u'lt‘,
s." ' 59w.

. )
a
v.
Ln}.
1!‘
m

r?

.L..
AS; u.-
.! EN

1'
$JI‘.;;~¥§.‘§
2.311333“:
blissful}???

:, g. '

['3

I
9
fig .4?

L‘

‘Q.
2“):
55'

"yllELal-J

ma, E;- “t ‘ 4:52;;-

"f.
,4

.1

.L'
- .5

V

'1..¢.'

.u'J-ht' Inga A

- u .apu‘l

 

'15? (
39%!

hr “Gig ‘

'“u. a;

' w 1‘
{13:15 t?“

.
.
’ t”:

.2

'fi;i~r‘.§.‘ .. _,

u I“:
4%.»

vs.

 

Eu
3'

THESIS

This is to certify that the
thesis entitled

IDENTIFYING SOURCES OF SURFACE WATER
CONTAMINATION BY DISCRIMINANT ANALYSIS OF
PATTERNS OF ANTIMICROBIAL RESISTANCE IN E. COLI

presented by

Raida Sayah Sayah

has been accepted towards fulfillment
of the requirements for the

M .8. degree in Large Animal Clinical Sciences
(Epidemiology)

 

 

y VNTajor Professor’s Signature
August 18, 2004

Date

MSU is an Affirmative Action/Equal Opportunity Institution

 

 

 

LIBRARY
Michigan State
University

 

 

 

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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRCJDatoDuepes-pts

IDENTIFYING SOURCES OF SURFACE WATER CONTAMINATION BY
DISCRIMINANT ANALYSIS OF PATTERNS OF ANTIMICROBIAL RESISTANCE
IN E. COLI

By

Raida Sayah Sayah

A THESIS

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

MASTER OF SCIENCE

Department of Large Animal Clinical Sciences
(Epidemiology)

2004

ABSTRACT
IDENTIFYING SOURCES OF SURFACE WATER CONTAMINATION BY
DISCRIMINANT ANALYSIS OF PATTERNS OF ANTIMICROBIAL RESISTANCE
IN E. COLI
By

Raida Sayah Sayah

A repeated cross-sectional study was conducted to investigate whether discriminant
analysis of antimicrobial resistance profiles of E. coli would be useful in identifying
sources of fecal contamination of surface water from the Red Cedar (Michigan)
watershed. Fecal samples were collected from livestock, companion animals, human
septic tanks, farmed deer and wild geese from the watershed, and water was collected
from several different sites on the Red Cedar River. Disc diffusion was used to test for
susceptibility to neomycin, gentamicin, streptomycin, chloramphenicol, ofloxacin,
trimethoprim/sulfamethoxazole, tetracycline, ampicillin, nalidixic acid, nitrofurantoin,
cephalothin, and sulfisoxazole. Both resubstitution and cross-validation discriminant
function analysis methods were applied to the data.

Resistance to at least to one antimicrobial was seen in isolates from domestic animals,
wildlife, surface water, and human septic tanks. Overall, E. coli isolates from food
animals showed resistance to the largest number of antimicrobials, followed by horses,
companion animals, human septic tanks, farmed deer, wild geese and surface water.
Discriminant analysis using the resubstitution method produced a higher but biased
average rate of correct classification (ARCC), while cross-validation produced lower but

valid ARCC. Also, the ARCC improved as the number of source categories decreased.

ACKNOWLEDGMENTS

I would first like to thank USDA and Population Medicine Center for their financial
support of this project.

I am grateful to my major advisor Dr. John B. Kaneene for all the encouragement and
tremendous support that he provided from the beginning to the end of this project.
Without his epidemiological experience, wise guidance and strong determination to keep
meeting the challenges and working through the obstacles that come along with this
venture, my success would not be possible.

I would like to thank Dr. Joseph Gardiner, Dr. Michellel Kopcha and Dr. Scott Witter
for being a part of participating on my committee.

A special thank you goes to Dr. Yvette Johnson for her enormOus support and help
throughout the study.

I would like to thank Dr. Mark Hansen; Ingham County Extention Agent and Dr.
Betsy Deerburger; Livingston county Extention Agents for assisting us reach out to the
farmers.

I would like to thank all the participated farmers in this study and the lab technicians:
Katie May at Michigan State University, Janice Donohoe at the University of Maryland
and all the lab staff at Michigan Department of Environmental Quality, I would like to
thank Mr. Barry Looper from the Department of Natural Resources, Mr. Randy Abbot
from Ingham County Drain Commissioner, Mr. Bob Godbold and Mr. Rene Franco from
Ingham County Environmental Health Department, Mr. Don Hayduk from Livingston
County Environmental Health Department, RoseAnn Miller from the Population

Medicine Center and Betty Wernette-Babian for all the help in collecting and processing

iii

the fecal and water samples, and analyzing the data. Without their assistance this project
would not be possible.

Finally, my deepest thank to my family especially my husband and children who give
me the strength on a day-to-day basis to undertake my responsibilities, and make

everyday worthwhile; they are my inspiration in all I do.

iv

TABLE OF CONTENTS

LIST OF TABLES .................................................................................. vii
LIST OF FIGURES ............................................................................... viii
KEY TO APPREVIATIONS ........................................................................ x
INTRODUCTION .................................................................................... 1
Purpose ........................................................................................ l
Hypotheses tested ........................................................................... 2
Objectives ..................................................................................... 2
Overview ..................................................................................... 2
CHAPTER 1
LITERATURE REVIEW: USE OF DISCRIMINANT ANALYSIS FOR THE
CLASSIFICATION OF ECOLOGICAL DATA ................................................ 4
Definition ..................................................................................... 4
History of discriminant analysis ........................................................... 5
Types of discriminant analysis ............................................................. 6
Measuring the performance of discriminant analysis ................................... 7
Use of discriminant analysis in ecological studies ...................................... 9
Assumptions ................................................................................. 21
Interpretation of discriminant analysis ................................................... 24
Logistic regression and discriminant analysis .......................................... 25
Cluster analysis and discriminant analysis ............................................. 25
MANOVA and discriminant analysis ................................................... 26
Conclusion .................................................................................. 27
CHAPTER 2

PATTERNS OF ANTIMICROBIAL RESISTANCE OBSERVED IN E. COLI
ISOLATES OBTAINED FROM DOMESTIC AND WILD ANIMAL FECAL
SAMPLES, HUMAN SEPTAGE, AND SURFACE WATER IN MICHIGAN, USA....3O

Abstract ...................................................................................... 30
Introduction ................................................................................. 31
Hypotheses tested .......................................................................... 35
Objectives ................................................................................... 35
Materials and Methods ..................................................................... 36
Results ....................................................................................... 47
Discussion ................................................................................... 58

CHAPTER 3

DISCRIMINANT ANALYSIS AS A TOOL FOR CLASSIFYING THE SOURCES OF

E. COLI CONTAMINATION OF SURFACE WATER IN MICHIGAN, USA ........... 80
Abstract ...................................................................................... 8O

Introduction ................................................................................. 81

Hypotheses tested ........................................................................... 87
Objectives ................................................................................... 88
Materials and Methods .................................................................... 88
Results ....................................................................................... 92
Discussion .................................................................................. 97
OVERALL SUMMARY AND CONCLUSIONS ............................................ 107
APPENDEX ........................................................................................ 110
Informed consent form ................................................................... 111
Letter from Population Medicine Center to farmers ................................. 112
Letter from Ingham County Extension Agent to farmers ........................... 113
Letter from Livingston County Extension Agents to farmers ..................... 114
REFERENCES .................................................................................... 115

vi

LIST OF TABLES

Table 2-1: Concentrations and diffusion zone break points for resistance for
antimicrobial agents tested in this study, by class of antimicrobial agent .................. 46

Table 2-2: Antimicrobial agents reported as having been used on the farms in the
different species of animals sampled ............................................................ 48

Table 2-3: Percentage of resistant isolates for specific antimicrobial agents, by species
exposure class ....................................................................................... 55

Table 2-4: Percentage of multi-drug resistant isolates, by species exposure class . ...56

Table 2-5: Most commonly identified combinations of agents in multi-drug resistant
isolates from food animals ......................................................................... 57

Table 3-1: Comparison of classification tables for classification rules generated by
non-parametric discriminant analysis of antimicrobial resistance profiles (neomycin,
streptomycin, tetracycline, ampicillin, sulfamethoxazole, cephalothin, sulfisoxazole)
using two different rule development methods, for five animal classes .................... 94

Table 3-2: Comparison of classification tables for classification rules generated by
non-parametric discriminant analysis of antimicrobial resistance profiles (neomycin,
streptomycin, tetracycline, ampicillin, sulfamethoxazole, cephalothin, sulfisoxazole)
using two different rule development methods, for two animal classes ..................... 95

Table 3-3: Comparison of average rate of correct classification (ARCC) values
between re-substitution and cross-validation rule development for different animal
classification systems .............................................................................. 95

Table 3-4: Classification of E. coli from water samples using discriminant analysis. . ....96

vii

LIST OF FIGURES

Figure 2-1: Michigan State map showing the location of Red Cedar Watershed .......... 37
Figure 2-2: The Red Cedar Watershed, Michigan, USA ...................................... 37
Figure 2-3: The sampler used to collect water samples ....................................... 42

Figure 2-4: Disc diffusion test plate showing the zones of growth inhibition of E. coli

around an antimicrobial agent ..................................................................... 45
Figure 2-5: Antimicrobial resistance of E. coli from cattle ................................... 69
Figure 2-6: Antimicrobial resistance of E. coli fi'om cattle farms ............................ 69
Figure 2-7: Antimicrobial resistance of E. coli from sheep ................................... 70
Figure 2-8: Antimicrobial Resistance of E. coli from sheep farms .......................... 70
Figure 2-9: Antimicrobial resistance of E. coli from swine ................................... 71
Figure 2-10: Antimicrobial resistance of E. coli from swine farms ......................... 71

Figure 2-11: Antimicrobial resistance of E. coli from horses ................................. 72
Figure 2-12: Antimicrobial resistance of E. coli fi'om horse farms .......................... 72
Figure 2-13: Antimicrobial resistance of E. coli from poultry ............................... 73
Figure 2-14: Antimicrobial resistance of E. coli from poultry farms ........................ 73
Figure 2-15: Antimicrobial resistance of E. coli from companion animals ................. 74
Figure 2-16: Antimicrobial resistance of E. coli from farmed deer and wild geese ....... 74
Figure 2-17: Antimicrobial resistance of E. coli from human septage ...................... 75
Figure 2-18: Antimicrobial resistance of E. coli from water ................................. 75

Figure 2-19: Antimicrobial resistance of E. coli from a farm with pigs and chickens. . ...76

Figure 2-20: Antimicrobial resistance of E. coli from a farm with cattle, horses and
sheep .................................................................................................. 76

viii

Figure 2-21: Antimicrobial resistance of E. coli isolates from chicken and food
animals ............................................................................................... 77

Figure 2-22: Antimicrobial resistance of E. coli isolates fiom companion animals and
food animals .......................................................................................... 77

Figure 2-23: Antimicrobial resistance of E. coli isolates from companion animals and

horse ................................................................................................... 78
Figure 2-24: Antimicrobial resistance of E. coli isolates from horse and food

Animals ............................................................................................... 78
Figure 2-25: Multi-drug resistant E. coli isolates .............................................. 79

ix

KEY TO ABBREVIATIONS

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

AF LP Amplified Fragment Length polymorphism

ANERR Apalachicola National Estuarine Research Reserve
ARCC Average Rate of Correct Classification

ATCC American Type Culture Collection

CFU Colony Forming Unit

DA Discriminant Analysis

DNR Department of Natural Resources

E. coli Escherichia coli

IMVIC Indole Methyle red Voges proskauer Simmon Citrate
LDA Linear Discriminant Analysis

MANOVA Multiple Analysis Of Variance

MAR Multiple Antibiotic Resistance

MIC Minimum Inhibitory Concentration

NARMS National Antimicrobial Resistance Monitoring System
NCCLS National Committee for Clinical Laboratory Standards
PFGE Pulsed field Gel Electrophoresis

RNA Ribonucleic Acid

TMP/SMX Trimethoprim Sulfamethoxazole

TSA Triple Sugar Iron

TSA Tryptic Soy Agar

TSB Tryptic Soy Broth

USA United States of America

USDA Untied States Department of Agriculture

USEPA United States Environmental Protection Agency
VRB Violet Red Bile

 

 

 

INTRODUCTION

Purpose

Water plays a major role in sustaining the natural systems on and under the earth's
surface. Contamination of surface water poses serious risks not only to human and
animal health but also to enviromnental health on a global scale.

Chemical fertilizers, livestock manure, and sewage may lead to deterioration of water
quality. However, equally threatening to both quality, and human and animal health is
the potential damage to water caused by bacterial contamination from manure and
sewage. Bacteria such as Salmonella, Campylobacter and E. coli are considered a major
threat to human health and are routinely isolated from animal and human feces.

Efforts to monitor water quality by determining the level of contamination with coliform
bacteria have been undertaken for more than a century. Attempts have also been made to
determine the source of fecal contamination. Discriminant function analysis has been
reported to provide a promising tool that could offer a low cost and statistically valid
means of identifying the source species of water contamination based on the antibiotic
sensitivity profile of the E. coli isolates from different sources. Since the magnitude of
threat posed by bacterial contamination of surface water in Michigan is unknown and
protecting water quality is a vital issue from the perspective of environmental protection
and human health, epidemiological research is needed to determine the extent to which
human, livestock, companion animals, and wildlife contribute to contamination of surface
water in the state. The overall goal of the studies in this thesis was to investigate whether
combining two techniques (determining antimicrobial resistance profiles of E. coli and

conducting discriminant function analysis on such profiles) would be useful in

identifying the source of fecal contamination of surface water in Michigan. The Red
Cedar Watershed was used as a case site. It is hypothesized that such techniques would
be useful tool in programs aimed at reducing the risk of surface water contamination.
Hypothesis Tested
The general hypothesis that was tested in this thesis was that,
“Domestic animal, human, and wildlife sources of fecal contamination of
surface water can be identified using discriminant function analysis of
antimicrobial resistance patterns of E coli isolated from such sources.”
Objectives
1. Identify patterns of antimicrobial resistance in E. coli isolates obtained fi'om
human septage, domestic and wild animals fecal samples in Red Cedar
Watershed- Michigan, USA.
2. Identify patterns of antimicrobial resistance in E. coli isolates obtained from
surface water samples in Red Cedar Watershed- Michigan, USA.
3. Apply discriminant fimction analysis of antimicrobial resistance patterns in
identifying the source of fecal contamination of surface water.
Overview
The thesis is arranged in chapters as outlined below. In addition to the overall
hypothesis stated above, specific hypotheses are presented for chapter 2 and chapter 3.
Chapter 1 is a literature review of the use of discriminant function analysis in ecological
studies. Particular attention is paid to its successful use in classification the ecological
data. Chapter 2 discusses patterns of antimicrobial resistance in E. coli isolates obtained

from domestic animal, human and wildlife fecal samples, and surface water samples in

Michigan, USA. Chapter 3 describes the application of the discriminant function analysis
to the anti-microbial resistance profiles of E. coli isolated from different sources in order
to classify and identify the source species of microbial surface water contamination in
Red Cedar Watershed in Michigan, USA. Chapter 4 presents an overall summary for the

thesis and recommendations for future research.

CHAPTER 1
LITERATURE REVIEW: USE OF DISCRIMINANT ANALYSIS FOR THE
CLASSIFICATION OF ECOLOGICAL DATA

Definition

Discriminant analysis or discriminant fimction analysis is a multivariate statistical
method entails separating sets of observations and allocating new observations to
previously defined groups (Tatsuoka, 1970; Morrison, 1990; Johnson and Wichem, 1992;
Lattin et al., 2003). It is exploratory in nature. As a separator procedure, it is often
employed on a one-time basis in order to investigate observed differences when causal
relationships are not well understood. A classification procedure is less exploratory in the
sense that it leads to well-defined rules, which can be used for assigning new
observations (Johnson and Wichem, 1992). The two goals for discriminant analysis are
to determine which variables discriminate between two or more naturally occurring
groups and to classify cases into the values of categorical dependent groups (Tatsuoka,
1970; Morrison, 1990; Johnson and Wichem, 1992; Lattin et al., 2003). For example, an
educational researcher may want to investigate which variables discriminate between
high school graduates who decide to go to college, to attend a trade or professional school
or seek no further training or education. For such an investigation, the researcher could
collect data on numerous variables prior to students’ graduation such as achievement
motivation, academic performance, and personality. After graduation, most students will
fall into one of the three categories. Discriminant analysis could be used to determine
which variables are the best predictors of the students’ subsequent educational choice.

Measurement of height in a random sample of 50 males and 50 females provides another

example. Females are, on the average, not as tall as males and this difference will be
reflected in the difference in means for the variable height. Therefore, the variable height
allows us to discriminate between males and females with a better than chance
probability: if a person is tall then he is likely to be a male, if a person is short then she is
likely to be a female. The basic idea behind discriminant function analysis is to determine
whether groups differ with regard to the mean of a variable, and then to use that variable
to predict group membership. So that rather than test the usual hypothesis of equal mean
vectors, we wish to construct a linear compound or index for summarizing observations
from the groups on a one-dimensional scale that discriminates between the populations

by some measure of maximal separation.

History of discriminant analysis

Discriminant analysis was first proposed by Fisher inl936 as a statistical tool for use
in taxonomic problems originally confined to two-group situations (Fisher, 1936). He
suggested finding a linear combination of observations that would maximize the
difference between groups relative to the standard deviation within groups. Then
discriminant analysis involved more than two groups in which more than one
discriminant function is found (Rao, 1948; Tukey, 1949). The number of discriminant
functions in multi-group case is one less than the number of groups (Tatsuuoka, 1970).
Wald (1944) and Anderson (1951), created the classification statistic (W), in order to
assign an observation to population I if classification statistic W is usually greater than 0,
otherwise to population 11. Mahalanobis (1936) originated the Mahalanobis squared

distance, which is the distance between a case and the centroid for each group of the

dependent. The smaller the Mahalanobis distance, the closer the case to the group
centroid, and the more likely it is to be classified as belonging to that group
(Mahalanobis, 1936). When the parameters are unknown, they may be replaced by their
unbiased sample estimators to give the sample quadratic discriminant function, which
was originally proposed by Smith (1947). This resulted in discriminant analysis being

widely used in many types of ecological studies.

Types of discriminant analysis

Data for discriminant analysis consist of observations for which there are a grouping
index and associated vector of measurements. One objective of the analysis is to predict
the group to which an observation belongs based on its measurement value. Such a
formulation is called predictive discriminant analysis, and the prediction equations are
called classification or discriminant functions (Lattin et al., 2003). Alternatively, the
objective may be to exhibit optimal separation of groups, based on certain linear
transforms of the measurement variables, and this is called descriptive discriminant
analysis (Lattin et al., 2003). The associated linear functions are known as canonical
variates, the structure of which is often of primary concern to ecologist (Hix, 1988). Both
methods have been used in ecological studies but most have used the descriptive
orientation (Williams, 1983, 1988). However, there are studies that used the predicative
methods (Rice et al., 1983; Vemer et al., 1986).

When the distribution within each group is assumed to be multivariate normal, a
parametric method can be used to develop a discriminant function. It is also known as

classification criterion and determined by a measure of generalized square distance (Rao,

1973). The classification criterion can be based on individual within group covariance
matrices yielding a quadratic function of the pooled covariance matrix producing a linear
function. When no assumption can be made about the distribution within each group, or
when the distribution is assumed not to be multivariate normal, nonparametric methods
can be used to estimate the group specific densities. These methods include the kernel

and k-nearest neighbor methods (Rosenblatt, 1956).

Measuring the performance of discriminant analysis

The performance of discriminant analysis can be evaluated by its ability to classify
future observations using error count estimates and posterior probability error rate
estimates. The error count estimate can be calculated by applying the classification
criterion derived from the training sample data set that discriminant analysis uses to
derive the discriminant function- to a test set, and then counting the number of
misclassified observations. The group specific error count estimates are the proportion of
misclassified observations in the group. It has an optimistic bias called “apparent error
rate.” The estimation of error rate is very important, especially in medical application.
For example, in the field of electrocardiography, a patient is diagnosed as being healthy
or unhealthy on the basis of the results of an electrocardiogram; for this particular
application of W it is vital not to underestimate the error of classifying a patient as
healthy when the patient actually does have a heart disease.

Many techniques are used to estimate the error rates of sample discriminant functions
and to reduce the bias (Lachenbruch and Mickey, 1968). Most of the time, these methods

are not satisfactory in cases where the sample sizes are smaller than the variables. The

first one, the hold out method, is often used for validation of the discriminant function
(Lattin et al., 2003; Wiggins et al., 2003). This is a split halves test, where a portion of the
cases are assigned to the analysis sample for purposes of training the discriminant
function, then it is validated by assessing its performance on the remaining cases in the
hold-out sample. In other words, the training sample is used to construct the classification
function, and the validation sample is used to evaluate it. The validation sample is not
biased and also possesses the same properties as the analysis sample. The proportion
misclassified in the validation sample determines the error rate. Although this method
overcomes the bias problem by not using the same data to both build and judge the
classification, there are drawbacks to this method.

First, in many applications, large samples are not available especially when the data
are expensive to obtain. Second, if the size of the holdout sample is large, a good estimate
of the performance of the discriminant function will be obtained, but that function is
likely to be poor. On the other hand, if the size of the holdout sample is small, the
discriminant function will be better, but the estimate of its performance will be highly
variable (Lachenbruch and Mickey, 1968). Third, the function evaluated is not the
function of interest. Ultimately, almost all of the data must be used to construct the
classification function; if not, valuable information may be lost (Johnson and Wichem,
1992). Another way to reduce the bias is cross validation (Lachenbruch and Mickey,
1968; Lattin et al, 2003; Wiggins et al., 2003). Cross validation treats n-l out of 11
training observations as a training set. It determines the discriminant functions based on
n-l observations and then applies them to classify the one observation left out. This is

done for each of the training observations. The misclassification rate for each group is

the proportion of sample observations in that group that are misclassified (Lattin et al.,
2003; Wiggins et al., 2003).

The third method is the resubstitution method in which the sample used to compute
the discriminant function would be reused to estimate the error (Smith, 1947; Lattin et al.,
2003; Wiggins et al., 2003). When the test set is independent of the training sample, the
estimate is unbiased. However, it can have a large variance, especially if the test set is
small. The good classification (low error rates) will depend upon separation of the
populations. The farther apart the groups, the more likely a useful classification rule can
be developed (Johnson and Wichem, 1992).

Several studies addressed the appropriate statistical assumptions (Williams, 1983) and
the potential prediction bias (V erbyla, 1986; Hix, 1988), involved when applying
discriminant analysis in ecological studies. The prediction bias is likely to occur when
the number of independent variables in the model is large relative to the sample size, or
when many different sets of independent variables are tested by a stepwise procedure
(V erbyla, 1986), or when multivariate normality cannot be assumed (Hix, 1988). Biased
results could occur by using categorical and continuous data sets; mixed data sets have
been used in several ecological studies (Pregitzer and Barnes, 1984; Spies and Barnes

1985a; Hix, 1988).

Use of discriminant analysis in ecological studies
Discriminant analysis is used widely in ecological studies in which multiple
measurements are made on samples of observations possessing an identifiable group

structure. Its application would focus on the structure of plant or animal communities

indexed by geographically distinct habitat. In this case, the samples in each habitat
consist of the abundance of species and the objective is to highlight differences in
community structure (Matthews, 1979; Tonn and Magnuson, 1982). On the other hand,
discriminant analysis could be used to highlight habitat differences separating different
animal species. In this case, the samples of each species would consist of multiple
habitat measurements and the objective is to highlight differences in habitat use (Munro
and Rounds, 1985; Seagle, 1985). Matthew (1979) applied multiple discriminant analysis
to a number of plant assemblage-types in order to differentiate, as clearly as possible, and
give information about within and between type variability. Multiple discriminant
analysis is the classification that derives discriminant functions which maximize the
between group variance and minimize the within group variance on the discrimiannt
function scores. Norris and Barkham (1970) found multiple discriminant analysis to be a
useful technique in the analysis of differences between some beechwood ground floras in
the Cotswold Hills, England, but did not extend their study to include an analysis of
within wood variability. Grigal and Golstein (1971) gave graphical representation of
variability within and between four types of woodland in the oak hickory forest of eastern
Tennessee, USA, and concluded that the four types were distinct.

In Tom and Magnuson’s (1982), study discriminant analysis was used to discern two
assemblage types of fish in eighteen lakes. All the Umbra-cyprinid and centrarchid —Esox
assemblages groups were correctly classified in the study, and each had a distinctive
species composition and seasonal change in composition. Also, discriminant analysis
was applied to the log transformed environmental data on two of the eighteen lakes. The

purpose was to evaluate the environmental distinctness between the two groups of lakes

10

to help identify environmental factors contributing to their separation (Tom and
Magnuson, 1982). This aided the ecologist to properly manage the lakes and maintain
these assemblies.

When forestland classification is needed to develop a framework for managing forests,
an ecological multifactor approach integrating climate, physiography (physical
geography or landforms), soils, and vegetation was used to develop a classification of the
upland hardwood forest ecosystems of the Kickapoo River Watershed of southwestern
Wisconsin (Hix, 1988). After applying discriminant analysis on the dataset, it was found
that a combination of physiographic, soil, and vegetational variables resulted in the
highest overall percentage of correct classification about 97%. It is similar to results
obtained by Pregitzer and Barnes (1984) and Spies and Barnes (1985a). The overall
probabilities of correct classification were 81% and 91%, respectively.

Discriminant analysis has been used to classify trees as decayed versus sound (LeMay
et a1, 1994). Age, size, and the presence or absence of external indicators of possible
internal decay are variables used in deriving various rules for classifying trees. The
success of each developed classification rule was evaluated using the misclassification
error rates.

Discriminant analysis has been used to differentiate between human and animal
sources of fecal pollution in natural waters (Wiggins 1996; Wiggins et al., 1999;
Hagedom et al., 1999; Harwood et al., 2000) using a set of antimicrobial resistance
profiles for fecal streptococci in the USA, and using a set of antimicrobial resistance
profiles in E. coli in one study in Canada (Guan et al., 2002) and another study in USA

(Parveen et al., 1998), and to a set of antibiotic resistance profile in Enterococcus isolates

11

in Virginia Watershed, USA (Graves et al., 2002). Discriminant analysis has also been
applied on ribotyping profile for E. coli isolates to differentiate human and non-human
source of fecal pollution of water (Parveen et al., 1999; Carson et al., 2001; Troy et al.,
2003). Wiggins (1996) first addressed the application of discriminant analysis of
antimicrobial resistance patterns of streptococci to identify fecal pollution sources of
water. A total of 1,435 streptococcus isolates from 17 samples of cattle, poultry, human
and wildlife were used to create the antibiotic resistance profile of five antibiotics
(chlortetracycline, halofuginone, oxytetracycline, salinomycin, and streptomycin) using
four concentrations of each antibiotic. After analysis, 74% of known isolates were
correctly classified into one of six possible sources (beef, chicken, dairy, human, turkey
and wildlife). Ninety-two percent of human isolates were correctly classified. When
isolates were pooled into four categories, cattle, human, poultry and wildlife, the average
rate of correct classification (ARCC) increased to 84%. Human versus animal isolates
were correctly classified on average of 95%. Human and wildlife had an ARCC = 98%,
cattle versus poultry had ARCC= 92%. However, 72% of isolates from surface waters
that received fecal pollution from unknown sources from one site were classified as cattle
isolates, and 68% isolates from the other sites were classified as cattle also. The data
strongly suggest that discriminant analysis can be used to differentiate among isolates
from several sources of fecal pollution and to determine the sources of fecal pollution in
natural waters. The use of several concentration of each antibiotic to establish
antimicrobial resistance profiles coupled with discriminant analysis, provides the
predictive power necessary to provide useful information about the sources of isolates

from surface water. Another study performed by the same author (Wiggins et al., 1999),

12

used a larger sample size. Four sets of isolates of fecal streptococci (fi'om 2,635 to 5,990
isolates per set) were obtained from 236 samples of human sewage, cattle and poultry
feces and pristine waters. The patterns of antimicrobial resistance of the isolates to each
four concentrations of up to nine antibiotics were analyzed by discriminant analysis. The
ARCC in four possible groups (human, cattle, poultry and wildlife) ranged from 64% to
78%. It is confirmed that there are measurable and consistent differences in the
antimicrobial resistance patterns of fecal streptococci isolated from various sources of
fecal pollution, and it could be used to classify and identify these sources. Hagedom et a1.
(1999) validated the Wiggins’s methods by using larger database (7,058 fecal
streptococcus isolates) of known sources including human, livestock and wildlife from
wide geographical region and thirteen antibiotics, the five reported by Wiggins plus
amoxicillin, ampicillin, chloramphinicol, erythromycin, neomycin sulfate, rifampin,
tetracycline and vancomycin hydrochloride. The correct fecal streptococcus
identification averaged 87% for the entire database and ranged from 84% for deer to 93%
for human isolates. In order to test the database, and additional 892 streptococcus isolates
were compared against the database and resulted in ARCC of 88%. When all animals’
isolates were combined the ARCC increased to greater than or equal to 95%. Stream
samples from three highly contaminated collection sites, found that 78% of the fecal
streptococcus isolates from these sites were classified as cattle. These results were
consistent and support the Wiggins’s results. Based on these results, the cattle access to
the river was prohibited by installation of fencing and in— pasture watering stations. These

interventions lead to less than 45% of fecal streptococcus isolates being classified as

13

cattle. Additionally, the fecal coliforms counts were reduced from 15,900 per 100ml to
960 per 100ml.

The use of thirteen antibiotics was necessary in Hagedom’s (1999) study in order to find
those that did provide levels of separation that were as high as possible. Hagedom et a1.
(1999) supported his results by using cluster analysis with discriminant analysis to get
additional confidence when the two methods provide the same answers.

Another study by Harwood et a1. (2000) also supported the research done by Wiggins
(1996) and Hagedom et a1. (1999). In his study, Harwood et a1. (2000) used two
databases; one for 4619 fecal streptococcus isolates and one for 6144 fecal coliform
isolates, from a large geographical area in Florida. The antimicrobial resistance profiles
were established using nine antibiotics (ampicillin, amoxicillin, cephalothin,
chlortetracycline, oxytetracycline, streptomycin, tetracycline, erythromycin, and
vancomycin). Only eight antibiotics were used for fecal coliforms because gram-
negative bacteria are not susceptible to vancomycin. In the previous studies, Wiggins
(1996) and Hagedom et a1. (1999), the highest correct classification rates were obtained
by the use of a subset of antibiotics tested for analysis while in the Harwood et a1. (2000)
study, omission of any antibiotic resistance data resulted in lower correct classification
rates. The ARCC for fecal streptococcus database was 62.3% and that for fecal coliform
database was 63.9%. These results are lower than Wiggins’ original study results in
1996, because the database was composed of bacteria isolated from sources within a
limited geographical area, and the sample sizes were relatively small. However, the
second study by Wiggins (1999) was less homogeneous (more sampling sites) and

samples sizes were larger and resulted in correct classification rates comparable to

14

Harwood’s (2000) results. The lower correct classification rate for some sources reflected
the geographic diversity. Hagedom’s et a1. (1999) study resulted in higher classification
rates than Harwood’s et a1. (2000) study because the sources of isolates designated
human in the two studies probably contributed to the differences in correct classification
rates. In Harwood’s et a1. (2000) study, the human isolates were obtained from domestic
wastewater, which provides a cross-section of human antibiotic resistance profile and,
thus, high variability in antibiotic resistance. In Hagedom’s et a1. (1999) study, the
human isolates were obtained from experimental domestic wastewater treatment systems
from individual homes providing lower sample variability. The sample is not likely to be
representative of a large human population, which yielded higher correct classification
rates because antibiotic resistance variability is lower.

A study done in Canada (Guan et al., 2002) in which a collection of 3 19 E. coli
isolates from feces of cattle, poultry, swine, deer, goose, and moose as well as human
sewage and clinical samples were used. Fourteen antibiotics, ampicillin, cephalothin,
streptomycin, neomycin, kanamycin, gentamicin, tetracycline, chloramphinicol,
sulfathiazole, cotrimoxazole, apramycin, ceftiofur, spectinomycin, and tilrnycosin were
used in this study. By applying discriminant analysis on all E. coli isolates, the ARCC
was 33.9% when all isolates were classified into nine groups (human, beef, dairy,
chicken, pig, turkey, deer, goose, and moose). The ARCC for moose was 100% and 80%
for chicken, while goose, beef - dairy cattle and deer were poorly classified (0,0, and
14.8% respectively). It was suggested that most goose and beef-dairy isolates were
misclassified into moose categories because these groups displayed similar multiple

antibiotic resistance (MAR) profiles. When deer, goose, and moose isolates were pooled

15

together as the wildlife group, and swine, turkey, chicken, and bovine isolates were
pooled together as livestock group, the ARCC for wildlife and livestock were 95% and
46%, respectively, and 55% for human isolates. When all nonhuman sources were
pooled and discriminant analysis applied on human and nonhuman isolates, the ARCC
for human increased to 56.25% and to 92.38% in nonhuman. These results are different
from Wiggins (1996) and Hagedom et a1. (1999) studies due to several factors. First, E.
coli is the microorganism that was investigated in this study, not streptococcus that was
investigated in the previous studies. Second, different antibiotics were used in all the
studies, and third, the collection and isolation of E. coli is different in the three studies.
In Guan et al. (2002), study samples were collected from farms and parks over a wide
geographical area, and only one E. coli isolate was selected from each animal’s fecal
sample so that repetitious selection of the same clone of E. coli was avoided. The
sampling protocol could produce heterogeneous collection of bacterial isolates than other
protocols in which multiple bacterial isolates were derived from each fecal sample.

In the Guan et al. (2002) study, three methods were evaluated for their ability to
differentiate E. coli isolates from various sources. Multiple antibiotic resistance (MAR),
amplified fragment length polymorphism (AF LP) and 16S r RNA sequences. It was
found that AF LP provided the greatest discriminatory power, the highest rate of correct
classification and ease of standardization and automation, but it is expensive.

Another study done by Parveen et a1. (1998) in which discriminant analysis was used on a
MAR profile of E. coli isolates from the Apalachicola National Estuarine Research
Reserve (ANERR) found that 82% of human source isolates and 68% of non-human

isolates were classified with an ARCC of 75%.

16

Grave et a1. (2002), built a library of 1174 known source Enterococcus isolates, and
then applied the discriminant analysis on antibiotic profiles from these isolates. She
found that the ARCC was 94.6% for 203 human isolates, 93.7% for 734 livestock isolates
and 87.8% for 237 wildlife isolates. The antibiotic resistance analysis of enterococcal
isolates recovered from the stream samples indicated that isolates of human origin
appeared throughout the stream, but it was small compared with the proportion of isolates
from the livestock, which was 50%. The results presented in this study, affirm the use
antimicrobial resistance analysis and discriminant analysis in the previous studies
(Wiggins, 1996; Wiggins et al.1999; Hagedom et al., 1999; and Harwood et al., 2000).
Today, with the fast development in the genotypic methods in determining the source of
fecal contamination in water, the applicability of ribotyping to predict the source of E.
coli pollution has been tested using discriminant analysis. (Parveen et al., 1999; Carson
et al., 2001; Troy et al., 2003). These studies strongly indicated that discriminant
analysis of ribotypes profiles could be used to differentiate human sources and non-
human sources of fecal pollution. Discriminant analysis of ribotype profiles of 238 E.
coli isolates showed that 97% of the nonhuman isolates and 100% of animal fecal isolates
were correctly classified, and the ARCC for both was 82% (Parveen et al., 1999).

The advantage of discriminant analysis is that it generates a classification rule based on
all isolates; that rule can then be used to classify each individual isolates into one of many
possible sources. Once a discriminant analysis classification rule is developed by using
ribotyping profiles of E. coli isolates from known sources, discriminant analysis can then
be used to classify an unknown isolates to one of the lmown sources by using unknown

organism’s ribotypes profiles (Parveen et al., 1999). According to Parveen et a1. (1999)

17

the MAR has disadvantages that antibiotic resistance patterns of bacteria are influenced
by selective pressure, and thus may be different in other geographical areas and may vary
over time. But ribotype profiles are a genetic characteristic and are not as easily
influenced by the selective pressure. The discriminant analysis of ribotype profiles
provides a strong method for differentiating human source and non-human source fecal
pollution, and may enhance efforts to improve the natural quality of estuarine ecosystems
and to assess the importance of upstream activities, local storm water runoff and marine
activities (Parveen et al., 1999).

In Carson’s et a1. (2001) study, a total of 287 known host ribotype patterns were
generated for E. coli strains isolated fiom human, cattle, pigs, horses, chickens, turkeys,
geese and dogs. Discriminant analysis of riboprint of human and nonhuman isolates
resulted in 95.0 and 99.2% correct classification, respectively. The ARCC for riboprints
compared to all eight hosts classes was 73.6% and when the comparison was made
between more limited numbers of classes, the ARCC improved.

Another study by Troy et a1. (2003), using discriminant analysis on ribotype of E. coli
isolates supported the idea that it may be possible to differentiate human from animal
derived E. coli over a broad geographic region via the single-enzyme (HindIII) ribotype
procedure. However, it is more difficult to differentiate between E. coli isolated from
multiple non-human sources by using the same method- ribotype profiles. It showed that
over all, the correct classification of animal derived E. coli isolates, as being either
human or animal - derived was greater than 78%, while human - derived isolates were
correctly classified greater than 85% of the time. Troy et a1. (2003) also found that an

overlap of ribotype profiles within and among animal groups was significant; the reasons

18

for the overlapping are unlmown and the overlapping make it hard to differentiate sources
of E. coli. It was thought that fecal material from birds could be present in the samples
and cause the overlapping. Troy et a1. (2003) concluded that the ribotyping procedure
continue to be a viable molecular tool to be used for determining whether the source of E.
coli is from human or animal sources. This conclusion supports the previous work
reported by Parveen et al., 1999; and Carson et al., 2001.

Troy’s et a1. (2003) study is different from Carson’s et a1. (2001) study because of the
diversity of sample collection and the type of the sample collected. In Carson’s study,
the samples collected were from central Missouri and flour a small number of animals
while in the Troy’s study, composite fecal samples were collected from poultry, swine,
dairy and beef cattle, over a wide geographic region. This type of sampling protocol was
chosen because it is more likely that these samples would contain isolates that have the
most environmental impact, subjected to external stressor, more likely to survive and
more representative of those one would expect to find in the environment (Troy et al.,
2003).

Although there is not an established standard of accuracy defined for any bacterial
source tracking method, any method with a correct rate of classification of over 50%
when there are five or more possible source categories has been considered as a
worthwhile tool for predicting the potential sources of fecal pollution in enviromnental
water (Harwood et al., 2000).

Discriminant analysis is used in human medicine to differentiate basal cell carcinoma
and other skin neoplasms from normal skin by using infrared spectroscopy as a screening

tool for cutaneous neoplasia with correct classification 93.5%. Linear discriminant

19

analysis characterized spectra, as arising from basal cell carcinoma, epidermis, or follicle
sheath was 98.7% accurate. In addition, linear discriminant analysis accurately classified
spectra as arising from epidermis overlying basal cell carcinoma versus epidermis
overlying nontumor-bearing skin in 98.0% of cases (McIntosh, 1999). In this study, non-
subjective classification of normalized spectra was performed using linear discriminant
analysis (LDA). The genetic algorithm was implemented to select the regions of spectra
that contained the most useful information. These spectral regions were then subjected to
LDA, which determines the boundaries that best separate classes by computing the
distance of each spectrum from all class centroids. It then allocates each spectrum to the
class whose centroid is nearest. Data were split into a training set, which was used to
train the LDA algorithm to find the discriminatory patterns in the data, and a test set to
assess the accuracy of the trained algorithm (McIntosh, 1999).

The applicability of discriminant analysis to the interpretation of skin images was
tested by Josef (2000) where the classification of tissue elements into 5 classes
(background, epidermis, papillary dermis, reticular dermis and dermal infiltration)
yielded a correct classification in 98.4% of all elements. Cluster analysis was used also,
and when reclassification of cluster analysis was done by discriminant analysis it yielded
100% correctly classified elements. Josef (2000) concluded that discriminant analysis
might be a helpful technique for an independent, and subjectively unbiased measurement

system of skin structures in digital images.

20

Assumptions

Ecologists almost always use linear discriminant analysis; therefore, it is important to
recognize the assumptions that are critical for its use.

Sample size: Unequal sample sizes are acceptable. The sample size of the smallest group
needs to exceed the number of predictor variables. As a rule, the smallest sample size
Should be at least 20 for a few (4—5) predictors. The total sample size should be at least
two or preferably three times as large as the number of variables measured (Tatsuoka,
1970). The maximum number of independent variables is n-2 where n is the sample size.
While this low sample size may work, it is not encouraged, and generally it is best to
have 4 or 5 times as many observations and independent variables (Johnson and Wichem,
1992)

The sample means and covariance must be estimated from the data. When the number
of parameters to be estimated approaches the number of samples, there is a good
likelihood that any patterns exhibited by individual coefficients are fortuitous, and
therefore, of no ecological consequence. Therefore, data splitting, balanced repeated
replication should always be used when sample size is small relative to dimensionality
(Williams, 1983).

Normal distribution: It is assumed that the data (for the variables) represent a sample
from a multivariate normal distribution (Johnson and Wichem, 1992; Lattin et al., 2003).
One can examine whether or not variables are normally distributed with histograms of
fiequency distributions. However, violations of the normality assumption are not fatal,
and the resultant significance tests are still reliable as long as non-normality is caused by

skewness and not outliers (Tabachnick and Fidell, 1996). Klecka (1980) points out that

21

dichotomous variables, which often violate multivariate normality, are not likely to affect
conclusions based on discriminant analysis.

Priori: Prior probabilities could influence the forms of discriminant functions so that
incorrect specification of priors can distort or obscure any underlying structure of the
data. In ecological studies, it is impractical or even impossible to sample in such a way
that reasonable estimates of priors can be obtained. Instances involving rare species
demonstrate such a lack and the need to use historical data to supplement samples, which
also entails high costs associated with obtaining random samples. The solution is
collecting systematic or stratified samples and priors are guessed, determined from
ancillary information, or assigned arbitrarily. When priors are replaced by relative
sample sizes that bear no direct relationship to them, an uncontrolled and largely
inscrutable amount of arbitrariness is introduced into the discriminant analysis, so that
initial decision of sample size can affect the resulting mathematical forms irrespective of
underlying statistical properties (Williams, 1983).

Homogeneity of variances/ covariances: Discriminant analysis is very sensitive to
heterogeneity of variance —covariance matrices (Lattin et al., 2003). It is assumed that
the variance —covariance matrices of variables are homogeneous across groups, but
researchers should perform some tests on the covariances to better understand their data.
This includes tests for homogeneity of within class covariance matrices before accepting
final conclusions for a study. In particular, a scatter plot matrix can be produced and can
be very usefirl for this purpose. Lachenbruch (1975) suggested that discriminant analysis

is relatively robust even when there is modest violation of this assumption.

22

Equal dispersion is a very important assumption since several properties results from it

 

including linearity of discriminant functions, improved efficiency of estimation, and
invariance of posterior probabilities in canonical space. This is important for
interpretation of canonical plots, as it enable one to display data in canonical space
without distortion. Most ecologists base their interpretation on canonical function, but
their data exhibit heterogeneous dispersions. Heterogeneity of dispersion is manifested by
the nonuniform patterns of dispersion in canonical space (Williams, 1983).

Outliers: Discriminant analysis is very sensitive to the inclusion of outliers. A test for
outliers should be performed for each group, and outliers should be eliminated. If one
group in the study contains extreme outliers that impact the mean, they will also increase
variability. Overall significance tests are based on pooled variances, that is, the average
variance across all groups. The significance tests of relatively larger means (with larger
variances) would be based on the relatively smaller pooled variances, resulting
erroneously in statistical significance.

Non-Multicollinearity: If one independent variable is very highly correlated with
another, or one is a function (e.g., the sum) of other independents, then the tolerance
value for that variable will approach 0, and the matrix will not have a unique discriminant
solution. Such a matrix is said to be ill-conditioned, especially if any one of the variables
is completely redundant with the other variables. There must also be low
multicollinearity of the independents to the extent that independents are correlated; the
standardized discriminant function coefficients will not reliably assess the relative

importance of the predictor variables. The tolerance value is computed as 1 minus R-

23

square of the respective variable with all other variables included in the current model.

 

Thus, it is the proportion of variance that is unique to the respective variable.

When several assumptions are violated, statistical and interpretive problems are
compounded. Discriminant analysis in this case, like any other mathematical technique,
becomes a data-exploratory procedure. It may provide fruitful insights into the data, but
without it, additional testing and evaluations are needed (Williams, 1983). However,
researchers should know the difference between exploratory and confirmatory analysis.

Statistical procedures can be used to explore data whether assumptions are met or not.

Interpretation of discriminant analysis

Ecologist are generally interested in parameters that separate populations, on the
assumption that the operation of natural selection will be reflected in among-group
differences of these parameters. A stepwise procedure is used to select variables that are
highly separating groups, and then canonical transformations of these variables are
determined. The canonical transformations are interpreted through the signs and
magnitudes of the associated coefficients (Williams, 1983; 1988). In other words, the
importance of a variable in a discriminant analysis is related to the magnitude and sign of
its function coefficients (Hix, 1988), and by means of their correlations with the original
variables (Williams, 1983). However, these coefficients do not tell us between which of
the groups the respective functions discriminate. We can identify the nature of
discrimination for each discriminant (canonical) function by looking at the means for the
functions across the groups. Also, we can visualize how the two or more functions

discriminate between groups by plotting the individual scores for the discriminant

24

fimctions (Williams, 1983). One should distinguish between discriminant analysis and

logistic regression, cluster analysis, and Multiple Analysis Of Variance MANOVA.

Logistic regression and discriminant analysis

Discriminant analysis is a close kin to logistic regression. Although logistic
regression answers the same questions as discriminant analysis, computing the functions
is quite different. Logistic regression is often a preferred as an alternative procedure to
discriminant analysis as it is more flexible in its assumptions, and the types of data to
which it can be applied. It can handle both categorical and continuous variables, and the
predictors do not have to be normally distributed, linearly related or of equal variances
within each group (Tabachnick and Fidell, 1996). In their study, Press and Wilson
(1978) compared logistic regression and parametric discriminant analysis, and concluded
that logistic regression is preferable to parametric discriminant analysis in cases for
which the variables do not have multivariate normal distributions within classes.
The interpretation of the results of a two-group problem is straightforward and closely
follows the logic of multiple regression: Those variables with the largest (standardized)
regression coefficients are the ones that contribute most to the prediction of group

membership.

Cluster analysis and discriminant analysis

It is important not to confuse discriminant analysis with cluster analysis. All varieties
of discrimiannt analysis require prior knowledge of the classes, usually in the form of a
sample fiom each class. In cluster analysis, the data do not include information on class

membership, and the purpose is to construct a classification. In other words, in

25

discriminant analysis the groups are determined beforehand, and the objective is to
determine the linear combination of independent variables which best discriminates
among the groups. In cluster analysis, the groups are not predetermined and in fact, the
purpose is to determine the best way in which cases may be clustered into groups

(Johnson and Wichem, 1992).

MANOVA and discriminant analysis

The close association between separation and multivariate analysis of variance has led
to confusion about the nature of discriminant analysis. At first glance, discriminant
analysis appears to do what MANOVA does, since both deals with differences among
group of observations. Discriminant analysis addresses a statistical mixture of population
and MAN OVA does not. In fact, it is multivariate analysis of variance reversed, and
computationally very similar to MAN OVA. In MANOVA, the independent variables are
the groups and the dependent variables are predictors. In discriminant analysis, the
independent variables are the predictors (discriminating variables) and the dependent
variables are the groups. The focus is on the groups as the outcome, and the main purpose
is to develop a firnction that will maximize the distance among the groups (Morrison,
1990). MANOVA can be used to see the effect on multiple dependents of a single
categorical independent, while discriminant analysis can be used to see the effect on a
categorical dependent of multiple interval independents
In this case, we have a matrix of total variances and covariances; likewise, we have a
matrix of pooled within-group variances and covariances. Those can be compared via
multivariate F tests in order to determine whether or not there are any significant

differences with regard to all groups and between groups. It is the same in MANOVA.

26

One could first perform the multivariate test and if it is statistically significant, proceed to
see which of the variables have significantly different means across the groups.

Another alternative to discriminant analysis is to perform a series of univariate, one - way
Analysis of Variance ANOVAs. However, the advantage of the multivariate approach is
that two or more classes that overlap considerably when each variable is viewed
separately may be more distinct when examined from a multivariate point of view.
According to Williams (1983), discriminant analysis has frequently been used improperly
in ecology. The reasons are that multivariates procedures are complex, and require

complex statistical analysis, and the documentation of these are highly technical.

Conclusion

Discriminant analysis is a multivariate statistical test used to perform three operations:
a) classify cases into groups; b) determine which variable discriminates between groups
by creating a classification rule based on the measurement of that variable; and c) use the
discriminating variable to predict the group membership for an unknown sample.
Discriminant analysis has two approaches, which are descriptive or predictive. The
objective of discriminant analysis is to exhibit the maximum separation between groups
based on linear transformation of the measurement variables (descriptive), and predicts
the group to which an observation belongs based on its measurement value (predictive)
respectively.

A reliable classification procedure should results few misclassifications and less error
rate, so that the performance of discriminant analysis can be evaluated by estimating the

error rate by either the hold-out, cross validation or resubstitution methods.

27

Discriminant analysis is usually applied to a case of two populations so that effective
allocation of an observation is probably not possible unless the populations are well
separated. The same is true for many populations, which means the populations’ mean
vectors differ significantly from one another. Although apparent and significant
differences do not automatically imply effective classification, testing is a necessary first
step. If no significant differences are found, constructing classification rules will be a
waste of time. Discrimination is often attempted with large number of variables, some of
which are qualitative. In this situation, the multivariate normality may not be a sensible
assumption, in which case the results of discriminant analysis may be affected. Therefore
testing for normality is a very important issue. The distribution should be multivariate
normal with equal covariance matrices. Discriminant analysis is very sensitive to sample
size, and the sample size of the smallest group needs to exceed the number of predictor
variables.

Usually, researchers deal with more than one population and many variables. This
results in a very complicated situation requiring computer software programs to perform
the calculations and have the capability for stepwise discriminant analysis. Selection of a
subset of variables on the basis of minimizing the apparent error rate or maximizing
discriminatory power may perform poorly in future samples, especially if there are large
correlations among variables.

According to the articles reviewed in this chapter, discriminant analysis has been
applied to a wide range of ecological problems in which multiple measurements are made
on samples of observations possessing an identifiable group structure. More research has

been conducted in the forestry industry where it focused on the structure of the plant, the

28

animal communities indexed by geographically distinct habitat, and classify trees into
decayed versus sound where forestland classification is needed in order to apply the
proper management plans and maintain the forest structure.

Discriminant analysis is a successful method to identify the sources of fecal pollution
in water. It has been applied to the antimicrobial resistance patterns of fecal
Streptococcus and E. coli bacteria that are considered indicator bacteria. The advantage
of discriminant analysis method is that it generates a classification rule based on all
bacterial isolates, then employs this rule to classify each isolate into one of many possible
sources. According to Wiggins (1996), who first introduce this method, the high
classification rates could be achieved when two sources are compared, supporting the
hypothesis that different animal species will harbor different bacteria with different
patterns of antibiotic resistance. Sometimes, one species of animal was misclassified into
other animal species categories. For example chicken isolates were misclassified as
turkey isolates indicating that pooling of the sources into poultry category would be
helpful for the classification process. Usually, there is higher ARCCs when discriminant
analysis is performed with pooled sources (poultry, cattle, wild animal and human) than
all sources (chicken, turkey, beef, dairy, wild animal and human), or human and animal
sources as two categories.

The use of discriminant analysis in research is simple, cost effective, and provides a
strong method for differentiating human source and non-human source of fecal pollution
of surface water. It will be suitable for surveillance purposes and routine monitoring that
enhance the efforts to improve the quality of water, and it promises to accurately evaluate

the risk to the public health.

29

 

CHAPTER 2

PATTERNS OF ANTIMICROBIAL RESISTANCE OBSERVED IN E. COLI
ISOLATES OBTAINED FROM DOMESTIC AND WILD ANIMAL FECAL
SAMPLES, HUMAN SEPTAGE, AND SURFACE WATER IN MICHIGAN, USA

Abstract

A repeated cross-sectional study was conducted to determine the patterns of
antimicrobial resistance in E. coli isolated from fecal samples collected from
domesticated animals, companion animals, farmed deer and wild geese. E. coli also
isolated from human septic tanks, environmental samples on farms and surface water
samples in Michigan, USA. Thirty-one farms, and three septic companies in the Red
Cedar Watershed participated in the study. Isolation and identification of E. coli were
conducted done using enrichment media, selective media, and biochemical tests.
Antimicrobial susceptibility testing was conducted using the disc diffusion method.
Antimicrobial agents tested included neomycin, gentamicin, streptomycin,
chloramphenicol, ofloxacin, trimethoprim/sulfamethoxazole, tetracycline, ampicillin,
nalidixic acid, nitrofurantoin, cephalothin, and sulfisoxazole. Resistance to at least one
antimicrobial was demonstrated in isolates from food animals, wild geese, farmed deer,
surface water, and human septic tanks. In general, E. coli isolates from food animal
species showed the highest resistance to most antimicrobial agents, followed by those
from horses, companion animals, wild geese, farmed deer, human septic tanks, and
surface water. The top three-antimicorbial agents for which resistance was demonstrated
were cephalothin, tetracycline, and streptomycin. At least one E. coli isolate from food

animal species, wild gees, farmed deer, human septic tanks and water samples were

30

resistant to cephalothin. Additionally, the antimicrobial resistance profiles in E. coli
isolated from companion animals and human septic tanks were closer to that from surface
water than those from domesticated animals, suggesting that companion animals and
human septic tanks may play a major role in contamination of surface water in the Red

Cedar Watershed than previously thought.

Introduction

Antimicrobial resistance has been recognized as an emerging worldwide problem, and
a serious problem in both human and veterinary medicine (N eu, 1992; Witte, 1998).
Antimicrobial use is considered the most important factor for emergence, selection and
dissemination of antimicrobial resistant bacteria in both human and veterinary medicine
(N eu, 1992; Witte, 1998). Antimicrobial agents are used in animals and humans for
therapy and control of bacterial infections. In food animals, it may be incorporated in
feed and fed continuously to the whole population rather than to an individual animal as
growth promoters, which may lead to selection of resistant bacteria. (Van Den Bogaard et
aL,2001)

Multiple antimicrobial resistance testing is based on detection of bacterial resistance to
a panel of antimicrobial agents. The principle behind developing resistance is that
bacterial flora in the gut of humans and animals are subjected to different types,
concentrations, and frequencies of antimicrobial agents. Over time, selective pressure
will select resistant bacteria that have specific fingerprints against the antibiotics that
have been used (Prescott et al., 2000; Troy et al., 2002). This test is simple, cost effective

and suitable for surveillance (Troy et al., 2002).

31

With the exception of the large number of studies that have dealt with the
antimicrobial resistance of E. coli isolated from food, various species of animals and
humans (Krumperman, 1983; Langlois et al., 1983; Ginns et al., 1996; Blanco et al.,
1997; Meng et al., 1998; Galland et al., 2001; Van Den Boggard et al., 2001; Schroeder et
al., 2002a; Schroeder et al., 2002b), little is known about the antimicrobial resistance
patterns of E. coli isolated fiom surface water.

Multiple antimicrobial resistance (MAR) profiling has been used to identify the
sources of fecal contamination in water (Kasper et al., 1990; Wiggins, 1996; Parveen et
al., 1997; Hagedom et al., 1999; Wiggins et al., 1999; Harwood et al., 2000; Graves et al.,
2002; Guan et al., 2002). The MAR profile may be a useful tool, from a public health
and environmental protection standpoint, in establishing standards to determine water
quality and facilitate detection of sources of water contamination. Efficient use of
resources for water quality improvement needs to be based on accurate identification of
the sources of fecal pollution. Hagedom et al., (1999) reported that antibiotic resistance
analysis proved to be accurate in identifying the source of fecal pollution in the Page
Brook Watershed in Virginia. He found that livestock contributed more than humans to
the contamination of the stream. Fencing part of the stream to reduce livestock’s access
to it reduced the fecal coliform population by an average of 94%.

The MAR profile is used to distinguish between E. coli that comes from point sources,
such as industrial, municipal effluents, and meat-processing plants wastes; as well as non-
point sources as in cases of soil erosion and runoff over a wide area of land. A study by
Parveen et al., (1997) found that more than 80% of E. coli strains isolated from municipal

waste, and river and estuarine water show antibiotic resistance. Kaspar et al., (1990)

32

showed that urban waters have higher percentage of resistant E. coli strains than rural
waters, and antibiotic resistance E. coli may offer an index of water quality related to the
source.

Multiple antibiotic resistance indices have been used to identify and differentiate E.
coli of high-risk sources (human, poultry and swine) of fecal contamination of food from
those of low-risk sources (wild animals) (Krumperman, 1983). Krumperrnan (1983)
showed that multiple antibiotic resistance index of E. coli from wild animals (low risk)
was generally low while human and poultry E. coli isolates had higher MAR indices
(high risk) suggesting that multiple antibiotic resistance E. coli exist in large numbers
within the major reservoirs of enteric diseases for humans while present in low number
elsewhere. It is interesting that MAR index for isolates from human raw sewage was
greater than the index found among isolates recovered by direct anal swabbing and this
could be due to plasmid exchange that occurs in sewage systems (Krumperman, 1983).

Bacteria could gain resistance to antimicrobial agents by many ways including 1)
acquisition of antibiotic resistance genes through mobile elements such as plasmids and
insertion sequences (Rubens et al., 1979), which tend to code for enzymes that
metabolize antibiotics (Prescott et a1, 2000); 2) mutations in genes responsible for
antibiotic uptake or binding sites (Spratt, 1994) and it often produces changes in bacterial
cell, and 3) activation of MAR locus (mar) in the bacterial chromosome (Hachler et al.,
1991; Alekshun and levy, 1999).

There are four mechanisms of resistance 1) enzymatic inactivation, or modification of

antibiotics; 2) impermeability of the bacteria cell wall or membrane; 3) active expulsion

33

of the drug by the cell efflux pump, and 4) alteration in target receptors (Prescott et al.,
2000).

The ability of resistance factor containing R+ bacteria, especially E. coli, to transfer
drug resistance is well known (Richmond, 1972), with a large number of studies that
discussed the antibiotic resistance patterns of E. coli isolated from fecal samples.
Information describing antibiotic resistance patterns of E. coli (or other bacteria) isolated
from the surface water in the environment is limited. Kelch and Lee (1978) suggested
that resistant fecal coliform bacteria survive better than sensitive bacteria in surface water
because R factor could increase the survival ability of coliform bacteria, supporting
Grabow’s findings (Grabow et al., 1973; Grabow et al., 1976). On the other hand,
Anderson (1974) suggested that R factor-mediated antibiotic resistance might reduce
survival ability of E. coli. Smith et al., (1974), however, suggested that R factor —
mediated antibiotic resistance had no effect on E. coli survival.

It is widely speculated that the use of antimicrobial agents in food animals may be
contributing to antimicrobial resistance in humans (Barton, 1998; Witte, 1998).
Antibiotics fed at low levels as part of the feed have been used as growth promoters in
domestic animals. Using antibiotics in this manner, will produce a substantial difference
in the gut microflora of these animals, increasing the incidence of resistant bacteria
especially when these bacteria are resistant to antibiotic used clinically for treating
diseases in humans (Van den Boggard and stobberingh, 1999; Casewell et al., 2003). One
strategy recommended in order to minimize this problem was to stop using antibiotics
needed for human treatment as food additives (Swarm Committee Report, 1969;

Richmond, 1972; World Health Organization, 1997). Nevertheless, there is ongoing

’34

debate whether and to what extent antibiotic use as feed additives contribute to resistance
development in human bacterial pathogens (Prescott et al., 2000; Casewell et al., 2003).
Antimicrobial resistant bacteria have been isolated from a variety of sources such as
hospitals, domestic sewage, drinking water, rivers and lakes (Mulamattathil et al., 2000;
McKeon et al., 1995; Kasper et al., 1990). The level of antibiotic resistance reported by
Mulamattathil et al., (2000) was 72% lower than that reported by McKeon et al., (1995)

which was 100 and 87% for fecal and non-fecal coliforms respectively.

Objectives

The two major objectives of this study were to 1) identify patterns of antimicrobial
resistance in E. coli obtained from human septage, domestic animals and wildlife living
in the Red Cedar Watershed in Michigan, USA; and 2) compare such patterns with that

fiom E. coli obtained from surface water of the same watershed.

Hypotheses tested
0 Prevalence of antimicrobial resistant strains of fecal E. coli isolates should be
significantly lower in isolates obtained from wildlife fecal samples than those
from domestic animals and humans.
0 Prevalence of antimicrobial resistant strains of fecal E. coli isolates should be
significantly lower in isolates obtained from domestic pets and human septage

than those from domestic livestock and poultry.

35

Materials and Methods
Study design

A repeated cross- sectional study design approach was used to collect animal and
human fecal, and water samples and all the data related to antibiotic use in the farm
during the four seasons (Spring, Summer, Fall, and Winter), starting in the winter of 2002
and ending in the winter of 2003.
Study area

The Red Cedar Watershed was chosen as the study area, encompassing an area of
293,000 acres and takes in portions of 16 townships in Ingham and Livingston counties
of the lower part of Michigan (figure 2-1). (Please notice that images in this thesis are
presented in color). The Red Cedar River arises in Cedar Lake in the south central portion
of the lower peninsula of Michigan and flows about 45 miles to its confluence with the
Grand River in the city of Lansing. The Grand River empties into lake Michigan, which
connects to the other great lakes flowing into the Atlantic Ocean. The Red Cedar River
has 12 tributaries and drains, and provides mid-Michigan residents with numerous
recreational opportunities. The river also serves as a source of water for the irrigation of
crops throughout the watershed. Swine and dairy are the predominant livestock in this
watershed (figure 22).

The study was conducted into two parts. Part one consisted of determining
antimicrobial resistance profiles of E. coli isolated from fecal samples of animals and
humans. The second part consisted of determining antimicrobial resistance in E. coli

isolated from surface water.

36

Figure 2-1: Michigan State map showing the location of Red Cedar Watershed

a ‘1‘)”

 

Figure 2-2: The Red Cedar Watershed, Michigan, USA (green area) map showing
the farms location (yellow dots) and the water sampling sites (blue dots)

 

*Arrows in figure 2-1 and figure 2-2 indicate north

37

Part I: Antimicrobial resistance profiling of E. coli from fecal specimens
Study population:
Enrollment of participating farmers: Farmers in the Red Cedar Watershed were sent a
letter through Ingham and Livingston Counties extension agents, asking whether they
would like to participate in the study, and asking them to indicate their interest in
participating by returning a pro-stamped post card to the Population Medicine Center at
Michigan State University.

A total of 31 farmers were contacted and farm visits were arranged through the winter,
spring, summer, and fall starting in the winter of 2002 until the winter of 2003.
Sample size: In order to detect at least one animal with E. coli on each farm, the general
formula used by (Smith, 1995) was used as shown below.

ninf = [loga]/ [log (l-Prevalence of E. coli)]
Where ninf = Sample size for a very large population.
or = Probability of Type I error (0.05).

We assumed the prevalence of E. coli to be 10%. Using the above equation and
assumptions stated; it was calculated that 29 animals per species would be the minimum

to be tested.

Data collection
Data relating to antimicrobial use and number of animal species on the farm were

collected during the time of collection of fecal samples. The data relating to

38

antimicrobial use were collected via a question asking about any antimicrobial agents
used for therapy, prevention or grth promotion during the previous 60 days.

Sample collection:

Animal fecal samples: Fecal samples were obtained from dairy and beef cattle, swine,
horses, sheep, goats, chicken, cats, dogs, farmed deer, ducks and wild geese. Fecal
samples from livestock (dairy cattle, beef cattle, swine, sheep and goats) were collected
directly from the rectum using the culturette swab system. However, due to difficulty in
capturing feedlot cattle on some farms, the fecal samples were collected from the fresh
drops using the culturette swab system. Fecal samples from horses, dogs and cats were
collected by rectal swabbing, using the culturette swab system. Chicken fecal samples
were collected by cloacal swab. Farmed deer and some wild geese fecal samples were
collected from the fresh drops on the ground. Also, some samples were collected directly
from the wild birds (ducks and some Canadian geese) during the geese-banding season
conducted by Department of Natural Resources (DNR) in the study area. All‘deer fecal
samples were collected from farmed deer, and not a free-range deer. Samples from the
manure storage facilities (i.e. lagoon, slurry pit, and manure pile) on the farms were
collected using the culturette swab system. Every swab was labeled with the farm
number and given a serial number from the farm it was collected and shipped overnight
to the lab at the University of Maryland for isolation and identification of E. coli. A total
number of 2,292 fecal samples were collected.

Human septage samples: Samples representative of human fecal material were collected
with the help of the local septic pumping companies in the study area. These companies

were asked to provide samples of human septage material pumped from the septic tanks

39

from homes in the Red Cedar River. It was determined that septic samples would
provide material most likely to affect water quality via leakage fi'om septic tanks or
improper disposal of pumped septic contents in the study area. A total of 34 human fecal

samples were collected from the study area.

Procedure for shipment of fecal samples to the laboratory

It was important to make sure that all culturette tubes were labeled properly and
legibly. Each individual culturette tube was sealed using clear packaging tape. Every 10
to 15 culturette tubes were bundle together, wrapped with absorbent cotton and placed in
a screw-top cardboard tube mailer. The cardboard tube was taped using clear packaging
tape. The Maryland laboratory address was placed on the screw-top cardboard tube
mailer and covered with clear tape. The cardboard tube mailer was placed in a small
cardboard box with Styrofoam packing peanuts to cushion it. A copy of the sample
information was placed in a Ziploc bag on top of the peanuts in the box. The box was
taped and another address label was placed on the outside of the cardboard box and

shipped overnight by FedEx.

Isolation and identification of E. coli from fecal samples

Standard methods were used for the enrichment, isolation, identification, and
biochemical confirmation of E. coli isolates (Ameican Public Health Association, 1998).
The culturettes containing the sample were placed in tubes of Tryptic Soy Broth (TSB)
and incubated at 35°C for 24 hours. This is an enrichment step to stimulate bacterial

growth. A loop (1011.13) of the turbid broth was transferred to a plate of Violet Red Bile

40

(V RB) media and streaked across the plates. The plates were incubated for 18 to 20
hours at 35°C. The VRB plates were examined for reddish purple colonies that fluoresce
under a black light. Five or six colonies were then selected from each plate and streaked
on a plate containing MacConkey’s media. This plate was incubated at 35°C for 18 to 20
hours. The MacConkey plate was examined for pink colonies that precipitate bile and
have a dark red center. One or two colonies were selected and streaked on a plate
containing Tryptic Soy Agar (TSA). The purpose of this plate was to make sure that there
were no transparent colonies being transferred from the MacConkey plate. This plate was
streaked and incubated for 18 hours. The TSA plate was then examined for single
colonies that are round, milk-colored, and slightly convex. One single colony was
selected and placed in a tube containing TSB and incubated for approximately 3—4 hours
until turbid (Difco, 1998). The broth was transferred into tubes for the biochemical
confirmation: indole, methyl red, Voges Proskauer, and Simmons Citrate (IMVIC) test
(American Public Health Association, 1998). Also, a test for sugar metabolism and gas
production was performed using a tube of Triple Sugar Iron (TSI) (Difco, 1998). The
results of the biochemical testing determine if the colony in the TSB tube is E. coli. The
broth must pass all the biochemical tests to be used for the antimicrobial susceptibility
testing. Once the colony of E. coli was confirmed then another tube of TSB was
inoculated with the broth and incubated until it reached a turbidity of 0.5 MacFarland

Standard (usually about 2-3 hours). That broth was used for the Kirby-Bauer procedure.

41

 

Part II: Antimicrobial resistance profiling of E. coli from water samples.

Sample collection: Water-sampling sites were determined with the help of the Ingham
county drain commissioner based on the direction of the rain flow from every single farm
enrolled in the study in order to get an accurate reading results (figure 2-2). The sampler
(figure 2-3) that holds a 100ml sterile plastic bottle was used to collect surface water
samples. These bottles contain 10mg sodium thiosulfate to neutralize any residual
chlorine in water. The water bottles were labeled with the sampling site, date, time of
collection and placed in a cooler and sent to the lab within 6 hours of collection for

isolation of E. coli. A total of 37 water samples were collected during the study period.

Figure 2-3: The sampler used to collect water samples

 

Isolation and identification of E. coli from water samples
The Membrane filtration method was used to isolate E. coli from water samples. The
original mTEC agar enumeration method (Dufour et al., 1981) for E. coli was used to

provide a direct count of E. coli in water based on the development of colonies that grow

42

on the surface of a membrane filter. In this procedure, water samples were filtered
through the membrane, sterile, white, grid marked 47-min diameter, with 0.45i0.02um
pore size, which retains the bacteria. After filtration, the membrane containing the
bacteria was placed on a selective and differential medium, in this case the mTEC
(Dufour et al., 1981; United State Environmental Protection Agency, 1985), and
incubated at 352t0.5°C for 2 hrs to resuscitate the injured or stressed bacteria, and then
incubated again at 44.5:t0.2°C for 22 hrs. The filter was transferred from mTEC agar to a
filter pad saturated with Urea Substrate Medium. After 15 to 20 minutes, yellow, yellow-
green, or yellow-brown colonies on mTEC were counted with the aid of a fluorescent
lamp and a glass lens (2—5x magnification) or stereoscopic microscope and used for
antimicrobial susceptibility testing. Non-E. coli colonies turned pink or purple on the
Urea Substrate Medium. The number of E. coli per 100 mL was calculated according to
the following general formula:

100 (number of E. coli colonies counted)
E. coli/ 100 mL =

 

(Volume of sample filtered, in mL)

The results were reported as E. coli / 100ml colony forming units (CFUs).

Antimicrobial susceptibility testing
Antimicrobial agents used for testing E. coli isolates

The following twelve antimicrobial agents were included in the tests: neomycin,
gentamicin, Streptomycin, chloramphenicol, ofloxacin, trimethoprim/sulfamethoxazole,
tetracycline, ampicillin, nalidixic acid, nitrofurantoin, cephalothin, and sulfisoxazole.

These antimicrobials were chosen on the basis of their importance in treating human or

43

animal E. coli infections, or used as feed additive to promote growth in animals, and to
provide diverse representation of antimicrobial classes. Additionally, two or more
antimicrobial agents from the same class were used to see if there is any cross-resistance
between them. We were also interested in seeing whether there would be any resistance
to the banned antimicrobial agents such as Chloramphenicol, which are restricted from
use in food animals (Krumperman, 1983)

Kirby-Bauer Method

The standard Kirby-Bauer disc diffusion method was used to develop the
antimicrobial sensitivity profile of E. coli isoltes (National Committee for Clinical
Laboratory Standards (N CCLS), 1997, 1999) for 12 antimicrobial agants (Table 2-1).

A 150 mm plate containing Mueller-Hinton media was swabbed with tryptic soy broth
that reached a turbidity of 0.5 MacFarland Standard and allowed to dry for 1 minute. The
Mueller-Hinton agar was formulated to have a pH between 7.2 and 7.4. The 12
commercially prepared, and standardized antibiotics discs, were dispensed onto the
inoculated plate and tapped lightly to adhere to the media surface. The plate was
incubated at 35°C for 18 to 20 hours. The diameter of the clear zones of growth inhibition
around the antibiotic discs including the 6-mm disc diameter was measured in m using
precision calipers (NCCLS, 1997; 1999). The principle of this test is that when an
antibiotic impregnated disc is placed on an E. coli growth, it will create a clearing zone
around the disc where the bacteria can’t grow. The size of this zone of inhibition
depends on the sensitivity of the E. coli to the specific antibiotic (figure 2-4). This test
was used for its flexibility in type and number of drugs that can be tested, and because of

its low cost.

44

The breakpoints. Breakpoints are the zones of inhibition at which an organism is
considered to be susceptible or resistant based on obtainable serum concentrations of the
drug and clinical trail. In other words, when a lab reports that an organism is susceptible
it implies the recommended dosage of the antimicrobial agent that will reach serum or
tissue concentration and sufficient to inhibit the bacterium’s grth in vivo (Prescott, et
al., 2000). The breakpoints used to categorize isolates as resistant or not resistant for each
antimicrobial agent for E. coli were those recommended by the National Antimicrobial
Resistance Monitoring System (N ARMS) (Table 2-1). The American Type Culture
Collection (ATCC) 25922 E. coli was used to evaluate the performance of in vitro

susceptibility tests.

Figure 2-4: Disc diffusion test plate showing the zones of growth inhibition of E. coli
around an antimicrobial agent.

 

45

Table 2-1: Concentrations and diffusion mne breakpoints for resistance for
antinn'crobial agents tested in this study, by class of antinn'crobial agent

 

Disk Drug Diffusion Zone
Class of Agent Drug Concentration Breakpoint
Antinu'crobial Agent Code (pg) (mm)

Aminoglycosides

Neomycin N30 30 pg _<_ 12 nm

Gentamicin GMlO 10 pg 3 12 mm

Streptomycin 810 10 pg 511mm
Phinicols

Chbrarrphenicol C30 30 pg 5 12 mm
Quinolones and Fluoroquinolones

Ofloxacin OFX5 5 pg 5 12 mm

Nalidixic Acid NA30 30 pg _<_ 13 rrrn

Sulfonamides and potentiated sulfiaonamides

23.75 pg/ 1.25

Trimethoprim/Sulfamethoxazole STX ”g 5 10 mm

Sulfisoxazole G.25 250 pg _<_ 12 mm
T etracyclines

Tetracyline TE30 30 pg _<_ 14 mm
Penicilllins

Anpicillin AM10 10 pg _<_ 13 mm
Nitrofiarans

Nitrofurantoin FM 300 300 pg 5 14 mm
Cephalosporins

Cephalothin CF30 30 “g 5 14 mm

 

’ The zones of bacterial growth inhibition less than or equal the break point value indicate
the bacteria was resistant, and the zone value of 6mm indicates no zone of resistance

46

Data Analysis

All the data were entered into computer spreadsheets (Access and Excel; Microsoft),
which were used to generate descriptive statistics including graphs and charts of
antimicrobial resistance. The prevalence of antimicrobial resistance was calculated as the
number of samples yielding E. coli with resistance to a given antimicrobial agent divided
by total number of samples tested.

To test the stated hypothesis, Fisher's exact test was used for data analysis (SAS ver

8.2. SAS Inc. Cary, NC).

Results

A total of 31 farms agreed to participate in the study and included 7 dairy farms, 7
beef farms, 6 sheep farms, 5 pigs farm, 2 horse farms, 2 chicken farms, and 2 deer farms.
However, some farms contained more than one species, for example, sheep and horses,
sheep and chicken, pigs and chicken, sheep and beef cattle, beef cattle and horses. The
antimicrobial agents mentioned as being used in the different species on farms are shown
in Table 2-2. No data were collected regarding antimicrobial used by humans from

whom septic tank samples were collected.

47

Table 2-2: Antimicrobial agents reported as having been used on the farms in the
different species of animals sampled

1. Dairy

2. Beef c

3. Sheep

4. Goat

cattle

Quarter master (penicillin)

Cefa—lak (cephapirin sodium)

Excenel

Naxcel (ceftiofur sodium)

Aureo S 700 (Chlorotetracycline and sulfamethazine)
Tetracycline

Oxytetracycline

Trimethoprim-sulfamethoxazole

attle

Excenel

Micotil

Baytril (enrofloxacin)

Aureo S 700 (Chlorotetracycline and sulfamethazine)
Tetracycline

Oxytetracycline

OR-E-O Krumbs A (chlorotetracycline)

Penicillin, injection
Tetracycline, injection
Oxytetracycline, injection

Penicillin

OR-E-O Krumbs A (chlorotetracycline)
Bacitracin Methylene Disalicylate
Tylan (Tylosin)

Lincomix (Lincomycin)

Pulrnotil (Tilrnicosin)

Tetracycline

CTC-8 (chlorotetracycline)

6. Horses

Sulfa drug

7. Poultry: data not provided
8. Companion animals (dogs and cats): data not provided

48

Source of E. coli isolation versus antimicrobial resistance

To answer the question whether samples from farmed deer, wild geese, water, human
and different domestic species of animal have the same or different antimicrobial
resistance profiles, the prevalences of such profiles were compared. (Please notice that
images in this thesis are presented in color). In general, the antimicrobial resistance
profiles of E. coli isolates from cattle, sheep and swine were largely similar. Resistance to
antimicrobial agents was found in E. coli isolates from all animal species, human, and
water samples. In cattle (n=4l7; figure 2-5) resistance was found to all antimicrobial
agents tested. The most prevalent forms of resistance seen were to cephalothin,
tetracycline, sulfisoxazole and streptomycin. Minute resistance (less than 1%) to
gentamicin, nitrofurantoin, nalidixic acid and ofloxacin was seen. In sheep (n=156;
figure 2-7) resistance was found to ten of twelve antimicrobial agents tested. The most
prevalent forms of resistance seen were to cephalothin, tetracycline, streptomycin, and
sulfisoxazole. No resistance to ofloxacin and gentamicin was noted. In pigs (n=176;
figure 2-9), resistance was found to ten of twelve antimicrobial agents, the most prevalent
forms of resistance seen were particularly in regards to tetracycline, sulfisoxazole,
streptomycin, arnpicillin and cephalothin; no resistance to nalidixic acid and
nitrofurantoin was seen. In horses (n=58; figure 2-11) resistance was found to seven of
twelve antimicrobial agents tested. The most prevalent resistance seen was to
cephalothin, sulfisoxazole, trimethoprim sulfamethoxazole, tetracycline, streptomycin,
and ampicillin, while no resistance was observed to neomycin, chloramphenicol,

ofloxacin, nalidixic acid and nitrofurantoin was seen. In poultry (n=87; figure 2-13)

49

resistance was found to eleven of twelve antimicrobial agents tested. The most prevalent
forms of resistance seen were to tetracycline, cephalothin, sulfisoxazole, streptomycin,
and neomycin. No resistance to ofloxacin was seen.

In companion animals n=23 (17dogs, 6 cats; figure 2-15) resistance was found to four
of twelve antimicrobial agents tested; cephalothin, streptomycin, tetracycline, and
ampicillin. No resistance was shown for the rest antimicrobial agents. In farmed deer
(n=34; figure 2-16) and wild geese (n= 54; figure 2-16) there was resistance only to
cephalothin in farmed deer and resistance to cephalothin and tetracycline in wild geese;
no resistance was recorded for the other antimicrobial agents. In humans (n=3; figure 2-
17) resistance was found to three of twelve antimicrobial agents tested; cephalothin,
tetracycline, and streptomycin, whereas no resistance to the rest antimicrobial agents was
observed. In water samples (n=26; figure 2-18) all the E. coli isolates were resistant only

to cephalothin.

E. coli isolated from environmental samples

To answer the question whether E. coli isolated from the environmental samples of
domestic animals have the same or different antimicrobial resistance profile than that
from domestic animals, the prevalence of such profiles was compared. In general, the
resistance profiles in isolates from feces collected directly from domestic animals and
from the environment on the farm were similar but with slight variation among species.
Most E. coli isolates showed resistance to cephalothin, tetracycline, streptomycin and
sulfisoxazole. In E. coli isolated from cattle farm (n=118; figure 2-6) the most prevalent
forms of resistance seen were to cephalothin, tetracycline, streptomycin and

sulfisoxazole, low resistance was seen to neomycin, ampicillin, chloramphenicol,

50

trimethoprim-sulfamethoxazole and nitrofurantoin. No resistance to gentamicin,
ofloxacin, and nalidixic acid was observed. In sheep farms (n=3 1; figure 2-8) most E.
coli isolates showed resistance to cephalothin, tetracycline, sulfisoxazole, streptomycin
and trimethoprim- sulfamethoxazole, low resistance was seen for gentamicin,
chloramphenicol, and no resistance was seen to neomycin, ofloxacin, ampicillin, nalidixic
acid, and nitrofurantoin. The E. coli isolated from pigs farms (n=38; figure 2-10) showed
the highest resistance to tetracycline, compared to other animal farms, then to
cephalothin, sulfisoxazole, streptomycin, ampicillin and neomycin. Small resistance to
gentamicin, chloramphenicol was seen. No resistance to ofloxacin, trimethoprim
sulfamethoxazole, nalidixic acid or nitrofurantoin was observed. In horse farms (n=16;
figure 2—12) the high resistance was seen to cephalothin, tetracycline, streptomycin,
sulfisoxazole, and trimethoprim sulfarnethoxazole. Slight resistance was seen to
gentamicin and ampicillin, no resistance to neomycin, chloramphenicol, ofloxacin,
nalidixic acid or nitrofurantoin was seen. Regarding poultry farms (n=19; figure 2-14)
the highest resistance to tetracycline was seen then cephalothin, and sulfisoxazole and
streptomycin and neomycin. No resistance was seen to the rest of antimicrobial agents.
Over all, the resistance of E. coli isolated from the environment was higher to cephalothin

than that isolated directly fi'om animals.

51

Source of E. coli isolation under same condition versus antimicrobial resistance
To answer the question whether animals under same condition in the same farm have

the same antimicrobial profiles, the following results answer the question.

Antimicrobial resistance of E. coli from a farm with pigs and chickens

By comparing the antimicrobial resistance profile for pigs and chickens from the same
farm we found that they have similar antimicrobial resistance profile except that of
chickens, which do not show any resistance to ampicillin. Both, however, showed high

resistance to tetracycline (figure 2-19).

Antimicrobial resistance of E. coli from a farm with cattle, horses and sheep
Under this farm condition the E. coli isolated from horse and sheep showed high
resistance to cephalothin only but the E. coli isolated from cattle showed resistance not

only to cephalothin but also to tetracycline, streptomycin, and sulfisoxazole (figure 2-20)

Antimicrobial resistance of E. coli isolates from chicken and food animals living on
the same farm

Over all, the resistance patterns were similar but the resistance to tetracycline in E.
coli isolated from food animals (n=132) was higher than that for E. coli isolated from
chicken (n=52). On the other hand, the resistance among food animal isolates to
cephalothin and streptomycin was less than that for E. coli isolated from chicken isolates.
The resistance to sulfisoxazole among food animal isolates was higher than that for

chicken isolates. Low resistance was observed to the following antimicrobial agents

52

among food animals and chicken isolates: neomycin, ampicillin and chloramphenicol

(Figure 2-21).

Antimicrobial resistance of E. coli isolates from companion animals and food
animals living on the same farm

The E. coli isolated fiom companion animals (n=8) and food animals (253) living on
the same farm showed resistance to cephalothin, tetracycline, streptomycin and ampicillin
but the level of resistance to these antimicrobial agents in food animals is less than that in

companion animals (Figure 2-22).

Antimicrobial resistance of E. coli isolates from companion animals and horses
living on the same farm

The prevalence of resistance among companion animals’ isolates, dogs and cats
(n=15), and horse (n=12) under the same farm was the same. Both were resistance to

cephalothin only and susceptible to the rest antimicrobial agents (figure 2-23).

Antimicrobial resistance of E. coli isolates from horses and food animals living on
the same farm
Both isolates from horses (n=21) and food animals (n=162) on same farm were

resistant to cephalothin only (figure 2-24).

Multi-drug resistant E. coli isolates
When comparing the E. coli isolates from domestic animals, wildlife (farmed deer and

wild geese), human, water and farms and the number of antimicrobial agents that

53

resistant to, we found that 25.4% of E. coli isolated from domestic animals was resistant
to one antimicrobial agent, 10.0% resistant to two antimicrobial agents, 7.0% to three
antimicrobial agents. In wildlife (farmed deer and wild geese) 13.3% resistant to one
antimicrobial agent, 33.3% of human E. coli isolates were resistant one and two
antimicrobial agents, 61.5% of E. coli from water samples were resistant to one
antimicrobial agent. According to E. coli isolated from farms 29.4% resistant to one
antimicrobial agent, 13.3% resistant to two antimicrobial agents, 9.7% resistant to three
antimicrobial agents, and 5.5% to four antimicrobial agents (figure 2—25). Table 2-3
shows the percentage of resistant isolates for specific antimicrobial agents, by species

exposure class.

54

Table 2-3: Percentage of resistant isolates for specific antimicrobial agents, by
species exposure class

 

Fisher's
Exact P
Humans & for all
Companion species
Overall Livestock Animals Equines Wildlife within
Agent (n=1,041) (n=863) (n=26) (n=60) (n=90) agent
Neomycin 4. 7 5.7 *** 0 0 0 "' .0089
Gentamicin .8 .7 0 3.3 O .2021
Streptomycin 13.1 14.7 *** 11.5 10.0 0 *** < .0001
Chlorarnphenicol 1. 1 l .3 0 O .85 10
Ofloxacin .2 .2 0 0 0 1.0
Trimethoprim 2.2 1.7 * 13.3 0 < .0001
Sulfamethoxazole ***
Tetracycline 30.7 35.7 *** 11.5 * 10.0 3.3 *** < .0001
*Ikik
Ampicillin 6.5 7.2 3.8 8.3 0 ** .0146
Nalidixic Acid .9 0 0 0 1.0
Nitrofirrantoin .9 1.0 O 0 0 1.0
Cephalothin 21.2 21.8 34.6 21.7 10.0 ** .0142
Sulfisoxazole 14.0 16.0 *** 0 * 13.3 0 *** < .0001

 

* - Fisher's Exact P < .05 for individual species exposure class within agent

** - Fisher's Exact P < .01 for individual species exposure class within agent

*** - Fisher's Exact P < .001 for individual species exposure class within agent

Table 2-4 shows the percentage of multi-drug resistance isolates by species exposure

class while table 2-5 shows the common combinations of antimicrobial agents in multi-

drug resistant isolates.

Table 24: Percentage of multi-drug resistant isolates by species exposure class

 

 

Fisher's
Exact P
for all
Humans & species
Companion within
# Agents Overall Livestock Animals Equines Wildlife #
Resistant (n=1,041) (n=863) (n=26) (n=60) (n=90) agents
0 53.2 48.6 61.5 68.33 86.67 < .0001
1 25.0 26.7 23.1 16.67 13.33 .0142
2 8.5 9.7 11.5 1.67 0 < .0001
3 or more 13.4 15.1 3.9 13.33 0 < .0001
Mantel
Hagensel
X test
for trend - 39.58 1.40 2.74 41.43
within
class
x2 p - < .0001 .2366 .0980 < .0001
Fisher's - < .0001 .4615 .0417 < .0001
Exact p

 

 

56

Table 2-5: Most commonly identified combinations of agents in multi-drug resistant
isolates from food animals

 

Percent of
Multi-resistant
Agents All isolates " isolates b
Tetracycline - Sulfamethazine 14.1 53.9
Tetracycline - Streptomycin 12.7 48.4
Streptomycin — Sulfamethazine 10.0 38.1
Tetracycline — Cephalothin 7.2 27.5
Tetracycline -- Ampicillin 6.5 24.9
Tetracycline — sulfamethazine — Streptomycin * 7.3 27.8

 

* Includes numbers from tetracycline-sulfamethazine, tetracycline-streptomycin, and
streptomycin-sulfamethazine.

To illustrate the calculation of a and b, tetracycline-sulfamethazine used as an example;
a = Number of isolates resistant to tetracycline-sulfamethazine / All isolates tested
b = Number of isolates resistant to tetracycline-sulfamethazine / Number of multiresistant
isolates

Within the 4 population subgroups, the livestock group including cattle, sheep, goats,
poultry and swine manifested the highest rate of antimicrobial resistance for all but 4 of
the individual antimicrobials tested (gentamicin 0.7%, sulfamethoxazole 1.7%, ampicillin
7.2%, and cephalothin 21.8%). Additionally, the livestock subpopulation had the highest
rate of resistance to 3 or more antimicrobials (13.5%). The rate of multi-drug resistance

by subpopulation group was also statistically significant (Fisher's exact p <0.0001).

57

 

Discussion

This study found antimicrobial resistance to a variety of agents in E. coli isolates from
food animals, companion animals, farmed deer and wild geese, human and water samples
in Michigan, United States. This study was not designed to determine the relationship
between antimicrobial use in animals on farms and the prevalence of patterns of
antimicrobial resistance. It is, however, noteworthy to mention that the majority of
antimicrobials for which resistance was demonstrated, were reported as being used in
different animal species on farms. Additionally, resistance in E. coli isolates from cattle,
pigs and sheep was reported to antimicrobial agents such as chloramphenicol that was
banned fi'om use in these animals. Such resistance may be due to persistence of resistant
E. coli on farms for a long time. The differences in the prevalence of resistance between
species may be due to different exposures to antimicrobial growth promoters and the time
that the animals were exposed.

The biological plausibility that antimicrobial use in animal feed and the selection
pressure for bacteria with resistance to the exposed agents was observed in many bacteria
(Witte et al., 2000). The resistance to tetracyclines, especially chlorotetracycline in swine
and poultry industries, may be due to the widespread use of this antimicrobial agent in
animals as feed additive, prophylactic or used clinically to treat diseases. Smith, (1975)
showed that feeding pigs rations contain tetracyclines resulted in the recovery of large
number of tetracycline- resistant E. coli from their feces. Enterotoxigenic E. coli usually
is resistant to tetracycline in swine, poultry and cattle since this drug and
chlorotetracycline has been used widely in medicated feed as growth promoters (Prescott

et al., 2000). Tetracyclines especially oxytetracycline were frequently used in animal

58

ration as feed additives in cattle, and sheep yet a substantial number of cattle and sheep
isolates were tetracycline resistant. The relationship between the degree of antimicrobial
use and the extent of resistance was explained by Linton (1977). Acquired resistance is
wide spread in tetracyclines and it is usually plasmid mediath that interfere with the
active transport of tetracycline into the cell and increased efflux from the cell (Prescott et
al., 2000).

Antimicrobial use in human facilitates resistance development in many pathogenic
bacteria (N eu, 1992). The E. coli isolated from human septic tank showed resistance to
tetracycline as a results of therapeutic or prophylactic application.

We found low resistance to tetracycline in E. coli isolated fiom horses’ fecal samples
compared to other animals in our study. It could be due to the little use of tetracycline in
horses, since it has high potential for toxic effects, including local necrosis after
intramuscular injection, or its broad spectrum suppression of intestinal microflora that
allow for super infection with resistant salmonella and clostridia that result in severe and
lethal diarrheic illness (Prescott et a1. 2000).

High percentages of E. coli isolates were resistant to cephalothin, which is a first
generation cephalosporin beta-lactam antibiotic. The clinical applications of first
generation cephalosporins have decreased with the development of beta lactamase stable
cephalosporins. Oral cephalosporin has been used extensively in small animal medicine
for treating skin and urinary tract infections especially cephalexin the drug of choice for
Klebsiella pneumoniae infections or as prophylaxis of surgical wound infections. It is
used for the same reasons in human. This explains the high resistance to this drug among

companion animals and humans in our study (Figure 2-15 and Figure 2-17).

59

During our visits to farms, we collected information about the antimicrobial agents
that the farmers used to treat their animals or given to their animals as a feed additive.
We found that most of the dairy farmers used Cefa-Lak that contains cephapirin sodium,
which is a first generation-group one cephalosporin as an intrarnarnmary infusion in dairy
cows to treat bovine mastitis. The other drug that has been used widely by veterinarians is
Naxcel. Naxcel which contains cefiiofur sodium, third generation-group four
cephalosporin, to treat systemic infections especially respiratory disease caused by gram-
negative aerobes such as E. coli, Salmonella, and Pasteurella in cattle, swine, sheep and
I horse. It is also been used to treat urinary tract infections associated with E. coli and
Proteous mirabilis in dogs. In poultry it has been used to control E. coli infections. The
use of first and third generations of cephalosporin may explain the high resistance to
cephalothin among these animals in our study (Figures 2-5, 2-7, 2-11, 2-13, 2-15).

Thirty three percent of E. coli isolated from human septic tank was resistant to
cephalothin. The use of first, third and fourth generation cephalosporins to treat skin and
urinary tract infections, in addition to respiratory tract infections and meningitis in
humans, may explain this resistance (Figure 2-17). 1

For susceptibility testing, cephalothin is the class drug to use. There are three
mechanisms of resistance to cephalosporins: first, reduced permeability, second,
enzymatic inactivation and third, absence of specific penicillin binding proteins.

In our study we noticed the high resistance to cephalothin, tetracycline, and streptomycin
and this could be due to the cross-resistance between these antimicrobial agents. We
noticed that some cephalosporin mutanted with altered outer membrane permeability that

involves the extra cellular expression of beta lactamases through efflux pumps, may show

60

cross resistance with aminoglycosides, chloramphenicol, fluoroquinolones, tetracyclines,
and trimethoprim (Prescott et al., 2000). Cross resistance means when one organism
becoming resistant to one antibiotic thereby becomes resistant to another.

Transferable, broad-spectrum plasmid mediated resistance to beta-lactamase stable
cephalosporin has increasingly been described; and consider a threat to the continued use
of these cephalosporins (Prescott et al., 2000).

The persistence of resistant bacteria is related to the persistent use of the antimicrobial
agent. The prolonged use of the antimicrobial agent is more likely associated with the
persistence of resistant E. coli, but it may not be readily reversed by withdrawal (Langlois
et al., 1983) this could explain the presence of small percentage of E. coli isolates that
resist chloramphenicol, even after the a banned use of this drug in food animal in the
USA. Chloramphenicol was banned because of its toxic effect in humans; it causes
suppression to the bone marrow leading to aplastic anemia.

Interestingly, a very small percentage of E. coli isolates in our study were resistant to
nalidixic acid, nitrofurantoin, and non resistant to ofloxacin. Because none of these drugs
are approved for use in food animals in the United States, the small resistance observed
may be due to the use of related approved bovine antimicrobial agents such as florfenicol
and enrofloxcacin (White et al., 2000).

Recent studies have suggested the use of cephalosporins, tetracyclines, sulfa drugs, to
be the major factor in the emergence and dissemination of antimicrobial resistance E. coli
in animals (Meng et al., 1998; Galland et al., 2001; Schroeder et al., 2002a; Schroeder et

al., 2002b). These studies were consistent with our findings that there was high

61

prevalence of resistance to cephalothin, tetracycline, and sulfisoxazole among E. coli
isolates from different species of animals.

Streptomycin is part of the aminoglycosides class that selectively binds to kidney
tissue. For that reason, the US. Food and Drug Administration do not approve it for use
in food animals. Although, for a long time it was used extensively in combination with
penicillin in food animals, to treat Corynebacterium renale and Arcanobacterium
pyogenes and actinobacillosis. Its combination with tetracycline is effective against
Brucella abortus in cattle and Brucella melitensis in sheep. Streptomycin used to control
leptospirosis in swine, but in horses it has little implication because of wide spread of
resistance (Prescott et al., 2000). In our study we found that E. coli isolated from
different animal species was resistant to streptomycin. Acquired resistance to
streptomycin is widespread in veterinary pathogens, and plasmid-mediated resistance is
commonly linked with sulfonamide, ampicillin, and tetracycline resistance genes. We
noticed that E. coli isolates that resistant to streptomycin were resistant also to
tetracycline and sulfonamide (Prescott et al., 2000).

The high resistance of E. coli isolated from horses fecal samples to trimethoprim-
sulfamethoxazole and sulfisoxazole due to the wide use of sulfa drugs to treat acute
respiratory infections, acute urinary tract infections, wounds, and abscesses.
Trimethoprim-sulfadiazine and sulfisoxazole are the drugs of choice to treat
Salmonellosis, Actinobacillus in foals, and coliform meningitis. We also found
resistance to trimethoprime sulfamethoxazole and sulfisoxazole in E. coli isolated fiom
different animals species and this may due to the wide spread use of the sulfonamides in

these species. For example, sulfadimethoxine is the only sulfonamide approved for use in

62

dairy cows over 20 months of age in the United States. Sulfonarnides have been used
with chlortetracycline in feedlot lambs to improve performance and prevent clostredial
enterotoxemias. In addition, sulfonamide mixed with chlortetracycline to promote
growth and control group B streptococcal infections, and atrophic rhinitis in pig. Because
of the problem of residues in carcasses in excess of legally permitted amount, there have
been moves to ban the use of sulfonamides in swine. In poultry sulfonamide has been
used to treat coccidiosis, and infectious coryza.

Multiple antimicrobial resistance in E. coli isolates may results from the spread of
genetic elements including plasmids, tranposons, and integrons (Jones et al., 1997),
which may also confer resistance to numerous antimicrobials. The R factor or R
resistance plasmid may code for resistance to between one and ten different antibiotics.
The linkage of resistance genes on the same plasmid means that the use of any one
antibiotic for which resistance was determined by the plasmid promotes resistance to all
the antibiotics. Withdrawing the use of all antibiotics in a herd may not results in the loss
of resistance by E. coli because such genes may be incorporated into bacterial
chromosome (Prescott et al., 2000) and because possession of R plasmids may not be
deleterious to bacterial survival (Smith et al., 1974).

Nonspecific resistance to a wide range of unrelated antibiotics associated with
mutations, leads to over expression of multi-antibiotic resistance locus, which controls
multidrug efflux pumps in bacteria that can be selected by low concentrations of an
antimicrobial drug (Prescott et al., 2000)

It is plausible that bacteria with resistance to antimicrobial agents could be developed

on farm and transmitted to human (Levy, 1987). The use of antimicrobial agents in food

63

animals may contribute to the antimicrobial resistance in human. For example, cattle are
considered a symptomatic carrier for E. coli, when exposed to antimicrobial agents may
serve as a reservoir of antimicrobial resistant E. coli (Barton, 1998; Witte, 1998). Since
first and third generations cephalosporins were used in cattle in our study, the observation
that htunan E. coli isolates resistant to cephalothin suggests the transfer of resistant E.
coli from food animals to humans (Zhao et al., 2001).

Our study was not designed to determine the contribution of antimicrobial use in
animals to the antibiotic resistance in humans. Rather, patterns of antimicrobial
resistance were observed, which may form a data —based basis for hypothesis
formulation.

Resistant bacteria may colonize the human population via food chain through many
pathways: first, the resistant pathogenic bacteria could be selected in the animal’s gut; it
could contaminate the meat during animal slaughter or meat preparation. If a human
ingested the contaminated meat, it would cause infection and the treatment will be
compromised. Second, the non-pathogenic antibiotic resistant bacteria will be selected in
the animal’s gut, leading to contamination of food that, if ingested, the bacteria could
transfer the resistance to other bacteria in the human gut. Third, residual remains of
antibiotic in animal products, may allow the selection of antibiotic resistant bacteria in
the consumer of the food (Piddock, 1996).

Resistant bacteria could also enter humans through occupational exposure; for
example, farmers, slaughterhouse workers and food handlers are more likely to have
resistant E. coli than the general population. Drinking or swimming in water

contaminated with animal fecal materials that contain resistant bacteria especially E. coli

64

and the direct contact with animals, especially companion animals, are another ways for
humans to be exposed to resistant bacteria.

Water becomes contaminated with resistant bacteria when waste run off from animal
production facilities, leakage from human septic tank or sewage dumped in the river
(Witte, 1998; Van den Bogaard and Stobberingh, 1999; Van den Bogaard et al., 2001).
The E. coli strain responsible for 1989 waterborne outbreak in Missouri was resistant to
streptomycin, sulfisoxazole, and tetracycline (Swerdlow etal., 1992).

Our results showed that E. coli isolates from domesticated animals showed resistance
to more antimicrobials tested than those isolates from human septic tanks, wild geese,
farmed deer or surface water. On the other hand, the level of resistance to cephalothin in
E. coli isolated from water was closer to that observed in E. coli isolated from human
septic tanks. These results suggested that surface water contamination by antimicrobial
resistant E. coli is a complex issue in terms of determining the source of the problem.
Even though E. coli from farm animals showed resistance to more antimicrobial agents
(including cephalothin), the percentage of isolates that were resistant to cephalothin was
much lower than that in isolates from human septic tanks or companion animals.
However, a previous study with sludge and septic tank wastes showed relatively high
levels of antimicrobial resistance in E. coli (Pillai et al., 1997).

We expected that E. coli isolated from wild geese would not show resistance to any of
the antimicrobial agents tested, because wild birds have never been treated with
antimicrobial agents. However, we found that 11% and 4% of E. coli isolates were
resistant to cephalothin and tetracycline respectively. The access of wild birds to the

medicated animal feed, and then developing the resistance in their gut, or the easy access

65

of wild birds to river water that may be contaminated with the resistant bacteria, may
explain these results. Resistant E. coli could reach the water from a failed septic tank
system of a nearby residence, or from runoff water over a wide field area that may have
been spread with manure of animals harboring resistant bacteria, or the water could have
the resistance factor from other bacteria due to the relative ease with which resistance
factors are exchanged among promiscuous bacteria (LeClerc etal., 1996).

Surprisingly, the E. coli isolates from water were 81% resistant to cephalothin only
(the highest in all E. coli isolates). The induction of cross-resistance by structurally
related compounds might explain the presence of the corresponding resistance in the
water system. Our results agree with the results of Mulamattathil et al., (2000) who
found the high level of resistance to cephalothin in closed water reticulation system. He
also found a correlation between the specific use of certain antimicrobial agents and the
prevalence of the corresponding resistant bacterial isolates. Our results are similar to the
results from a recent survey of US. rivers found that cefotaxime a third generation
cephalosporin resistant gram negative bacterium to range from 16%-96% across 22 rivers
(Ash et al., 2002).

Grabow et al., 1973; Grabow et al., (1975) suggested that R—factor mediated
antimicrobial resistance increased the survival ability of coliform bacteria; on the other
hand, (Anderson, 1974) suggested that R-factor mediated may reduce survival ability in
E. coli but Smith et al., (1974) indicated that R-factor antimicrobial resistance had no
effect on E. coli survival. Once multiresistance bacteria develop, they may persist in the
host or the environment in the absence of antibiotic selection and may act as reservoirs of

resistance genes that may spread to other bacteria (Prescott et al., 2000).

66

There were trends in patterns of antimicrobial resistance indicating common sources
of resistance factors for different types of samples and for different species. First, similar
patterns of resistance from fecal and environmental samples classified by animal species
were similar, indicated common sources of resistant bacteria. It is possible that livestock
function as a reservoir of resistant bacteria for environmental contamination, particularly
in cases where higher levels of resistance were seen in fecal isolates compared to
environmental isolates (e.g. tetracycline, sulfisoxazole). However, in cases where higher
levels of resistance were seen in environmental samples (e. g. cephalothin, trimethoprim-
sulfarnethoxazole) it is also possible that contaminated environments serve as the major
reservoirs for resistant bacteria for livestock. This does not support the once widely held
theon that the presence and expression of resistance genes impairs viability and survival
outside the host. It also suggests that, while collection of environmental samples from a
farm may not be valid means of assessing the prevalence and distribution of antimicrobial
resistance patterns in animals residing on the farm, it is a more accurate measure of
exposure to resistance from farm runoff into watershed.

It was interesting and noteworthy to observe in our study that animal species that lived
on the same farm had similar antimicrobial resistance profiles. Thus, food animal and
chickens, living on the same farm as well as different combinations such as horses and
companion animals, food animals and horses had similar antimicrobial resistance
patterns, these observations suggest that bacteria which share a common environment
also share a common mode for developing antimicrobial resistance (Kelch and Lee,

1978)

67

Because of logistical reasons, we were unable to collect samples in a manner that
allowed us to evaluate whether there was seasonal variation in the frequency of E. coli in
samples collected and hence antimicrobial resistance profile. This issue needs to be

addressed in future research.

68

Figure 2-5: Antimicrobial resistance of E. coli from cattle

 

 

 

 

 

 

 

 

30% I Neomycin
I Gentana'cin
25% I Streptomycin
... 20% I Chloranmhenicol
g D Ofloxacrn
'§ 15% I TMP/SMX
9: I Tetracycline
°\ 10% u Anpicillin
I Nalidixic Acid
0 _
5 A E] Nitrofurantoin
0% fl I Cephalothin
I Sulfisoxamle

n=417

Figure 2-6: Antimicrobial resistance of E. coli from cattle farms

 

 

 

 

 

 

 

30% I Neomycin
I Gentam'cin
25% I Streptomycin
... 20% I Chloramphenicol
s D Ofloxacm
9: I Tetracycline
°\ 10% D Anpicillin
I N alidixic Acid
5% - -
D Nitrofurantoin
0% p I Cephalothin
I Sulfisoxamle
n = 118

Figure 2-7: Antimicrobial resistance of E. coli from sheep

 

 

 

 

25% I Neomycin
I Gentana'cin
20% —_ I Streptomycin
I Chloramphenicol
g 15% - TMP/SMX
E I Tetracycline
: 10% El Anpicillin
° Nalidixic Acid
5% _ E1 Nitrofurantoin
I Ceplmlothin
0% _ I Sulfisoxazole

 

 

n=156

Figure 2-8: Antimicrobial resistance of E. coli from sheep farms

 

 

 

 

 

 

30% I Neomycin
I Gentamicin
25% I Streptomycin
... 20% I Chloranphenicol
g I TMP/SMX
g 15% I Tetracycline
5: I D Anpicillin
°\ 10% - n Nalidixic Acid
.1 - I a W...
0 _
I1 I - I '°""“'°"""
I
0% _ Sulfisoxaznle
n = 31

70

Figure 2-9: Antimicrobial resistance of E. coli from swine

80% I Neomycin
70% I Gentam'cin
60% I Streptomycin
.. I Chloramphenicol
50"
g A’ m Otloxacin
E 40% I TMP/SMZ
°\e 30% I Tetracycline
20% E] Anpicillin
I Cephalothin
10% I Sulfisoxamle

 

0%

n=176

Figure 2-10: Antimicrobial resistance of E. coli from swine farms

80% I Neomycin
70% I Gentamicin
60% I Streptomycin
,_, I Clrloranphenicol
00
g 5 A I] Ofloxacin
Z. 40% I TMP/SMZ
°\e 30% I Tetracycline
[:1 . . .
20% “mm”
I Cephalothin
10% l Sulfisoxazole
0%

 

71

Figure 2-11: Antimicrobial resistance of E. coli from horses

 

 

 

 

25%
20% I Gentamicin
I Streptomycin

E 15% I TMP/SMx
.‘g I Tetracycline
0
a: 10% _ El Ampicillin
°\ I Cephalothin

50/ I Sulflsoxaznle

o A
0% i

 

 

Figure 2-12: Antimicrobial resistance of E. coli from horse farms

 

 

20%
I Gentanicin
15% I Streptomycin
E I TMP/SMX
.‘g 10% I Tetracycline
é - Ampicillin
g I Cephalothin
5% I Sulfisoxamle
0%

 

 

72

Figure 2-13: Antimicrobial resistance of E. coli from poultry

 

 

 

 

 

 

 

50% I Neomycin
I Gentamicin
40% I Streptomycin
I Chloramphenicol
E 30% I TMP/SMX
g I Tetracycline
5° 20% , W A I] Ampicillin
° I Nalidixic Acid
10% U Nitrofurantoin
I Cephalothin
0% _ I Sulfisoxamle
n = 87

Figure 2-14: Antimicrobial resistance of E. coli from poultry farms

 

 

 

 

50% -E I Neomycin
I Gentamicin
40% I Streptomycin
I Chlorarrphenicol
g 30% j__ I TMP/SMX
'5 I Tetracycline
0
a: 20% D Anprcrlhn
°\ I:I Nalidixic Acid
10% 7 El Nitrofurantoin
I Cephalothin
Sulfrsoxazole
0% ~

 

 

 

73

Figure 2-15: Antimicrobial resistance of E. coli from companion animals

40%
35%
0

30 A I Streptomycin
E 25% I Tetracycline
E 200/ El Ampicillin

0

g l Cephalothin
°\° 15%

10%
5%

 

0%
n = 23 (17 dogs, 6 cats)

Figure 2-16: Antimicrobial resistance of E. coli from farmed deer and wild geese

 

 

15%
.5 10%
a ITetracycIine
E ICephalothin
,\°

50/0 '

0% '

 

 

 

Deer (n=34) Geese (n=54)

74

Figure 2-17: Antimicrobial resistance of E. coli from human septage

 

40%

 

35%
30% 1
25% r
20% ~

% Resistant

15% —
10% r

5%4

 

0%*

 

I Streptomycin
I Tetracycline
I Cephalothin

 

Figure 2-18: Antimicrobial resistance of E. coli from water

100%

75%

50%

% Resistant

25%

0%

 

 

 

75

 

I Cephalothin

Figure 2-19: Antimicrobial resistance of E. coli from a farm with pigs and chickens

 

 

 

 

 

 

 

 

 

 

100%
_ 3 I Neomycin

75% . ..; E1 Streptomycin
iii . 1“ I] Chloramphenicol
,3 1 1~1 El Tetracycline
g 50% El Ampicillin
f 1 I Cephalothin
°\ 25% p I Sulflsoxamle

0% -

Pigs Chickens
(n=40 of 61) (n=l6 of 22)

Figure 2-20: Antimicrobial resistance of E. coli from a farm with cattle, horses and
sheep

 

 

 
  
 

 

100%

75% I Streptomycin
E D Tetracycline
g I Cephalothin
E 50% ‘ I Sulfisoxazole
m
e\°

25%

0% ‘

 

 

 

Cattle Horses Sheep
(n=10 of 24) (n=6 of 9) (n=4 of 14)

76

Figure 2-21: Antimicrobial resistance of E. coli from chicken and food animals

% Resistant

 

 

 

 

 

 

 

 

Chicken (N=52)

 

 

 

Food Animals (N=132)

I Ampicillin

I Cephalothin

E] Snlflsoxazole

El Neomlcln

I Chloramphlnleol
D Streptomycin

I Tetracycllne

Figure 2-22: Antimicrobial resistance of E. coli from companion animals and food

% Resistant

10%

9%~

8%

7% — _,
6% —
5% —— -—
4% — ——
3% -
2% —
1%
0% -

 

 

 

 

 

 

Companlon Anlmals (N=8)

 

Food Anlmals (#253)

77

I Amflcillln
I Ceflralothin

El Strepomycin
I Tetracycline

Figure 2-23: Antimicrobial resistance of E. coli from companion animals and horse

% Resistant

35%

 

30%

 

25% .
20% -

15% ~

 

 

 

10% - -- ,

5%«

 

0%-

 

w

 

 

Companion Animals

(n=15)

Horse (n=12)

ICephalothin

Figure 2-24: Antimicrobial resistance of E. coli from horse and food animals

% Resistant

60%

 

50%

40% -

30%

20%

10% ~

 

0%

 

2 ICephalothin

 

 

Horse (N=21)

Food Animals (N=162)

78

Figure 2—25: Multi-drug resistant E. coli isolates

 

100%
80%
m I Domestic
0
E 60% I3 Wildlife
0
E o I Human
,\° 40 A. I Water
[I Farms
20%
0%

0 l 2 3 4 5 6 >6

# Antimicrobial agents

79

CHAPTER 3

DISCRIMINANT ANALYSIS AS A TOOL FOR CLASSIFYING THE SOURCES OF
E. COLI CONTAMINATION OF SURFACE WATER IN MICHIGAN, USA
Abstract
Discriminant analysis of patterns of antimicrobial resistance profile in E. coli was used

to differentiate between different animals’ sources of fecal pollution in water. The disc
diffirsion test was used to conduct in vitro antimicrobial susceptibility testing of 1169 E.
coli isolates from cattle, pigs, poultry, sheep, farmed deer and wild geese fecal samples
and human septic tank samples. Seven antimicrobial agents (neomycin, streptomycin,
trimethoprim-sulfamethoxazole, tetracycline, ampicillin, cephalothin, and sulfisoxazole)
were tested for susceptibility. When five source species were analyzed by discriminant
analysis, the Average Rate of Correct Classification (ARCC) was 80.6% based on
resubstitution method and 38.4% based on cross-validation method. The ARCC
increased in both methods when less animal categories were analyzed; 96.8% and 60.8%
in resubstitution and cross-validation methods, respectively, when animals pooled into
domestic animals and compared with wildlife (wild geese) animals. When E. coli
isolated from surface water receiving fecal pollution from unknown sources were
analyzed, more than 19.2% of the isolates were classified as cattle, 26.9% as pig, 19.2%
as sheep, 0.0% as poultry and 30.8% as wildlife, However, when animals were pooled as
domestic and compared with wildlife animals 73.1% of the isolates were classified as
domestic and 26.9% as wildlife animals. Based on the resubstitution discriminant
analysis results, discriminant analysis should be considered further as one of the methods

to determine the source of fecal pollution in water. The use of the resubstitution method

80

 

(each isolates is classified based on the patterns in the entire library, including its own
patterns) as a measure of the performance of discriminant analysis gave higher ARCC
than the more conservative method, the cross- validation (an individual isolate was
removed from the library one at a time, then the removed isolate is classified based on the
library comprised of the remaining isolates). The ARCC for two categories is higher than

that of more than two in both methods.

Introduction

Fecal contamination of water is a widespread problem in the United States (United
State Environmental Protection Agency, 1986). The high levels of fecal coliform bacteria
in many lakes and rivers have impaired the quality of recreational and drinking water.
Public exposure to pathogens in recreational and drinking water increased human health
risks of acquiring pathogenic bacteria, resulting in reports of waterborne outbreaks of
diseases involving fecal organisms (Swerdlow et al., 1992; Keene et al., 1994). In
addition, increased levels of nutrients (phosphorus and nitrogen) in a receiving body of
water can results in eutrophication that led to economic losses in shellfish industries
(Crane and Moore, 1986). Knowing the source of fecal contamination is necessary to
determine the degree of risk associated to human health in order to develop effective
control strategies and resources management. In particular, it is important to determine
whether the source of fecal contamination is of human, livestock, or wildlife origin as
microorganisms of human origin are regarded as having greater potential to cause

diseases in humans.

81

 

Water contamination can originate from point or non-point sources. The point source
refers to single, identifiable points of origin, for example, municipal effluents or
industrial discharge. On the other hand, non-point sources have diffuse origins such as
surface runoff manure treated agricultural land or farm animal feedlots, failing or
inadequate sewage treatment plants, leaking septic systems, sewer overflow, and wildlife
waste.

In order to identify the source of water contamination whether it is from animals or
humans, we need to have an indicator of fecal pollution. In our study, we used E. coli as
an indicator bacterium of fecal contamination, because it is the type of fecal coliform
bacteria that is commonly found in the intestinal tract of animals and humans. Also, it
can be rapidly detected and easily numerated in the laboratory. E. coli is increasingly
recognized as a water borne pathogen, because many outbreaks are associated with
consuming drinking water (Swerdlow et al., 1992; Jones and Roworth, 1996) or coming
into contact with recreational water (Keene et al., 1994; Ackrnan et al., 1997) had
occurred. Total coliforms are used as an indicator of fecal contamination in water since
they inhabit the intestine of warm-blooded animals, or are found naturally in the soil,
vegetation and water. But E. coli is a species of total coliform that is always found in the
feces and, therefore, a more direct indicator of human sewage or animal waste
contamination of water.

During rainfall or snowmelt, E. coli can get into rivers, streams, lakes or groundwater
and contaminate them. Using untreated water or inadequately treated water as a source
for drinking water has a detrimental impact on human health. The ability of E. coli to

survive in feces, on pasture land, and water systems has implications for its spread to

82

crops by direct application of manure, irrigation with infected water or by direct contact
with animals or contaminated soil. It appears that E. coli can remain viable in soil for
greater than 4 months (Jones, 1999). A study by Jones (1999), found that the survival of
E. coli was greatest in soil cores containing rooted grass and the viable number decreased
after 130 days, but when the organism was inoculated into cattle feces it remained
detectable at high levels for more than 50 days. The organism survived less in cattle
slurry and river water in which it fell to undetectable levels in 10-27 days. The survival
on stainless steel surfaces also was investigated and it was found that E. coli could
survive for 60 days, and most stable at a temperature of 4 oC. The viability reduced at 18
0C. It also survives for extended periods on domestic (plastic) cutting boards in both
temperatures (Maule, 2000).

There have been several attempts to differentiate between human and nonhuman
sources of fecal pollution. Initially, the ratio of fecal coliform to fecal streptococcus was
used as an indicator of the source: the ratio of 4.0 or more indicates human source
pollution while a ratio less than or equal to 0.7 would suggest nonhuman source of fecal
pollution (Feachem, 1975). The rational behind the use of this method was that human
feces contain higher fecal coli form counts while animal feces contain higher fecal
streptococcus count. However, this method is not reliable anymore due to the variability
in the survival rates of fecal streptococcus and variability in detection methods. Other
investigators have proposed bacteriophages as indicators of the source of fecal pollution.
Furuse et al., (1981) and Osawa et al., (1981) noticed that animal and human waste
contained different serotypes of RNA coliphages. Tartera et al., (1989) suggested the use

of Bacteriodes fragilis as an indicator of human source. However, the usefirlness of these

83

approaches is limited because only a small percentage of human fecal samples contain
phages, in addition to the difficulty in performing the assay (Troy et al., 2002).

An immuonological method, which is based on the presence of somatic O antigenic
determinants, has been used to differentiate E. coli from different sources (Parveen et al.,
2001). Different serotypes of E. coli are associated with different animal species but
many serotypes are shared among humans and animals (Bettelheim et al., 1976), even
though this overlap is not significant. The need for a large bank of antisera and the use of
this method in conjunction with others to get valid results are the limitations to this
approach (Parveen et al., 2001).

Recently, researchers used molecular methods such as DNA fingerprinting techniques
to differentiate between human and animal source of fecal contamination of water. Such
methods include ribotyping (Samadpour and Chechowitz, 1995; Hartel et al., 1999;
Parveen et al., 1999; Carson et al., 2001; Hartel et al., 2002 and Troy et al., 2003), PCR
of repetitive DNA sequences (Dombek et al., 2000), pulsed field gel electrophoresis
(Kariuki et al., 1999; Parveen, et al., 2001), and 16S ribosomal DNA length heterogeneity
PCR with terminal restriction fragment length polymorphisim (Bernhard and Field,
2000). These methods are expensive, labor intensive and need special experience in
molecular era. Other researchers used chemical methods, for example, looking for
caffeine in water since the human is the major consumer of this product through coffee,
tea and other beverages (Troy et al., 2002), or searching for coprostanol, which is the
fecal stanol that formed during the catabolism of cholesterol by the gut bacteria of

humans. It is considered as a chemical indicator of human fecal pollution (Leeming et

84

al., 1996; Chan et al., 1998). The major limitation of this approach is that it requires an
expensive chemical analysis.

Multiple antimicrobial resistance (MAR) approach is another way to identify the
source of fecal contamination in water. The principle behind it is that bacterial flora in
the gut of humans and animals are subjected to different types, concentrations, and
frequencies of antimicrobial agents. Over time, selective pressure will select resistant
bacteria that posses specific fingerprints against the antibiotics that have been used (Troy
et al., 2002). Several attempts have been made to compare patterns of antimicrobial
resistance in fecal coliforms with the source of isolates. Krumperrnan (1983) showed that
multiple antibiotic resistance index of E. coli from wild animals (low risk) was generally
low while human and poultry E. coli isolates had higher MAR indices (high risk),
suggesting that multiple antibiotic resistance E. coli exist in large numbers within the
major reservoirs of enteric diseases for humans, while present in low number elsewhere.
Kaspar and Burgess, (1990) demonstrated that urban waters have higher percentage of
resistant E. coli strains than rural waters, and antibiotic resistance E. coli may offer an
index of water quality related to source of fecal contamination. The MAR profile is used
to distinguish between E. coli that comes from point sources such as industrial, municipal
effluents, and meat processing plant wastes; and non-point sources such as soil erosion
and runoff over a wide area of land. A study by Parveen et al., (1997) found that more
than 80% of E. coli strains isolated from municipal waste, river and estuarine water
showed antibiotic resistance. The MAR test is relatively simple, cost effective and

suitable for surveillance since a simple technique (Kirby-Bauer method or disc diffusion

85

test) is employed. This technique is easily standardized so that results will be highly
reproducible across laboratories.

While multiple antimicrobial resistance patterns in E. coli can differ depending on the
source of E. coli, more research is needed in techniques that will be accurate in separating
these MARS into the three major potential sources of water fecal contamination,
specifically, domestic animals, wildlife and human. One method that has shown promise
is the discriminant analysis.

Discriminant analysis (DA) or discriminant function analysis is a multivariate
statistical method concerned with separating sets of observations and allocating new
observations to previously defined groups (Tatsuoka, 1970; Morrison, 1990; Johnson and
Wichem, 1992). The main purpose of discriminant analysis is to determine which
variables discriminate between two or more naturally occurring groups and to classify
cases into the values of categorical dependent groups (Johnson and Wichem, 1992;
Tatsuoka, 1970; Morrison, 1990). In our study discriminant analysis can be used to
analyze the data of the isolates from known source libraries and generate a classification
rule which then can be used to classify E. coli from unknown sources (water) into the
closest source class in the database. It has been used successfully to classify fecal
streptococcus, fecal coliforms and E. coli isolates based on the sources. Wiggins (1996)
first demonstrated the use of antimicrobial resistance patterns in fecal streptococci and
discriminant analysis to differentiate between human and animal sources with more than
90% correct classification and 84% correct classification when 6 species population were
being classified. Other investigators used this approach successfully to differentiate

human versus animal source of fecal contamination in water, which help in the direction

86

of water quality improvement (Hagedom et al., 1999; Parveen et al., 1999; Wiggins et al.,
1999; Harwood 2000; Carson 2001;Grave et al., 2002; Guan et al., 2002). Discriminant
fimction analysis of antimicrobial resistance profile can offer a very low cost and
statistically valid means of identifying the most probable species as a source for fecal
contamination of surface water compared with new molecular methods of source
identification such as RNA ribotyping and pulsed field gel electrophoresis (PFGE). The
use of individual animal sampling, including domestic and wild animals, would minimize
the risk of misclassification of E. coli isolates from fecal samples resulting in a more
accurate discriminating function.

The overall goal of our study is to apply the discriminant analysis technique to classify
E. coli isolates from a known source (e.g. cattle, pigs, sheep, etc.) by creating a
classification rule on the basis of their patterns of antimicrobial resistance, and then to
classify E. coli isolated from surface water contaminated with unknown source on the

basis of the patterns of the known isolates.

Hypothesis tested
Application of the discriminant analysis technique on antimicrobial resistance profiles
of E. coli isolated fiom domestic animals, wildlife (wild geese), and human septage is an

accurate method for determining the source of fecal contamination of surface water.

87

Objectives

The objectives of this study were: 1) To determine the accuracy of discriminant analysis
as a multivariate statistical technique to patterns of antimicrobial resistance of E. coli
isolated from different species of animal’s fecal samples and human septage.

2) Use this method to identify the source of E. coli contamination of surface water in Red

Cedar Watershed in Michigan.

Materials and Methods:
Study design

A repeated cross-sectional study design approach was used to collect the animal and
human fecal, and water samples and all the data related to antibiotic use on the farm
during the four seasons (Spring, Summer, Fall, and Winter), starting in the winter of 2002
and end in the winter of 2003.
Study area

The Red Cedar Watershed was chosen as the study area, encompassing an area of
293,000 acres and takes in portions of 16 townships in Ingham and Livingston counties
of the lower part of Michigan (figure 2-1). (Please notice that images in this thesis are
presented in color). The Red Cedar River arises in Cedar Lake in the south central portion
of the lower peninsula of Michigan and flows about 45 miles to its confluence with the
Grand River in the city of Lansing. The Grand River empties into Lake Michigan, which
connects to the other great lakes, the water of which collects in the Atlantic Ocean. The
Red Cedar River has 12 tributaries and drains and provides mid-Michigan residents with

numerous recreational opportunities, which include angling, canoeing, kayaking,

88

photography and bird watching. The river also serves as a source of water for the
irrigation of crops throughout the watershed (figure 2-2). Swine and dairy are the
predominant livestock in this watershed.

The study was conducted into two parts. Part one consisted of determining
antimicrobial resistance profiles of E. coli isolated from fecal samples of animals and
humans. The second part consisted of determining antimicrobial resistance in E. coli

isolated fi'om surface water.

Part I: Antimicrobial resistance profiling of E. coli from fecal specimens

Study population, enrollment of participating farmers, sample size, collection of fecal
samples and data regarding antimicrobial use, isolation and identification of E. coli from
fecal samples and antimicrobial susceptibility testing using Kirby Bauer method as
described in chapter 2. We used the following seven antimicrobial agents in this study
(neomycin, streptomycin, trimethoprim-sulfamethoxazole, tetracycline, ampicillin,

cephalothin, and sulfisoxazole) rater than twelve.

Part II: Antimicrobial resistance profiling of E. coli from water samples

Water sample collection, isolation and identification of E. coli from water samples and
antimicrobial susceptibility testing using Kirby Bauer method as described in chapter 2,
except that we used seven (neomycin, streptomycin, trimethoprim-sulfamethoxazole,
tetracycline, ampicillin, cephalothin, and sulfisoxazole) antimicrobial agents in this study

rater than twelve.

89

Data analysis

Data were entered into an Excel spreadsheet. F isher's exact test was conducted to
identify significant differences in the rate of resistance to each antimicrobial agent and in
the rate of multi-drug resistance by subpopulation, distributional characteristics and tests
for homogeneity of covariance matrices, and multiple analysis of variance (MANOVA)
were conducted (SAS v. 8.2, SAS Institute, Cary, NC). Those antimicrobial agents with
a statistically significant difference in the percentage of resistance for one or more sub-
populations at p<0.05 level were presented to the discriminant function model for
multivariate analysis. The data were analyzed by SAS v. 8.2 (SAS institute, Cary, NC) by
using the procedure DISCRIM. The nonparametric option was used for the development
of the discriminant function model because the data were not distributed in a multivariate

normal fashion and the covariance matrices were not homogeneous.

Assessing the validity of the discriminant analysis model

The classification table produced by discriminant analysis from the library of known
source E. coli isolates was used to calculate the percentage of misclassified isolates and
determine the average rate of correct classification. The table is a source-by-source
matrix, in which the number and percentage of correctly classified isolates are found on
the diagonal. The ARCC was computed by averaging the percentages of correctly
classified isolates for each source along the diagonal. The percentage of misclassified
isolates for a given source was determined by adding the percentages of misclassified

isolates in the appropriate row of the table, excluding the value in the diagonal (Wiggins,

90

1996, Harwood et al., 2000). Once an acceptable classification rule is developed,
discriminant analysis applies that rule to a set of isolates from an unknown source species
(in this case the water samples) against the database of known source isolates and then
classifies each water isolate into the most probable source species population.

Two techniques were applied to the development of the discriminant function
classification rule: resubstitution and cross-validation. The default method for the
software is the resubstitution method. With the resubstitution method each isolate is
classified based on the patterns in the entire dataset including its own pattern. As a result,
the ARCC from this method may overestimate the validity of the classification rule
because the exact same dataset is used for both development of the rule and evaluation of
its accuracy. The most conservation approach is to select the option to use the cross-
validation method. This method of developing the classification rule is called the leave-
one-out method. An individual isolate is removed from the dataset one at a time. The
classification rule is developed fiom the remainder of the dataset and then the removed
isolate is classified based on the rule created by n-l observations.

The known species source library was stratified in two ways. The first consisted of 5
separate sub-populations: cattle, pigs, poultry, sheep, and wildlife (wild geese). The
second consisted of only two subpopulations domestic animals (cattle, pigs, poultry,
sheep) in one population and wildlife (wild geese) in another. Isolates from pets, and
humans were not used to develop the discriminant analysis rule. Low rates of isolation of
biochemically confirmed E. coli from these samples resulted in sample sizes that were
too small for inclusion in the analysis. Discriminant fimction analysis is very sensitive to

imbalances in sample size for the population being analyzed.

91

Results

A total of 2,292 fecal and septage samples were collected. The patterns of
antimicrobial use reported by the study participants are detailed in table 2-2 (shown in
chapter 2). The results of the Fisher's exact test, for the selection of antimicrobials to
include in the multivariate analysis are detailed in table 2-3. Based on these findings the
antimicrobial resistance profiles for the following drugs were used to develop the
discriminant function: neomycin, streptomycin, sulfamethoxazole, tetracycline,
ampicillin, cephalothin, and sulfisoxazole. Data from 1169 biochemically confirmed E.
coli were used to develop the discriminant function model. The MANOVA tests the null
hypothesis that the observations are all from the same population. Failure to reject the
null hypothesis of the MANOVA would indicate that there is not adequate differentiation
between the two populations to conduct discriminant analysis. The results of the
MAN OVA found a significant difference in population distributions for both the 5-
species stratification scheme and the two-species stratification scheme (Wilk's lambda
<0.001). The data were then presented to the discriminant function model. Tables 3-1
and 3-2 detail the average rate of correct classification obtained using the two methods of
classification rule creation. Table 3-3 shows the comparison of the average rate of
correct classification (ARCC) values between re-substitution and cross-validation rule
development for different animal classification systems. The average rate of correct
classification for the 5-source population classification scheme was unacceptably low for
application to the water samples. However, for 2 sources classification 73.1% of E. coli

isolated from water samples were classified as domestic including livestock (cattle, pig,

92

sheep) and poultry, and 27.0% as wildlife as shown in table 3-4.

93

 

 

snow 2;»

coumommmmflo 82.80 mo 8am own—03V. NOON: .

 

«.mmubom... asthma... $3 .38.
4.9. we 3 e: 2m 32 as as ca 2 8 age;
«.2 SN 3: gm 0: we as as 0.3 as :N Seam
3 a: «.9. SN 3 3 3 ma 3 ca 2: EBB
3 0.2 ZN mam 4.2 as as a: 3w so am as
as as ea 3: 0.: 2: S: a. we 2e 1% memo

.0222», Seam $38 ea 2:8 SEE Seam .338 am 2:8 2 828%

3% 3%

 

momma—o 1:55 25 ..8 .3232: 23592966 0:... 320.5“. 95 win: £33853
.Efiefiaaoo ionaxefiofiaaa .5539: .05—05958 .fiozfieunohm ESE—85 85o...— oogmmoea Engages
we arm—2:. Esau—5.8:. 95089895: .3 @3935» 8:... 5.635%“? .8.“ 833 2535880 no Sauna—:80 “Tm «Ben.

94

Table 3-2: Comparison of classification tables for classification rules generated by
non-parametric discriminant analysis of antimicrobial resistance profiles
(neomycin, streptomycin, tetracycline, ampicillin, sulfamethoxazole, cephalothin,
sulfisoxazole) using two different rule development methods, for two animal classes

 

 

 

Re-substitution Method Cross-validation Method
Species N Domestic Wildlifea Domestic Wildlifea
Domestic 1,070 93.5 6.5 87.8 12.2
Wildlife“ 74 0.0 100.0 66.2 33.8
Total 1,144 ARCC‘ = 96.8 ARCC‘ = 60.8

 

" ARCC = Average Rate of Correct Classification
3’ Wild geese
Table 3-3: Comparison of average rate of correct classification (ARCC) values

between re-substitution and cross-validation rule development for different animal
classification systems

 

 

Animal classification Re-substitution Cross-validation
ARCC (%) ARCC (%)

Cattle, Pig, Poultry, Sheep, Wildlifea 80.6 38.4

Cattle, Pig, Poultry, Sheep 81.1 43.5

Cattle, pig, sheep 81.0 50.4

Domestic, Wildlifea 96.8 60.8

 

3" Wild geese

95

 

Table3-4: Classification of E. call from water samples using discriminant analysis

 

Number of groups Species (% classified)

 

 

5 Cattle (23.1), sheep (19.2), pig (27.0), poultry (0.0), wildlife“(30.8)_
4 Cattle (19.2) sheep (23.1), pig (27.0), wildlife"(30.8)
2 Domestic (73.1), wildlifea (27.0)

 

a : Wild geese

Discriminant analysis using separate versus pooled sources

When five sources (cattle, pig, poultry, sheep, and wild geese) were analyzed by
discriminant analysis using the resubstitution method, based on the resistance to seven
anitrnicrobial agents (neomycin, streptomycin, trimethoprim-sulfamethoxazole,
tetracycline, ampicillin, cephalothin, and sulfisoxazole), the average rate of correct
classification (ARCC) was 80.6%. However, when the same sources and antimicrobial
agents were analyzed by discriminant analysis based on the cross-validation method the
ARCC was 38.4% (Table3-1). When cattle, pig, poultry, and sheep were pooled and
renamed domestic animals and compared with wild geese the ARCC increased to 96.8%
and 60.8% by using the resubstitution and cross- validation methods respectively (table
3-2). The high classification rates were achieved when only two sources were compared
strongly supports the conclusion that different species of animals harbor E. coli bacteria

with different patterns of antimicrobial resistance profiles.

Analysis of unknown isolates from stream water
The antimicrobial resistance profiles of a total of 26 E. coli isolates from water

samples were measured and classified by discriminant analysis (the known isolates were

96

 

used as reference); in other word, discriminant analysis was used to assign each isolate to
a source category based on the comparison of its antimicrobial resistance patterns to
those isolates from known sources. Using 5-source species (cattle, pig, poultry, sheep,
and wild geese), 23.1% of E. coli isolates from water samples were classified as cattle,
19.2% as sheep, and 27% as pig, 0.0% as poultry and 30.8% wild geese (table 3-4).
When 4-source species cattle, pig, sheep, and wild geese were analyzed, 19.2% of E. coli
isolates from water samples were classified as cattle, 23.1% as sheep, 27.0% as pig and
30.0% as wild geese (table 3-4) suggesting that among domesticated animals, pigs
contributed the greatest fecal pollution to the water stream. But when the isolates from
known sources were grouped as both domestic and wild geese, 27% were classified as
wild geese and 73.1% as domestic animals (table 34). This suggests that the water

stream polluted with animal fecal material mainly from domestic animals.

Discussion

Several researchers have used discriminant analysis of antimicrobial resistance profile
to identify fecal pollution sources (Wiggins, 1996; Hagedom et a1. 1999; Wiggins et al.,
1999; Harwood etal., 2000; Graves et al., 2002; Guan et al., 2002; Whitlock et al., 2002;
Wiggins et al., 2003). The advantage of discriminant analysis method is that it generates
a classification rule based on all the isolates, then uses this rule to classify each individual
isolates into one of many possible sources. Wiggins (1996), who first used this method,
reported that the ARCC was 84% when streptococcal isolates were pooled into four
categories (cattle, human, poultry, and wild life). Moreover, the ARCC increased to 92%

when all animals were pooled together and compared with human. However, Wiggins

97

 

 

used the resubstitution method to measure the performance of discriminant analysis in his
study and five antibiotics with different concentrations. Our findings are similar to
Wiggins’s (1996) findings based on the resubstitution method as a measure of the
performance of discriminant analysis. We found that the ARCC for 4 source species was
81.1%, for 3 source species was 81.0% and for 2 source species was 96.8% (table 3-3).
There were consistently more correctly classified isolates (higher ARCC) when
discriminant analysis was performed using two sources than when all four or five sources
were used, because there are fewer possible categories. ARCC reflects both the false
negative from that source and false positive from other sources (Wiggins, 1996); it is a
better measure of the ability of a given analysis to classify the isolates. In addition,
Wiggins (1996) tested a total of 1,435 isolates from 17 samples from cattle, poultry,
human and wild life (multiple isolates from the same sample) leading to increase
similarity of isolates within samples (isolates from same sample might have similar
patterns) than similarity between samples which would make the library seem more
representative than it actually is. Such an approach leads to high ARCC. In our study,
we overcame this issue by using one isolate from one sample and not many isolates from
the same sample that might have similar antimicrobial resistance patterns. We tested a
total of 1169 E. coli isolates from 1169 fecal samples from different animal species. The
drawback of sample level analysis is the assumption that all of the isolates in a given
sample are from the same source, and this is valid for known homogeneous samples but
not for a sample which was contaminated by more than one source (Wiggins et al., 1999).
In a recent study, Wiggins et al., (2003) avoided this problem by using pulled-sample

cross-validation analysis, where he removed all of the isolates from each sample and then

98

classified them against the remaining isolates. By comparing the difference between the
ARCC of the resubstitution analysis, and the ARCC of the pulled-sample analysis, the
representativeness of the library can be estimated. If the difference is small (less than
5%) then the library can be considered representative. In other words, new isolates are
classified almost as well as the isolates in the library, but in this a case large number of
isolates is needed to be in the library. Hagedom et a1. (1999) followed Wiggins’s steps
but he used a larger sample size of 7,058 fecal streptococcus isolates fi'om humans,
livestock, and wildlife (3 sources species) and demonstrated that the ARCC was 87%.
Combining all animal isolates and comparing it with humans (2 source species) increased
the correct classification rate to more than 95%. These results are similar to ours when we
used the resubstitution analysis (ARCC for 3 species was 81.0% and for 2 species was
96.8%). Harwood et a1. (2000) used discriminant analysis of antimicrobial resistance
profiles but in his study using two separate datasets one fiom fecal streptococcus (4,619
isolates) and the other from fecal coliform (6,144 isolates) isolated from wild bird, cattle,
chicken, dogs, pigs, and raccoons. He used nine antimicrobial agents to determine which
one is the best to classify and identify the source of surface water contamination. He
found that fecal streptococcus and fecal coliform databases classified isolates from
known sources with similar accuracies, and the ARCC for fecal streptococcus was 62%
and for fecal coliform was 63.9% by applying discriminant analysis on two categories
animal and human using the cross-validation analysis. These results are similar to ours;
when we applied the cross-validation discriminant analysis for two categories only

(domestic animals and wild geese) and found the ARCC was 60.8%.

99

Guan et al., (2002) applied discriminant analysis on a total of 319 E. coli isolates
using 14 antimicrobial agents and found that the ARCC when the data were pooled into
three categories (human, livestock, and wildlife) was 64.5%, this low ARCC comparing
with Wiggins and Hagedom results may be due to several factors: first, fecal
streptococcus was investigated by Wiggins and Hagedom rather than E. coli so that the
antimicrobial resistance profile in E. coli bacteria will be different from that in
Streptococcus. Second, different antimicrobial agents were used, and the method of
performing the antimicrobial susceptibility test was different between study by Guan et
al., and studies by Wiggins (1996) and Hagedom et a1. (1999) and third, in Guan’s study
samples were collected fi'om animals over a wide geographical area leading to more
heterogeneous collection of bacterial isolates than other studies sampling protocol. When
we applied the discriminant analysis on all E. coli (1169 isolates) from cattle, pigs,
poultry, sheep and wild animals the ARCC was 38.4% based on the cross-validation
results of discriminant analysis performance. Our results were similar to the results
obtained by Guan et al., (2002) when applied discriminant analysis on multiple
antimicrobial resistance (MAR) profile of E. coli isolated from nine host groups (human,
beef, dairy, chicken, pig, turkey, deer, goose, moose); the ARCC was 33.9% when the
cross-validation method was used.

Over time, as patterns of antibiotic use change, so do bacterial patterns of antibiotic
resistance, and because the selective pressure of antibiotic treatment on microflora of
animals and humans is an important determinant of the prevalence of antibiotic resistance
in a population, the database that developed for discriminant analysis will require

periodic updating (Harwood et al., 2000). The other issue is that it is not known if the

100

discriminant analysis of antimicrobial resistance profile fi'om one geographical location
can be used to predict the source of isolates from another since the patterns of
antimicrobial agents may vary from one geographical area to another.

There are several issues that should be taken into consideration in order for the dataset
or library to reliably identify fecal sources. First, the library needs to be representative of
the sources that are present in the watershed. In other words, it should contain examples
of all of the patterns found in the bacteria from each of the sources types that are found in
the watershed with attention to the choice of antimicrobial agents used and variability of
animal husbandry practices in different regions. Second, the library should be able to
classify isolates from other geographical areas, and should be stable overtime so that new
libraries do not need to be continually created. Wiggins et a1. (2003) were able to obtain
a representative, temporally stable merged multi-watershed library (6,587 enterococci
isolates), from six Virginia Watersheds. When isolates fiom the contributed watersheds
approximately one year later were analyzed with this library, they were classified as well
as the isolates in the library suggesting that the resistance patterns are temporally stable
for at least one year, but more samples will be needed to determine the extent of that
variability (Wiggins et al., 2003).

Wiggins et al., (2003) found that the larger the multi-watershed library the lower the
ARCCs and this is due to variability in the resistance patterns of the isolates within each
source type in the watershed. The more isolates of each source type that are contained in
the library (the more representative it is) the greater the chance they will vary in their
resistance patterns, which would result in lowering the classification success. If there is

variability in the patterns in a watershed, then the ARCC of a small library could be

101

misleading because it would be unable to classify the large number of unknown isolates
that have a pattern that is not included in it. Therefore, it is unwise to relay on the ARCC
of a library without also knowing its representativeness (Graves et al., 2002).

The best way to determine if the library is representative is to regularly add samples of
known sources to it, if the ARCC and or the individual correct classification do not
change significantly as new samples are added then the library should be representative
(Hagedom et al., 1999; Graves et al., 2002)

Discriminant analysis can assign each unknown isolate to one of the known sources
according to the baseline data and the unknown organism’s resistance pattern. The
ability of the library to predict the source of the unknown isolates should be determined
by assessing the classification accuracy of isolates from known sources that are not
included in the library (Whitlock et al., 2002).

When we classified the E. coli isolates fiom the water samples (unknown) based on 5
categories (cattle, pigs, poultry, sheep and wild geese) 23.1% of E. coli isolates from
water samples were classified as cattle, 19.2% as sheep, and 27% as pig, 0.0% as poultry
and 30.8% wild geese (table 3-4). These results suggested that E. coli in the water
samples came from animal fecal material mainly wild geese. When combining cattle,
pig, sheep and poultry as domestic animals then compare the domestic animals with wild
geese, 73.1% of the E. coli isolated from water samples were classified as domestic
animals and only 27.0% classified as wild geese, this could be due to the similarity of
antimicrobial resistance profile of E. coli isolated from water and the wild geese. The E.
coli isolated from both the water samples and wild geese samples were resistant to

cephalothin.

102

Although there is no established standard of accuracy that has been defined for any
bacterial source tracking method, any method with a correct rate of classification of over
50% when there are five or more possible source categories, has been considered as a
worthwhile tool for predicting the potential sources of fecal pollution in environmental
water (Harwood et al., 2000). Based on this, and our results, the resubstitution
discriminant analysis method gave higher ARCC (80.6%) than cross-validation
discriminant analysis (38.4%) when 5 possible source categories were analyzed. This is
not surprising since the same dataset that was used to create the classification rule was
used to test it in the resubstitution analysis. Our findings support the work done by
Wiggins et a1. (2003) where he found that the ARCC based on the resubstitution analysis
was higher than the cross-validation analysis.

The water quality authorities are interested in first, discriminating between human and
animal sources of fecal pollution; second, determining the major sources of animal’s
contamination. It would be desirable to identify which type of animal was causing the
pollution especially when the streams run through mixed agricultural areas. The results
obtained by Hagedom et a1. 1999 study were very helpful for the watershed authorities in
which cattle access to the river was prohibited by installation of fencing, and in- pasture
watering stations, which lead to a reduction in fecal coliforms counts from 15,900 per
100ml to 960 per 100ml.

Antimicrobial resistance analysis requires several days to obtain results, because
growth of bacteria is required to apply the susceptibility test. This time period may be
critical for public health officials who need to make rapid decisions on closure of

recreational water contaminated by fecal material. However, the antimicrobial resistance

103

test is relatively inexpensive, and technicians can quickly be taught how to perform the
assay and apply it on several hundreds of isolates per week. Comparing with the
molecular methods that is expensive, needs experience, and could only be performed on a
small number of isolates.

Although we did not have enough samples fiom human source to be classified, E. coli
could be classified well since the antimicrobial resistance profile of E. coli isolated from
human septic tanks was similar to that from wild geese and water. In addition, sampling
the septic tanks rather than treatment plant has an advantage because it is more likely that
influent in many plants could have included overland flow fi'om agricultural land, which
should have introduced contaminating bacteria from animal sources and thus reduced the
classification success.

In the future, it will be worthwhile to combine discriminant analysis of antimicrobial
resistance profiles with that for molecular methods, the ribotyping, on a large dataset to
cross validate both approaches and to assess wither one method might be more suitable
than the other.

In conclusion, the results of our study suggest that discriminant analysis of
antimicrobial resistance profile could be used to differentiate among isolates from several
sources of fecal pollution. It provides a strong method for classifying and identifying
fecal E. coli and will serve to help identify the non-point sources of fecal pollution in
water, which will aid in the evaluation of risk to the public health. The underlying
assumption of antimicrobial resistance analysis is that the differential use of antimicrobial
agents in humans, domestic animals or in wild animals lack thereof allows for

discrimination of indicator organism isolates from those hosts. The use of resubstitution

104

method (each isolates is classified based on the patterns in the entire library, including its
own patterns) as a measure of the performance of discriminant analysis gave higher
ARCC than the more conservative method, the cross-validation (an individual isolate was
removed from the library one at a time, then the removed isolate classified based on the
library comprised of the remaining isolates). The ARCC for 2 categories is higher than
the one of more than two in both methods.

Moreover, our study not only classify and identify the sources of fecal contamination
in water but also provided us with information about the antimicrobial agents that E. coli
isolated fiom different animals species in the watershed showed resistance. This
information would be useful for farmers and veterinarians to know for future use.

The size of the library is critical to the success of discriminant analysis in predicting the
sources of fecal contamination in water. It has been demonstrated that small libraries
generally give higher correct classification and this due to under sampling of the true
population diversity in fecal material (Wiggins et al., 1999; Harwood et al., 2000).

The ability of the library to predict the source of indicator organisms should be
determined by assessing the classification accuracy of isolates from known sources that
are not included in the library (Whitlock et al., 20002). In order to classify a new isolates
from another geographical area the library used should be representative, large and
temporally stable for at least one year. In order for discriminant analysis to be valid it
should show its applicability in the field rather than simply showing high correct
classification rates.

The sampling protocol should reflect the type of the watershed under study; for

example, companion animals could be added as a category for more urban watersheds,

105

and wildlife removed if necessary. On the other hand, sampling should be taken more
from domestic animals such as cattle, pigs, poultry, sheep and wild animals in addition to
human in rural watershed. By doing that, the samples will be representative to the type of
source species that live in the watershed.

Knowing the sources of fecal pollution in water is a very important step in
determining the degree of risk for humans exposed to contaminated water, assessing the
development of best management practices to reduce the fecal loading which include
stream fencing, establishing riparian-zone buffers, installing in pasture watering stations,
and improving waste treatment facilities and amending leakage in a septic tank.

We concluded that antimicrobial resistance analysis of E. coli isolates analyzed with
discriminant analysis was a suitable method to differentiate and identify sources of fecal
pollution in Red Cedar River where the classified isolates came from multiple sources, if
our results based on taking the resubstitution method as a measure of performance of
discriminant analysis, but this method is biased and it is recommended to take cross-

validation method into consideration.

106

OVERALL SUMMARY AND CONCLUTIONS

A repeated cross—sectional design was used to conduct studies to investigate whether
combining two techniques (determining antimicrobial resistance profiles of E. coli, and
conducting discriminant function analysis on such profiles) would be useful in
identifying sources of fecal contamination of surface water in Michigan. The Red Cedar
watershed was used as a case site. Fecal samples were collected from livestock and
companion animals living on farms that drained into the Red Cedar watershed, human
septic tanks from homes in the watershed, and wildlife living in the watershed. Water
from the Red Cedar was collected from several different sites for the study. The study
was conducted in two parts, presented here as chapters.

There were two objectives for the first part of the study (Chapter 2): 1) determining
patterns of antimicrobial resistance in E. coli obtained from human, domestic animals,
and wildlife living in the Red Cedar watershed in Michigan, USA; and 2) comparing
patterns of antimicrobial resistance from these animals with patterns found in E. coli
obtained from surface water samples from the same watershed. Using disc diffusion
assay, the following antimicrobial agents were tested for susceptibility: neomycin,
gentamicin, streptomycin, chloramphenicol, ofloxacin, trimethoprim/sulfamethoxazole,
tetracycline, ampicillin, nalidixic acid, nitrofurantoin, cephalothin, and sulfisoxazole. The
study found:

1. Resistance to at least one antimicrobial was demonstrated in isolates from food

animals, wildlife, surface water, and human septic tanks.

107

2. In general, E. coli isolates from food animals showed resistance to the largest
number of antimicrobial agents, followed by E. coli from horses, companion
animals, human septic tanks, wildlife, and surface water.

3. The three antimicrobial agents to which resistance was demonstrated most
frequently in this study were cephalothin, tetracycline, and streptomycin.

4. Further, the antimicrobial resistance profiles of E. coli isolated from companion
animals and human septic tanks were closer to those from surface water isolates,
than from food animal species.

5. Within the groups tested, food animal species (cattle, sheep, goats, poultry and
swine) manifested the highest rates of antimicrobial resistance for all but 4 of the
individual antimicrobials tested (gentamicin 0.7%, sulfamethoxazole 1.7%,
ampicillin 7.2%, and cephalothin 21 .8%). Additionally, food animals had
statistically significantly (F isher's exact p <0.0001) higher rates of resistance to 3
or more antimicrobials (13.5%).

In the second part of the study (Chapter 3), the objectives were: 1) to test the validity
of discriminant function analysis of antimicrobial resistance profiles to identify the
source of fecal E. coli isolates, and 2) use this method to identify the most probable
sources of fecal E. coli contamination of surface water in the Red Cedar watershed in
Michigan. After conducting statistical analyses to determine which antimicrobial agents
had the greatest probability of resistance, two methods for conducting discriminant
function analysis (resubstitution and cross-validation) were applied to the data. The major

findings from this part of the study include the following:

108

1.

Our studies agree with the literature, in that the resubstitution method is biased, it
produces higher, but invalid ARCC. In contrast, the cross-validation method
produces lower but valid ARCC.

Using the cross-validation method, it was found that the ARCC increased as the
number of source categories was reduced. As an example, the ARCC for five
source categories was 38.4%, compared to 60% when two source categories were

used.

Recommendations

The use of discriminant function analysis of antimicrobial resistance profiles to

determine the source of fecal contamination of surface water has shown great promise,

but some limitations still remain. For this method to serve as a very useful tool future

studies should include:

1.

2.

Using a larger number of samples from human and animal sources.

Use fewer numbers of antimicrobial agents during the development of the
classification rule.

Evaluate the use of MICs instead of binary outcomes (resistant or not resistant) in
discriminant analysis.

Evaluate additional statistical methods, such as cluster analysis, prior to
conducting the discriminant function analysis to reduce the number of variables
being assessed.

Compare the performance of discriminant analysis to ribotyping and other

molecular techniques.

109

APPENDIX

110

Surface Water Study
Informed Consent Form

The overall goal of this study is to identify various sources of surface water
contamination. To complete this goal, the study will develop a low-cost screening tool
for the idartification of the source species for fecal contamination of surface water, and
then identify factors associated with fecal contamination of surface water by comparing
bacteria from surface water with bacteria collected from deer, waterfowl, pets, household
septic systems, and livestock

A data collector from MSU will contact you, and administer a questionnaire designed to
describe some of the different livestock and manure management practices being used by
Michigan livestock operators. This questionnaire will be completed through and in-
person interview. The interview itself should last less than half an hour. A sample
collector from MSU will schedule an appointment to come to your farm to collect manure
samples from manure storage wits and a small percentage of your livestock. These
samples will be sent to Michigan State University for laboratory testing, to compare the
bacteria fiom your livestock to those bacteria found in local bodies of water.

All information collected by this study will be kept confidential: n_o identification will be
kept that will tie your identity with questionnaire and laboratory test results. Only
researches immediately involved with the study will have access to these data: no outside
private or governmental groups will be able to use these data. Only summaries of the
data will be used to generate reports, and informtion will be reported so that it will not
be possible to identify any specific individual participating in the study. Your privacy
will be protected to the maximum extent allowable by law.

Again, your participation in this study is vohmtary: refirsal to participate will involve no
penalty, and you may discontinue participation at any time.

 

Participant Data Collector

 

Date Date

To discuss any questions regarding the research, please contact:
Dr. John B. Kaneene

Phone- (517) 353-5941, Fax (517) 423-0976

Email-

To discuss questions about your role as a subject of research, please contact:

Dr. Ashir Kumar, Chairman of the miversity Committee on Research Involving Human
Subjects, MSU (517) 355-2180

111

January 30, 2002
Dear Michigan Farmer:

The college of Veterinary Medicine at Michigan State University will be conducting a
study to determine the source of bacterial contamination of surface water. Bacterial
surface water contamination, such as E. coli, can be from a number of sources, including
non-agricultural sources. Your participation will help us to identify these sources

We would like to ask for your cooperation in part of this study. Ifyou agree, your
participation in the study will be in two steps. First, you will be asked to complete a short
questionnaire. This can be done on your farm or at your local cormty Extension office.
After the questionnaire is completed, a sample collector item MSU will schedule an
appointment with you, to come to your firm to collect manure samples fiom your manure
storage units and a small percentage of your livestock. These samples will be sent to
MSU for laboratory testing, to compare the bacteria types from your livestock to those
bacteria found in local bodies of water.

All information collected will be kept grim confidential. Ne identifying information
about your specific operation will be collected during the study, so it cannot be released
to any governmental agency. All results of the study will be presented as summaries, so
that no specific information about your farm will be identifiable. If you would life to see
the results of bacterial testing for your operation, your specific test results will be
released M to you upon your request through your county Extension agent.

Your participation is optional and voluntary. Ifyou would like to help with our research
by participating in this study, please fill out the enclosed, pro-stamped post card with
your name and contact information. We will work through your county Enension agent
to schedule a time for you to come to the office to ask any questions you have, sign an
informed form, and complete the questionnaire.

Ifyou have any questions or concerns about the study, or your participation in the study,
please contact Dr. John B. Kaneene at (517) 353-5941.

Thank you for your time and assistance.

John B. Kaneene, DVM, MPH, PhD
Professor of Epidemiology and
Director, Population Medicine Center

112

March 12, 2002

Dear Dairy & Livestock Producer:

Recartly, you received an invitation to take part in a study regarding E. coli and
livestock The study is being performed by the College of Vetemiary Medicineat
Michigan State University. The E. coli bacteria can come from waterfowl, septic
systems, and many other types ofanimals The research will attempt to link E.coli strains
with firm animal types. In part, this research will help us protect the animal industry in
Michigan from being associated with E. coli contamination in surface water, when it may
be from non-agricultural sources.

We have heard fi'om a number of Ingham County producers so fir, and all of these
produces who have livestock have indicated their desire to participate. However, we
have not yet heard fiom you.

Of course, your participation is completely optional and voluntary. If you would

like to participate in this study, please out the enclosed, pre-stamped post card
with your name and contact information. If not, please indicate on the post card

that you are not interested and return it. By doing so, we will remove your name

from the list for any further reminders.

If you have any questions or concerns about the study, or your participation in the study,
please contact Dr. John B. Kaneene at (517) 353-5941. The attached letter and form will
explain the study.

Thank you for your time and assistance.

Sincerely

Mark F. Hansen
Extension Coordinator, and
Extension Agriculture & Natural Resources Agent

Cc: Dr. John B. Kaneene
Dr. Raida Sayah

113

March 25, 2002

Dear Agricultural Producer:

The College of Vetemiary Medicineat at Michigan State University is performing a study
that involves E. coli and livestock waste. Many people are not aware that E. coli bacteria
can come fi'om different sources inchrding livestock, birds, pets, wildlife and humans.

The goal of this study is to identify strains of E. coli to better understand the sources of E.
coli, which are found in surfice wares in the mid-Michigan area.

As your Extension Agriculture Agent, I have agreed to send out this information. In this
way, we were able to help the research project without revealing your name.

You have the opportunity to participate in this research project. Your participation is
completely optional and voluntary. Information colleted will be kept confidential. This
study will further our knowledge of bacterial contamination of surfice water, which in tur
help us demonstrate agriculture’s quest for environmental quality. The research team has
enclosed a cover letter, consent form and return postcard

Ifyou have questions, please call me at (517) 546-3950. You may also call the research
leader, Dr. John B. Kaneene directly at the College of Veterinary Medicine, for more
study details Dr. Kaneene on be reached at (517) 355-5941.

Sincerely,

Betsy Dierberger
Extension Agriculture and Natural Resources Agent

114

REFERENCES

115

REFERENCES

Ackrnan, D., S. Marks, P. Mack, M. Caldwell, T. Root and G. Birkhead. 1997.
Swimming-associated haemorrhagic colitis due to Escherichia coli 0157:H7 infection:
evidence of prolonged contamination of fresh water lake. Epidemiol. Infect. 119, 1-8.

Alekshun, M. N. and S. B. Levy. 1999. The mar regulon: multiple resistance to
antibiotics and other toxic chemicals. Trend Microbiology. 71410-413.

American public health association. 1998. Standard Methods for Examination of Water
and Wastewater, 20th ed. Clesceri, Greenberg, And Eaton (ed.), American Public Health
Association; American Water Works Assn; and Water Environment Federation.
Washington DC Methods, Number 9225.

Anderson, J. D. 1974. The effect of R-factor carriage on the survival of Escherichia coli
in the human intestine. J. Med. Microbiol. 7:85-90.

Anderson, T. W. 1951b. Classification by multivariate analysis. Psychometrika. 16:31-
50.Application of sedimentary fecal stanols and sterols in tracing sewage pollution in
coastal water. Water Res. 32:225-235.

Ash, R. J ., B. Mauck, and M. Morgan. 2002. Antibiotic resistance of gram-negative
bacteria in rivers, United States. 8(7):713-716.

Barton, M. 1998. Does the use of antibiotic in animals affect human health? Aust. Vet.
Journal. 76(3): 177-1 80.

Bernhard, A. E., and K. G. Field. 2000. Identification of nonpoint sources of fecal
pollution in coastal waters by using host-specific 16S ribosomal DNA genetic markers
fiom fecal anaerobes. Applied and Environmental Microbiology. 66(4):1587-1594.

Bettelheim, K. A., N. Ismail, R. Shinebaurn, R. A. Shooter, E. Moorhouse and S.
O’Farrel. 1976. The distribution of serotypes of Escherichia coli in cow pats and other

animal materials compared with serotypes of E. coli isolated from human sources.
Journal of Hygiene. 76:403-406.

Blanco, J. E., M. A. Mora and J. Blanco. 1997 . Prevalence of bacterial resistance of
quinolones and other antimicrobial among avian Escherichia coli strain isolated from
septicemic and healthy chickens in Spain. Journal of Clinical Microbiology. 35(8):2184-
2185

Carson, A., B.L. Shear, M. R. Ellersieck, and A. Asfaw. 2001. Identification of fecal

Escherichia coli fi‘om humans and animals by ribotyping. Applied and Environmental
Micobiology. 67(4): 1503-1507.

116

Casewell, M., C. Friis, E. Marco, P. McMullin, and 1. Phillips. 2003. The European ban
on growth- promoting antibiotics and emerging consequences for human and animal
health. Journal of Antimicrobial chemotherapy. 52:159-161.

Chan, K. H., M. H. W. Lam, K. F. Poon, H.Y. Yeung, and T. K. T.Chui. 1998.

Crane, S. R., and J .A. Moore. 1986. A management strategy to reduce bacterial pollution
in shellfish areas: a case study. Environ. Manag. 10:41-51.

Difco Manual. 11th ed. 1998. Difco Laboratories a Division of Becton Dickinson and Co.
Sparks, MD USA. pages 527, 556, 288, 525 for isolation of E. coli . Page 521 for triple
sugar iron test.

Dombek, P.E., L. K. Johnson, S. T. Zimmerly, and M. J. Sadowsky. 2000.Use of
repetitive DNA sequences and the PCR to differentiate Escherichia coli isolates from
human and animal sources. Applied and Environmental Microbiology. 66(6):2572-2577.

Dufour, A.P., E.R. Strickland, V.J. Cabelli. 1981. Membrane filter teqnique for
enumerating Escherichia coli . Applied and Environmental Microbiology. 41:1152-1 158.

F eachem, R. 1975. An improved role for fecal coliform to fecal streptococci ratios in the
differentiation between human and nonhuman pollution sources. Water Research. 9:689-
690.

Fisher, R. A. 1936. The use of multiple measurements in taxonomic problem. Annals of
Eugenics. 7:179-188.

F uruse, K. A., S. Osawa, and I. Watanabe. 1981. Distribution of ribonucleic acid

colipghages in raw sewage fi'om treatment plants in Japan. Applied and Environmental
Microbiology. 41 :1 139-1 143.

Galland, J. C., D. R. Hayatt, S. S. Crupper, and D. W. Acheson. 2001. Prevalence,
antibiotic susceptibility, and diversity of Escherichia coli 0157 :H7 isolates from a

longitudinal study of beef cattle feedlots. Applied and Environmental Microbiology.
67(4):l619-l627.

Ginns, C. A., G. F. Browning, M. L. Benham, G.A. Anderson and K. G. Whithear. 1996.
Antimicrobial resistance and epidemiology of Escherichia coli in broiler breeder chicken.
Avain Pathology. 25:591-605.

Grabow, W. O. K., I. G. Middendroff, and O. W. Prozesky. 1973. Survival in maturation

ponds of coliform bacteria with transferable drug resistance. Water Research. 7: l 589-
l 597.

117

Grabow, W. O. K., M. Van Zyl, and O. W. Prozesky. 1976. Behavior in conventional
sewage purification process of coliform bacteria with transferable or non-transferable
drug- resistance. Water Research. 10:717-723.

Graves, A. K., C. Hagedom, A.Teetor, M. Mahal, A. M. Booth, and R. B. Reneau, Jr.
2002. Antibiotic resistance profiles to determine sources of fecal contamination in a rural
Virginia watershed. Journal of Environmental Quality. 31:1300—1308.

Grigal, D. F. and R. A. Goldstein. 1971. An integrated Ordination —classification analysis
of an intensively sampled oak-hickory forest. Journal of Ecology. 59:481-492.

Guan, S., R. Xu, 8. Chen, J. Odumeru, and C. Gyles. 2002. Development of a procedure
for discriminating among Escherichia coli isolates fi‘om animal and human sources.
Applied and Environmental Microbiology. 68(6):2690—2698.

Hachler, H., S. P. Cohen, and S. B. Levy. 1991. marA, a regulated locus which controls
expression of chromosomal multiple antibiotic resistance in Escherichia coli. Journal of
Bacteriology. 173:5532-5538.

Hagedom, C., S. L. Roberson, J .R. Filtz, S. M. Grubbs, T. A. Angier and R. B. Reneau,
JR. 1999. Determining sources of fecal pollution in a rural Virginia watershed with
antibiotic resistance patterns in fecal streptococci. Applied and Environmental
Microbiology. 65(12):5522-5531.

Hartel, P. G., J. D. Summer, J. L. Hill, J. V. Collins, J. A. Entry, and W. I. Segars. 2002
Geographic variability of Escherichia coli ribotypes fi'om animals in Idaho and Georgia.
Journal of Environmental Quality. 31 :1273-1278.

Hartel, P. G., W. I. Segars, N. J. Stern, J. Steiner and A. Buchan. 1999. Ribotyping to
deterime host origin of Escherichia coli isolates in different water samples, P.377-382.
In D.S. Olsen and J .P. Potyondy (ed.), Wildland hydrology. Technical publication Series
TPS-99-3. American Water Resources Asssociation, Hemdon, Va.

Harwood, V. J ., J. Whitlock, and V. Withington. 2000. Classification of antibiotic
resistance patterns of indicator bacteria by discriminant analysis: use in predicting the
source of fecal contamination in subtropical waters. Applied and Environmental

Microbiology. 66(9):3698-3704.
Hix, D. M. 1988. Multifactor classification and analysis of upland hardwood forest
ecosystems of the Kickapoo River watershed, Southwestern Wisconsin. Can. J. For. Res.

18:1405-1415.

Johnson, R. A. and D. W. Wichem. 1992. Applied Multivariate Statistical Analysis, 3rd
edition. Chapter 11, pages 493-556, New Jersey: Englewood Cliffs.

118

Jones, D. L. 1999. Potential health risk associated with the persistence of Escherichia coli
0157 in agricultural environments. Soil Use and Management, 15:76-83.

Jones, I. G. and M Roworth. 1996. An outbreak of Escherichia coli 0157 and
campylobacteriosis associated with contamination of drinking water supply. Public
health (London) 110:277-282.

Jones, M. E., E. peters, A. M. Weersink, A. Fluit, and J. Verhoef. 1997 . Widespread
occurrence of integrons causing multiple antibiotic resistance in bacteria. Lancent.
349: 1742-3.

Josef, S. 2000. Computer recognition of skin structures using discriminant and cluster
analysis. Skin Research and Technology. 6(2):58.

Kariuki, S., C. Gilks, J. Kimari, A. Obanda, J. Muyodi, P. Waiyaki, and C. A. Hart. 1999.
Genotype analysis of Escherichia coli strains isolated from children and chickens living
in close contact. Applied and Environmental Microbiology. 65:472-476.

Kasper, C. W., J. L. Burgess. 1990. Antibiotic resistance indexing of Escherichia coli to
identify sources of fecal contamination in water. Can. J. Microbiol. 36: 891-894.

Keene, W. E., J. M. McAnulty, F.C. Hoesly, et a1. 1994. A swimming-assiciated outbreak
of hemorrhagic colitis caused by Escherichia coli 0157:H7 and shigella Sonnei. New
England Journal of Medicine. 331(9):579-584.

Kelch, W. J. and Lee J. S. 1978. Antibiotic resistance patterns of gram-negative bacteria
isolated from environmental sources. Applied and Environmental Microbiology.
36(3):450—456.

Klecka, W. R. (1980). Discriminant Analysis. Quantitative Applications in the Social
Sciences Series. No. 19. Thousand Oaks, CA: Sage Publications.

Kruperman, P. H.1983. Multiple antibiotic resistance indexing of Escherichia coli to
identify high-risk sources of fecal contamination of foods. Applied and Environmental
Microbiology. 46(1): 165-170.

Lachenbruch, P. A. and M. A. Mickey.l968. Estimation of error rates in discriminant
analysis. Technometrics. 10:1-10.

Lachenbruch, PA. 1975. Discriminant Analysis. NY: Hafi'ler.

Langlois, B.E., G. L. Cromwell, T. S. Stahly, K. A. Dawson, and V. W. Hays. 1983.
Antibiotic resistance of fecal coliforms after long-term withdrawal of therapeutic and

subtheraputic antibiotic use in swine herd. Applied and Environmental Microbiology.
46(6): 1433-143.

119

Lattin, J ., J. Douglas Carroll, P. E. Green. 2003. Analyzing Multivariate Data. Chapter
12, pages 426-473.

LeClerc, J. E., B. Li, W. L. Payne, and T. A. Cebula. 1996. High mutation frequencies
among Escherichia coli and Salmonella pathogens. Science. 274:1208-1211.

Leeming, R., A. Ball, N. Ashbolt, and P. Nichols. 1996. Using faecal sterols from
humans and animals to distinguish faecal pollution in receiving waters. Water Research.
Pages 2893-2900.

LeMay, V. M., Tait D. E., Van Der Kamp B. J. 1994: Classification of cedar, aspen, and
tru fir trees as decayed versus sound. Can. J. For. Res. 24:2068-2077.

Levy, S. B. 1987. Antibiotic use for grth promotion in animals: Ecologic and public
health consequences. Journal of food protection. 50(7):616-620.

Linton, A. H. 1977. Antibiotic resistance: the present situation reviewed. Vet. Rec.
100:354-360.

Mahalanobis, P. C. 1936. On the generalized distance in statistics. Proceeding of the
national institute of sciences of India. 68: 176-179.

Matthews, J. A. 1979. A study of the variability of some successional and climax plant
assemblage-types using multiple discriminant analysis. Journal of Ecology. 67:255-271.

Maule, A. 2000. Survival of verocytotoxigenic Escherichia coli 0157 in soil, water and
on surfaces. Journal of Applied Microbiology. 88:718-788.

Mckeon, D.M., J. P. Calabrese, G. K. Bissonnette. 1995. Antibiotic resistant gram-
negative bacteria in rural ground water supplies. Water Research. 29: 1902-1908.

McIntosh, L. M., M. Jackson, H. H. Mantsch, M. F. Stranc, D. Pilavdzic and A. Neil
Crowson. 1999. Infrared spectra of basal cell carcinoma are distinct from non-tumor
bearing skin components. Journal of Investigative Dermatology. 112 (6):951.

Meng, J ., S. Zhao, M. P. Doyle, S. W. Joseph. 1998. Antibiotic resistance of Escherichia
coli 0157:H7 and 0157:NM isolated fiom animals, food and humans. Journal of Food
Protection. 61(11):511-1514.

Morrison, D. F. 1990. Multivariate Statistical Methods. 3rd edition. Chapter 6, pages 269-
287, McGraw-Hill Publishing Co.

Mulamattathil, S. G., H. A. Esterhuysen, P. J. Pretorius. 2000. Antibiotic resistant gram-

negative bacteria in a virtually closed water reticulation system. Journal of Applied
Microbiology. 88:930-937.

120

Munro, H. L., and R. C. Rounds. 1985. Selection of artificial nest sites by five sympatric
passerines. Journal of wildlife management. 49:264-276.

National Committee for Clinical Laboratory Standards. 1997 Approved Standard M2-A6.
Performance standards for antimicrobial disc susceptibility tests. 6th ed. NCCLS, Wayne,
Pa.

National Committee for Clinical Laboratory Standards. 1999. NCCLS document M100-
S9. Performance standards for antimicrobial susceptibility testing, 9“1 ed. Informational
supplement. NCCLS, Wayne, Pa.

Neu, H. C. 1992. The crisis in antibiotic resistance. Science. 257: 1064-73.

Norris, J. M. and J. P. Barkham. 1970. A comparison of some Costwold Beechwoods
using multiple discriminant analysis. Journal of Ecology. 58:603-619.

Osawa, S., K. Furuse, and I. Watanabe. 1981. Distribution of ribonucleic acid
colipghages in animals. Applied and Environmental Microbiology. 41:164-168.

Parveen, S., and M. L. Tamplin. 1998. Methods for measuring levels of human and
nonhuman sources of fecal pollution in water, abstr. Q-108, P. 439. In abstracts of the
98th general meeting of the American Society for Microbiology 1998. American Society
for Microbiology, Washington, DC.

Parveen, S., K. M. Portier, K. Robinson, L. Edmiston and M. Tamplin. 1999.
Discriminant analysis of ribotype profiles of Escherichia coli for differentiating human

and nonhuman sources of fecal pollution. Applied and Environmental Microbiology.
65(7):3142-3147.

Parveen, S., N. C. Hodge, R. E. Stall, S. R. Farrah and M. L. Tamplin. 2001. Genotypic
and phynotypic characterzation of human and non human Escherichia coli . Water Res.
35:379-386.

Parveen, S., R. L. Murphree, L. Edmiston, C. W. Kaspar, K. M. Portier, and M. L.
Tamplin. 1997. Association of multiple antibiotic resistance profiles with point and
nonpoint sources of Escherichia coli in Apalachicola Bay. Applied and Environmental
Microbiology. 63(7):2607-2612.

Piddock, L. 1996. Does the use of antimicrobial agents in veterinanry medicine and
animal husbandry select antibiotic-resistant bacteria that infect man and compromise
antimicrobial chemotherapy? Journal of Antimicrobial Chemotherapy. 38: 1-3.

Pillai, S. D., K. W. Widmer, K.G. Maciorowski, S.C. Ricke. 1997. Antibiotic resistance
profiles of Escherichia coli isolated fi'om rural and urban environments. Journal of
Environ. Science. Health. 32:1665-1675.

121

Pregitzer, K. S. and B. V. Barnes. 1984. Classifiacation and comparison of upland
hardwood and conifer ecosystems of the Cyrus H. McCormick experimental forest, upper
Michigan. Can. J. For. Res. 14:362-375.

Prescott, J. F., J. D. Baggot, and R. D. Walker. 2000. Antimicrobial Therapy in
Veterinary Epidemiology. 3rd edition. Iowa State University Press, Ames, IA.

Press, S. J. and S. Wilson. 1978. Choosing between logistic regression and discriminant
analysis. Journal of the American Statistical Association. 73:699-705.

Rao, C. R. 1948. The utilization of multiple measurements in problems of biological
classification. Journal of the royal Statistical Society, Series B, 10:159-193.

Rao, C. R. 197 3. Linear Statistical Inference and Its Applications, 2"d Edition, New York:
John Wiley& Sons, Inc.

Rice, J. R., D. Ohmart, and B. W. Anderson. 1983. Habitat selection attributes of an
avian community: a discriminant analysis investigation. Ecological Monographs. 53:263-
290.

Richmond, M. H. 1972. Some environmental consequences of the use of antibiotics: or
“what goes up must come down”. Journal of Applied Bacteriology. 35:155-176.

Rosenblatt, M.1956. Remarks on some nonparametric estimates of a density function.
Annals of Mathematical Statistics. 27:832-837.

Rubens, C. E., W. F. McNeil], and W. E. Farrar, Jr. 1979. Evolution of multiple
antibiotic resistance plasmids mediated by transposable plasmid deoxyribonucleic acid
sequences. Journal of Bacteriology.140(2):713-719.

Samadpour, M., and N. Chechwtiz. 1995. Little Soos Creek microbial source tracking: a
Survey. University of Washington, Dep. of Environ. Health, seattle, WA.

SAS institute, Inc. 1999, SAS/STAT User’s Guide Version 8.2 Cary, NC: SAS institute,
Inc.

Schroeder, C. M., C. Zhao, C. DebRoy, J. Torcolini, S. Zhao, D. G. White, D. D.
Wagner, P. F. McDermott, R. D. Walker, and J. Meng. 2002a. Antimicrobial resistance
of Escherichia coli 0157 isolated fi'om humans, cattle, swine, and food. Applied and
Enviromental Microbiology. 68(2):576—58 1.

Schroeder, C. M., J. Meng, S. Zhao, C. DebRoy, J. Torcolini, C. Zhao, P. F. McDermott,
D. D. Wagner, R. D. Walker, and D. G. White. 2002b. Antimicrobial resistance of
Escherichia coli 026, 0103, 01 11, 0128, and 0145 from animals and humans.
Emerging Infectious Diseases. 8(12):1409-1414.

122

Seagle, S. W. 1985. Patterns of small mammal microhabitat utilization in cedar glade
and deciduous forest habitats. Journal of Mammalogy. 66:22-35.

Smith, C., A. B. 1947. Some examples of discrimination. Annals Eugenics. 18:Page 272.

Smith, H. W. 197 5. Persistence of tetracycline resistance in pig E. coli. Nature 258:628-
639

Smith, P. R., E. Farrel, and K. Dunican. 1974. Survival of R+ Escherichia coli in sea
water. Applied Microbiology. 27(5):983-984.

Smith, R. D.1995. Veterinary Clinical Epidemiology, 2nd Edition. pages 130-131. CRC
Press, Ann Arbor.

Spies, TA, and B. V. Barnes. 1985a. A multifactor ecological classification of the
northern hardwood and conifer ecosystems of Sylvania Recreation Area. Upper
Peninsula, Michigan. Can. J. For. Res. 15:949-960.

Spratt, B. G. 1994. Resistance of antibiotics mediated by target alterations. Science
264:388-393.

Swarm Committee. 1969. Report of Joint Committee on the Use of Antibiotic in Animal
Husbandry and Veterinary Medicine. Her Majesty’s Stationary Office. London,
September.

Swerdlow, D.L., B. Woodruff, R. Brady, et a1. 1992. A waterborne outbreak in Missouri
of Escherichia coli 0157:H7 associated with bloody diarrhea and death. Annals of
Internal Medicine. 1 17:812-819.

Tabachnick, B. G. and L. S. F idell. 1996. Using Multivariate Statistics. Harper Collins
College Publishers: New York.

Tartera, C., F. Lucena, and J. Jofre. 1989. Human origin of Bacteroides fragilis
bacteriopgages present in the environment. Applied and Environmental Microbiology.
55 :2696-2701.

Tatsuoka, M. M. 1970. Selected Topics in Advanced Statistics, Number 6, Illinios.
Tonn, W. M., and J. J. Magnuson. 1982. Patterns in the species composition and richness
of fish assemblages in northern Wisconsin lakes. Ecology. 63:1149-1166.

Troy, M. S., J. B. Rose, T. M. Jenkins, S. R. Farrah, and J. Lukasik. 2002. Microbial
source tracking: current methodology and future directions. Applied and Environmental
Microbiology. 68(12):5796-5803.

Troy, M. S., S. Parveen, K. M. Portier, J. B. Rose, M. L. Tarnplin, S. R. Farrah, A. K00,
and J. Lukasik. 2003. Geographical variation in ribotype profiles of Escherichia coli

123

isolates from humans, swine, poultry, beef and dairy cattle in Florida. Applied and
Environmental Micobiology. 69(2): 1089-1092.

Tukey, J. W. 1949. anova, an analysis of variance for vectors. Huma biology. 21 :65-1 10.

United States Environmental Protection Agency. 1985. Test methods for Escherichia
coli and enterococci in water by the membrane filter procedure. Environmental
Monitoring and Support Laboratory, Cincinnati, OH. EPA-600/4-85/076.

United States Environmental Protection Agency. 1986. Ambient water quality criteria for
bacteria, 1986. Report EPA440/5-84-002. United States Environmental Protection
Agency, Washington, DC.

Van den Bogaard, A. E, N. London, C. Driessen and E. E. Stobberingh. 2001. Antibiotic
resistance of faecal Escherichia coli in poultry, poultry farmers and poultry slaughterers.
Journal of Antimicrobial Chemotherapy. 47:763-771.

Van den Bogaard, A. E., and E. E. Stobberingh. 1999. Antibiotic usage in animals:
impact on bacterial resistance and public health. Drugs 58:589-607.

Verbyla, D. 1986. Potential predection bias in regression and discrimiannt analysis. Can.
J. For. Res.16:1255-1257.

Vemer, J ., M.L. Morrison and C. J. Ralph. 1986. Wildlife 2000-modeling habitat
relationships of terrestrial vertebrates. University of Wisconsin press, Madison,
Wisconsin, USA.

Wald, A. 1944. On a statistical problem arising in the classification of an individual into
one of two groups. Annals of Mathematical Statistics. 44:229-233.

White, D.G., C. Hudson, J. J. Maurer, S. Ayers, S. Zhao, M. D. Lee, L. Bolton, T. Foley,
and J. Sherwood. 2000. Characterization of chlorarnphinicol and florfenicol resistance in
Escherichia coli associated with bovine diarrhea. Journal of Clinical Microbiology.
38(12):4593-4598.

Whitlock, J. E., D. T. Jones, V. J. Harwood. 2002. Identification of the sources of fecal
coliforms in an urban watershed using antibiotic resistance analysis. Water Research.
36:4273-4282.

Wiggins, B. 1996. Discriminant analysis of antibiotic resistance patterns in fecal
streptococci, a method to differentiate human and animal sources of fecal pollution in
natural waters. Applied and Environmetal Microbiology. 62(11):3997-4002.

Wiggins, B. A., P.W. Cash, W. S. Creamer, S. E. Dart, etal., 2003. Use of antibiotic

resistance analysis for representativeness testing of multiwatershed libraries. Applied and
Environmental Microbiology. 69(6):3399-3405.

124

Wiggins, B. A., R. W. Andrews, R. A. Conway, C. L. Corr, E. J. Dobratz, D. P.
Dougherty, J. R. Eppard, S. R. Knupp, M. C. Limjoco, J. M. Mettenburh, J. M.
Rinehardt, J. Sonsino, R. L. Torrijos and M. E. Zimmerman. 1999. Use of antibiotic

resistance analysis to identify nonpoint sources of fecal pollution. Applied and
Environmental Microbiology. 65(8):3483-3486.

Williams, B. K. 1983. Some observations on the use of discriminant analysis in ecology.
Ecology. 64:1283-1291.

Williams, B. K. 1988. Assessment of sampling stability in ecological applications of
discriminant analysis. Ecology. 69(4):1275-1285.

Witte, W. 1998. Medical consequences of antibiotic use in agriculture. Science. 279: 996-
997.

Witte, W., H. Tschape, I. Klare and G. Werner. 2000. Antibiotic in animal feed. Acta
Veterinary Scandinavica. 93 (supplement):37-45.

World Health Organization. 1997. The medical impact of the use of antimicrobials in
food animals. Report of a WHO meeting. WHO/EMC/ZOO/97.4.

Zhao, S., D.G. White, B. Ge, S. Ayers, S. Friedman, L. English, D. Wagner, S Gaines, J.
Meng. 2001. Identification and characterization of Integron-mediated antibiotic resistance
among shiga toxin producing Escherishia coli isolates. Applied and Environmental
Microbiology. 67(4): 155 8-1564.

125

   

i0111111111