholerae O1 and O139. Infect Immun 66: 3752–3757.

54. Butler SM, Camilli A (2004) Both chemotaxis and net motility greatly influence

the infectivity of Vibrio cholerae. Proc Natl Acad Sci U S A 101: 5018–5023.

55. von Kruger WM, Humphreys S, Ketley JM (1999) A role for the PhoBR

regulatory system homologue in the Vibrio cholerae phosphate-limitation responseand intestinal colonization. Microbiology 145 (Pt 9): 2463–2475.

56. von Kruger WM, Lery LM, Soares MR, de Neves-Manta FS, Batista eSilva CM, et al. (2006) The phosphate-starvation response in Vibrio cholerae O1

and phoB mutant under proteomic analysis: disclosing functions involved inadaptation, survival and virulence. Proteomics 6: 1495–1511.

57. Leigh JA, Dodsworth JA (2007) Nitrogen regulation in bacteria and archaea.Annu Rev Microbiol 61: 349–377.

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 14 October 2008 | Volume 4 | Issue 10 | e1000187

58. Merrell DS, Tischler AD, Lee SH, Camilli A (2000) Vibrio cholerae requires rpoS

for efficient intestinal colonization. Infect Immun 68: 6691–6696.59. Tischler AD, Camilli A (2004) Cyclic diguanylate (c-di-GMP) regulates Vibrio

cholerae biofilm formation. Mol Microbiol 53: 857–869.

60. Tischler AD, Lee SH, Camilli A (2002) The Vibrio cholerae vieSAB locus encodesa pathway contributing to cholera toxin production. J Bacteriol 184: 4104–4113.

61. Sun F, Chen J, Zhong L, Zhang XH, Wang R, et al. (2008) Characterizationand virulence retention of viable but nonculturable Vibrio harveyi. FEMS

Microbiol Ecol 64: 37–44.

62. Du M, Chen J, Zhang X, Li A, Li Y, et al. (2007) Retention of virulence in aviable but nonculturable Edwardsiella tarda isolate. Appl Environ Microbiol 73:

1349–1354.63. Nesper J, Schild S, Lauriano CM, Kraiss A, Klose KE, et al. (2002) Role of Vibrio

cholerae O139 surface polysaccharides in intestinal colonization. Infect Immun70: 5990–5996.

64. Chiang SL, Mekalanos JJ (1999) rfb mutations in Vibrio cholerae do not affect

surface production of toxin-coregulated pili but still inhibit intestinal coloniza-tion. Infect Immun 67: 976–980.

65. de Haan JR, Wehrens R, Bauerschmidt S, Piek E, van Schaik RC, et al. (2007)

Interpretation of ANOVA models for microarray data using PCA. Bioinfor-

matics 23: 184–190.

66. Barrangou R, Azcarate-Peril MA, Duong T, Conners SB, Kelly RM, et al.

(2006) Global analysis of carbohydrate utilization by Lactobacillus acidophilus using

cDNA microarrays. Proc Natl Acad Sci U S A 103: 3816–3821.

67. van der Werf MJ, Pieterse B, van Luijk N, Schuren F, van der Werff-van der

Vat B, et al. (2006) Multivariate analysis of microarray data by principal

component discriminant analysis: prioritizing relevant transcripts linked to the

degradation of different carbohydrates in Pseudomonas putida S12. Microbiology

152: 257–272.

68. Ravantti JJ, Ruokoranta TM, Alapuranen AM, Bamford DH (2008) Global

transcriptional responses of Pseudomonas aeruginosa to phage PRR1 infection.

J Virol 82: 2324–2329.

69. Poranen MM, Ravantti JJ, Grahn AM, Gupta R, Auvinen P, et al. (2006) Global

changes in cellular gene expression during bacteriophage PRD1 infection. J Virol

80: 8081–8088.

Cholera Transmission Is Blocked by Lytic Phage

PLoS Pathogens | www.plospathogens.org 15 October 2008 | Volume 4 | Issue 10 | e1000187

Transmissibility of cholera: In vivo-formed biofilmsand their relationship to infectivity and persistencein the environmentShah M. Faruque*†, Kuntal Biswas*, S. M. Nashir Udden*, Qazi Shafi Ahmad*, David A. Sack*, G. Balakrish Nair*†,and John J. Mekalanos†‡

*Molecular Genetics Laboratory, International Centre for Diarrhoeal Disease Research, Bangladesh, Dhaka-1212, Bangladesh; and ‡Department ofMicrobiology and Molecular Genetics, Harvard Medical School, 200 Longwood Avenue, Boston, MA 02115

Contributed by G. Balakrish Nair, February 15, 2006

The factors that enhance the waterborne spread of bacterialepidemics and sustain the epidemic strain in nature are unclear.Although the epidemic diarrheal disease cholera is known to betransmitted by water contaminated with pathogenic Vibrio chol-erae, routine isolation of pathogenic strains from aquatic environ-ments is challenging. Here, we show that conditionally viableenvironmental cells (CVEC) of pathogenic V. cholerae that resistcultivation by conventional techniques exist in surface water asaggregates (biofilms) of partially dormant cells. Such CVEC can berecovered as fully virulent bacteria by inoculating the water intorabbit intestines. Furthermore, when V. cholerae shed in stools ofcholera patients are inoculated in environmental water samples inthe laboratory, the cells exhibit characteristics similar to CVEC,suggesting that CVEC are the infectious form of V. cholerae inwater and that CVEC in nature may have been derived from humancholera stools. We also observed that stools from cholera patientscontain a heterogenous mixture of biofilm-like aggregates andfree-swimming planktonic cells of V. cholerae. Estimation of therelative infectivity of these different forms of V. cholerae cellssuggested that the enhanced infectivity of V. cholerae shed inhuman stools is largely due to the presence of clumps of cells thatdisperse in vivo, providing a high dose of the pathogen. The resultsof this study support a model of cholera transmission in which invivo-formed biofilms contribute to enhanced infectivity and envi-ronmental persistence of pathogenic V. cholerae.

cholera epidemic � Vibrio cholerae � conditionally viable environmentalcells � waterborne disease

Toxigenic Vibrio cholerae of the O1 or O139 serogroup are thecausative agents of cholera and belong to a group of organ-

isms whose major habitat is aquatic ecosystems (1–3). Epidemicsof cholera occur frequently in many developing countries ofAsia, Africa, and Latin America (4, 5), and these occurrencescoincide with increased prevalence of the causative V. choleraestrain in the aquatic environment (6). However, the concentra-tion of toxigenic V. cholerae usually detected in surface water bystandard culture techniques is far less than that required toinduce infection and cause clinical disease under experimentalconditions in volunteers challenged with V. cholerae (7, 8). Thisdiscrepancy between the required infectious dose and the ap-parent concentration of V. cholerae in surface water fostering anepidemic has not been adequately explained.

Various hypotheses have been proposed to explain the modeof persistence of pathogenic V. cholerae in the aquatic environ-ment and to suggest that the true prevalence of the pathogen innatural water may be considerably higher than that observed bystandard culture methods. For example, it has been proposedthat V. cholerae can exist in the aquatic environment in a viablebut nonculturable state, which by definition is a condition inwhich cells are incapable of forming a colony on commonly usedmedia and thus cannot be recovered from the water by standardculture techniques (1, 2). This hypothesis stems from observa-

tions that water carries vibrio-shaped cells that appear to beantigenically related to V. cholerae O1 strains, as determined bya fluorescent anti-O1 antibody-based procedure (9), but the cellsare apparently not culturable. However, without a verifiedresuscitation procedure, it has not been possible to ascertainwhether the same strain producing viable but nonculturable cellcounts in the water is responsible for cholera epidemics. Both V.cholerae O1 and the non-O1�non-O139 strains that do not carrythe major virulence genes also exist in the environment, furthercomplicating this challenge (10).

Another explanation for the low recovery of V. cholerae fromenvironmental water by conventional culture techniques mightrelate to the cellular organization of the bacteria in the water.Environmental bacteria are known to form biofilms, which aresurface-associated communities of bacterial cells, to enhance theirsurvival under adverse conditions (11). V. cholerae also has beenpostulated to form biofilms in association with animate and inan-imate surfaces (12, 13). Because the transition between a planktonicform of the bacteria and biofilms is a complex and highly regulatedprocess, conventional culture techniques may not permit accurateestimation of V. cholerae that exist as biofilms in the environment.The role of such biofilms in the epidemiology of cholera is alsounclear, but presumably these cells can reconvert to active bacteriaand cause subsequent outbreaks of the disease.

We recently showed that a significant proportion of toxigenic V.cholerae O1 cells in the aquatic environment exist in a ‘‘condition-ally viable’’ state and require appropriate growth conditions fortheir detection, and we designated these cells ‘‘conditionally viableenvironmental cells’’ (CVEC) (14). However, the public healthimportance of this survival form of V. cholerae O1 depends onwhether these cells are infectious and naturally reconvertible toactive bacteria. Here, we demonstrate that V. cholerae CVEC areaggregates of partially dormant cells that resuscitate to normal,viable bacteria, both under appropriate in vitro conditions and in theintestinal environment of adult rabbits. We further show that theCVEC are derived from biofilm-like multicellular clumps of V.cholerae shed in human stools, and that such biofilms are respon-sible for enhanced infectivity of V. cholerae. This study thus providesinsight into factors that promote the rapid spread of choleraepidemics, as well as into the persistence in the environment of V.cholerae with epidemic potential.

ResultsCharacterization of V. cholerae CVEC. Our method for demonstrat-ing the presence of CVEC of V. cholerae in water samples that

Conflict of interest statement: No conflicts declared.

Abbreviations: CVEC, conditionally viable environmental cells; AST, antibiotic selectiontechnique; SmR, streptomycin resistance; NalR, nalidixic acid resistance; cfu, colony-formingunits; BP, bile�peptone; TTGA, taurocholate–tellurite–gelatin agar.

†To whom correspondence may be addressed. E-mail: faruque@icddrb.org, gbnair@icddrb.org, or shannon�humphreys@hms.harvard.edu.

© 2006 by The National Academy of Sciences of the USA

6350–6355 � PNAS � April 18, 2006 � vol. 103 � no. 16 www.pnas.org�cgi�doi�10.1073�pnas.0601277103

only rarely yielded viable V. cholerae by conventional cultureexploited the antibiotic-resistant property of the bacteria andwas termed the ‘‘antibiotic selection technique’’ (AST). Thestrain of V. cholerae known to have caused recent choleraepidemics in Bangladesh carries the SXT element (15, 16), whichencodes resistance to multiple antibiotics, including streptomy-cin, sulfamethoxazole, and trimethoprim. This strain is alsoconstitutively resistant to nalidixic acid. We used a selectivemedium containing streptomycin (70 �g�ml) to isolate V. chol-erae O1 with epidemic potential from surface water samples thatappeared negative for the organisms in conventional assays.However, we observed that a prerequisite to the success of theAST procedure was that the water sample be cultured for at least5 h in bile�peptone (BP; 1% peptone�0.5% taurocholic acid�1%NaCl, pH 9.0) medium (enrichment), in the absence of strep-tomycin, before aliquots were plated on the antibiotic-containingplates. When streptomycin was added directly to the enrichmentmedium, very few (0.03–0.8%) V. cholerae O1 colonies could berecovered (see Table 3, which is published as supporting infor-mation on the PNAS web site). This finding led us to explorewhether V. cholerae CVEC were impaired in expressing strep-tomycin resistance (SmR) prior to being revitalized by initialculturing in the absence of streptomycin.

In the epidemic strain of V. cholerae, SmR is mediated throughenzymatic modification of the antibiotic by enzymes encoded bythe strAB genes of the SXT element (15, 16). Cells that aredefective in protein synthesis are unlikely to produce the re-quired enzymes, despite carrying their genetic determinants, andhence are unable to express SmR phenotype. Furthermore,because streptomycin itself is an inhibitor of protein synthesis, atemporary or partial impairment of SmR is likely to be furtheraugmented by the presence of streptomycin, resulting in com-plete inhibition of protein synthesis and eventual cell death. Ourresults indicated that, in the CVEC state, V. cholerae cellspossibly undergo at least a partial impairment in protein syn-thesis and hence cannot express SmR unless they are firstrevitalized by growing in nutrient medium, in the absence of theantibiotic. We verified this assumption by exploiting the nalidixic

acid resistance (NalR) of the same bacteria, because NalR is notdirectly dependent on protein synthesis, but is associated withmutations in the gyrA gene encoding the GyrA subunit of DNAgyrase (17). We conducted assays in which nalidixic acid wasused in the enrichment media instead of streptomycin, beforeselection of the organism on antibiotic-containing plates. Asexpected, use of nalidixic acid during the enrichment did notdrastically reduce the recovery of cells (Table 1), suggesting that,in contrast to SmR, the NalR phenotype remained largely un-changed in the CVEC form of V. cholerae. These results furthersupported our assumption that the CVEC form of V. choleraefound in water represents a metabolically impeded state of thebacteria, as evidenced by their lack of expression of SmR.

V. cholerae CVEC Exist as Multicellular Clumps. We further examinedthe cellular organization and metabolic status of CVEC byconducting a series of experiments that included in vitro growthkinetics assay and fluctuation analysis. In the growth kineticsassay, samples were removed from the enrichment mixture in BPmedium at regular intervals and plated on antibiotic-containingselective medium for the target V. cholerae strain. We sought tounderstand whether the rate of increase in colony counts withtime is comparable with that predicted on the basis of an increasein cell number by binary fission, or whether the increase isconsiderably higher, suggesting possible activation of dormant V.cholerae cells. The results showed a minimal increase in viablecell number during the first 4 h of enrichment; subsequently,however, viable cells increased drastically faster than would beconsistent with normal bacterial growth kinetics (Table 2). Thisinitial lack of viable cell counts was also inconsistent with theduration of the lag phase of normal bacterial growth, as wasobserved with laboratory-grown V. cholerae in control experi-ments. Thus, the growth kinetics study suggested that V. choleraeCVEC are living cells, but they require at least 4 h of enrichmentto assume normal growth and colony formation. The resuscita-tion kinetics are also consistent with results reported by Bakeret al. (18) on the properties of nutrient-starved V. cholerae cells,which required incubation for �5 h in rich medium to reassume

Table 1. Differential response of environmental V. cholerae O1 cells to streptomycin andnalidixic acid in enrichment culture, despite carrying genetic determinants for resistanceto both antibiotics

Date ofsamplecollection

SampleID

V. cholerae O1 counts (cfu�ml) on TTGA platescontaining streptomycin, after enrichment for 6 h

under various conditions*

BP mediumwithout

antibiotic

BP mediumcontaining

streptomycin†

BP mediumcontaining

nalidixic acid‡

Sept. 6, 2005 1714 3.3 � 105 0 1.8 � 104

Sept. 11, 2005 1724 8.6 � 104 0 8.4 � 103

Oct. 2, 2005 1753 1.2 � 104 0 2.6 � 104

Oct. 2, 2005 1754 1.2 � 104 0 1.8 � 104

Oct. 9, 2005 1764 2.0 � 104 0 1.2 � 104

Nov. 28, 2005 1824 1.6 � 104 2 � 101 1.5 � 104

Nov. 28, 2005 1825 8.0 � 102 0 1.1 � 103

Dec. 6, 2005 1834 2.4 � 103 0 1.7 � 103

Dec. 13, 2005 1845 2.0 � 102 0 2.7 � 102

Dec. 13, 2005 1847 6.8 � 103 0 9.2 � 103

cfu, colony-forming units; TTGA, taurocholate–tellurite–gelatin agar.*Each value represents the mean of at least three independent observations with different aliquots of the samesample.

†A colony count of zero means that no colonies were detected on plating 100-�l aliquots of enriched culture. Intheory, the concentration of bacteria should be �10 cfu�ml.

‡All V. cholerae strains selected on nalidixic acid plates were later found to be resistant to streptomycin and tocarry the SXT element.

Faruque et al. PNAS � April 18, 2006 � vol. 103 � no. 16 � 6351

MIC

ROBI

OLO

GY

normal size and shape under laboratory conditions. Thus, theCVEC present in water samples characterized in the currentstudy appear to be viable V. cholerae cells that are perhapssimilar to the nutrient-starved cells described by Baker et al.

Dispersion of clumped viable cells present in biofilms might alsoexplain the sudden increase in viable counts that occurred after 4–5h of enrichment cultivation of CVEC. Thus, we further investigatedwhether CVEC are planktonic cells or are present in clumps thatdisperse into planktonic cells during the AST procedure. Fluctua-tion analysis (19) was used to determine whether there was anasymmetric spatial distribution of CVEC in the water samples. Inthis assay, water samples were diluted in BP medium, and thediluted samples were broken down further into multiple smallaliquots. Each aliquot was assayed for the appearance of viable cellsduring the AST procedure by using the growth kinetics assaydescribed above. Although most of the aliquots were found to benegative for V. cholerae, the assays revealed ‘‘jackpots’’ of V.cholerae cells in small number of aliquots, and detectable cellsincreased dramatically in the tubes containing these aliquots (Table4, which is published as supporting information on the PNAS website). Therefore, it appears that the CVEC exist as clumps ofpotentially viable cells that disperse into viable planktonic cellsduring the enrichment process. Taken together, these results sug-gest that CVEC are aggregates of metabolically impeded cells, andthat the resuscitation process involves both dispersion of clumps andactivation of cells.

To observe the presence of clumped cells, we conductedfluorescent anti-O1 antibody-based microscopic examination ofaliquots of water samples, and this assay showed the presence ofclumps of cells that reacted with the antibody (Fig. 1). Interest-ingly, some of these cells showed a coccoid morphology, inagreement with the morphology of starved V. cholerae cellsreported by Baker et al. (18). To further document that theclumped cells were alive, we conducted assays involving enrich-ment on agar plates instead of in broth medium. We platedaliquots of the same water on a nylon membrane placed on BPagar plates; after 5 h of enrichment, the membrane was trans-ferred to the selective plate [taurocholate–tellurite–gelatin agar

(TTGA) containing streptomycin]. Large ‘‘macrocolonies’’formed on the plates (Fig. 4, which is published as supportinginformation on the PNAS web site) and were confirmed to becomposed of V. cholerae O1 cells by biochemical tests andagglutination with V. cholerae O1-specific antiserum. Ribotypingand antibiotic resistance profiles confirmed that these cells wereidentical to the strain responsible for recent epidemics of chol-era. We conclude that the unusual morphology of these coloniesreflects the attachment of large number of vibrios to macro-scopic material present in the aquatic samples.

CVEC Convert to Infectious Bacteria in the Rabbit Intestine. In viewof previous studies suggesting that the epidemiological cycle of

Table 2. Growth kinetics assays for V. cholerae O1 CVEC in environmental waters

Date ofsamplecollection

SampleID

V. cholerae O1 counts (cfu�ml)*† on TTGA plates containingstreptomycin, after enrichment for 1–6 h in BP medium

1 h 2 h 3 h 4 h 5 h 6 h

Aug. 2, 2005 1664 0 0 0 8.9 � 102 2.2 � 104 7.5 � 104

Aug. 9, 2005 1674 0 0 0 1.2 � 103 4.1 � 104 1.2 � 105

Aug. 14, 2005 1684 0 0 0 2.4 � 103 7.6 � 104 4.0 � 105

Oct. 9, 2005 1753 0 0 2.0 � 101 2.5 � 102 2.5 � 103 5.4 � 103

Oct. 9, 2005 1754 0 0 0 5.0 � 101 6.0 � 102 3.3 � 103

Oct. 17, 2005 1774 0 0 6.0 � 101 1.2 � 103 1.1 � 104 3.1 � 104

Nov. 28, 2005 1824 0 0 2.1 � 101 3.9 � 102 7.4 � 103 2.7 � 104

Nov. 28, 2005 1825 0 0 0 2.4 � 101 3.3 � 102 1.2 � 103

Dec. 6, 2005 1834 0 0 0 1.1 � 102 2.6 � 103 1.5 � 104

Dec. 13, 2005 1845 0 0 0 1.6 � 102 1.9 � 103 7.3 � 103

Dec. 13, 2005 1847 0 0 2.0 � 101 2.4 � 102 3.6 � 103 1.2 � 104

Laboratory-grownenvironmental straindiluted in PBS

8.0 � 101 3.6 � 102 1.7 � 103 7.4 � 103 2.9 � 104 1.5 � 105

Laboratory-grownclinical strain dilutedin PBS

2.5 � 101 9.0 � 101 4.2 � 102 2.2 � 103 9.2 � 103 4.5 � 104

*A colony count of zero means that no colonies were detected on plating of 100-�l aliquots of enriched culture.In theory, the concentration should be �10 colonies per ml.

†The rate of increase in viable cells during enrichment of V. cholerae in environmental water samples wasconsiderably faster than would be consistent with normal bacterial growth kinetics between 3 and 5 h,suggesting that cells might have generated by activation of dormant V. cholerae cells.

Fig. 1. Fluorescent antibody-based detection of V. cholerae O1 CVEC inwater samples. Shown are aggregates of CVEC reacting with V. choleraeO1-specific monoclonal antibody (A–C), as well as planktonic cells of V. chol-erae O1 (B and D). C shows high magnification of a CVEC clump that appearsto contain both coccoid (starved) and rod-shaped V. cholerae O1 cells. Thecorresponding water samples produced viable V. cholerae O1 in the AST assay,as well as in the ileal loop assay in rabbits, and further analyses confirmed thatthese isolates were identical to the strain responsible for recent epidemics ofcholera in Bangladesh.

6352 � www.pnas.org�cgi�doi�10.1073�pnas.0601277103 Faruque et al.

pathogenic V. cholerae is supported by both environmental andhost factors (4), we examined whether the CVEC form of V.cholerae would convert to viable pathogenic bacteria in themammalian intestinal environment. Adult rabbits have longbeen used as a suitable model to study the pathogenesis of V.cholerae (20). We inoculated rabbit ileal loops with aliquots ofwater sample that were negative for V. cholerae O1 in conven-tional culture but appeared to contain CVEC in the AST assay.After 18 h of incubation, 46.7% of the ileal loops were found tocontain between 1.2 � 102 and 5.7 � 106 cells of viable toxigenicV. cholerae O1 (Table 5, which is published as supportinginformation on the PNAS web site), and 38.8% of the V.cholerae-positive ileal loops also showed fluid accumulation,which is characteristic of enterotoxin production by the bacteria.Thus, it appears that CVEC are converted to viable virulentbacteria in the mammalian intestinal environment. These resultssuggest that CVEC forms of environmental V. cholerae are likelyto be infectious in humans, and because CVEC exist as multi-cellular aggregates, human victims are likely to receive a signif-icantly high dose of the pathogen by ingesting CVEC. Theobservation that only some aliquots of the same water are ableto cause a fluid-accumulation response in the rabbit intestine(Table 5) may also reflect the asymmetric spatial distribution ofthe CVEC clumps in the water.

CVEC-Like Properties of V. cholerae Excreted in Human Stools. Therapid spread of cholera epidemics is known to be fostered bysurface waters through contamination of water by fecal materialfrom cholera patients. To understand the possible origin of theCVEC found in water, we designed experiments to test whetherV. cholerae shed in cholera stools show CVEC-like propertieswhen inoculated in otherwise V. cholerae O1-negative environ-mental water samples. We collected stools from cholera patientsand initially examined the stools by dark-field microscopy for thepresence of characteristic highly motile V. cholerae organisms.Aliquots of the stools were inoculated in environmental watersamples that had previously tested negative for V. cholerae O1and O139. The pH of the water samples was between 7.0 and 7.5.After the samples had been maintained for 6 h and 72 h at roomtemperature, aliquots of each water sample were analyzed withthe same AST procedure and growth kinetics assay we used totest environmental water samples. The results suggested that atleast a portion of the V. cholerae organisms shed in stools hadconverted to CVEC after inoculation in water (Table 6, which ispublished as supporting information on the PNAS web site), asevidenced by their loss of streptomycin resistance while retainingthe NalR phenotype. Interestingly, growth kinetics of cells inthese samples also resembled those of naturally occurring CVEC(Table 7, which is published as supporting information on thePNAS web site). However, this transformation of V. cholerae instools to the CVEC form was not observed when stools left for6 h without dilution in water were subjected to the same assay

(data not shown). Similarly, when stools were diluted in PBS (pH7.2) instead of water, the conversion to the CVEC form was notobserved (Table 7). Because environmental fresh water is mostlyhypotonic, the conversion to CVEC may have been driven, atleast in part, by hypoosmotic stress. Lack of optimum NaClconcentration might also have contributed to the process, be-cause many V. cholerae metabolic functions are powered by Na�

gradient (21, 22).

In Vivo-Formed Biofilms Are Responsible for Enhanced Infectivity ofV. cholerae Shed in Human Stools. Because V. cholerae in humanstools, when inoculated in environmental water samples, showedcharacteristics identical to CVEC, which are multicellular aggre-gates, we conducted microscopic examination of 18 stool samplesfrom different cholera patients to determine the cellular organiza-tion of V. cholerae present in the stools. Although a typical cholerastool is generally described as watery (i.e., ‘‘rice-water’’ stool), allstools we examined also appeared to contain particulate matter.Phase-contrast microscopy confirmed that, in addition to plank-tonic cells, each of these stools contained aggregates of cellsattached to debris, forming biofilm-like structures of various sizes(Fig. 2). The presence of these biofilms generally resulted in anasymmetric distribution of V. cholerae cells in the stools.

Merrell et al. (23) reported that V. cholerae present in cholerastools appeared to be hyperinfectious for infant mice, comparedwith a laboratory strain grown in vitro to stationary phase. Theyused microarray analysis to suggest that the effect might be relatedto transcriptional changes in the organism. Because our studiesindicated that the stools of cholera victims contain vibrios in morethan one morphological state, we assumed that the apparenthyperinfectivity of stool vibrios could also be due to factors otherthan genetic regulation of virulence-associated factors. For exam-ple, the presence of clumped cells might result in underestimationof input colony counts during competition assays with the referencelaboratory-grown strain to determine infectivity. To begin to in-vestigate this possibility, we estimated the relative infectivity of thedifferent phenotypic forms of V. cholerae present in cholera stoolsby using differential centrifugation to fractionate stools accordingto the size of suspended particles and cell clumps (Fig. 2). We thenconducted a competition assay with the same reference strain usedby Merrell et al. (23), to determine the relative infectivity of theplanktonic cells and biofilm-like cells. The reference strain wasgrown to both logarithmic phase and stationary phase. Interestingly,the average infectivity of the biofilm-like clumped cells fromdifferent stools was 12- to 145-fold higher than for the correspond-ing planktonic cells, even when the stool vibrios competed with thereference strain grown to logarithmic phase (Fig. 3). The highestcompetitive index (infectivity) observed in a particular assay for V.cholerae clumps in stools was 1,560, compared with the logarithmic-phase reference strain, and this infectivity was 487-fold higher thanthat of the corresponding planktonic cells from the same stool.When they competed with the reference strain grown to a station-

Fig. 2. Micrographs of different fractions of human cholera stools. Both planktonic cells and biofilm-like clumps of V. cholerae cells are present inunfractionated stool (A). The planktonic cells (B) were separated from the clumps (C) by differential centrifugation, and the precipitated clumps were furtherwashed to remove free cells. Numerous V. cholerae cells in multiple layers were found tightly attached to debris in stools, forming biofilms (C).

Faruque et al. PNAS � April 18, 2006 � vol. 103 � no. 16 � 6353

MIC

ROBI

OLO

GY

ary phase, the general infectivity of stool vibrios was 5- to 10-foldhigher than that observed with the reference strain grown tologarithmic phase (Table 8, which is published as supportinginformation on the PNAS web site). We conclude that the reportedhyperinfectivity of V. cholerae shed in human stools is largely dueto the presence of biofilm-like clumps of cells that disperse in vivo,providing a high dose of the pathogen. However, stools fromcholera patients contain complex mixtures of cells, presumably withvarious mutually exclusive transcriptional programs, that have notbeen fully characterized.

Conclusions and ImplicationsWe predict that the results described herein will have profoundsignificance for understanding cholera epidemiology. Our resultsindicate that the surface waters in a cholera-endemic area carryhigh concentrations of pathogenic V. cholerae as clumps or biofilmsof CVEC. Although the V. cholerae concentration determined bystandard culture may generally appear to be less than the requiredinfectious dose, we predict that human victims are likely to contractcholera by ingesting clumps of CVEC, because these clumps mayprovide a sufficiently high dose to cause infection. We have shownthat environmental water that is apparently negative for virulent V.cholerae by conventional enrichment methods, but contains CVEC,can be introduced into rabbit ileal loops and yield virulent V.cholerae strains after 18 h of in vivo incubation. Similarly, humanswho ingest clumps of CVEC may resuscitate V. cholerae and evenamplify the strain. Such an individual could become an indexcholera case that begins a cholera epidemic.

Studies by Merrell et al. (23) provided evidence that passagethrough the human host alters critical phenotypic properties thatpromote the transmission of pathogens. Our studies confirm thatthe enhanced infectivity of V. cholerae in human stools is largelydue to the presence in the stools of V. cholerae biofilms formedin vivo. Our studies also suggest that (i) when stools from cholerapatients are mixed with environmental water, CVEC are formedand (ii) CVEC are infectious in the rabbit model. Regardless ofthe mechanisms causing it, the fact that V. cholerae cells derivedfrom cholera stools can have enhanced infectivity has significantepidemiological implications. A recent study in Bangladeshreported that filtering environmental water through four layersof old sari cloth material before consumption reduced the

incidence of cholera among villagers by 48% (24). The authorsof this study proposed that the reduction in cholera cases occursbecause the filtration removes large planktonic material to whichV. cholerae is likely attached (24). We assume that a similarreduction in the incidence of cholera would be observed if V.cholerae in the environment exist as CVEC, which are largeclumps of cells possibly derived from human stools, because intheory such clumps can also be eliminated by sari cloth filtration.

V. cholerae may have at least three different means for adheringto surfaces, depending on whether the organism is within its humanhost or in an aquatic environment. The toxin-coregulated pilus isimportant for colonizing the gut of animals and is an importantvirulence factor (4, 5). A second type-IV pilus, mannose-sensitivehemagglutinin, is not required for gut colonization and pathogen-esis, but was shown to be required for biofilm formation on abioticsurfaces (25). V. cholerae is also known to colonize the chitin (apolymer of N-acetyl-D-glucosamine) surfaces of crustaceans (2, 26),and toxin-coregulated pilus has recently been suggested to have arole in forming biofilms on chitin (12). Thus, V. cholerae appears toadopt different strategies for biofilm formation, depending on itssurroundings. Our studies suggest that cholera patients shed highconcentrations of V. cholerae as in vivo-formed biofilms, and thatthe shedding of these clumps of cells, which disperse in the gut ofthe next human victim, appears to be an efficient way of deliveringan infectious dose of the pathogen to the victim. The survival of V.cholerae, both in the natural environment and in the acidic stomachenvironment of the human host, is also likely to be enhanced if cellsexist in biofilms (13).

Because cholera is a waterborne disease, the better adapted thisbacterium is to survival in the aquatic environment the more likelyit is to infect the next victim and even cause subsequent outbreaksof disease. Because our study showed that environmental V. chol-erae exist largely as CVEC biofilms, we propose that the cells inbiofilms may be better protected than the corresponding planktoniccells against the hypoosmotic stress of environmental water. Thus,in vivo biofilm formation appears to be an efficient strategy for boththe spread of an epidemic and the persistence of the pathogen inwater. In agreement with this assumption, our data also suggest thatV. cholerae can persist in water, both during and between epidemicperiods, as clumps of cells that are conditionally viable and canrevive into virulent bacteria in the mammalian intestine. Thesensitivity or resistance of CVEC to killing by environmentalvibriophages represents an area for future study, given that theselytic viruses modulate the dynamics of cholera epidemics (27).Thus, the prediction of cholera epidemics can likely be facilitatedthrough environmental monitoring of vibriophages, CVEC, andfactors that control CVEC resuscitation. This study constitutes thelatest refinement in our perception of natural phenomena associ-ated with the transmission of cholera by water.

MethodsCulture of Environmental Water Samples. Previously establishedsampling sites (14) along three major rivers, one lake, andmarshy lands in and around Dhaka city were used. Watersamples were collected from these sites every week betweenJanuary and December 2005 by using sterile containers and weretransported to the laboratory for processing within 2 h ofcollection. V. cholerae were isolated from the samples by usingvarious enrichment and culture techniques, described below, andby using modifications of the AST described in ref. 14. An aliquot(5.0 ml) of each water sample was added to 2.5 ml of 3�concentrated BP medium and incubated for 6 h for enrichmentof V. cholerae. In another set of BP-enrichment cultures, anantibiotic (streptomycin or nalidixic acid) was added at anappropriate concentration. Dilutions of the enrichment culturewere plated on TTGA (28) containing streptomycin (70 �g�ml),as well as on TTGA plates devoid of any antibiotic. SuspectedVibrio colonies were picked and subjected to biochemical and

Fig. 3. Infectivity of V. cholerae shed in human cholera stools. Shown aremean competition indices of V. cholerae O1 present in different fractions ofstools from 10 different cholera patients, against a reference strain grown tologarithmic phase. Each mixture of test and reference strains was assayed in5–10 different mice. Results shown here are for stools that were negative forthe presence of any lytic vibriophage.

6354 � www.pnas.org�cgi�doi�10.1073�pnas.0601277103 Faruque et al.

serological tests to identify V. cholerae O1 or O139 (29). Theisolates were also saved for additional genetic analyses, includingvirulence gene content and ribotype.

Growth Kinetics and Fluctuation Tests. In the growth kinetics assay,100-�l aliquots of samples were removed from the enrichmentmixture of a typical AST procedure at regular intervals andplated on antibiotic-containing selective medium for the targetV. cholerae strain, to determine the concentration of viable V.cholerae cells. In the fluctuation assay, water samples wereinitially diluted 10-fold in BP medium, and the diluted sampleswere broken down further into small aliquots. Each aliquot wasassayed for the appearance of viable cells during the ASTprocedure by using the growth kinetics assay. A laboratory-grown V. cholerae O1 strain, diluted in PBS, was used as apositive control. The control strain was similarly inoculated in areaction mixture, in parallel with each set of assay.

Assays in Rabbits. Multiple aliquots (1 ml) of each water samplewere inoculated directly into different ileal loops of adult NewZealand White rabbits as described (11). After 18 h, the rabbitswere killed and the loops were examined for fluid accumulation.Ileal loop fluids were collected, and dilutions of the fluids in 10mM PBS (pH 7.5) were plated on TTGA plates containingstreptomycin. Rabbit ileal loops showing no fluid accumulationwere also washed internally with PBS, and the washings werecultured. Suspected Vibrio colonies were picked and tested bybiochemical and serological methods (29).

Assays in Cholera Stools. Freshly collected cholera stools frompatients who did not receive antibiotics before reporting to thehospital were examined by dark-field microscopy to ascertainthe presence of typical highly motile vibrio-shaped organisms.The stools were then diluted in environmental water that hadpreviously tested negative for V. cholerae O1 or O139 by the ASTprocedure. An aliquot of this dilution, as well as a dilution ofstools in PBS, was cultured immediately to calculate the initialnumber of V. cholerae organisms present in the stools and toisolate the strain for further analysis. Because stools fromcholera patients are known to contain lytic vibriophages, thepresence of phage was also tested as described (14). Resultsshown in this study are from stools that were found negative forany lytic phage. The stool-spiked water was left for 6–16 h atroom temperature before conducting the AST assays. Aliquots

of the same water, not inoculated with stools, were used asnegative controls. Human studies protocols were reviewed andapproved by the Research Review Committee and EthicalReview Committee at the International Centre for DiarrhoealDisease Research, Bangladesh.

Infectivity Assay in Infant Mice. Stools from cholera patients werefractionated according to the size of suspended particles by cen-trifugation initially at 2000 � g to precipitate cells bound to debris;the supernatant contained mostly planktonic cells. The precipitatewas further washed in PBS (pH 7.2) to remove any loosely adheringcells and was then resuspended in fresh PBS. The relative infectivityof V. cholerae cells present in different fractions of cholera stoolswas determined by the competition assay, in accordance withmethods described by Merrell et al. (23). Competition assays wereconducted by mixing the lacZ-negative reference strain DSM-V984(23), grown in LB medium to stationary phase and logarithmicphase, with stool bacteria in a ratio of 1:10 (vol�vol). A total of 50�l of a 1,000-fold dilution of this mixture, representing 104 to 105

colony-forming units (cfu), was administered to 3- to 5-day-oldSwiss Albino mice. After 20–24 h, the animals were killed and thesmall intestines were removed, homogenized, and plated on me-dium containing 1% peptone, 1% NaCl, 3% gelatin, and 1.5% agar,supplemented with X-Gal to allow for subsequent enumeration of�-galactosidase-positive and �-galactosidase-negative V. choleraecolonies. Ratios of output bacteria were corrected for deviations ininput ratios from 1:1, and the corrected ratios were termed the‘‘competitive index’’ of the strains in question.

Probes and PCR Assays. The presence of virulence genes wasdetermined by using specific PCR assays for the tcpA, tcpI, andacfB genes of the toxin-coregulated pilus pathogenicity island,and for the ctxA and zot genes of the CTX prophage (10). TheSXT probe and the rRNA gene probe used in ribotyping havebeen described (10, 15). Southern blots were prepared andhybridized with the radioactively labeled probes, in accordancewith standard methods (30).

This work was supported by National Institutes of Health ResearchGrant GM068851, under subagreements between the Harvard MedicalSchool and the International Centre for Diarrhoeal Disease Research,Bangladesh (ICDDR,B). The ICDDR,B is supported by countries andagencies that share its concern for the health problems of developingcountries.

1. Colwell, R. R. & Spira, W. M. (1992) in Cholera, eds. Barua, D. & Greenough,W. B., III (Plenum, New York), pp. 107–127.

2. Colwell, R. R. & Huq, A. (1994) in Vibrio cholerae and Cholera: Molecular toGlobal Perspectives, eds. Wachsmuth, I. K., Blake, P. A. & Olsvik, O. (Am. Soc.Microbiol., Washington, DC), pp. 117–133.

3. Pascual, M., Rodo, X., Ellner, S. P., Colwell, R. R. & Bouma, M. J. (2000)Science 289, 1766–1769.

4. Faruque, S. M., Albert, M. J. & Mekalanos, J. J. (1998) Microbiol. Mol. Biol.Rev. 62, 1301–1314.

5. Kaper, J. B., Morris, J. G. & Levine, M. M. (1995) Clin. Microbiol. Rev. 8, 48–86.6. Khan, M. U., Shahidullah, M., Haque, M. S. & Ahmed, W. U. (1984) Trop.

Geogr. Med. 36, 335–340.7. Cash, R. A., Music, S. I., Libonati, J. P., Snyder, M. J., Wenzel, R. P. & Hornick,

R. B. (1974) J. Infect. Dis. 129, 45–52.8. Levine, M. M., Kaper, J. B., Herrington, D., Losonsky, G., Morris, J. G., Clements,

M. L., Black, R. E., Tell, B. & Hall, R. (1988) Infect. Immun. 56, 161–167.9. Roszak, D. B. & Colwell, R. R. (1987) Appl. Environ. Microbiol. 53, 2889–2893.

10. Faruque, S. M., Chowdhury, N., Kamruzzaman, M., Dziejman, M., Rahman,M. H., Sack, D. A., Nair, G. B. & Mekalanos, J. J. (2004) Proc. Natl. Acad. Sci.USA, 101, 2123–2128.

11. O’Toole, G., Kaplan, H. B. & Kolter, R. (2000) Annu. Rev. Microbiol. 54, 49–79.12. Reguera, G. & Kolter, R. (2005) J. Bacteriol. 187, 3551–3555.13. Zhu, J. & Mekalanos, J. J. (2003) Dev. Cell 5, 647–656.14. Faruque, S. M., Islam, M. J., Ahmad, Q. S., Faruque, A. S. G., Sack, D. A., Nair,

G. B. & Mekalanos, J. J. (2005) Proc. Natl. Acad. Sci. USA 102, 6119–6124.15. Waldor, M. K., Tschape, H. & Mekalanos, J. J. (1996) J. Bacteriol. 178, 4157–4165.

16. Hochhut, B., Lotfi, Y., Mazel, D., Faruque, S. M., Woodgate, R. & Waldor,M. K. (2001) Antimicrob. Agents Chemother. 45, 2991–3000.

17. Walsh, C. (2003) Antibiotics: Actions, Origins, Resistance (Am. Soc. Microbiol.,Washington, DC), pp. 89–156.

18. Baker, E. M., Singleton, F. L. & Hood, M. A. (1983) Appl. Environ. Microbiol.46, 930–940.

19. Luria, S. & Delbruck, M. (1943) Genetics 28, 491–511.20. Spira, W. M., Sack, R. B. & Froehlich, J. L. (1981) Infect. Immun. 32, 739–747.21. Hase, C. C. (2000) Trends Microbiol. 8, 490–491.22. Dibrov, P. (2005) Biochemistry (Moscow) 70, 150–153.23. Merrell, D. S., Butler, S. M., Qadri, F., Dolganov, N. A., Alam, A., Cohen, M. B.,

Calderwood, S. B., Schoolnik, G. K. & Camilli, A. (2002) Nature 417, 642–645.24. Colwell, R. R., Huq, A., Islam, M. S., Aziz, K. M., Yunus, M., Khan, N. H.,

Mahmud, A., Sack, R. B., Nair, G. B., Chakraborty, J., et al. (2003) Proc. Natl.Acad. Sci. USA 100, 1051–1055.

25. Watnick, P. I., Fullner, K. J. & Kolter, R. (1999) J. Bacteriol. 181, 3606–3609.26. Meibom, K. L., Li, X. B., Nielsen, A. T., Wu, C. Y., Roseman, S. & Schoolnik,

G. K. (2004) Proc. Natl. Acad. Sci. USA 101, 2524–2529.27. Faruque, S. M., Naser, I. B., Islam, M. J., Faruque, A. S. G., Ghosh, A. N., Nair,

G. B., Sack, D. A. & Mekalanos, J. J. (2005) Proc. Natl. Acad. Sci. USA. 102,1702–1707.

28. Monsur, K. A. (1961) Trans. R. Soc. Trop. Med. Hyg. 55, 440–442.29. World Health Organization (1974) World Health Organization Guidelines for the

Laboratory Diagnosis of Cholera (Bacterial Disease Unit, W.H.O., Geneva).30. Maniatis, T., Fritsch, E. F. & Sambrook, J. (1982) Molecular Cloning: A

Laboratory Manual (Cold Spring Harbor Lab. Press, Woodbury, NY).

Faruque et al. PNAS � April 18, 2006 � vol. 103 � no. 16 � 6355

MIC

ROBI

OLO

GY