X‘L
. ‘ 4:...

film

(Seat: 5 tulfi: .1
a\ ...v . I
9» ‘ I... in. 5'

Mn... 3115?... an. w»:~ia.o.fi.nr.t$x . .
V . . , . a. .3 a
x - tr! 3.5» . .

‘¥.ruht..3.uufit:eu ”a I.

“Wasting... iv? .2

 

. it
z.:~ahulz§v......urfl
Jfilhln!..§ :3!
. .‘r. iii-£215.:

. ta

.

.2...
:15.
33!!

x: 2!.
{2.1 .
.141.
i ..

t$03..

u .X.’

.Iiuet:
. ,nail‘fil‘
25.172}, .

X.

unfit”: .2...
t a K. .‘hug‘t2ll. ”slam.
.pl.!ilriial...$.lial.h:l¢|lfle
- 1 11.1.5. .

3‘

a)
‘37er
.5...
1.3.8.”;

. (buli‘In:
in,” u» I} I ;§M.hu.uuflhflh.lh.0.
1.5.x “ennui;

Hartman?” c 1.! yd

 

.3...(.:.l.c... ‘9 TVA. .rLL,l.

 

 

’5‘ ll filllllll\llmll\\\\\\\\\\ll§l\ ,

. 319060302
1351’?" “
LIBRARY l
Michigan mate r
l University \

 

 

This is to certify that the
thesis entitled
DEVELOPMENT AND EVALUATION OF A GAS SENSOR BASED INSTRUMENT

FOR THE DETECTION AND DIFFERENTIATION OF E_. _(_I_OLI 0157: H7
FROM NON- 0157. H7 E. C_O__LI_

presented by

Spring Marie Younts

has been accepted towards fulfillment
of the requirements for

M.S. degree in Animal Science

DJ/mm
MIA Led/l @Ls

Vajor professor

 

 

 

Date 12/17[99

0-7639 . MS U is an Affirmative Action/Equal Opportunity Institution

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

001
591109712: [I 3;

 

W

t<=

 

 

 

 

 

 

 

 

 

 

 

 

 

11m chIRCIDatoOmpGS—p.“

. _aii,

DEVELOPMENT AND EVALUATION OF A GAS SENSOR BASED INSTRUMENT
FOR THE DETECTION AND DIFFERENTIATION OF E. COLI OlS7:H7
FROM NON—OlS7:H7 E. COLI

BY

Spring Marie Younts

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

Master of Science

Department of Animal Science

. 1999

ABSTRACT

DEVELOPMENT AND EVALUATION OF A GAS SENSOR BASED INSTRUMENT
FOR THE DETECTION AND DIFFERENTIATION OF E. COLI OlS7:H7
FROM NON-OlS7:H7 E. COLI

BY

Spring Marie Younts

Rapid and economical detection of human pathogens in animal
and food production systems would enhance food safety
efforts. The objective of this research was to develop
a gas sensor based instrument, coupled with an eutificial
neural network (ANN), which is capable of differentiating
the human pathogen E. coli 0157:H7 from non—0157zH7 E. coli
isolates. The production of gases from eight laboratory
isolates anui 20 field isolates of ll. coli were nonitored
during growth in laboratory conditions, and a unique gas
signature for each isolate was generated. An ANN was used
to analyze the gas signatures, and classify the bacteria as
OlS7:H7 or :non—0157zH7 E3 coli. Detectable differences
were observed between the gas signatures of the E. coli
OlS7:H7 and non-0157zH7 isolates and the ANN classified the
isolates with a high degree of accuracy. Based on this
work, gas sensor based technology has promise as a
diagnostic tool for pathogen detection in pre—harvest and

post—harvest food safety.

Copyright by
Spring Marie Younts
1999

This thesis is dedicated to Mom and Dad, Richard and
Melodee Younts, for their love, support, and encouragement.

iv

ACKNOWLEDGMENTS

Special thanks to. . .

Dr. Wes Osburn and Dr. Dan Grooms for working together
to create the Opportunity for me to focus my Master's
program on animal production food safety.

Dr. Dan Grooms for the opportunity to work in his lab
in the College of Veterinary Medicine. I am grateful for
the support, encouragement, research mentoring,
availability, and innumerable hours he has given to me.

Dr. Wes Osburn for the assistantship position in
Animal Science and mentoring in Meat Science that has
greatly enhanced my exposure and experience.

Dr. Evangelyn Alocilja for this research opportunity
and for enthusiastic encouragement.

Dr. Harlan Ritchie for serving as a committee member
and providing insight into the impact of food safety issues
in the beef industry.

Dr. Mel Yokoyama for serving as a committee member and
for contributing to the Microbiology aspects of this
research.

Steve Marquie for technical assistance.

Nikki Ritchie for lab assistance.

Drs. David Smith, Laura Hungerford, and Jeff Gray,
University of Nebraska-Lincoln, for research contributions
and additional animal production food safety research
opportunities.

Dr. Qijing Zhang for laboratory isolates of E. coli

Additional Acknowledgements:
Dr. Rodney Holcomb, Oklahoma State University
Dr. Fred Ray, Oklahoma State University
Dr. Dee Griffin, University of Nebraska
Roy Radcliff, Graduate Student, Michigan State
University
Erick Harp, Graduate Student, Oklahoma State
University

Mom and Dad, Richard and Melodee Younts

Funding for this research project was provided by the

Animal Industry Coalition.

vi

TABLE OF CONTENTS

 

 

 

 

 

 

 

 

 

 

List of Tables viii
List of Figures ix
Introduction 1
Chapter 1

Review Of The Literature 3
Chapter 2

Development And Evaluation Of A Gas Sensor Based
Instrument For Identifying E. coli OlS7:H7 In A
Laboratory Setting 96
Chapter 3

Differentiation Of Escherichia coli 0157:H7

From Non-OlS7:H7 E. coli Serotypes Using A

Gas Sensor based, Computer-Controlled

Detection System 54
Chapter 4

Experimental Use Of A Gas Sensor Based

Instrument For Differentiation Of E. coli OlS7:H7

From Non-0157zH7 E. coli Field Isolates 74
General Discussion 95
Appendices

Appendix A: Research Protocols 103
Appendix B: Representative Gas Signatures from

Each E. coli Isolate 107
Bibliography 138

 

vii

LIST OF TABLES

Table 2.1 Serotypes and sources of E. coli isolates ...... 35

Table 3.1 Sensors employed in instrument with

detectable compound specificity and sensitivity

levels

Table 3.2 Serotypes and sources of E. coli isolates ......

Table 3.3 Scenarios for training and testing the ANNm

Table 3.4 Sensitivity and specificity of the gas
sensor instrument based on the artificial
neural network interpretation of the ammonia
sensor output

 

Table 3.5 Sensitivity and specificity of the gas
sensor instrument based on the artificial
neural network interpretation of the air
contaminants sensor output

 

Table 3.6 Sensitivity and specificity of the gas
sensor instrument based on the artificial
neural network interpretation of the alcohol
sensor output

 

Table 4.1 Sensors employed in instrument with
detectable compound specificity and
sensitivity levels

 

Table 4.2 Contingency tables showing the results
of ANN classification of field isolates using
non-normalized data

Table 4.3 Contingency tables showing the results

of ANN classification of field isolates using
normalized data

viii

57

58

6O

66

67

68

77

87

88

LIST OF FIGURES

Figure 2.1 Picture of the sensor system 29

Figure 2.2 Picture of gas sensor array in sensor
Chamber 31

Figure 2.3 Representative gas signature generated
by non-0157zH7 E. coli in BHI broth 40

Figure 2.4 Representative gas signature generated
by E. coli 0157:H7 in BHI broth 41

Figure 2.5 Representative gas signature generated
by E. coli 0157:H7 in nutrient broth 42

Figure 2.6 Representative gas signature generated
by non—0157:H7 E. coli in nutrient broth 43

Figure 2.7 Average time in hours (+/— standard
deviation) for initial increase in gas
concentration to occur, as measured by voltage
increase in gas sensors, at different initial
bacteria concentrations 45

Figure 2.8 Control gas signature from monitoring
dry block heater 47

Figure 2.9 Control gas signature from monitoring
nutrient broth 48

Figure 2.10 Representative growth curve for E. coli
OlS7:H7 plotted against a typical gas signature.49

Figure 2.11 Representative growth curve for non—
Ol75:H7 E. coli plotted against a typical gas

signature 50

Figure 3.1 Representative gas signature generated
by E. coli 0157zH7 62

Figure 3.2 Representative gas signature generated
by non—0157zH7 E. coli 64

Figure 4.1 Gas signatures from the ammonia sensor
for each of the E. coli 0157:H7 field isolates ...... 83

ix

Figure 4.2 Gas signatures from the ammonia sensor
for each of the non-0157:H7 E. coli field
isolates 84

 

Figure 4.3 Gas signatures from the ammonia sensor
for E. coli 0157:H7 field isolates from outbreak
of human illness 85

 

Figure 4.4 Gas signatures from the ammonia sensor
for E. coli 0157:H7 isolates from the same
feedyard 86

 

INTRODUCTION

Food safety concerns are currently impacting’ public

health, the meat industry, and animal production
agriculture . Animal agriculture has been under increasing
scrutiny as a source of foodborne pathogens. In this

study, an initial investigation was conducted to develop a
new technology that could be applied to pre—harvest food
safety efforts, particularly for identifying and monitoring
a potential human pathogen “on the farm”.

Escherichia coli (E. coli) 0157:H7 has been recognized
as a significant bacterial pathogen belonging to a group of
enterohemorrhagic E. coli associated with bloody diarrhea.
It is an important public health concern because of its
association with commonly consumed foods, such as ground
beef. Infection with this organism can cause hemorrahagic
colitis, hemolytic uremic syndrome, and thrombotic
thrombocytopenic purpura. The association of E. coli
0157:H7 with ground beef has led to the identification of
cattle as a reservoir for the organism. Recent pre-
harvest food safety efforts have emphasized identifying
factors within cattle production systems for the monitoring

and control of E. coli 0157:H7.

Computer controlled gas sensor based instruments,
referred to as artificial olfactory technology, are finding
increasing application in the food industry. The sensors
are designed to detect volatile compounds that result from
spoilage, rancidity, or other “off” odors. Promising
results have been shown when this technology was applied to
differentiating between different species of bacteria and
spoilage fungi.

The hypothesis for this investigation is that
artificial olfactory technology can be used for detecting
and differentiating E. coli OlS7:H7 from various other E.
coli strains based on the pattern of gas emissions. The
objective of this research was to develop and evaluate a
gas sensor based instrument that was capable of
differentiating E. coli 0157:H7 from non-0157:H7 E. coli
serotypes by evaluating gas emissions. The instrument was
employed to monitor the volatile breakdown products of
bacterial metabolism as the organisms grew. As the gas

emissions were monitored the measurements were plotted to

generate gas signatures or patterns. Analytical computer
programs were used for pattern recognition and
interpretation. The long-term goal of this investigation

is to develop a diagnostic tool for identifying E. coli

OlS7:H7 in cattle production systems.

Chapter 1

REVIEW OF THE LITERATURE

I. Background and Significance of Escherichia coli 0157:H7

Escherichia coli (E. coli) OlS7:H7 has become
recognized as ea significant public health concern because
of its virulence as a foodborne bacterial pathogen
(Buchanan and Doyle, 1997). E. coli 0157:H7 is identified
as one of the most serious foodborne pathogens due to a low
infectious dose and potential severity of symptoms (Doyle
et al., 1997). Though designated by O (somatic) and H
(flagella) antigens, it is the specific virulence factors
that separate E. coli OlS7:H7 from generic E. coli (Doyle
et al., 1997). Current scientific research efforts and
regulatory strategies emphasize obtaining a greater
molecular understanding, developing effective control
strategies, and enhancing detection, identification, and
monitoring techniques for this organism.

The virulence factors that distinguish E. coli 0157zH7
from generic E. coli, found in the gastrointestinal tract
of healthy animals and humans, include specific genes
encoding for the ability to attach to host cell membranes
and produce specific toxins (Doyle et al., 1997). E. coli

0157zH7 is the prominent serotype in the group referred to

as Enterohemorrhagic E. coli (EHEC) which. possess these
virulence genes. .E. coli 0157:H7 is the leading cause of
EHEC associated disease in the United States (Buchanan and
Doyle, 1997). The pathogenesis of E. coli 0157:H7 relies on
attachment to epithelial cell walls and production of
cytotoxins (Doyle et al., 1997). The attachment/effacement
mechanism is due to the presence of the “eae” gene (E. coli
attaching and effacing gene) located on the organism’s
chromosome. Although this gene alone does not provide
virulence, it is characteristic of pathogenic EHEC strains
(Buchanan and Doyle, 1997) . The cytotoxins produced were
identified after E. coli 0157:H7 was determined to be a
human pathogen. The toxins are referred to as verotoxin 1
and verotoxin 2, because of their toxicity to African green
monkey kidney tissue cells (Vero cells), or Shiga toxin 1
and Shiga toxin 2, because the ability to produce these
toxins was obtained from a bacteriophage originating from
Shigella (Doyle 6H: al., 1997). These 'virulence factors
indicate that genetically coded differences exist between
E. coli 0157zH7 and other serotypes of E. coli. In
developing a new technology for identifying E. coli
0157:H7, we proposed that there could be detectable
differences in the metabolic activity of E. coli OlS7:H7

due to genetic differences.

Hemorrhagic colitis is the most common human illness
resulting from an E. coli OlS7:H7 infection (Buchanan and
Doyle, 1997). The symptoms of hemorrhagic colitis include;
mild to overtly bloody diarrhea, extremely severe abdominal
cramps, and dehydration. The onset time for symptoms of
hemorrhagic colitis ranges from 1-5 days following
ingestion of the bacteria, with the symptomatic phase
lasting 4-10 days. Systemic complications of hemorrhagic
colitis patients can be life threatening. The most common
sequelae is hemolytic uremic syndrome which is the leading
cause of acute renal failure in children. Tarr (1995),
stated that approximately 10% of younger children develop
hemolytic uremic syndrome after infection with E. coli
0157:H7. Around 15% of hemolytic uremic syndrome cases
lead to chronic kidney failure and there is a 3—5%
mortality rate associated with hemolytic uremic syndrome
(Buchanan and Doyle, 1997) . Another complication
associated with E. coli 0157:H7 infection is thrombotic
thrombocytopenic purpura, which causes deterioration of the
central nervous system (Boyce et al., 1995). Thrombotic
thrombocytopenic purpura generally affects the elderly and
is considered a more rare sequelae of E. coli 01572H7
infection, however the mortality rate among those afflicted

is 50%. The potential severity of symptoms, particularly

the mortality rate in children, necessitates research

focused on enhancing food safety.

II. Epidemiology

Incidence rates for E. coli 0157zH7 related illness
were estimated to be 2.8 cases per 100,000 people in 1998.
In the United States, with a population of around 272.6
million, an estimated 7,626 cases of illness due to E. coli
OlS7:H7 infection occur annually (USDA-FSIS. 1998). These
estimates are obtained through the Foodborne Diseases
Active Surveillance Network (FoodNet) system (USDA-FSIS.
1998). Hospitalization is required in approximately 32% of
E. coli 0157:H7 infections. Surveillance for the incidence
of hemolytic uremic syndrome is also conducted by FoodNet,
through pediatric nephrologists. For children less than 15
years of age, the overall rate of hemolytic uremic syndrome
is 8.1 cases/1,000,000 population, or approximately 2,206
cases a year in the United States (USDA-F818. 1998).
Deaths in children, associated with hemolytic uremic
syndrome, have drawn the most attention to promoting the
importance of enhancing human food safety (USDA-FSIS.
1998).

The FoodNet systeni (USDA-FSIS. 1998) implicated

undercooked ground beef as the principal food source for E.

coli O157:H7 infections. Epidemiological links established
between outbreaks of human disease and foods of bovine

origin led to the identification of cattle as a reservoir

for the organism (Padhye and Doyle, 1992). To enhance food
safety, research efforts have expanded to focus on
establishing “farm txn table” control strategies. Gaining

an understanding of the ecological association of E. coli
0157zH7 with cattle and their environment and being able to
identify cattle that are carriers of the organism is
essential.

Cattle have been identified as asymptomatic carriers
of E. coli 0157:H7 (Cray and Moon, 1995; Garber et al.,
1995). 'The absence of adverse health effects and lack of
clinical signs in cattle make the identification of cattle
carrying the organism a challenge. Generic E. coli is
found normalLy in the gastrointestinal tract of ruminants,
along with large populations of various other
microorganisms (Brown et al., 1997). To isolate E. coli
0157:H7 from cattle, not only does it have to be separated
from the normal microflora but it must also be
differentiated from other non-pathogenic serotypes of E.
coli.

Reported prevalence rates of E. coli OlS7:H7 in bovine

feces have varied across studies due to the type of cattle

and production systems evaluated, the time of year, and
the type of detection and culturing methods used (Buchanan
and Doyle, 1997; Dargatz et al., 1997; Hancock et al.,
1998; Hancock et al., 1997b). However the prevalence has
been reported to be increasing over the years, primarily
due to increasing sensitivity of culturing methods (Hancock
et al., 1997b). Recent estimates demonstrate that E. coli
OlS7:H7 is widely distributed throughout the United States
(Garber et al., 1995; Hancock et al., 1997b) with 1.1-6.1%
of cattle shedding the organism in their feces (Hancock et
al., 1998) on approximately 75% of cattle operations
(Hancock et al., 1997a; Hancock et al., 1997b). Sheep and
deer have also been shown to serve as natural hosts for E.
coli 0157:H7 while remaining healthy (Buchanan and Emyle,
1997; Kudva et al., 1996; Rice et al., 1995). Companion
animals have been implicated as carriers in cases of human
illness as well (Trevena et al., 1996). Studies indicating
that other species of animals may serve as hosts for E.
coli 0157:H7 imply that non-beef meat products can be
contaminated by their pre-harvest source rather than solely
by cross-contamination from beef products (Kudva et al.,
1996). Food safety can be enhanced by methods to detect
carriers of the pathogen and identification of pre-harvest

intervention strategies.

Transmission of E. coli 0157:H7 is typically by the
fecal—oral route, and illness can result from a very low
dose of less than a hundred bacteria (Buchanan and Doyle,
1997; Doyle et al., 1997). Foods of bovine origin were
found to be the leading vehicle in almost 40% of E. coli
0157:H7 outbreaks in.tflu3 United States from 1982 tx> 1994
(Doyle et al., 1997). Other vehicles and routes of
transmission include; vegetables, apple cider, cantaloupe,
mayonnaise, deer jerky, drinking and recreational water,
and person to person contact (Buchanan and Doyle, 1997;
Doyle et al., 1997; Padhye and Doyle, 1992). Although
these other vehicles exist, popular press leads us to
believe that the ultimate source of contamination is
contact with contaminated beef products or bovine feces.
The association of E. coli 0157zH7 with cattle and beef
products has a negative impact on the beef industry,
strengthening the need for research efforts in pre-harvest
food safety. Research focused on developing monitoring and
control strategies in meat production is important to

enhance public perception of beef.

III. Current Control Strategies and Diagnostic Techniques
Food safety has become a significant focus of both the

government and the scientific community, largely due to the

media’s attention to deaths in children associated with E.
coli 0157:H7 (Buchanan and Doyle, 1997). The Food Safety
and Inspection Service of the USDA has suggested a “zero—
tolerance” policy for E. coli 0157:H7 including the testing
of slaughter bound cattle. A Pathogen Reduction and Hazard
Analysis Critical Control Points (HACCP) rule was published
by the USDA in 1996 (Stevenson and Bernard, 1995),
mandating that all USDA-inspected meat and poultry plants
develop and implement HACCP plans (Stevenson and Bernard,
1995). HACCP is a systematic, preventative, process
control strategy for food safety that is based on 7
principles (Stevenson euxi Bernard, 1995). The pminciples
involve the identification of hazards, critical control
points, critical limits, monitoring strategies, corrective
actions, record keeping, and verification procedures. The
potential for implementing HACCP puinciples ‘xni the farm”
has surfaced due to the regulations placed on packing
plants and the association of E. coli 0157:H7 with live
ruminants.

HACCP Principle #4 is “Establish critical control
point (CCP) monitoring requirements". In meat processing
this may include monitoring the product temperature to
ensure a specific internal temperature was reached or

maintained. On the farm it may mean monitoring the E. coli

10

0157:H7 carrier status of cattle prior to shipment for
slaughter. Particularly, it would mean monitoring of
prevalence of the organism following an intervention
(control) strategy. Monitoring is defined as a planned
sequence of observations or measurements to assess whether
a previously identified CCP is under control and to produce
an accurate record for future use in verification
(Stevenson and Bernard, 1995). Monitoring is essential to
an effective HACCP system, however the cost of
detecting/monitoring a hazard can be high (Unnevhr and
Jensen, 1996).

Currently, establishing the epidemiology of E. coli
0157:H7 in live ruminants is important in efforts to
identify critical control points and study the
effectiveness of intervention strategies in production
systems before an on—farm HACCP system can be implemented.
Current techniques for determining' the prevalence of E.
coli 0157:H7 in cattle usually involve the collection and
culturing of feces. In: identify E. coli 0157:H7 in feces
it must be selected from the normal microbial populations
and be differentiated from other E. coli (Sanderson et al.,
1995). Developing a rapid and economical technique for
detecting and differentiating E. coli OlS7:H7 would greatly

enhance pre-harvest food safety efforts.

ll

Isolating E. coli 0157:H7 from feces or food requires
selective enrichment and culturing media, usually involving
several steps and incubation periods. These traditional
laboratory methods usually require hands-on preparation and
24-48 hours before suspect colonies can be identified. For
E. coli 0157:H7 the selectivity of the culture nethods is
usually based on differences in sugar fermentation (March
and Ratnam, 1986; Sanderson et al., 1995; Zadik et al.,
1993). Selective culturing for E. coli 0157:H7 often

includes the addition of sorbitol, rhamnose, or 4—
methylumbelliferyl-B-D—glucuronide to the culture media. E.
coli 0157:H7 is unable to ferment these sugars and lacks B-

glucuronidase to hydrolyze 4-methylumbelliferyl—B-D—
glucuronide (Ratnam et al., 1988) (Sanderson et al., 1995).
The addition of certain antibiotics not inhibitory to E.
coli 0157:H7, such as cefixime, are also used to inhibit
the: growth. of (other' organisms (Sanderson. et al., 1995).
Following selective culturing, suspect colonies are often
subjected to further testing for serotype confirmation.
The biochemical, genetic, and immunologic techniques
currently used have both advantages and disadvantages.
Immunoassays have been. developed. using selected

antibodies known to react with. particular antigens

12

associated with a specific metabolite or biomass, often the
toxin associated. with. a pathogen (Doyle et al., 1997).
Latex agglutination and enzyme—linked immunosorbent assay
(ELISA) are two readily available immunologic methods used
for confirmation of E. coli 0157:H7 (Doyle et al., 1997).
These techniques are widely accepted, however they require
that a critical mass of the metabolite or biomass exists to
give a positive test result, thus requiring culturing of
the organism before testing. Another disadvantage of
immunoassays, based on the binding of specific antibodies
to antigens, is that the organism is not isolated, so
further typing is not possible (Doyle et al., 1997).

The polymerase chain reaction (PCR) method has emerged
recently as a genetic technique for pathogen detection
based on DNA hybridization (Doyle et al., 1997) . PCR has
greatly enhanced confirmation of the presence of foodborne
pathogens, however it is not used routinely. Disadvantages
of this technique include the inability to distinguish
between live and dead bacteria, the need for pre-enrichment
of samples to reduce polymerase inhibitors and other
organisms, and the lack of isolation of the organism for
further characterization. Genetic based assays are

primarily limited to research laboratories because of the

13

tedious and exacting nature of the reaction setup (Doyle et
al., 1997).

New detection technologies can aid.i11 the development
and evaluation. of intervention strategies to reduce the
number of cattle carrying E. coli O157:H7. Several
researchers have addressed potential intervention
strategies that could be incorporated into production
systems (Diez—Gonzalez et EH”, 1998; Zhao ex: al., 1998).
The validity and efficiency of intervention strategies must
be established by monitoring the presence or reduction of
E. coli 0157:H7. Rapid and economical detection methods
are important for complementing these studies.

The development of a detection method that is rapid,
less labor intensive, and more economically feasible would
greatly enhance food safety nonitoring efforts. In field
research or management systems, it is not always as
important to gain an understanding of the immunological or
genetic properties of the organism as it is to identify the
presence of the pathogen. Gas sensors can detect and
identify specific compounds instantaneously and monitor
theni over time. Incorporated into artificial olfactory
technology, gas sensors can potentially provide a

convenient and inexpensive monitoring tool for certain

14

compounds or volatile gases, such as volatile breakdown

products of bacterial metabolism.

IV. Principles and Applications of Artificial Olfactory
Technology
Artificial olfactory technology, referred to as an
electronic nose, is finding increasing application for
differentiating odors and various volatile compounds
(Bartlett et al., 1997). An electronic nose is a device
usually consisting of metal oxide gas sensors coupled with
an artificial neural network. Analysis of compounds using
this technology has been shown to be rapid, nondestructive,
economical and continuous (Bartlett et al., 1997). The
metal oxide sensors are based on the principle that the
electrical resistance established in the sensor is
decreased in the presence of specific volatile compounds.
The specificity of the sensor is determined by the metal
oxide used in the sensor. Sensor resistance will drop very
quickly in the presence of a specific gas and recover to
its original level in the absence of the gas. A simple
electrical circuit can convert the change in conductivity
to an output signal that corresponds to the gas
concentration (Figaro USA, 1996). The output signal is

reported as a voltage reading that is transferred to a

15

computer software program for continuous plotting,
generating a gas signature or pattern. An artificial
neural network (ANN) is used for data analysis or pattern
recognition. An ANN is an information processing system
that functions similar to the way the brain and nervous
system process information (Alocilja, 1998; Tuang et al.,

1999). The ANN must be trained for the analysis and then

tested to validate the system. In the training process, an
ANN can be configured for pattern recognition, data
classification, and forecasting. Commercial software

programs are available for this instrument of data
analysis. Recent advances with electronic nose technology
have found applications in the food industry for enhancing
traditional quality control techniques, based on the
ability to detect rancidity, spoilage, and “off” odors
(Bartlett et al., 1997).

Gardner et a1. (1998) investigated the use of
electronic nose technology to predict the type and growth
phase of bacteria. In this study, a sensor chamber was
designed that contained six metal oxide sensors chosen by
their sensitivity to known products of bacterial
metabolism. Two bacteria, Staphylococcus aureus and
Escherichia coli, were cultured and the headspace gas of

each was monitored for 12 hours in each experimental run.

16

The gas concentration or voltage measurements were taken
every eight minutes. A back-propagation neural network was
used for data analysis and prediction of bacteria type.
Results showed that this technology accurately classified
100% of the S. aureus samples, and correctly classified 92%
of E. coli samples. An accuracy of 81% was also seen for
predicting the growth phase of the bacteria. The
researchers concluded that there was considerable promise
for the use of electronic nose technology to rapidly detect
the type and growth phase of pathogenic organisms.

Interest in the potential of using dominant odor
volatiles produced by fungi for its detection, spurred an
investigation of the use of gas sensors for this purpose.
Keshri et a1. (1998) used an electronic nose to monitor the
patterns of volatile gas production to detect activity of
spoilage fungi, prior to visible growth, and differentiate
between species. Six different fungi were monitored and
good replication was seen among the gas patterns generated
by the same species. 'The results indicated that early
detection and differentiation of fungi species was possible
using electronic nose technology to monitor the patterns of
gas emissions.

The potential for field use of electronic nose

technology in animal production was demonstrated in a study

17

by Lane and Wathes (1998). An electronic nose was used to
monitor the perineal odors and predict estrus in the cow.
Detectable differences in the perineal odors of cows in the
midluteal phase and cows in estrous were observed.
However, more research. was needed to find sensors more
sensitive to the specific emitted 'volatile compounds to
enhance prediction of stage in estrous. The goal of
ongoing studies is to develop an electronic nose device for
use in cattle operations to enhance estrus detection.
Applications of electronic nose technology for the
detection of microorganisms are based on the ability to
sense the volatile products resulting from metabolism
(Gardner et al., 1998; Keshri et al., 1998). Current
selective culturing methods for identifying E. coli 0157:H7
are based on differences in physiological processes or
biochemical reactions. Differences iJI sugar fermentation
are seen in E. coli 0157:H7 which are used to differentiate
this serotype from other E. coli strains (Padhye and
Doyle., 1992). The inability to ferment sorbitol and
rhamnose and the lack of B-glucuronidase production are
known to be indicative of E. coli 0157:H7 (Ratnam et al.,
1988; Sanderson et al., 1995; Thompson et al., 1990).
These biochemical characteristics and the ability to

produce specific cytotoxins indicate that genetically

l8

encoded differences could exist in the cellular physiology
and metabolism between pathogenic E. coli 0157:H7 and other
strains of E. coli.

Enterobacteriaceae, including E. coli, carry out mixed
acid fermentation resulting in the end product formation of
ethanol, acetate, succinate, formate, ‘molecular' hydrogen,
and carbon dioxide (Atlas, 1995). In this study, it was
hypothesized that an electronic nose could be used to
detect the volatile compounds produced by various E. coli
strains and differentiate serotype 0157:H7 based on a
unique pattern of gas emissions. An instrument was
designed that contained biosensors sensitive to known end
products of microbial metabolism: ammonia and nitrogenous
compounds; methane, ethanol, and isobutane; and. hydrogen
sulfide (Atlas, 1995; Gardner et al., 1998; Moat and
Foster, 1995). Selecting several sensors reactive to
various compounds was important for later studies involving
other types cflf bacteria. Based CH1 the detectable
differences observed between the gas patterns of generic E.
coli and E. coli 0157:H7 and further' evaluation. of sgas
sensors it may be possible to identify a single gas sensor
capable of demonstrating metabolic differences between

strains of bacteria (Younts et al., 1999).

19

Conclusion

In the midst of current efforts to reduce human
exposure to foodborne pathogens, animal production has come
under scrutiny as a potential source of these organisms.
The government, scientific community, and producers are
aware of a need to study the epidemiology and control of
pathogens “on the farm”. Electronic nose technology has
the potential to enhance efforts addressing pre-harvest
food safety concerns involving IL. coli 0157:H7, by
providing a convenient, economically feasible, and less
labor intensive tool for identifying carrier cattle or
other environmental sources/reservoirs of the organism.
Advantages of an electronic nose as a diagnostic tool would
include the identification of live bacteria and monitoring
of their growth, no requirement for reagents, and the
capability of being automated. The purpose of this study
was to conduct an. initial evaluatitwi of electronic nose
technology for detecting and differentiating E. coli
0157:H7 from various other E. coli strains in vitro through
gas emissions in a laboratory setting. The long-term goal
of this research is to develop a non-invasive, easy-to-use—
screening test for E. coli 0157:H7 for applications in
enhancing pre—harvest food safety programs. Electronic

nose technology is gaining attention throughout the food

20

industry and nedical fields. .Applications of this
technology for identifying E. coli 0157:H7 and other

microbial pathogens may’ be possible throughout the food

chain.

21

REFERENCES

Alocilja, E. 1998. Personal Communications.

Atlas, R. M. 1995. Principles of Microbiology. Mosby-Year
Book, Inc .St Louis, MO.

Bartlett, P. N., J. M. Elliot, and J. W. Gardner. 1997.
Electronic noses and their application in the food
industry. Food Technology 51:44-48.

Boyce, T. G., D. L. Swerdlow, and P. M. Griffin. 1995.
Escherichia coli 0157:H7 and the hemolytic uremic
syndrome. New England Journal of Medicine 333:364~368.

Brown, C. A., B. G. Harmon, T. Zhao, and M. P. Doyle. 1997.
Experimental Escherichia coli 0157:H7 carriage in
calves. Applied and Environmental Microbiology 63:27-
32.

Buchanan, R. L., and M. P. Doyle. 1997. Foodborne disease
significance of Escherichia coli 0157:H7 and other
enterohemorrhagic E. coli. Food Technology 51:69—76.

Cray, W. C., and H. W. Moon. 1995. Experimental infection
of calves and adult cattle with Escherichia coli
0157:H7. Applied and Environmental Microbiology
61:1586-1590.

Dargatz, D. A., S. J. Wells, L. A. Thomas, D. D. Hancock,
and L. P. Garber. 1997. Factors associated with the
presence of Escherichia coli 0157 in feces of feedlot
cattle. Journal of Food Protection 60:466-470.

Diez-Gonzalez, F., T. R. Callaway, M. G. Kizoulis, and J.
B. Russell. 1998. Grain feeding and the dissemination
of acid-resistant Escherichia coli from cattle.
Science 281:1666-1668.

Doyle, M. P., L. R. Beuchat, and T. J. Montville, eds.
1997. Food Microbiology: Fundamentals and Frontiers.

Washington, DC: American Society for Microbiology.

Figaro USA, I. Product Information Guide Figaro USA, Inc,
February, 1996.

22

USDA—FSIS. 1998. Report to Congress, FoodNet: An Active
Surveillance System for Bacterial Foodborne Diseases
in the United States. Washington, D.C.

Garber, L. P., S. J. Wells, D. D. Hancock, M. P. Doyle, J.
Tuttle, J. A. Shere, and T. Zhao. 1995. Risk factors
for fecal shedding of Escherichia coli 0157:H7 in
dairy calves. Journal of the American Veterinary
Medical Association 207:46—49.

Gardner, J. W., M. Craven, C. Dow, and E. L. Hines. 1998.
The prediction of bacteria type and culture growth
phase by an electronic nose with a multi-layer

perceptron network. Measurement Science and Technology
9:120-127.

Hancock, D. D., D. H. Rice, D. E. Herriott, T. E. Besser,
E. D. Ebel, and L. V. Carpenter. 1997a. Effects of
farm manure-handling practices on Escherichia coli
0157:H7 prevalence in cattle. Journal of Food
Protection 60:363-366.

Hancock, D. D., D. H. Rice, L. A. Thomas, D. A. Dargatz,
and T. E. Besser. 1997b. Epidemiology of Escherichia
coli 0157 in feedlot cattle. Journal of Food
Protection 60:462-465.

Hancock, D. D., T. E. Besser, D. H. Rice, E. D. Ebel, D. E.
Herriott, and L. V. Carpenter. 1998. Multiple sources
of Escherichia coli 0157 in feedlots and dairy farms
in the Northwestern USA. Preventive Veterinary
Medicine 35:11-19.

Keshri, G., N. Magan, and P. Voysey. 1998. Use of an
electronic nose for the early detection and
differentiation between spoilage fungi. Letters in
Applied Microbiology 27:261-264.

Kudva, I. T., P. G. Hatfield, and C. J. Hovde. 1996.
Escherichia coli 0157:H7 in microbial flora of sheep.
Journal of Clinical Microbiology 34:431-433.

Lane, A. J. P., and D. C. Wathes. 1998. An electronic nose
to detect changes in perineal odors associated with
estrus in the cow. Journal of Dairy Science 81:2145-
2150.

23

March, S. B., and S. Ratnam. 1986. Sorbitol-MacConkey
medium for detection of Escherichia coli 0157:H7
associated with hemorrhagic colitis. Journal of
Clinical Microbiology 23:869-872.

Moat, A. G., and J. W. Foster. 1995. Microbial Physiology
(Third ed.). Wiley-Liss, New York, NY.

Padhye, N. V., and M. P. Doyle. 1992. Escherichia coli
0157:H7: epidemiology, pathogenesis, and methods for
detection in food. Journal of Food Protection 55:555-
565.

Ratnam, S., S. March, R. Ahmed, G. Bezanson, and S.
Kasatiya. 1988. Characterization of Escherichia coli
serotype 0157:H7. Journal of Clinical Microbiology
26:2006-2012.

Rice, D. H., D. D. Hancock, and T. E. Besser. 1995.
Verotoxigenic E. coli 0157:H7 colonization of wild
deer and range cattle. The Veterinary Record 137:524.

Sanderson, M. W., J. M. Gay, D. D. Hancock, C. C. Gay, L.
K. Fox, and T. E. Besser. 1995. Sensitivity of
bacteriologic culture for detection of Escherichia
coli in bovine feces. Journal of Clinical Microbiology
33:2616-2619.

Stevenson, K. E., and D. T. Bernard, eds. 1995. HACCP
Establishing Hazard Analysis Critical Control Point
Programs: A Workshop Manual. Second ed. The Food
Processors Institute. Washington, D.C.

Tuang, F. N., J. L. Rademaker, E. C. Alocilja, F. J. Louws,
and F. J. de Bruijn. 1999. Identifiation of bacterial
rep-PCR genomic fingerprints using a backpropagation
neural network. FEMS Microbiology Letters 177:249—256

Tarr, P. I. 1995. Escherichia coli 0157:H7: Clinical,
diagnostic and epidemiological aspects of human
infection. Clinical Infectious Disease 20:1-10.

Thompson, J. S., D. S. Hodge, and A. A. Borczyk. 1990.
Rapid biochemical test to identify verocytotoxin-
positive strains of Escherichia coli serotype 0157.
Journal of Clinical Microbiology 28:2165-2168.

24

Trevena, W. B., R. S. Hooper, C. Wray, G. A. Willshaw, T.
Cheasty, and G. Domingue. 1996. Vero cytotoxin—
producing Escherichia coli 0157 associated with
companion animals. The Veterinary Record 142:400.

Unnevhr, L. J., and H. H. Jensen. 1996. HACCP as a
regulatory innovation to improve food safety in the

meat industry. American Journal of Agricultural
Economics 78:764-769.

Younts, S. M., D. Grooms, W. Osburn, S. Marquie, and E.
Alocilja. 1999. Differentiation of Escherichia coli
0157:H7 from non-0157:H7 E. coli serotypes using a gas
sensor based, computer controlled detection system.
Paper presented at Institute of Biological
Engineering, June, Charlotte, NC.

Zadik, P. M., P. A. Chapman, and C. A. Siddons. 1993. Use
of tellurite for the selection of verocytotoxigenic

Escherichia coli 0157. Journal of Medical Microbiology
39:155-158.

Zhao, T., M. P. Doyle, B. G. Harmon, C. A. Brown, P. O. E.
Mueller, and A. H. Parks. 1998. Reduction of carriage
of enterohemorrhagic Escherichia coli 0157:H7 in

cattle by inoculation with probiotic bacteria. Journal
of Clinical Microbiology 36:641-647.

25

Chapter 2

DEVELOPMENT AND EVALUATION OF A GAS SENSOR BASED INSTRUMENT
FOR IDENTIFYING E. COLI 0157:H7 IN A LABORATORY SETTING

INTRODUCTION

A rapid, easy to use, diagnostic tool for the
detection of E. coli 0157:H7 would greatly enhance pre-
harvest food safety efforts. To reduce the incidence of
human exposure to this foodborne pathogen, it is important
to establish monitoring and control strategies throughout
meat production and processing (Buchanan and Doyle, 1997).
Currently, there is still a need for research focused on
the ecological association of E. coli 0157:H7 with cattle
and production facilities (Hancock et al., 1998; Hancock et
al., 1997). Methods to easily monitor E. coli 0157:H7 “on
the farm” are important for the development and evaluation
of intervention strategies to control this organism.

Metabolic and physiological differences between
strains of bacteria allow for their selection and
identification in current culturing methods (Doyle et al.,
1997; Moat and Foster, 1995). Many of these methods are
based on the ability or inability of the organism to
breakdown or ferment specific compounds. We proposed that

differences in the breakdown products produced by certain

26

bacteria may be detectable by monitoring their gas
emissions during growth. Volatile compounds can. be
monitored using artificial olfactory technology based on
gas sensors (Bartlett et al., 1997; Gardner et al., 1998).
The objective of this research was to develop and evaluate
a gas sensor based instrument capable of detecting and
differentiating E. coli 0157:H7 from non-0157:H7 E. coli

isolates through gas emissions in laboratory cultures.

MATERIALS AND METHODS
Instrumentation

An instrument was assembled for collecting,
monitoring, and recording the gas emissions from various
growing E. coli cultures. Several considerations were
addressed 1J1 designing tflua instrument. The first
consideration was the need for a culturing system or a way
to grow and maintain bacteria within the instrument. The
next consideration was a method to capture or collect the
gas emissions in a confined space. Detection of the
presence of the gas and identification of the type of
volatile compounds being emitted must be available. The
final consideration was a means of recording the data or
gas measurements automatically. Construction involved

assembly of the sensor chamber and interconnections between

27

chamber and data collection system (computer). A sensor
chamber was designed to sit on a dry-block heater, which
could hold a culture vial and maintain a temperature
supportive of bacterial culture growth. The chamber was
rectangular in shape, approximately 10cm in height X 12.5cm
in length X 10cm in width. The chamber was constructed out
of plexiglass and sealed to capture or contain the volatile
compounds and prevent permeation of odors from the outside
environment into the sensor chamberu Gas sensors, for
detecting the presence of specific compounds, were mounted
in the ceiling of the chamber, directly above the opening
of the culture vial in the dry block heater. The gas
sensors were linked to a circuit board placed on the top of
the chamber, which was connected to the power source. A
data acquisition module (model 2328DA12, B & B Electronics,
Ottawa, IL) was used to convert the output from the gas
sensors to digital output for recording. This module was
also positioned on the chamber and directly connected to a
computer housing the software for data collection. Ports
for tubing were drilled into either side of the chamber; on
one side the tubing was connected to a vacuum pump and the
other side had tubing open to the outside. These tubes
were used to evacuate and draw air through the chamber

between experiments. Figure 2.1 shows the overall system

28

 

Eoumxm nomcmm wen wo endowed H.m mnzmfim

condo
3: ..& .

   

w Hmzom

j:tflh..»;-LI=V

29

and Figure 2.2 views the gas sensors placed in the chamber
ceiling.

Metal oxide gas sensors were acquired from a
proprietary vendor (Figaro USA, Inc., Glenview, IL) to
detect, measure, and monitor the volatile gases released
from the bacterial cultures. The following description of
the sensor operating principle was obtained from the
“General Information for TGS Sensors” (Figaro USA, 1996).
In these sensors, a chemical reaction occurs between the
metal oxide, usually SnOz, in the sensor and the deatile
gas it is designed to detect. An electrical current flows
between connected micro crystals of metal oxide within the
sensor. The sensing material, metal oxide, has a negative
charge on the surface and absorbs oxygen, which accepts
electrons, leading to a positive charge. The resulting
surface potential can act as a potential barrier against
electron transfer, increasing the electrical resistance
within the sensor. The volatile compound for which the
sensor is specifically designed to detect serves as a
reducing gas. When this compound is present, the
negatively charged oxygen density on the surface between
the metal oxide crystals is decreased. The height of the

barrier against electron transfer is reduced and there is a

30

HODEMSU Homflwm CH \AMHHM Homcmm

new no mnduofld N.N shaman

 

31

decrease in sensor resistance. The amount of decrease in
sensor resistance is proportional to gas concentration; the
higher the gas concentration the greater the increase in
electron flow. The decrease in sensor resistance, or
increase in electrical conductivity, is converted to a
change in voltage by the circuit board. The voltage
readings are fed to the data acquisition board and
transferred to the computer for continuous plotting.

The sensors employed in our instrument were chosen
based on their ability to detect volatile metabolites known
to be produced from bacterial metabolism (Moat and Foster,
1995). Four gas sensors were used, specific for the
following: amines (sensitivity of 30 ppm ammonia in air,
Figaro TGS 826), alcohol (SO-5,000 ppm, Figaro TGS 822),
air contaminants (1-10 ppm, Figano TGS 800), and hydrogen
sulfide (5 ppm, Figaro TGS 825). The amine sensor is very
sensitive to ammonia and amine compounds; the alcohol
sensor to methane, iso-butane, and ethanol; and the air
contaminants sensor txa similar alcohol compounds an: lower
concentration (Figaro ‘USA, 1996). Two additional sensors
were used. to monitor the ambient temperature (Figaro D
Thermistor) and relative humidity (Figaro NHU-3) within the
instrument. Monitoring the stability of temperature and

humidity is critical due to their effects on the

32

sensitivity of the sensors. A change in temperature or
relative humidity can affect the rate of the chemical

reaction as it occurs within each sensor (Figaro USA,

1996) .

A. data acquisition software progranl (MeterBOSS,
Teramar Group, Inc., El Paso, TX), was used to collect and
record each sensor response. This program controlled the

rate of gas sampling and plotted the voltage readings
generating a pattern or gas signature during the length of
each experimental run. The gas patterns could then be
analyzed for differences and similarities for

classification of the bacterial strains.

Artificial Neural Network Selection for Data Analysis

An artificial neural network (ANN) was chosen for the
analysis and interpretation of the gas signatures. An ANN
is an information processing system that is patterned after
the way the brain and nervous system process information
(Alocilja, 1998; Tuang et al., 1999). For this
investigation, we employed a back-propagation neural
network (BPN) algorithm (BrainMaker, California Scientific
Software, 1998). The standardized. data front each
experiment, the gas signature data points, serves as the

input vectcmn The desired. output vector‘ is the

33

classification of the organism, “0” for non-0157:H7 E. coli
and “1” for E. coli 0157:H7. Training is accomplished by
using a standardized data set (standard gas signatures) and
associating the input or gas signature with the desired
output or classification. The program compares the data
and computes network output with the desired output until
an acceptable level of recognition is achieved. Another
set of data is used for testing the predictive capability
of the trained BPN. In testing, the BPN is exposed to the
input vectors not labeled with the bacteria type or desired
output classification. Evaluation of the training is based
on the ability of the BPN to recognize and accurately
classify the bacteria type from the input gas signature.
The efficacy of the sensing instrument for differentiating
E. coli 0157:H7 from non-0157:H7 isolates is determined by
the ability of the BPN to distinguish between gas

signatures and correctly classify the bacteria type.

Bacteria Isolates and Culturing

Characterized strains of E. coli, four isolates of E.
coli 0157:H7 and four non-0157:H7 serotypes, from various
sources were obtained for use in the investigation (Table
2.1). Two of the isolates were obtained from Michigan

State University and the remaining six isolates were

34

 

Table 2.1 Serotypes and sources of E. coli isolates

 

 

 

 

 

 

 

 

 

 

 

Isolate Serotype Source
Lab Non-0157:H7 Non-0157:H7 VetefInary Medical Center,

Michigan State University
E47411/O OS:H- Dr. Qijing Zhang

The Ohio State University
80—2572 0157:H13 Dr. Qijing Zhang

The Ohio State University
SD89-3143 Olll:NM Dr. Qijing Zhang

The Ohio State University
Lab 0157:H7 0157:H7 Veterinary Medical Center,

Michigan State University
ATCC 43895 0157:H7 Dr. Qijing Zhang

The Ohio State University
CDC B8038-MSl/0 0157:H7 Dr. Qijing Zhang

The Ohio State University
E29962 0157:H7 Dr. Qijing Zhang

The Ohio State University

 

 

obtained from The Ohio State University. These isolates
were independently ‘verified as E. coli 0157:H7 or non-
0157:H7 E. coli by the Bacteriology Laboratory; at the
Veterinary Diagnostic Center, University of Nebraska,
Lincoln, Nebraska. For verification as E. coli 0157:H7,
the isolates were subject to PCR for the presence of the
eae gene, Shiga toxin (STX) structural gene and the
0 antigen biosynthesis (rfb) loci. All experiments were
performed 111 .a certified Biological Safety Level II

laboratory.

35

Media Testing

Two types of bacterial culture media, Brain Heart

Infusion Broth (BHI) and Nutrient Broth (Difco
Laboratories, Detroit, MI), were evaluated for their use in
the investigation. Two isolates (HE E. coli, one CU57zH7

serotype and one non-OlS7—H7 serotype, were used for the
comparison. Eknfli isolates were grown individually in each
medLa. For each experiment, 10ml of the nedia was placed
in a sterile 14ml polystyrene vial then inoculated with 100
colony forming units (CFUs) of one of the isolates. The
vial was centrally placed in the dry block heater,
maintained at 37:0.2WZ, and monitored over time within the
sensor chamber. The gas readings were collected at a one
minute sampling rate, plotted over 20 hours and a gas

signature generated for each experiment.

Bacteria Concentration Testing

To determine if there were differences in the gas
signatures based on the presence of different
concentrations of the same bacteria, a study was conducted
using <different initial concentrations (Hf bacteria. The
concentrations of the bacteria stock cultures were
determined by serial dilution and viable plate counts.

Bacterial concentrations of 102, 103, 104, 105, 106, and 107

36

per ml, were used for the initial inoculum and monitored
over time to determine the occurrence of the initial
voltage increase. For each experiment, the desired
concentration of bacteria was introduced into 10ml of
nutrient broth and the vial was centrally placed in the dry
block heater, maintained at 37i0.2°C, within the sensor
chamber. Both E. coli 0157:H7 and non-0157:H7 E. coli
isolates were assayed at the different inoculum
concentrations to determine the time each concentration
required to reach the initial voltage increase. Gas
patterns or signatures were identified starting at the
initial voltage increase and ending when the voltage
readings decreased to levels equivalent or less than those

prior to the initial increase.

Control Testing

The dry block heater and uninoculated media were
monitored. over' tine to determine if detectable volatile
compounds, not associated with bacterial growth, were being
released. The sensor chamber was placed over the dry block
heater with nothing in it. The sensor readings were taken
at a one minute sampling rate for 20 hours. For monitoring
the volatile compounds from the media, 10ml of nutrient

broth was placed in a emerile 14ml polystyrene vial. The

37

vial was placed.ix1tflu3 dry block heater at 37iO.2° C with
the sensor chamber in place and monitoned at a one ndnute

sampling rate for twenty hours.

Growth Curves

The growth activity of the ndcroorganisms in nutrient
broth within the gas sensor instrument was monitored to
investigate the relationship between bacterial growth and
gas emissions. All eight isolates of E. coli were used in
this experiment. Cultures were grown and maintained in
nutrient broth to establish a stock culture of each
isolate. There were two separate experimental runs on each
isolate, making a set of 16 growth curves. For each
isolate, a pmedetermined concentration of 1.05 CFUs/ml, was
introduced to a sterile polystyrene vial containing 10ml of
nutrient broth. The vial was then placed in the dry block

heater within the sensor chamber. At 2-hour intervals the
sensor chamber was lifted and 100u1 of the sample culture
was drawn out of the vial using a pipette over a 16 hour
period. The 100ul samples were serially diluted and viable

plate counts were performed. The results from the plate
counts were plotted over the 16 hour time period to

establish standard growth curves for each isolate.

38

RESULTS
Media Testing

The results of the experiments in BHI media
demonstrated that gas emissions could be detected from the
growing cultures. A distinct increase in voltage readings
was seen over time for each of the gas sensors. In the BHI
broth, the voltage readings dramatically increased
initially, peaked, then tapered. off. No obvious
differences were observed between the gas emissions from
the 0157:H7 and the non-0157:H7 isolates. Figures 2.3 and
2.4 show representative gas signatures for E. coli 0157:H7
and non-0157:H7 E. coli in BHI. Visually detectable
differences were observed between the gas signatures of the
E. coli 0157:H7 isolate and the non-0157:H7 isolate when
grown in nutrient broth. The initial increase in voltage in
the nutrient broth was not as dramatic as observed in the
BHI media. The gas pattern observed for the E. coli
0157:H7 isolate grown in nutrient broth showed an initial
increase and a period of stabilization followed by a
gradual decrease in.tflu3 voltage readings (Figure 2.5). .A
binary increase in voltage was observed with the non4fl£flzH7
E. coli isolate followed again by a period of tapering off
(Figure 2.6). Excellent reproducibility was seen in the

pattern of gas emissions between the replicate experiments

39

nDOHD Hmm GA HHOU .m bmubmaolcoc >3 pmumnmcwm UHSDMGmflm mom m>fiumDnmmmHQmm

AHm>H®ucfl mcflHmEmm mudnae Hv MODESZ mHmEmm

EST

8 L L 9 S C... .V E C... E .L
C.— 6 z 9 6 z 9 6 E 9 6
6 E L T. S 6 E L I S 6

T.-
o 6 6
S 6 Z
L I C._

 

‘L9

m.N mudmfim

SQIOA

40

SDOHD Hmm Ga bmubmao wHoo .m kg pmumumcmm oudumnmflm mmm m>HDManmmHQmm v.m wusmflm

AHm>HmuGa mGHHQEmm mundafi Hy Hmnfisz DHQEmm

_.L
0
S
L

6
6
T—

6
z
S

8
C._
6

L
6
E

L
z
L

9
9
T.-

C.—
6
C...

C.—
3
6

.7
9
CL

CL
6
L

E
E
TL

7...
9
C._

T.
6
6

T.
E
CC

 

L9

quoA

41

anonn unmflnudc EH bmubmflo wHoo .m >D pmumnmnmm Unnamcmflm mmm w>fiumunmmmnmmm m.m mhsmflm

AHm>kucfi mCHHdEmm cusses av HUQESZ wamsmm

TEE
S93
66I
EEI
L9

6 8 L L 9 S S .V E
z S 6 z 9 6 z 9 6
S 6 E L .L S 6 E L

I

0 6
S 6
L Tb

quoA

 

42

noose uswflhusc CH wHoo .m

bmubmaounoc SQ poumumcom unnumsmfim mmm o>flumunwmwnmmm m.m musmfim

AHm>kucH mcflHQEmm Dundee HV
6 6 8 L L 9 C._ C._ .V
6 Z S 6 Z 9 6 Z 9
I S 6 Ce L T. C... 6 cc

T.
0
S
L

 

HmnESZ wHQEmm

E E Z I T
6 Cc 9 6 to
L T. S 6 E

quoA

43

for each isolate. Based on these observations, we decided
to employ nutrient broth as the growth media for further

investigation of the instrument.

Bacteria Concentration Testing

The presence of a detectable level of gas
concentration was reached sooner with a higher initial
concentration of bacteria. The gas patterns for the same
bacteria. were similar' in. shape~ over the different
concentrations. However, the initial voltage change
occurred later for each decrease in initial bacteria
concentration. Figure 2.7 shows the initial bacteria
concentration and the average time in hours required for
the initial voltage increase to be observed. To establish
repeatable standard gas signatures for E. coli 0157:H7 and
non—0157:H7 E. coli isolates a standard initial
concentration of loscolony forming units (CFU’s) per ml and
a monitoring time of 16 hours was used for further
experiments. A concentration of 10S CFU’s/ml was chosen
because it optimized the length of time in which a

consistent gas signature could be obtained.

44

mcofiumuuswucoo mflhwuomn HmHuHcH quHmMMHU um
.mHOmcwm mmm CH wmmwuocH meuH0> xn Umwsmmwe mm .Hzooo Ou coHuMHucmocoo
mmm CH mmmwuocfl HmHuHEH HOw AGOHumH>wU pumpGMum -\+v mudon CH mEHu mmmhm>¢ h.m wudmflm

HE\mDmU "coHumuucmocou mHHmuumm HmHuHcH

 

bo+MOO.H mo+moo.H mo+moo. H ¢o+moo. H mo+moo. H No+moo. H
-;2: o

H

O

n

I

N S

3.

O

I:

w u

I.

A4

I.

P

_L

m m

m.m I
a

m

m 3

T.

U

m

3. a

P

S

8

PH. OH

45

Control Testing

No voltage change was observed. when the dry' block
heater was monitored for release of volatile compounds over
time (Figure 2.8). Monitoring of uninoculated nutrient
broth initially showed a slight increase in voltage over

time (Figure 2.9). This increase was expected as the media

was warmed to 37° in the heater and volatile compounds could
be detected. The decrease in sensor resistance was even
and. eventually' stabilized. indicating that 'volatiles from
the media did not impact the gas signatures seen with the

bacteria cultures.

Growth Curves

Representative growth curves plotted against gas
signatures for E. coli 0157:H7 and for non—0157:H7 E. coli
are shown. in IFigures. 2.10 and. 2.11, respectively. iflma
figures demonstrate the relationship between the lag, log,
and stationary phases of microbial growth and the
occurrence of gas emissions within the sensing system. It
was repeatedly observed that the initial voltage change or
detection of gases occurred during the mid to late log
phase of bacterial growth. It was also observed that the

voltage stabilized during the stationary growth phase.

46

Hmummn xooHQ >HU mGHHOUHGOE EOHm endumcmHm mmm Houucoo m.m wusmHm

AHm>HmucH mcHHmEmm muscHE HV HmnESZ meEmm

6 6 8 L L 9 S S b E E Z I I
6 79 S .0 Z a, 6 7v 9 ,o r. 9 ,o E
I c; 6 c. .L I p: 6 pt L TL S ,o E

.L
0
S
L

 

L9

sqtoA

47

buoun ucmfiuudc mCHHOUHGOE EOHw eunumcmflm mmm Honucoo m.m mHQme

AHm>kucH mGHHQEmm muSGHE HV Honesz mHQEmm

EST

6 6 8 L L 9 S C.- .7 CL CL z I
6 Z S 6 z 9 6 z 9 6 E 9 6
.L S 6 E L T. S 6 CL L T. C._ 6

T.
0
C.—
[—

 

L9

SQIOA

48

musumcmflm mam Hmufim>u m

umcflmmm Umuuon bmubmHO HHOU .m HON w>Hsu nuBOHm w>HuMqum0Hmwm 0H.N mudmflm

=snunoo anatd SIQEIA

 

Im/SHJD

AmHm>HmucH w>Hso museum H50: m .mHm>umucH
mGHHQEmm mam muncHE mv uwnESZ mHmEmm mmo

I I I I I I I
8 L S b E I O 6
E O L b I 8 S 8

LSSLVZT.
69€0L.V

mo+moo.H
¢O+MOO.H §§:%elf€...%. .qa;., a?:;:&5§i§§§%&
mo+moo.H
mo+Moo.H
bO+M00.H

wo+moo.H

mo+floo.H

 

TL

SQIOA

49

umGHmmm pmuuoHQ HHOU

=saunoo anatd GIQEIA

 

Im/sndo

musumcmHm mmm Hmufla%u m
.m bmnbmHoncoc Hem w>Hso cusoum 0>HuMucwm0HQ®m HH.N mMDmHm
AmHm>HousH m>udo Suzonm H50: N .mHm>kucH
mcHHQEmm mmm muDGHE mv Hmnfisz mHmEmm mmw

T.—
8LS.V€T.06L9S.V
EOLbISSZGQ’sO

z
L

fI
I

mo+floo.H

vo+moo.H

mo+floo.H

mo+moo.H

squA

bo+moo.H

wo+moo.H

 

mo+moo.H

50

rL

 

 

FL

A.»

DISCUSSION

A gas sensor based instrument was developed for use in
investigating the potential of identifyimg E. coli 0157:H7
based on the pattern of volatile gases released during
growth. The instrument was capable of detecting the gas
emissions from growing E. coli cultures. Differences in
the gas patterns were seen based on the media and bacteria
concentration employed. The variations in gas patterns
based on the type of media used are most likely due to
differences le the nutrient composition of tine media that
resulted in different metabolic breakdown. products. INo
obvious visual differences in the gas patterns produced by
E. coli 0157:H7 and non-0157:H7 isolates were observed when
cultured iJIIBHI brothd However, recognizable differences
were observed in the gas patterns when cultured in nutrient
broth. This suggests that some component of nutrient broth
is metabolized differently by the two types of bacteria,
resulting in different patterns of gas production. The
amount of time that it took to first detect gas production
was dependent on the initial bacterial concentration
introduced into the test system. Initial detection of gas
occurred faster when a higher concentration of bacteria was
used. This suggests that a critical mass of bacteria must

be present to produce detectable levels of the gases.

51

Control testing established that the media and sensor
apparatus do not give off volatile gases which may be
interpreted as bacterial gas production.

Preliminary' observations allowed. for" the Idefining' of
appropriate protocols for standard experiments to be used
in investigating the use of the electronic nose for
differentiating E. coli 0157:H7 from non—0157:H7 E. coli.
Based.cn1 these results, standard experiment protocols were
developed to include: using nutrient broth as the growth
medium, starting with an initial bacteria concentration of
11? CFUs/ml, monitoring the gas emissions for a period of 16
hours, and analyzing the gas signatures using an ANN

trained with standardized data sets.

52

REFERENCES
Alocilja, E. 1998. Personal Communications.

Bartlett, P. N., J. M. Elliot, and J. W. Gardner. 1997.
Electronic noses and their application in the food
industry. Food Techology 51:44—48.

Buchanan, R. L., and M. P. Doyle. 1997. Foodborne disease
significance of Escherichia coli 0157:H7 and other
enterohemorrhagic E. coli. Food Technology 51:69-76.

Doyle, M. P., L. R. Beuchat, and T. J. Montville, eds.
1997. Food Microbiology: Fundamentals and Frontiers.
Washington, DC: American Society for Microbiology.

Figaro USA, I. Product Information Guide Figaro USA, Inc,
February, 1996.

Gardner, J. W., M. Craven, C. Dow, and E. L. Hines. 1998.
The prediction of bacteria type and culture growth
phase by an electronic nose with a multi-layer
perceptron network. Measurement Science and Technology
9:120-127.

Hancock, D. D., T. E. Besser, D. H. Rice, E. D. Ebel, D. E.
Herriott, and L. V. Carpenter. 1998. Multiple sources
of Escherichia coli 0157 in feedlots and dairy farms
in the Northwestern USA. Preventive Veterinary
Medicine 35 (l998):11—19.

Hancock, D. D., D. H. Rice, L. A. Thomas, D. A. Dargatz,
and T. E. Besser. 1997. Epidemiology of Escherichia
coli 0157 in feedlot cattle. Journal of Food
Protection 60 (5):462-465.

Moat, A. G., and J. W. Foster. 1995. Microbial Physiology
(Third ed.). Wiley-Liss. New York, NY.

Tuang, F. N., J. L. Rademaker, E. C. Alocilja, F. J. Louws,
and F. J. de Bruijn. 1999. Identifiation of bacterial
rep-PCR genomic fingerprints using a backpropagation
neural network. FEMS Microbiology Letters 177:249-256

53

Chapter 3

DIFFERENTIATION OF ESCHERICHIA COLI 0157:H7 FROM NON-
OlS7zH7 E. COLI SEROTYPES USING A GAS SENSOR BASED,
COMPUTER-CONTROLLED DETECTION SYSTEM

INTRODUCTION

A. great deal. of ‘media and. regulatory' attention. has
been focused on E. coli 0157:H7 because of potential human
pathogenicity and association with ground beef and other
commonly consumed foods (Buchanan and Doyle, 1997). Human
illness associated with the consumption of contaminated
beef has led to the identification of cattle as a reservoir
for 11. coli 0157:H7 (Buchanan and Doyle, 1997; Padhye and
Doyle, 1992). Detecting and controlling pathogenic E. coli
in beef production management is being proposed, yet little
is currently known about “on the farm" environments

affecting the presence, magnitude, and duration of this

organism (Brown et al . , 1997) . E. coli are part of the
natural intestinal flora of cattle (Padhye and Doyle,
1992). Rapid differentiation of pathogenic E. coli is

essential for determining prevalence and monitoring the
efficacy of intervention strategies in farm and processing
environments. Currently, detection and differentiation
techniques are often time consuming, expensive, and lack

sensitivity. Developing £1 rapid anxi economical technique

54

n:

C;

 

 

r c.

a

elk
.1

his.

for detecting and. differentiating E. coli 0157:H7 would
greatly enhance pre-harvest food safety efforts.

Artificial olfactory technology, referred to as an
“electronic nose”, is finding increasing' application for
differentiating' odors and. 'various 'volatile compounds
(Gardner et al., 1998; Keshri et al., 1998; Lane and
Wathes, 1998). An electronic nose is a device usually
consisting of metal oxide gas sensors coupled with an
artificial neural network (ANN). The gas sensors detect
volatile compounds and generate a gas signature, which is
interpreted by the ANN. Recent advances with this
instrumentation have found application in the food industry
for detecting rancidity, spoilage, and “off” odors
(Bartlett et al., 1997). The use of an electronic nose
shows promise for monitoring the odor quality of food
products throughout the food chain. This technology has
also been studied for its application in differentiating
various types of bacteria. Gardner et al. (1998) showed
that an1 electronic nose could differentiate between
Staphylococcus areus and generh: E. coli with almost 100%
accuracy. Detection and differentiation of species of
fungi in early phases of growth has also been successful

(Keshri et al., 1998).

55

 

rv

(N

(W

(D

r»-
\r‘\

A gas sensor based instrument could provide an
economically viable, easy-to—use tool for identifying
possible sources of contamination in cattle production
before infection spreads and enters the food supply. The
objective of this investigation was to evaluate a sensor
based instrument for detecting and differentiating E. coli
0157:H7 from non-0157:H7 isolates through gas emissions in
laboratory cultures. The long term objective of this
research is to develop a diagnostic tool for identifying E.
coli 0157:H7, thus enhancing pre-harvest food safety

efforts.

MATERIALS AND METHODS
Instrumentation

A sensor chamber was built containing an array of 4
metal oxide gas sensors (Figaro USA, Inc., Glenview, IL), a
temperature sensor, and a humidity sensor. The metal oxide
gas sensors were chosen based on their capability to detect
common volatile breakdown products of bacterial metabolism
(Moat and Foster, 1995). Table 3.1 shows the sensor type,
target compounds it detects, and sensitivity. The sensor

detects the specific volatile compound which causes

electrical conductivity within the sensor to

56

Table 3.1 Sensors employed in instrument with detectable
compound specificity and sensitivity levels

 

increase. The electrical signals generated by this
increased conductivity are acquired karaa data acquisition
board, connected to a computer. A computer software
program (MeterBOSS, Teramar Group, Inc., El Paso, TX), was
used to record and continuously plot the voltage readings,

generating the gas signatures.

Culturing and Collection of Gas Emissions

Characterized strains of E. coli, four isolates of E.
coli 0157:H7 euui four non—0157:H7 serotypes, from various
sources were obtained for testing (Table 3.2). The
isolates were verified as being E. coli 0157:H7 by the
Bacteriology' Laboratory' at the ‘Veterinary' Diagnostic
Center, University' of Nebraska, Lincoln, Nebraska. 'The
isolates were grown and maintained in multipurpose nutrient
broth (Difco Laboratories, Detroit, MI). All culturing was

performed in a certified Biological Safety Level II

57

 

C»

(I;

a:

«C

AL

«3

S.

Table 3.2 Serotypes and sources of E. coli isolates

 

 

 

 

 

 

 

 

 

Isolate Serotype Source
Lab Non-0157:H7 Non-0157:H7 Veterinary Medical Center,

Michigan State University
E474ll/O OS:H— Dr. Qijing Zhang

The Ohio State University
80-2572 0157:Hl3 Dr. Qijing Zhang

The Ohio State University
SD89—3143 Olll:NM Dr.ICijing Zhang

The Ohio State University
Lab 0157:H7 0157:H7 Veterinary Medical Center,

Michigan State University
ATCC 43895 0157:H7 Dr. Qijing Zhang

The Ohio State University
CDC B8038-MSl/O 0157:H7 Dr. Qijing Zhang

The Ohio State University
E29962 0157:H7 Dr. Qijing Zhang

The Ohio State University

 

 

 

 

 

laboratory. Based on previous studies (Younts, 1999), four
standardized experimental runs were performed on each
isolate making a total set of 32 experimental runs or gas
signatures. First, 10ml of nutrient broth was placed into
a sterile 14ml polystyrene vial. .A set concentration of
bacteria, 105 colony forming units (CFU) per ml (Younts et

al., 1999), was introduced into the vial from culture
stocks. The vial was centrally placed in a 37i0.2° C dry

block heater and grown within the sensor chamber. Each

experiment ran for 16 hours with gas sampling every five

minutes. The gas readings or voltage measurements were
continuously plotted, generating a gas signature.
Preliminary studies identified the initial cell

58

concentration and time interval most appropriate for

experimental standardization (Younts et al., 1999).

Pattern Interpretation by the Artificial Neural Network
Each of the four experimental runs on every E. coli
isolate generated. a standardized. gas signature for that
isolate, providing four gas signatures for each isolate.
Data set “1” consisted of the signatures from the first
experimental run on each isolate. Data sets “2”, “3”, and
“4” were made up of the gas signatures from each subsequent
experimental run. The data were divided equally into
training and testing sets for the neural network analysis.
In the training process the ANN was configured for data
classification. The data sets were used in different
combinations as part of the training and testing of the
ANN. For example, data sets 1 and 2 were used as the
training set and sets 3 and 4 were used as the testing set
for one train—test scenarho. The next scenario used data
sets 3 and 4 for training and 1 and 2 for testing. The
third scenario involved data sets 1 and 3 for training and
sets 2 and 4 for testing. There were a total of six
scenarios for each responding sensor type (Table 3.3). The
recognition/classification by the ANN is based on the shape

of the gas pattern, not specific time-data points. Although

59

Table 3.3 Scenarios for training and testing the ANN

cenar o a n ng et est ng et
& 4
4 2
4

 

the shapes of the gas signatures are similar there is
fluctuation in the voltage readings at specific times due
to differences in gas concentration intensity. This
fluctuation in voltage level affects the ability of the ANN
to recognize unseen patterns and accurately classify them.
By dividing the data into testing and training sets, the
specific patterns used to “test” the ANN analysis have not
been seen before. The ANN is programmed to recognize a gas
pattern shape based on the training set. When tested, the
ANN calculates the probability that the previously unseen
patterns in the testing set are indicative of a desired
classification. For example, the ANN compares each gas
signature in the testing set with the patterns it was
“trained” to recognize from the training set. The
resulting output from the ANN is the probability for each
testing pattern, or isolate gas signature, that it is E.
coli 0157:H7 or non—0157:H7 E. coli. For each training and

testing scenario the previous training/testing scenario was

60

deleted and the ANN was re-trained and tested. The
sensitivity and specificity of detecting E. coli 0157:H7
for each scenarho was calculated auxi then averaged

together.

Test Accuracy

The sensing system was evaluated for its value as a
screening test for E. coli 0157:H7. Based on the
differences in the gas patterns of the two E. coli groups,
0157:H7 and. non-0157:H7, the .ANN'Igenerated. probabilities
that individual gas signatures were representative of E.
coli 0157:H7 or‘ not. iBased. on the correctness of the
classification from the pmobabilities, the sensitivity and
specificity of the instrument were calculated (Smith,

1995).

RESULTS
Gas Signatures

Detectable differences were observed between the gas
signatures of the E. coli 0157:H7 and the non-0157:H7
isolates. The gas pattern observed for the E. coli 0157:H7
showed an initial increase and a period of stabilization
followed by’ a gradual decrease in the voltage readings

(Figure 3.1). A binary increase in voltage was observed

61

hmubmHo HHOU .m xn pmumuwcmm musuwcme mmm m>HumucmmmHQmm H.m mudem

AHm>HmuGH mcHHmEmm muscHE mv HDQEDZ wHQEmm

HmH mmH hmH mvH MMH HNH moH hm mm Mb Hm mv hm mm MH

&

1.

.3; gdfigfli
§i§h§|_§.§§?ufl wmwm‘tgnti

 

H

.\

I...

sntoA

62

CI

3.

C

I

CC

HE

$L

.hl‘
an

.Q
.3

with the non—0157:H7 E. coli isolate followed again by a
period of tapering off (Figure 3.2). Subjectively, there
was reliable reproducibility observed between the gas
patterns of replicate experiments on each isolate and
within the two groups. The same overall signature shape
was seen for the E. coli 0157:H7 isolates. There was
greater variation in the shape of the gas patterns from the
non-0157:H7 isolates. Although four sensors were used in
monitoring gas production, only the ammonia, air
contaminant, and alcohol sensors showed a response over
time. No gas pattern resulted from the hydrogen sulfide
sensor, as was anticipated because hydrogen sulfide is not
a normal byproduct of E. coli metabolism. The temperature

and humidity measurements remained consistent over time.

Artificial Neural Network Analysis

The outputs of the three sensors (ammonia, air
contaminant, and alcohol) were used to train and test the
neural network for classifying E. coli 0157:H7. Based on
the evaluation of test accuracy (Smith, 1995), the ANN had
high predictive capability for accurately classifying the
bacteria based on the output of individual sensors. The

results of the sensitivity and specificity analysis for the

63

HHOU .m bmnbmHo-QOG >2 UDDMMmcmm DMDDmsmHm mmm 0>Humunmmmhmmm N.m mudem

AHm>HmusH mGHHQEmm MDSGHE mv Mwnfidz mHQEmm

I I I I I I I
8 9 S .7 E Z 0 6 8
I 6 L S E I 6 L S

L 9 .7 C...
Cc I 6 L

SZ

 

sqtoA

64

three sensors and scenarios are presented in Tables 3.4,
3.5 and 3.6. Sensitivity and specificity varied depending
on the probability cut off used to classify the gas
signatures as 0157:H7 and non-0157:H7 E. coli. As an
example, for the first cut off point any signature with a
50% or greater probability of being E. coli 0157:H7 was
considered “positive”. For all sensors, as the probability
cut off point was reduced, the ability to correctly
classify E. coli 0157:H7 increased while the rate of

misclassification of non—0157:H7 E. coli also increased.

DISCUSSION

An analytical instrument has been developed capable of
detecting and. differentiating E. coli 0157:H7 front non-
OlS7:H7 11. coli isolates ill a. laboratory' setting. Gas—
specific sensors were used. to detect volatile compounds
produced by bacteria during normal metabolic activity. The
gas patterns generated are most likely due to the presence
of amines, nitrogenous compounds, and alcohols, which are
common metabolic breakdown products known to be associated
with E. coli (Moat and Foster, 1995). The hydrogen sulfide
sensor did not show a response over time due to the fact
that hydrogen sulfide is not a normal by—product of E. coli

metabolism (Moat and Foster, 1995). However, inclusion of

65

 

 

 

 

 

 

 

 

 

 

00Huom m.mb 00Huom m.o> wom
00Huom mb 00Huom mo.bb mom
00H-om m6 00Huom m.mm wow
00Hum.bm m.on 00Hum.mm -m.mm wow
00Hum.>m mw.o> 00Humb b.Hm wom
m>wuwmom pmhmpHmGOU
«Symmnmm «woquonHome chmmcmm «woqu>HuHmfimm mmuumHO UM
kuwowmflomuw EMDE. kuw>wuwmmmm Emma. MO.ADHHHQMQOHQ
52 on... IE
Houuom an0554 GOHumuonhoucH

 

 

 

 

usmuso Homcmm mHGOEEm man mo coHumDmHQumncH xnozumc Haydon
HmHoHMHuHm mnu so pmmmn ucmESHumsH Homcmm mmm wan wo >DHUHMHoQO paw >DH>HuHmem v.m DHQMH

66

 

 

 

 

 

 

 

 

 

 

 

ooHum.mm mm.mm m.>mnm.hm m.mm wom
m.hmuom mm.Nb m.bwuom m.mm mwow
m.bw-om mm.o> m.bmnom m.mb won
m.hmuom mb.mm 00Hum.mm m.mm wow
m.bmuom mb.wm OOHumh ¢.mm wom
m>HuHmom pmhmpwmnou
Sommumm STE .833QO «Scmmnmm «.35 ES .356. $7.38 om
_xuwowmflomflm cdmz. .kufl>HuHmem udmz. MO kuHHHQMQOHm
22¢ an» 3
Random mundnflfiuuaoo HHd nOHumuOHmhmunH

 

 

 

DSQSO Homcmm mucmcHEmusoo HHm mnu mo CoHumumHmHmucH xHOBDDC HMHSDC
HMHUHMHDHM man so pmmmn ucmESHumcH Homcmm mmm map mo quoHMHommm paw >DH>HuHmem m.m mHQMH

67

 

 

 

 

 

 

 

 

 

 

 

m.>m-om o.ob ooHuom m.mm wom
m.om-om m.mm ooH-om m.oo woo
m.om-om m.mo ooHuom m.ob won
m.bm-om m.¢m ooHuom m.ob wow
m.bm-m.bm m.mm ooHum.mm m.Hm wom
m>HuHmom pmhmpwmnpu
«wommcmm STE .4on3QO $36an «the EU 32mm Enumeo om
kuHUHMHomQW EmmE Sowbwuwmcmm Gmmx. MO ADHHHQDQOHA
22¢ ma... \3
Random HonooHd QOHumuwumuounH

 

 

 

udmuso Hemcmm Honoon man mo GOHDMDDHQADDCH xhozumc HMHSDG
HmHoHMHuHm map so ommmn ucmEDMumcH Homcmm mom man mo quoHMHommm paw >DH>HuHmcmm m.m mHQMH

68

this sensor may be important in future investigations using
other organisms. The difference seen between the gas
patterns of the E. coli 0157:H7 isolates and the non-
0157:H7 isolates is likely due to genetically coded
differences in metabolic pathways. Edfferences lJ1.E. coli
metabolism are already taken advantage of in routine
differentiation of E. coli 0157:H7 from non-0157:H7 E. coli
by' biochemical assays (Moat. and. Foster, 1995; Ratnam (a:
al., 1988) .

The sensitivity and specificity of differentiating EL
coli 0157:H7 from. non—0157:H7 E. coli could be altered
depending on what probability level was used as a cut off
point. For each gas sensor, as the probability cut off
point was lowered the sensitivity' of detecting E. coli
0157:H7 increased (Tables 3.4, 3.5, 3.6). However,

specificity decreased resulting in more non—0157:H7 E. coli

being misclassified as E. coli 0157:H7. Sensitivity is the
number of “true positives”, or signatures from E. coli
0157:H7, correctly identified, while specificity is

determined by correct classification of “true negatives” or
non-0157:H7 E. coli gas signatures. With a greater
sensitivity, there is a greater probability of correctly
identifying E. coli 0157:H7 isolates, but with a lower

specificity there is an increased occurrence of “false

69

positives" or incorrect classification. of :non-OlS7zH7 E.
coli isolates. Deciding which probability cut off is most
appropriate is dependent on the goal of the screening
procedure. If it is important to detect as many E. coli
0157:H7 isolates as possible, even if some non-0157:H7 E.
coli isolates are falsely classified, then setting the
probability at a point which maximizes sensitivity is
warranted. If ndsclassification of non-0157:H7 E. coli as
E. coli 0157:H7 is undesirable, then setting the
probability at a point which maximizes the specificity is
most appropriate.

There are a number of limitations involved with this
initial study which include: isolates were grown and
monitored in only one type of media; only laboratory
isolates were obtained for experimental runs; a limited
number of isolates were monitored; only pure cultures were
monitored; and a second sensing instrument was not used to
reproduce and validate the results. Expanded studies, with
further refinement of the sensor instrument, may prove that
electronic nose technology is beneficial in monitoring
multiple E. coli cultures and identifying isolates as E.
coli 0157:H7 based on the pattern of gas emissions.

This ‘work. demonstrates the jpotential application. of

electronic nose technology to enhance pre—harvest food

70

safety efforts and aid in the rapid and economical
identification of E. coli 0157:H7. Because generic E. coli
is part of the normal microbial flora in the intestinal
track of cattle, one of the difficulties in studying the
relationship between cattle and pathogenic E. coli is the
differentiation of 0157:H7 strains from the numerous other
strains. 131 addition t1) pre-harvest. applications, there
are opportunities for this type of technology in both the

food industry and human medicine.

71

REFERENCES

Bartlett, P. DL, (1. M. Eflliot, and QL ‘W. Gardner. 1997.
Electronic noses and their application in the food
industry. Food Technology 51:44-48.

Brown, C. A., B. G. Harmon, T. Zhao, and M. P. Doyle. 1997.
Experimental Escherichia coli 0157:H7 carriage in
calves. Applied and Emvironmental Microbiology 63:27-
32.

Buchanan, R. L. , and M. P. Doyle . 1997 . Foodborne disease
significance of Escherichia coli 0157:H7 and other
enterohemorrhagic E. coli. Food Technology 51:69-76.

Gardner, J. W., M. Craven, C. Dow, and E. L. Hines. 1998.
The prediction. of bacteria type and culture growth
phase by an electronic nose with a nmlti-layer
perceptron network. Measurement Science and Technology
9:120-127.

Keshri, G., N. Magan, and P. Voysey. 1998. Use of an
electronic nose for the early detection and
differentiation between spoilage fungi. Letters in
Applied Microbiology 27:261—264.

Lane, A. J. P., and D. C. Wathes. 1998. An electronic nose
to detect changes in perineal odors associated with
estrus in the cow. Journal of Dairy Science 81:2145-
2150.

Moat, A” (L, emui.J. W. Foster. 1995. Microbial Physiology
(Third ed.). Wiley-Liss. New York, NY.

Padhye, N. V., and In. It Doyle. 1992. Escherichia coli
0157:H7: epidemiology, pathogenesis, and methods for
detection in food. Journal of Food Protection 55:555-
565.

Ratnam, S., S. March, R. Ahmed, G. Bezanson, and S.
Kasatiya. 1988. Characterizaticnl of Escherichia coli
serotype 0157:H7. Journal of Clinical Microbiology
26:2006-2012.

72

Smith, R. D. 1995. Evaluation of Diagnostic Tests. In

Veterinary Clinical Epidemiology: A Problem—Oriented
Approach. pp 31—52. Ann Arbor, MI: CRC Press.

Younts, S. M. 1999. Chapter 2: Development and Evaluation
of a Gas Sensor Based Instrument for Identifying E.
coli 0157:H7 in a Laboratory Setting. M. S.

thesis.
Michigan State Universtiy, East Lansing, MI.

Younts, S. M., D. Grooms, W. Osburn, S. Marquie, and E.
Alocilja. 1999. Differentiation of Escherichia coli

0157:H7 From Non—0157:H7 E. coli Serotypes Using a Gas

Sensor based, Computer Controlled Detection System.
Paper' presented. at Institute of

Biological
Engineering, June, Charlotte, NC.

73

Chapter 4

EXPERIMENTAL USE OF A GAS SENSOR BASED INSTRUMENT FOR
DIFFERENTIATION OF E. COLI 0157:H7 FROM NON-0157:H7 E. COLI
FIELD ISOLATES

INTRODUCTION
Escherichia coli (E. coli) 0157:H7 has been recognized
as a significant bacterial pathogen associated with

potentially' severe illness le humans (Padhye anui Doyle.,

1992). The association of E. coli 0157:H7 with commonly
consumed foods, such as ground beef, has made it an
important public health concern (Doyle et al., 1997). The

association of E. coli 0157:H7 with ground beef has led to
the identification of cattle as a reservoir for the
organism (Buchanan and Doyle, 1997; Padhye and Doyle.,
1992). Recent pre-harvest food safety efforts have
emphasized identifying the ecological association of E.
coli 0157:H7 with cattle or within cattle production
systems (Hancock et al., 1998). Because generic E. coli is
part of the normal intestinal flora of ruminants (Gyles,
1994), E. coli 0157:H7 must be differentiated in research
efforts. Currently, detection and differentiation
techniques are often time consuming, expensive, and lack
sensitivity (Doyle et al., 1997). The development of a

diagnostic tool that is more economically feasible, easy to

74

use, and time and labor efficient could prove valuable in
enhancing pre-harvest food safety research.

Artificial olfactory technology i1; finding increasing
application for differentiating odors and various volatile
compounds (Gardner et al., 1998; Keshri et al., 1998; Lane
and Wathes, 1998). Artificial olfactory technology is
usually based on the use of metal oxide gas sensors to
detect and measure volatile compounds coupled with an
artificial neural network (ANN) or pattern recognition
program for data interpretation (Bartlett et al., 1997).
The gas sensors detect volatile compounds and generate a
gas signature, which is interpreted by the ANN. Recent
advances with this instrumentation have found application
in the food industry for detecting rancidity, spoilage, and
“off” odors (Bartlett en: al., 1997). This technology has
also been studied for its application in differentiating
various species of bacteria and fungi (Gardner et al.,
1998) (Keshri et al., 1998).

In a previous investigation, a gas sensor based
instrument was developed and evaluated for use as a tool
for differentiating E. coli 0157:H7 from. non—0157:H7 E.
coli (Younts, 1999a). This investigation involved the

development of a gas sensor based instrument for the

differentiation of E. coli 0157:H7 from non—0157:H7 E. coli

75

by detecting unique gas emission patterns. Initial
evaluation of this technology involved monitoring the gas
emissions of eight E. coli isolates, four isolates of E.
coli 0157:H7 and four non—0157:H7 E. coli isolates,
cultured in a laboratory setting. Standard gas signatures
were generated from these isolates and analyzed by an
artificial neural network (ANN) (Younts et al., 1999a).
The ANN was used to recognize and classify the gas
signatures as E. coli 0157:H7 or non—0157:H7 E. coli. The
systenl was evaluated based on its ability to correctly
classify the organisms. Based on visually observable
differences between the gas signatures of E. coli 0157:H7
and non—0157:H7 E. coli and the accuracy of the ANN in
classifying the bacteria, this technology showed potential
for further development. A limitation in the initial
investigation was that only lab isolates were monitored.
The purpose of this study was to further test the combined
ability of the gas sensor instrument and the ANN to
differentiate isolates cflf.E. coli 0157:H7 from non—0157:H7
E. coli using field isolates from cattle, cattle

environments, and human clinical outbreaks.

76

MATERIALS AND METHODS
Instrumentation

In ea previous investigation, an instrument was
assembled for collecting, monitoring, and recording the gas
emissions from various growing EH coli cultures (Younts et
al., 1999b). This instrument was designed to allow for
culturing of bacteria, collection or capture of gas
emissions, detection and identification of the gases or
volatile compounds, and recording of the data. A sensor
chamber was built containing an array of 4 metal oxide gas
sensors (Figaro USA, Inc., Glenview, IL), a temperature
sensor, and a humidity sensor (Younts et al., 1999b). The
sensors in this chamber or instrument were chosen based on
their capability to detect common volatile breakdown
products of bacterial natabolism (Moat and Prater, 1995).
Table 4.1 shows the sensor type, target compounds it
detects, and sensitivity.

Table 4.1 Sensors employed in instrument with detectable
compound specificity and sensitivity levels

50-5000

1-10
5

 

77

The gas sensors are designed to detect specific
volatile compounds; the presence of the specific compounds
causes electrical conductivity within the sensor to
increase. The electrical signals generated by this

increased conductivity are acquired by a data acquisition

board and converted to voltage readings. A computer
software program, (MeterBOSS, Teramar Group, Inc., El Paso,
TX), was used to record and continuously plot the voltage

readings, generating the gas signatures.

Field Isolate Collection

Twenty E. coli isolates were obtained from the
Bacteriology' Laboratory' at the ‘Veterinary’ Diagnostic
Center, University of Nebraska. Most of the isolates were
collected as part of an ongoing animal production food
safety investigation.:n1 Midwestern feedyards. Additional
isolates were obtained from an outbreak of human illness
due to E. coli 0157:H7 and contaminated venison. These
isolates had been characterized using biochemical reactions
in selective culturing, latex agglutination, and polymerase
chain reaction (PCR). Of the twenty isolates, 12 were

confirmed as E. coli 0157:H7.

78

Culturing and Collection of Gas Signatures

Procedures for the bacteria culturing and collection
of gas signatures were performed as previously defined
(Younts, 1999b). All isolates were grown in nutrient broth
to create stock cultures. The bacteria concentration in
the stock cultures was determined by viable plate count
procedures. All culturing was performed in a certified
Biological Safety Level II laboratory. One experimental
run, generating a gas signature, was completed for each
isolate using previously described procedures (Younts
1999a). For each run, 10ml of nutrient broth was placed in
a sterile 14nd polystyrene ‘vial and inoculated. with 105
colony forming units/ml of the isolate. The vial was
centrally placed in a 37i0.2° C dry block heater and the
sensor chamber positioned on the heater over the culture

vial. Each isolate was grown for 16 hours with gas

measurements taken every 5 minutes.

Gas Signature Interpretation

The gas signatures were interpreted by 'visual
observation and computer analysis. Based on the general
shape, the gas patterns were visually evaluated for
characteristic differences and similarities compared to the

original gas signatures from eight laboratory isolates

79

previously studied (Younts, 1999b). For artificial neural
network (ANN) (BrainMaker, California Scientific Software,
1998) interpretation, EMZ.E. coli gas signatures generated
from the previous study (Younts et al., 1999a) were used to
train the ANN for pattern recognition. In the training
process the ANN was configured for pattern recognition and
data classification” Gas signatures from all 20 field
isolates were subject to interpretation and classification
by the trained ANN. Each of the gas signatures, in both
the training and testing data, were then normalized using

the following equation:

,2:. 2 _ E: .
xmax—Xmin

y:

Xi = voltage data point

i = l,m,n for all data for each sensor
xmax = the highest voltage point
xmin = the lowest voltage point

This method of normalization was used to reduce variation
in the gas patterns due to background voltage levels or
pattern height. Following normalization, the ANN was
retrained. with the original 32 gas signatures and then
tested with the 20 field samples. The ANN determines a

probability that the isolate being tested is E. coli

80

0157:H7 or non-0157:H7 E. coli. For this study, an isolate
was classified as E. coli 0157:H7 or non—0157:H7 E. coli
based on which probability was highest. For example, if
the isolate being tested had a greater probability of being
E. coli 0157:H7 than non-0157:H7 E. coli, it was classified

as E. coli 0157:H7.

Test Accuracy

Based on the results of the gas signature
interpretation by the artificial neural network using both
the normalized and non-normalized data, the sensitivity and
specificity of the instrument for differentiating E. coli
0157:H7 from.:non-0157:H7 E. coli ‘was determined (Smith,

1995).

RESULTS
Gas Signature Observations

As seen previously, the ammonia, air contaminants and
alcohol sensors detected. gases over time, indicative of
volatile bmeakdown products of bacterial growth and
metabolism. Many of the gas signatures shared shape
characteristics similar to either the standard E. coli
0157HT7 or non—0157flf7 E. coli isolates initially tested.

However, there was a greater variation in the overall form

81

of the gas signatures, presumably due to strain variation.
The greatest variation in gas signatures was observed among
the non-0157:H7 isolates. All of the gas signatures from
E. coli 0157:H7 isolates shared some general
characteristics; ihowevery 'visuallyr discernible <differences
in the E. coli 0157:H7 gas signatures were observed.
Figures 4.1 and 4.2 show the gas signatures from the
ammonia sensor for each of the E. coli 0157:H7 and non-
OlS7:H7 E. coli field isolates, respectively.
Interestingly, E. coli 0157:H7 isolates obtained from
similar sources produced gas signatures that were visually
most closely' alike. Fbm‘ example, the isolates obtained
from the outbreak of human illness had very similar
signatures (Figure 4.3). Isolates that were obtained from
the same feedlots, at different times and different
locations, also showed the same pattern of gas emissions

(Figure 4.4).

82

mmumHOmH UHOHM bmubmHO HHOU .m
on» no some no“ Homcom «HGOEEM ms» Eonm mmH5umcmHm mmw

AHm>HmucH mcHHmEmm mudsHE mo HDQESZ DHQEmm

NNH HHH 00H mm mo >0 mm mv vm mm

[c

.H J.,.
/, /
I/JI/ /.. (

IL) 1
It'll \

 

H.¢ musmfim

NH H

squA

83

mmumHOmH prHu HHOU .m bmnhmHoucoc
can no some MOM Homcmm mHGOEEm msu EOHM mmhsumcmHm mmw N.v mhdem

AHm>kucH mGHHQEmm muDCHE mo HmnEdz mHQEmm

hm mm mv vm MN NH H

NNH HHH OOH mm mo

SQIOA

 

84

mmmsHHH cmEss no xmmunudo ECHm

.m new Homcmm MHGOEEM map Eoum mwusumcmHm mmo m.¢ mnsmHm

mmumHomH cHme bmnbmao HHOU
AHm>HmucH mcHHmEmm muscHE mo HmnEsz wHQEmm

00H mm mb hm mm mv wm MN NH H

NNH HHH

 

m 3.633. . .

sntoA

KN manomHllL

 

 

a. 330,31

 

 

85

Unm>pmww wEmm wan 50H“

mmumHomH omubmHo HHOU .m MOM Homcmm MHGOEEM man Eon“ mwhsumcmflm mmw ¢.¢ wnsmflm

AHm>hmucH mGHHQEmm muncHE mo HDQESZ DHQEmm

Mb hm mm mv vM MN NH H

NNH HHH 00H mm

 

 

HH_wumaomH.
_ m mumHOmHull
. m mumflpmflull

SQIOA

 

 

 

86

Artificial Neural Network Analysis

Contingency' tables showing the frequency’ of correct
classification of the E. coli (EC) isolates by the ANN
based on the gas signatures from the ammonia, air
contaminants and alcohol sensors using non—normalized data

are shown in Table 4.2.

TABLE 4.2 Contingency tables showing the results of ANN
classification of field isolates using non-normalized data

 

 

 

 

 

 

 

 

 

 

 

 

 

 

True Type
Ammonia 0157:H7 Non-0157:H7
0157:H7 6 4 Sensitivity 50%
Non-0157:H7 6 4 Specificity 50%
Air True Type
Contaminants 0157:H7 Non-0157:H7
0157:H7 5 4 Sensitivity 41.70%
Non-0157:H7 7 4 Specificity 50%
True Type
Alcohol 0157:H7 Non-0157:H7
0157:H7 5 4 Sensitivity 41.70%
Non-0157:H7 7 4 Specificity 50%

 

 

 

 

87

The frequency of correct classification of E. coli isolates
using normalized data are shown in the contingency tables

in Table 4.3.

TABLE 4.3 Contingency tables showing the results of ANN
classification of field isolates using normalized data

 

 

 

 

 

 

 

 

 

 

 

 

 

 

True Type
Ammonia 0157:H7 Non-0157:H7
0157:H7 ll 4 Sensitivity' 91.7%
Non-0157:H7 1 4 Specificity 50%
Air True Type
Contaminants 0157:H7 Non-0157:H7
0157:H7 12 5 Sensitivity 100%
Non—0157:H7 0 3 Specificity 37.5%
True Type
Alcohol 0157:H7 Non-0157:H7
0157:H7 ll 4 Sensitivity' 91.7%
Non-0157:H7 1 4 Specificity 50%

 

 

 

 

DISCUSSION

Gas sensor based technology, in conjunction with an
ANN, has previously been used to differentiate between
classes of bacteria (Gardner et al., 1998). In a previous

study (Younts et al., 1999b), a gas sensor instrument was

88

developed to differentiate E. coli 0157:H7 from non-0157:H7
E. coli based on unique gas signatures generated during
bacterial growth in laboratory cultures. Using a limited
number of characterized E. coli 0157:H7 and non-0157:H7
isolates (n=8), gas signatures were generated and analyzed
by an ANN. The sensitivity and specificity of this system
ranged from 81-92% and 63-71% respectively, depending on
the types of gas sensor signature analyzed.

In this study the gas sensing instrument was evaluated
for its ability to aid in the identification of E. coli
0157:H7 and non-0157:H7 isolates obtained from various
field situations, including those associated with an
outbreak of clinical human illness and from multiple cattle
production systems. Greater variation in the bacteria
strains and. patterns of gas emissions made the correct
classification of the field isolates using the ANN less
accurate. Although the overall shape of the gas signatures
showed some variation, the isolates of E. coli 0157:H7
shared some general visual characteristics. Greater
conformity of the gas signatures of the E. coli 0157:H7
isolates was seen when the isolates were sorted by source.
For example, isolates originating from an outbreak of human
illness had virtually identical gas signatures. Isolates

obtained from the same feedlot, at different times and from

89

different environmental samples, also had visually similar
gas signatures. Similarities in gas patterns of E. coli
0157:H7 obtained from the same source may be an indication
of relatedness. Based on this observation, unique gas
signatures generated by individual strains of E. coli
0157:H7 may have value as an epidemiological tool for
determining the relatedness of different E. coli 0157:H7
isolates. The non-0157:H7 isolates generated a greater
variety of gas signature patterns as more serotypes were
represented. The differences between gas signatures from
different serotypes could result from the presence or
absence of various metabolic processes.

Using an ANN to analyze the gas signatures, a much
lower sensitivity and specificity was seen for predicting
the class of the field isolates than was observed
previously using a limited number of laboratory isolates.
However, the sensitivity of the system greatly improved
when the data was normalized. Pattern recognition by the
ANN is accomplished by comparing voltage readings at each
time point during the culture period. Normalizing the data
eliminates wide variation in voltage levels that may
confuse the ANN. By normalizing the data, interpretation
of the gas signatures can be made based more on the shape

of the gas curves rather than on specific voltage levels.

90

From the results of the pattern interpretation, it was
determined that a larger training set representing more
non-0157df7 E. coli serotypes was needed for training the
ANN to more accurately classify non—0157:H7 E. coli
isolates. However, with the limited training set, E. coli
0157:H7 can be detected with a high degree of sensitivity,
indicating greater similarity cflftflue E. coli 0157:H7 gas
signatures.

Further refinement. of the instrument and. parameters
for pattern interpretation may increase the sensitivity and
specificity of the instrument and AN for classifying E.
coli isolates. Pattern recognition needs to be focused on
determining the most distinctive characteristics of the gas
signatures of E. coli 0157:H7 isolates. Additional methods
of data normalization for the output from gas sensor
instruments may exist that will improve the accuracy of the
ANN. These means of normalizing the data may serve to
eliminate specific types of differences between. the gas
signatures, making pattern recognition by the ANN easier.
An analytical program also needs to allow for greater
variatrmn in the gas patterns seen with the numerous non-
0157:H7 serotypes. Through further development of an

analytical tool for interpreting the gas signatures and

91

ways of normalizing the data, the diagnostic value of the

gas sensor based technology could be greatly improved.

92

REFERENCES

Bartlett, P. N., J. M. Elliot, and J. W. Gardner. 1997.
Electronic noses and their application in the food
industry. Food Technology 51:44-48.

Buchanan, R. L., and M. P. Doyle. 1997. Foodborne disease
significance of Escherichia coli 0157:H7 and Other
enterohemorrhagic E. coli. Food Technology 51:69—76.

Doyle, M. P., L. R. Beuchat, and T. J. Montville, eds.
1997. Food Microbiology: Fundamentals and Frontiers.
Washington, DC: American Society for Microbiology.

Gardner, J. W., M. Craven, C. Dow, and E. L. Hines. 1998.
The prediction of bacteria type and culture growth
phase by an electronic nose with a multi-layer

perceptron network. Measurement Science and Technology
9:120-127.

Gyles, C. L., ed. 1994. Escherichia coli in Domestic
Animals and Humans. Wallingford: CAB International.

Hancock, D. D., T. E. Besser, D. H. Rice, E. D. Ebel, D. E.
Herriott, and L. V. Carpenter. 1998. Multiple sources
of Escherichia coli 0157 in feedlots and dairy farms
in the Northwestern USA. Preventive Veterinary
Medicine 35:11-19.

Keshri, G., N. Magan, and P. Voysey. 1998. Use of an
electronic nose for the early detection and
differentiation between spoilage fungi. Letters in
Applied Microbiology 27:261—264.

Lane, A. J. P., and D. C. Wathes. 1998. An electronic nose
to detect changes in perineal odors associated with
estrus in the cow. Journal of Dairy Science 81:2145-
2150.

Moat, A. G., and J. W. Foster. 1995. Microbial Physiology
(Third ed.). Wiley-Liss. New York, NY.

Padhye, N. V., and M. P. Doyle. 1992. Escherichia coli
0157:H7: epidemiology, pathogenesis, and methods for
detection in food. Journal of Food Protection 55:555—
565.

93

Smith, R. D. 1995. Evaluation of Diagnostic Tests. In
Veterinary Clinical Epidemiology: A Problem-Oriented
Approach. pp 31—52. Ann Arbor, MI: CRC Press.

Younts, S., E. Alocilja, W. Osburn, S. Marquie, and D.
Grooms. 1999a. Development of electronic nose
technology as a diagnostic tool in detection and
differentiation of Escherichia coli 0157:H7. Journal
of Animal Science 77 (Suppl. l):129 (Abstr.).

Younts, S. M., D. Grooms, W. Osburn, S. Marquie, and E.
Alocilja. 1999b. Differentiation of Escherichia coli
0157:H7 From non-0157:H7 E. coli serotypes using a gas
sensor based, computer controlled detection system.
Paper presented at Institute of Biological
Engineering, June, Charlotte, NC.

Younts, S. M. 1999a. Chapter 2: Development and evaluation
of a gas sensor based instrument for identifying E.
coli 0157:H7 in a laboratory setting. M. S. thesis.
Michigan State Universtiy, East Lansing, MI.

Younts, S. M. 1999b. Chapter 3: Differentiation of
Escherichia coli 0157:H7 From non—0157:H7 E. coli
serotypes using a gas sensor based, computer-
controlled detection system. M.S. thesis. Michigan
State University, East Lansing, MI.

94

General Discussion

A gas sensor based instrument was developed and
evaluated for its ability to demonstrate differences
between the gas emissions of E. coli 0157:H7 and non—
0157:H7 E. coli. This initial investigation shows that
this technology has promise for enhancing food safety with
further refinement. The instrument was capable of
detecting' gas emissions and. establishing’ gas signatures.
Based on visually discernible differences in the gas
signatures generated, it was possible to accurately
classify the laboratory isolates as either E. coli 0157:H7
or non-0157:H7 E. coli. The greater strain variation in

the field isolates made correct classification of the

isolates rmnxa challenging. However, similarities ill the
gas signatures of the E. coli 0157:H7 isolates were
observed. 13m: greatest similarities were seen. among E.

coli 0157:H7 isolates from the same source or geographical
location.

There are a number of limitations involved with this
initial investigation that need to be addressed in further
studies. This study was initiated with no previous
related research for defining protocols. Experiments were

conducted to develop protocols and procedures in a

95

laboratory setting with characterized E. coli cultures to
develop initial operating principles. Further research can
be focused on addressing the identified limitations.

As seen with the results from the field isolates, only
a limited number of E. coli isolates, strains, and
serotypes were represented. It may be difficult to group
isolates as either E. coli 0157:H7 or non-0157:H7 E. coli
because of strain variation and differences in the numerous
non-0157:H7 serotypes. In the future, as more gas
signatures are collected and analyzed from more isolates of
E. coli more classifications or more appropriate groupings
may be identified.

Pure cultures of E. coli grown in the sensor
instrument generated. distinct results (n: gas signatures.
In mixed cultures it may be more challenging to distinguish
the presence of E. coli 0157:H7 from the gas emissions of
other organisms. Further studies need to be carried out
with mixed cultures containing various serotypes of E. coli
and different species of bacteria.

Only one instrument was built and used for this study.
Results were not confirmed with an additional instrument.
Parallel studies with similar instruments incorporating new
methods and technologies are Ibeing ‘undertaken. (Alocilja,

1998). The results of this study and those of other

96

investigations will be used to evaluate the most
appropriate instrument refinement and. methods to enhance
this technology.

Addressing the limitations in the further development
of the gas sensing instrument is essential to determine the
practicality and most appropriate applications of this
technology in food safety. This research showed that
consistent observable differences do exist between the gas
signatures of E. coli 0157:H7 and non-0157:H7 serotypes.
Based on this finding there is a basis for future
investigations that focus on developing and refining this
technology. With more refinement, there could be numerous
applications for this technology in animal production food
safety, food processing, and human medicine.

As a diagnostic tool in pre—harvest food safety
efforts there could be several potential applications. A
gas sensor based instrument could aid in the identification
of cattle that carry E. coli 0157:H7 or to'determine the
location of environmental sources of the pathogen within
production systems. The technology may also find use as a
monitoring device to help determine the efficacy of
pathogen. control (n: intervention. strategies at time pre-
harvest level. As an economical and less labor intensive

screening test, it could also be used to enhance initial

97

laboratory identification of E. coli 0157:H7. Similarities
seen in the gas signatures of isolates from the same source
or geographical location indicate the potential use of the
instrument as an epidemiological tool for identifying
related isolates.

Future applications of the artificial olfactory
technology could include identification of the pathogen in
food. and humans. A. gas sensor* based instrument could
potentially be developed to monitor meat products, such as
ground beef, for E. coli 0157:H7 contamination. In human
medicine, this technology could enhance clinical diagnosis
and determination of treatment effectiveness.

This study served as an initial investigation into the
use of the gas sensor based technology for identifying E.
coli 0157:H7. Based on the results and apparent
limitations, there are several recommendations to be
considered to determine the practical value of this
technology. First, more bacteria types need to be assayed
to determine: if gas signature differences exist between
different species of bacteria. This is particularly
important for evaluating the results from mixed bacteria
cultures. Second, additional instruments need to be
constructed and tested for their ability to yield

comparable and consistent results. Third, it is important

98

to) determine the ‘most appropriate type of jpattern
recognition program that recognizes characteristbc E. coli
0157:H7 gas signatures while allowing for greater variation
from other E. coli serotypes. Fourth, the use of different
types of growth media should be evaluated. For example, is
it possible to identify the presence of E. coli 0157:H7 in
bovine fecal matter or ground beef using a gas sensor based
instrument? Fifth, by obtaining more field isolates it may
be possible to compare similarities or differences between
strains of E. coli 0157:H7 depending on source. Possible
use as an epidemiological tool for screening strains of
bacteria could exist if differences were found between
cattle versus human isolates. Lastly, continued evolvement
of the instrumentation to incorporate new developments in
gas sensors and related technologies is important.

This study' demonstrated. that discernible» differences
exist in the patterns of gas emissions from E. coli 0157:H7
and non—0157:H7 E. coli. An instrument was developed
capable of detecting these gas emissions that result from
bacterial metabolism, and generating repeatable gas
signatures. Expanded studies may prove that computer
controlled. gas sensor' based technology' is beneficial in

monitoring nmltiple Eh coli cultures and identifying

99

isolates as E. coli 0157:H7 based on the pattern of gas

emissions.

100

REFERENCES

Alocilja, E. 1998. Personal Communications.

101

APPENDICES

102

APPENDIX A

RESEARCH PROTOCOLS

Culturing and collection of E. coli gas signaturesmmmm104

Establishing growth curves 105

 

Creating stock cultures and determining concentrationm106

103

Culturing and collection of E. coli gas signatures:
To start

1. Turn on dry block heater and set to 37°C

2. Turn on computer and open MeterBOSS (Teramar Group,
Inc., El Paso, TX) program

3. In MeterBOSS, go to Data Record, click Data Plot

4. On screen, click Filename field, enter file name for
experiment

5. Next click Time Log Interval field, enter desired rate
of gas sampling (5 minutes for standard E. coli
experiments)

Working in a Biohazard hood
6. Aseptically transfer 10ml of nutrient broth (Difco

Laboratories, Detroit, MI) to a sterile 14ml
polystyrene vial (Falcon 352057, Becton Dickinson
Labware, Franklin Lakes, NJ)

7. Inoculate vial with desired bacteria concentration (105

CFU’s/ml for standard E. coli experiments) using a
sterile pipette

8. Place the open, inoculated vial in the rear center well
of the dry block heater
9. Settle the sensor chamber’ centrally' over the heater

block with the center hole in the chamber base plate
over the vial

To start sampling
10. On screen, click Setup Complete to start gas sampling

*Do not exit MeterBOSS Data Display Screen or sampling will
be discontinued

11. Let system run for desired length of sampling (16 hours
for standard E. coli experiments)

To End

12. Exit MeterBOSS, moving pointer to top of screen will
display menu, EXIT

13. Lift sensor chamber and remove vial

14. Add nolvassan solution to the vial, seal and discard in
biohazard waste container

15. Turn off dry block heater

104

Establishing growth curves:

To Start
1” Turn on dry block heater and set to 37°C
:2.Aseptically transfer 10ml of nutrient broth (Difco
Laboratories, Detroit, MI) to a sterile 14ml polystyrene
vial
3.Inoculate vial with desired bacteria concentration (105
CFU’s/ml for standard E. coli experiments) using a
sterile pipette
4. Place the open, inoculated vial in the rear center well
of the dry block heater
5.USing a sterile pipette, pull lOOpl of sample culture and
transfer to 10ml of nutrient broth in a polystyrene vial
(1:100 dilution)
—vortex vial approx. 20 seconds
-transfer lOul to each of 2 plates, spread, incubate
inverted at 37°C
EL Settle the sensor chamber centrally over the heater
block with the center hole in the chamber base plate over
the vial
At 2 & 4 HOURS
'7.Prepare 2 sterile polystyrene vials with 10 ml of
nutrient broth each
8.1ift the sensor chamber, Pull 100ul from sample and
transfer to first vial (1:100 dilution)
—vortex and transfer 100ul to second vial (serial
dilutions) (1:10,000 dilution)
-and transfer 10ul to each of 2 plates, spread
59.From second vial, transfer 10pl to each of two plates,
spread, and incubate
10. Replace sensor chamber
At 6, 8, 10, 12, 14, & 16 HOURS
11. Prepare 3 vials with 10 ml of nutrient broth each
12. Pull 100ul from sample and transfer to first vial(1:100
dilution)
-vortex and transfer 100pl to second vial
(1:10,000 dilution)
13. Vortex second vial and transfer lOOpl to third vial and
lOpl to each of 2 plates(1:1,000,000 dilution), spread
14. Vortex third vial and transfer lOpl to each of 2
plates, spread, incubate
15. Count plates 12—24 hours after incubation

105

Creating stock cultures and determining concentration:

To create stock cultures of E. coli

2.

Aseptically transfer E. coli isolate to 10ml of
nutrient broth (Difco Laboratories, Detroit, MI) in a
sterile 14 ml polystyrene vial ()

Incubate inoculated vial for 12—24 hours at 37°C

To determine stock culture concentration

Prepare 3 vials with 10 ml of nutrient broth each

Pull lOOpl from stock culture and transfer to first
vial (1:100 dilution)
-vortex and transfer 100ul to second vial
(1:10,000 dilution)
Vortex second vial and transfer 100ul to third vial and
lOpl to each of 2 plates (l:1,000,000 dilution), spread
Vortex third vial and transfer lOul to each of 2
plates, spread, incubate
Count plates 12-24 hours after incubation
Back calculate 10—fold. dilutions for every step to
determine original stock culture concentration

106

APPENDIX B

REPRESENTATIVE GAS SIGNATURES FROM EACH E. COLI ISOLATE

 

 

 

 

 

Figure B.1 Representative gas signature from
laboratory isolate #1, non-0157:H7
E. coli

Figure B.2 Representative gas signature from
laboratory isolate #2 (SD89—3143, E. coli
Olll:NM

Figure B.3 Representative gas signature from
laboratory isolate #3 (80—2575), E. coli
0157:H13

Figure B.4 Representative gas signature from
laboratory isolate #4 (E47411/0), E. coli
05:H-

Figure B.5 Representative gas signature from
laboratory isolate #5 (Lab-0157:H7), E. coli
0157:H7

Figure B.6 Representative gas signature from

laboratory isolate #6
E. coli 0157:H7

(CDC BBOBB-MSl/O),

 

 

 

Figure B.7 Representative gas signature from
laboratory isolate #7 (ATCC 43895), E. coli
0157:H7

Figure B.8 Representative gas signature from
laboratory isolate #8 (E29962), E. coli
0157:H7

Figure B.9 Representative gas signature from
field isolate #1 (human source), E. coli
0157:H7

 

Figure B. 10

0157:H7

Representative gas signature from
field isolate #2 (human source), E.

coli

 

107

110

111

112

113

114

115

.116

117

118

119

Figure B. 11 Representative gas signature from
field isolate #3 (human source), E. coli
0157:H7

 

Figure B. 12 Representative gas signature from
field isolate #4 (venison source), E. coli
0157:H7

 

Figure B. 13 Representative gas signature from
field isolate #5 (cattle source), E. coli
0157:H7

 

Figure B. 14 Representative gas signature from
field isolate #6 (cattle source), E. coli
0157:H7

 

Figure B. 15 Representative gas signature from
field isolate #7 (cattle source), E. coli
0157:H7

 

Figure B. 16 Representative gas signature from
field isolate #8 (cattle source), E. coli
0157:H7

 

Figure B. 17 Representative gas signature from
field isolate #9 (cattle source), E. coli
0157:H7

 

Figure B. 18 Representative gas signature from
field isolate #10 (cattle source), E. coli
0157:H7

 

Figure B. 19 Representative gas signature from
field isolate #11 (cattle source), E. coli
0157:H7

 

Figure B. 20 Representative gas signature from
field isolate #12 (cattle source), E. coli
0157:H7

 

Figure B. 21 Representative gas signature from

field isolate #13 (cattle source), non—0157:H7

E. coli

 

108

120

121

122

123

124

125

126

127

128

129

130

Figure B. 22 Representative gas signature from
field isolate #14 (cattle source), non-0157
E. coli

:H7

 

Figure B. 23 Representative gas signature from
field isolate #15 (cattle source), non-0157
E. coli

:H7

 

Figure B. 24 Representative gas signature from
field isolate #16 (cattle source), non—0157
E. coli

:H7

 

Figure B. 25 Representative gas signature from
field isolate #17 (cattle source), non-0157
E. coli

:H7

 

Figure B. 26 Representative gas signature from
field isolate #18 (cattle source), non-0157
E. coli

:H7

 

Figure B. 27 Representative gas signature from
field isolate #19 (cattle source), non-0157
E. coli

:H7

 

Figure B. 28 Representative gas signature from
field isolate #20 (cattle source), non-0157
E. coli

:H7

 

109

131

132

133

134

135

136

137

HHOU .m hm"PMHOIGOC
.H# mumHowH >Houmnoan EOHm wndumcmHm mom m>HDMDGwm®HQ®m H.m musmHm

AmHm>M0ucH mGHHQEmm ouncHE mo HDQEDZ mHmEmm

hm mm mm mm MN NH H

MMH NNH HHH 00H mm mo

sntoA

 

110

.Ameflm-mmomo

Fa"HHHO HHOO .m
mt oumHOmH NHODMHOQMH Bonn manumsmHm mam 0>Humucmmmummm

AmHm>h0usH mcHHmEMm wuscHE my HDQESZ mHoEmm

MMH NNH HHH 00H mm Mb hm mm mm vM MN

 

N.m DHDmHm

NH H

ERICA

111

MHmuhmHO HHOU .W

.Amhmmnowv mt wumHOmH NHODmHoan EOHM DHDDMGmHm mom w>Hu8uamm®Hmmm m.m mhsmHm

AmHm>hmucH mGHHQEmm mussHE mo HDQEDZ onEmm

mm Mb hm mm mm mm MN NH H

MMH NNH HHH 00H

squA

 

112

.AO\HH¢bvmo

umumO HHOU .W
6* mumHOmH NHODMHOQMH Eon“ chaunmmHm mmm m>HumudmmmHQmm

AmHm>uwusfl mnHHmEmm muscHE mo HDQESZ mHmEmm

MMH NNH HHH 00H MM Mb bm mm mv mm MN

 

v.m musmHm

NH H

sqtoA

113

.Lsmubmflo-nmqo

bmubMHO HHOU .W
mt mumHomH >HODMHOQMH Eoum mHSumcmHm mmm m>HuchmmmHmwm

AmHm>HmucH msHHQEmm mudsHE my MDQEDZ DHQEmm

MMH NNH HHH 00H MM Mb bm mm mm mm MN

 

m.m ousmHm

NH H

sqtoA

114

smusmao :00 .m .8\Hmz-mmomm oooo

mt mumHomH Nuoumnoan Eouu mudumcmHm mam w>Hu6ucomMHQmm m.m mnsmHm

AmHm>HoucH mGHHQEmm muchE my HmnEsz mHQEmm

NNH HHH 00H MM Mb bw mm mm mm MN NH H

MMH

8310A

 

115

bmubMHO HHOO .m
.Ammmme UUH<V b¢ DDMHomH NHODMHOQMH Sosa wudumcmHm mom m>Humucwmmummm

AHm>HmucH mcHHQEmm mudsHE my anEDZ mHQEmm

MMH NNH HHH 00H mM Mb bm mm mm mm MN

 

b.m whomHm

NH H

squA

116

bmubmHo HHOU .m
.ANmmmva on mumHOmH >HODMHOQMH Eouw wudumcmHm mom 0>HDMDGmmeQmm

AmHm>HmucH mcHHmEmm DDDGHE my Hwafisz mHQEmm

MMH NNH HHH OOH mM Mb bm mm mm mm MN

m.m mnsmfim

NH H

 

snIoA

117

bmubMHO HHOU .m
.Awoudom smeacv Ht DDMHomH onHu Scum mnsumanm mom m>Humusmmanwm m.m mudem

AmHm>HmucH mcHHmEmm mussHE mo Honesz mHmEmm

MMH NNH HHH OOH MM Mb bm mm mm mm MN NH H

 

sntoA

118

.Amohdom cmescv

 

bmnbMHO HHOU .m
m# wumHOmH prHm Eouu manumcmflm mom m>HDMDcmmmHQmm oH.m DHSMHM

AmHm>kucH MGHHQEmm muncHE my Honesz mHmEmm

MMH NNH HHH 00H MM Mb bm mm mm mm MN NH H

squA

119

bmubMHO HHOU .W
.ADUHSOm cmEdcv mt mumHOmH onHw Eouw wudumcmHm mom 0>Humucmmmumwm

AmHm>HmuGH MGHHQEmm muscHE my Honesz mHmEmm

MMH NNH HHH 00H MM Mb bm mm mm MM MN

 

HH.m musmfim

NH H

8310A

120

.ADUHSOm GOmHnw>o

 

bmubmHo HHOU .m
on wumHomH pHmHm Eouw DHSDMGMHm mmm m>Hu6ucmmmHmwm NH.m DHDMHM

AmHm>HmucH MCHHQEmm DDSGHE mo HDQEDZ mHmEmm

MMH NNH HHH 00H MM Mb bm mm mm MM MN NH H

8310A

121

bmubMHO HHOU .m
.ADUHDOm mHuumoo mt wumHomH onHm Eouu musumcmHm mmm m>HumucommHQmm MH.m DHSMHM

AmHm>HmucH MGHHQEmm muddHE my HDQESZ mHQEmm

MMH NNH HHH OOH MM Mb bm mm mm vM MN NH H

 

sqtoA

122

bmnbMHO WHOU .W
.Awoudom oHuumov wt mumHOmH onHu Eon“ DHSDMGMHm mmm m>HumucmmmHmmm

AmHm>HmucH MGHHQEmm wuchE mo HDQESZ meEmm

MMH NNH HHH 00H MM Mb bw mm mm MM MN

 

va.m wusmfim

NH H

sqtoA

123

.Aounsom mHuumuo

bmnbMHO HHOU .W
b# wumHomH pHmHu Scum DHSDMCMHm mam m>Humucmmmhmmm

AmHm>HmucH MCHHQEmm ODSGHE mo HmQESZ DHQEMM

MMH NNH HHH 00H MM Mb bM mm mv vM MN

 

MH.m DHDMHM

NH H

SIIOA

124

bm" bMHO HHOU .m
.ADUHDOm mHuumov Mt mumHOmH onHm EOHM DHSDCMMHm mom 0>Hu0ucmmmnmwm MH.m deMHm

AmHm>hwusH MGHHQEmm muchE mo HDQESZ mHmEmm

mm MM MN NH H

MMH NNH HHH 00H MM Mb bm Mm

sntoA

 

125

bmubMHO HHOO .m
.Amoudom mHuumoo ma wumHOmH onHw Eonm wusumcmHm mmm 0>HumDsmmmHmmm

AmHm>HmucH MGHHQEmm muDGHE mo HDQESZ wHQEmm

MMH NNH HHH 00H MM Mb bM mm mm mm MN

 

FH.m musmflm

NH H

sqtoA

126

bmubMHO HHOU .m
.ADUHDOm wHuumoo oH# mumHomH UHDHM Eoum wndumcMHm mmm 0>Humucommnmmm

AmHm>HoucH MGHHmEmm DDDGHE mo Hondz mHQEmm

MMH NNH HHH 00H MM Mb bw mm MM MM MN

 

mH.m mssmfim

NH H

sqtoA

127

bmubMHO HHOO.M
.ADUHSOm mHuumov HHM mumHomH pHmHu EOHu ondumcmHm mom m>Humusmwwummm MH.m mudmfim

AmHm>kucH MGHHQEmm DDSGHE mo MDQEDZ mHQEmm

MMH NNH HHH 00H MM Mb bw mm MM MM MN NH H

 

snIoA

128

bmnbMHO HHOU .W
.ADUHDOm mHuumov «Ht mumHOmH onHu EOHM mhsumanm mam m>Humucwmmummm

AmHm>HmucH MGHHQEMm mudsHE mo HDQEDZ DHQEMM

MMH NNH HHH 00H MM Mb bM mm MM MM MN

 

oN.m mHSMHm

NH H

3310A

129

.ADUHSOm mHuumov

HHOU .W bmubMHOIGOG
MHM mumHOmH UHDHM Eouu DHDDMGMHm mmm w>HDMDGmmmHQ0m

AmHm>h0ucH MGHHQEmm muDCHE Mo HwnEDZ DHQEMM

MMH NNH HHH OOH MM Mb bM MM MM MM MN

 

Hm.m ousmflm

NH H

sqtoA

130

HHOU .W bmubMHOIGOG
.Amoudom wHuumov MHM mumHOmH onHu EOHM DHSDMCMHm mom w>Hu0ucwmeme NN.m DMDMHM

AmHm>hmucH MGHHQEmm muscHE mo HDQESZ mHQEmm

MMH NNH HHH OOH MM Mb bM MM MM MM MN NH H

 

8310A

131

HHOU .m bmubMHOIGOG

.Amosom mHuumoo MHM muMHomH onHM EOHM mHSDMEMHm mom 0>Humucmmwsmwm mm.m DHSMHM

AmHm>MmusH MGHHQEMm ouscHE My HDQEDZ meEmm

HHH 00H MM Mb bm MM MM MM MN NH H

MMH NNH

sqtoA

 

132

HHOU .W bmubMHOuGOC

.ADUHDOm wHuumov MHM mumHomH UHDHM Eon“ mhdumCMHm mmm 0>Humudmmmhmmm Mm.m whdem

AmHm>HoucH MCHHoEmm muchE Mo HDQEDZ wHQEmm

Mb bM MM MM MM MN NH H

MMH NNH HHH OOH MM

sntoA

 

133

HHOU .m bm" bMHOIGOG
.ADUHSOm wHuumuv bHu mumHomH UHDHM EOHM mudumcMHm mmm m>HumucmmmHQmm mm.m DHDMHM

AmHm>hmucH MGHHQEmm DDDGHE my Hwnfidz mHoEmm

MMH NNH HHH OOH MM Mb bM MM MM MM MN NH H

 

sntoA

134

HHOU .m bmubMHOICOC
.AmoHSOm mHuumoo MHM mumHomH onHM Eon“ mndumcmHm mmm w>HumuswmmHmmm MN.m DHDMHm

AmHm>HmusH MGHHMEmm muscHE Mo Honesz mHmEmm

MMH NNH HHH OOH MM Mb bM MM MM MM MN NH H

 

8310A

135

HHOU .m bmnbMHOIGOG
.Amonsom mHuumoo MHM mumHomH onHM EOHM eunumanm mam m>HDMDGmmemwm bm.m DHDMHM

AmHm>hmucH MGHHQEmm wuscHE Mo Hwnssz mHmEmm

MMH NNH HHH OOH MM Mb bM MM MM MM MN NH H

 

sqtoA

136

HHOU .m bmubMHOIQOHH
.Amossom mHuumuo omt wumHOmH UHDHM EOHM wudumcMHm mom m>HumucmmmHme mm.m musmHm

AmHm>kunH MGHHQEmm DDDGHE My umnfisz meEmm

MMH NNH HHH OOH MM Mb bM MM MM MM MN NH H

 

squA

137

BIBLIOGRAPHY

138

Bibliography

Alocilja, E. 1998. Personal Communications.

Atlas,, R. M. 1995. Principles of Microbiology. Mosby-Year
Book, Inc. St Louis, MO

Bartlett, P. N., J. M. Elliot, and J. W. Gardner. 1997.
Electronic noses and their' application. in the food
industry. Food Technology 51:44-48.

Boyce, T. G., D. L. Swerdlow, and P. M. Griffin. 1995.
Escherichia coli 0157:H7 and the hemolytic uremic
syndrome. New England Journal of Medicine 333:364-368.

Brown, C. A., B. G. Harmon, T. Zhao, and M. P. Doyle. 1997.
Experimental Escherichia coli 0157:H7 carriage in
calves. Applied and EMvironmental Microbiology 63:27-
32.

Buchanan, R. L. , and M. P. Doyle. 1997. Foodborne disease
significance of Escherichia coli 0157:H7 and other
enterohemorrhagic E. coli. Food Technology 51:69-76.

Cray, VI. C., and IL 14. Moon. 1995. Experimental infection
of calves and adult cattle with Escherichia coli
0157:H7. Applied and Environmental Microbiology
61:1586—1590.

Dargatz, D. A., S. J. Wells, L. A. Thomas, D. D. Hancock,
and L. P. Garber. 1997. Factors associated with the
presence of Ebcherichia coli 0157 in feces of feedlot
Cattle. Journal of Food Protection 60:466—470.

Diez—Gonzalez, FR, if. R. Callaway, NL (3. Kizoulis, and J1
B. Russell. 1998. Grain feeding and the dissemination
of acid-resistant Escherichia coli from. (cattle.
Science 281:1666-1668.

Doyle, M. P., L. R. Beuchat, and T. J. Montville, eds.

1997. Food lMicrobiology: Fundamentals and. Frontiers.
Washington, DC: American Society for Microbiology.

139

Figaro USA, I. Product Information Guide Figaro USA, Inc,
February, 1996.

USDA-FSIS. 1998. Report to Congress, FoodNet: An. Active
Surveillance System for Bacterial Foodborne Diseases
in the United States. Washington, D.C.

Garber, L. P., S. J. Wells, D. D. Hancock, M. P. Doyle, J.
Tuttle, J. A. Shere, and T. Zhao. 1995. Risk factors
for fecal shedding of Escherichia coli 0157:H7 in
dairy calves. Journal of the American Veterinary
Medical Association 207:46-49.

Gardner, J. W., M. Craven, C. Dow, and E. L. Hines. 1998.
The prediction. of Ibacteria type and culture growth
phase by an electronic nose with a multi-layer
perceptron network. Measurement Science and Technology
9:120-127.

Gyles, C. L., ed. 1994. Escherichia coli in Domestic
Animals and Humans. Wallingford: CAB International.

Hancock, D. D., T. E. Besser, D. H. Rice, E. D. Ebel, D. E.
Herriott, and In I]. Carpenter. 1998. Multiple sources
of Escherichia coli 0157 in feedlots and dairy farms
in the Northwestern USA. Preventive Veterinary
Medicine 35:11-19.

Hancock, D. D., D. H. Rice, D. E. Herriott, T. E. Besser,
E. D. Ebel, and L. V. Carpenter. 1997a. Effects of
farm manure-handling practices on Escherichia coli
0157:H7 prevalence in cattle. Journal of Food
Protection 60:363-366.

Hancock, D. D., D. H. Rice, L. A. Thomas, D. A. Dargatz,
and T. E. Besser. 1997b. Epidemiology of Escherichia
coli 0157 in feedlot cattle. Journal of Food
Protection 60:462-465.

Keshri, G., N. Magan, and P. Voysey. 1998. Use of an
electronic nose for the early detection and
differentiation between spoilage fungi. Letters in
Applied Microbiology 27:261-264.

Kudva, I. T., P. G. Hatfield, and C. J. Hovde. 1996.

Escherichia coli 0157:H7 in ndcrobial flora of sheep.
Journal of Clinical Microbiology 34:431-433.

140

Lane, A. J. P., and D. C. Wathes. 1998. An electronic nose
to detect changes in perineal odors associated with
estrus in the cow. Journal of Dairy Science 81:2145-

2150.

March, 8. B., and S. Ratnam. 1986. Sorbitol-MacConkey
medium for detection of Escherichia coli 0157:H7
associated. with. hemorrhagic colitis. Journal of

Clinical Microbiology 23:869-872.

Moat, A” CL, enui.J. W. Foster. 1995. Microbial Physiology.
(Third ed.). Wiley—Liss. New York, NY.

Padhye, N. 'V., and 14. ER Doyle. 1992. Escherichia coli
0157:H7: epidemiology, pathogenesis, and methods for
detection in food. Journal of Food Protection 55:555-
565.

Ratnam, S., S. March, R. Ahmed, G. Bezanson, and S.
Kasatiya. 1988. Characterization of Escherichia coli
serotype 0157:H7. Journal of Clinical Microbiology
26:2006-2012.

Rice, D. H., D. D. Hancock, and T. E. Besser. 1995.
Verotoxigenic E. coli 0157:H7 colonization of wild
deer and range cattle. The Veterinary Record 137:524.

Sanderson, M. W., J. M. Gay, D. D. Hancock, C. C. Gay, L.
K. Fox, and T. E. Besser. 1995. Sensitivity of
bacteriologic culture for detection of Escherichia
coli in bovine feces. Journal of Clinical Microbiology
33:2616-2619.

Smith, R. D. 1995. Evaluation of Diagnostic Tests. In
Veterinary Clinical Epidemiology: A Problem-Oriented
Approach. pp 31-52. Ann Arbor, MI: CRC Press.

Stevenson, K. E., and D. T. Bernard, eds. 1995. HACCP
Establishing Hazard. Analysis Critical Control Point
Programs: A Workshop Manual. Second ed. The Food
Processors Institute. Washington, D.C.

Tarr, P. I. 1995. Escherichia coli 0157:H7: Clinical,

diagnostic and epidemiological aspects of human
infection. Clinical Infectios Disease 20:1-10.

141

Thompson, J. S., D. S. Hodge, and A. A. Borczyk. 1990.
Rapid biochemical test to identify verocytotoxin-
positive strains of Escherichia coli serotype 0157.
Journal of Clinical Microbiology 28:2165-2168.

Trevena, W. B., R. S. Hooper, C. Wray, G. A. Willshaw, T.
Cheasty, and G. Domingue. 1996. Vero cytotoxin—
producing Escherichia coli 0157 associated with
companion animals. The Veterinary Record 142:400.

Tuang, F. N., J. L. Rademaker, E. C. Alocilja, F. J. Louws,
amd F. J. de Bruijn. 1999. Identifiation of bacterial
rep-PCR genomic fingerprints using a backpropagation
neural network. FEMS Microbiology Letters 177:249-256

Unnevhr, L. J., and H. H. Jensen. 1996. HACCP as a
regulatory innovation to improve food safety in the
meat industry. .American. Journal of .Agricultural

Economics 78:764-769.

Younts, S., E. Alocilja, W. Osburn, S. Marquie, and D.
Grooms. 1999a. Development of electronic nose
technology as a diagnostic tool in detection and
differentiation of Escherichia coli 0157 :H7 . Journal
of Animal Science 77 (Suppl. l):129 (Abstr.).

Younts, S. M., D. Grooms, W. Osburn, S. Marquie, and E.
Alocilja. 1999b. Differentiation of Escherichia coli
0157:H7 From non-0157:H7 E. coli serotypes using a gas
sensor based, computer controlled detection system.
Paper' presented. at Institute of Biological
Engineering, June, Charlotte, NC.

Younts, S. M. 1999a. Chapter 2: Development and evaluation
of a gas sensor based instrument for identifying E.
coli 0157:H7 in a laboratory setting. M. S. thesis.
Michigan State Universtiy, East Lansing, MI.

Younts, S. M. 1999b. Chapter 3: Differentiation of
Escherichia coli 0157:H7 From non-0157:H7 E. coli
serotypes using' a «gas sensor based, computer-

controlled detection system. M.S. thesis. Michigan
State University, East Lansing, MI.

142

Zadik, P. M., P. A. Chapman, and C. A. Siddons. 1993. Use
of tellurite for the selection of verocytotoxigenic

Escherichia coli 0157. Journal of Medical Microbiology
39:155-158.

Zhao, T., M. P. Doyle, B. G. Harmon, C. A. Brown, P. O. E.
Mueller, and A. H. Parks. 1998. Reduction of carriage
of enterohemorrhagic EBcherichia coli 0157:H7 in
cattle by inoculation with probiotic bacteria. Journal
of Clinical Microbiology 36:641-647.

143

MICHIGAN STATE UNIV. LIBRARIES
\lHlWWllWlWWW“WNWWNW)
31293020608505