15
ORIGINAL ARTICLE Analytical Epidemiology of Genomic Variation among Pan Asia Strains of Foot-and-Mouth Disease Virus R. B. Garabed 1,2 , W. O. Johnson 3 and M. C. Thurmond 1 1 FMD Modeling and Surveillance Laboratory, University of California, Davis, CA, USA 2 Department of Veterinary Preventive Medicine, College of Veterinary Medicine, The Ohio State University, Columbus, OH, USA 3 Department of Statistics, Donald Bren School of Information and Computer Sciences, University of California, Irvine, CA, USA Introduction Foot-and-mouth disease (FMD) is an important animal pathogen on the global scale because it is easily transmit- ted, has a host range that includes many mammalian spe- cies most important to global agriculture and human survival, and has formidable economic consequences for domestic livestock production and international trade. FMD is caused by a single-stranded RNA virus of the Apthovirus genus in the Picornaviridae family called the foot-and-mouth disease virus (FMDV) (Belsham, 1993). Poliovirus, the most well-known close relative of FMDV, and rinderpest, an unrelated RNA virus with similar host range and clinical signs, have been eradicated from much of the world’s susceptible populations through global vac- cination campaigns (Aylward et al., 2003; OIE, 2006; Couacy-Hymann et al., 2007). FMD control, however, presents different challenges because of FMDV’s host range, which includes several wildlife reservoirs (Vosloo et al., 2002), and antigenic variability (Paton et al., 2005). Whereas vaccines have been used effectively to control FMD worldwide for many years (Kitching, 1998; Sutmol- ler et al., 2003), FMDV antigens are sufficiently variable that, at great expense, circulating viruses must be monitored and vaccines must be updated to match cur- rent field strains (Feigelstock et al., 1996; Barnett et al., 2001; Haydon et al., 2001b; Paton et al., 2005). Despite the variability found in the FMDV genome (Haydon Keywords: FMD, molecular epidemiology, Bayesian, Pan Asia, molecular ecology, disease evolution Correspondence: R. B. Garabed, Department of Veterinary Preventive Medicine, College of Veterinary Medicine, The Ohio State University, A100G Sisson Hall, 1920 Coffey Road, Columbus, OH 43210, USA. Tel.: +614 247 1842; Fax: +614 292 4142; E-mail: [email protected] Received for publication December 2, 2008 doi:10.1111/j.1865-1682.2009.01068.x Summary Genetic data from field isolates of foot-and-mouth disease virus (FMDV) have been used to trace the source of recent outbreaks of FMD, to design better vac- cines and diagnostic tests for FMDV, and to make conclusions regarding the general variability in the FMDV genome. Though epidemiologic data associated with FMDV isolates are available, these data have been used rarely to explore possible associations of epidemiologic factors with evolution or variation of the FMDV genome. In this study, factors associated with variation in the primary immunogenic peptide gene of FMDV (VP1), for a sample of 147 serotype O, Pan Asia strain sequences were explored using traditional analytical epidemio- logic methods: logistic regression and multinomial-response logistic regression. Hypothesized factors included host type (bovine, ovine, buffalo, or porcine) and geographical region (Middle East, South Asia, East Asia, Southeast Asia, and Europe). Results of two regression analyses suggest that host type and region, considered to be possible surrogates for host management, may be associated with selection in the VP1 amino acid sequence in FMDV. For example, isolates from cattle and sheep in South Asia appear to converge with a proposed ances- tor sequence, whereas isolates from the same species in the Middle East and Southeast Asia appear to diverge. The methods demonstrated here could be used on a more detailed dataset to explore the selective pressure of host immu- nity on the evolution of FMDV antigens in an endemic setting. More broadly, epidemiologic methods could be applied extensively to molecular data to explore the causes of genomic variation in disease-causing organisms. Transboundary and Emerging Diseases 142 ª 2009 Blackwell Verlag GmbH Transboundary and Emerging Diseases. 56 (2009) 142–156

Analytical Epidemiology of Genomic Variation among Pan Asia Strains of Foot-and-Mouth Disease Virus

  • Upload
    osu1

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

ORIGINAL ARTICLE

Analytical Epidemiology of Genomic Variation among PanAsia Strains of Foot-and-Mouth Disease VirusR. B. Garabed1,2, W. O. Johnson3 and M. C. Thurmond1

1 FMD Modeling and Surveillance Laboratory, University of California, Davis, CA, USA2 Department of Veterinary Preventive Medicine, College of Veterinary Medicine, The Ohio State University, Columbus, OH, USA3 Department of Statistics, Donald Bren School of Information and Computer Sciences, University of California, Irvine, CA, USA

Introduction

Foot-and-mouth disease (FMD) is an important animal

pathogen on the global scale because it is easily transmit-

ted, has a host range that includes many mammalian spe-

cies most important to global agriculture and human

survival, and has formidable economic consequences for

domestic livestock production and international trade.

FMD is caused by a single-stranded RNA virus of the

Apthovirus genus in the Picornaviridae family called the

foot-and-mouth disease virus (FMDV) (Belsham, 1993).

Poliovirus, the most well-known close relative of FMDV,

and rinderpest, an unrelated RNA virus with similar host

range and clinical signs, have been eradicated from much

of the world’s susceptible populations through global vac-

cination campaigns (Aylward et al., 2003; OIE, 2006;

Couacy-Hymann et al., 2007). FMD control, however,

presents different challenges because of FMDV’s host

range, which includes several wildlife reservoirs (Vosloo

et al., 2002), and antigenic variability (Paton et al., 2005).

Whereas vaccines have been used effectively to control

FMD worldwide for many years (Kitching, 1998; Sutmol-

ler et al., 2003), FMDV antigens are sufficiently variable

that, at great expense, circulating viruses must be

monitored and vaccines must be updated to match cur-

rent field strains (Feigelstock et al., 1996; Barnett et al.,

2001; Haydon et al., 2001b; Paton et al., 2005). Despite

the variability found in the FMDV genome (Haydon

Keywords:

FMD, molecular epidemiology, Bayesian, Pan

Asia, molecular ecology, disease evolution

Correspondence:

R. B. Garabed, Department of Veterinary

Preventive Medicine, College of Veterinary

Medicine, The Ohio State University, A100G

Sisson Hall, 1920 Coffey Road, Columbus, OH

43210, USA. Tel.: +614 247 1842; Fax: +614

292 4142; E-mail: [email protected]

Received for publication December 2, 2008

doi:10.1111/j.1865-1682.2009.01068.x

Summary

Genetic data from field isolates of foot-and-mouth disease virus (FMDV) have

been used to trace the source of recent outbreaks of FMD, to design better vac-

cines and diagnostic tests for FMDV, and to make conclusions regarding the

general variability in the FMDV genome. Though epidemiologic data associated

with FMDV isolates are available, these data have been used rarely to explore

possible associations of epidemiologic factors with evolution or variation of the

FMDV genome. In this study, factors associated with variation in the primary

immunogenic peptide gene of FMDV (VP1), for a sample of 147 serotype O,

Pan Asia strain sequences were explored using traditional analytical epidemio-

logic methods: logistic regression and multinomial-response logistic regression.

Hypothesized factors included host type (bovine, ovine, buffalo, or porcine) and

geographical region (Middle East, South Asia, East Asia, Southeast Asia, and

Europe). Results of two regression analyses suggest that host type and region,

considered to be possible surrogates for host management, may be associated

with selection in the VP1 amino acid sequence in FMDV. For example, isolates

from cattle and sheep in South Asia appear to converge with a proposed ances-

tor sequence, whereas isolates from the same species in the Middle East and

Southeast Asia appear to diverge. The methods demonstrated here could be

used on a more detailed dataset to explore the selective pressure of host immu-

nity on the evolution of FMDV antigens in an endemic setting. More broadly,

epidemiologic methods could be applied extensively to molecular data to

explore the causes of genomic variation in disease-causing organisms.

Transboundary and Emerging Diseases

142 ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156

et al., 2001b), studies of the major antigenic gene of

FMDV (named ‘VP1’) suggest that its variation is limited

and is probably under some evolutionary pressure (Martı-

nez et al., 1992; Haydon et al., 2001a; Tami et al., 2003;

Tully and Fares, 2006). A greater understanding of the

selective pressures driving FMDV antigens to vary would

be useful in reducing the cost of FMDV vaccination and

might suggest inexpensive management changes to com-

plement disease control through vaccination.

The major antigenic gene of FMDV is called the ‘VP1’

gene (also called the ‘1D’ gene) (Haydon et al., 2001a;

Mason et al., 2003a; Carrillo et al., 2005). A previous

study (Samuel and Knowles, 2001a) of FMDV suggested

that nucleotide sequences of the VP1 gene in different

field isolates are similar to the sequences in geographically

neighbouring isolates. Samuel and Knowles (2001a) used

the term ‘topotypes’ to refer to specific phylogenetic

groupings of FMDV isolates based on their VP1

sequences and apparent geographical proximity. One

might infer from several recent reports (Samuel and

Knowles, 2001b; Knowles and Samuel, 2003; Knowles

et al., 2005) that differences in topotypes are related to

differences in ancestral lineages and random mutation.

However, conclusions of several in vitro studies suggest

that these differences also may be related to the selective

pressure of vaccines and of population immunity in dif-

ferent regions of the world and in different host species

(Feigelstock et al., 1996; Tami et al., 2003).

The testing of hypotheses, originally conceived by

in vitro studies, as to whether geographical differences in

host management influence selection in the VP1 gene of

FMDV necessarily must be undertaken using field data.

Unfortunately, the possible influence of host population

factors on FMDV evolution has not been studied using

actual field data. Whereas published in vitro studies of

FMDV have considered different hypothesized selective

mechanisms, such as circulating antibodies and host

cell tropisms (Feigelstock et al., 1996; Tami et al., 2003;

Ruiz-Jarabo et al., 2004), observational phylogenetic tree

studies have been concerned with forensics of outbreaks

(Samuel et al., 1997; Knowles et al., 2001; Samuel and

Knowles, 2001b; Feng et al., 2003; Cottam et al., 2006,

2008; Klein et al., 2006), or with descriptions of differ-

ences in a small group of related isolates (Oem et al.,

2005; Nunez et al., 2006), and general variation in the

FMDV genome (Haydon et al., 2001a,b; Carrillo et al.,

2005; Oem et al., 2005; Klein et al., 2006), and not with

understanding the epidemiology of genomic variation. In

contrast, epidemiological methods for risk estimation, like

contingency table analysis and regression, are designed to

identify risk factors for disease outcomes using field data.

These methods could be applied as well in testing mecha-

nistic hypotheses to explore how various factors might be

associated with genomic variation and evolutionary

changes in the virus.

The objective of this study was to apply analytical epi-

demiologic methods in testing the effect of hypothesized

explanatory factors (or ‘risk’ factors) on phenotypic selec-

tion in the VP1 gene of the FMDV Pan Asia strain.

Methods

General approach

Previously published (Knowles et al., 2005) FMDV sero-

type-O Pan-Asia-strain VP1 sequences were used in this

study. Several statistical methods used frequently in epi-

demiology literature were applied to explore possible

explanatory factors for the phenotypic distance and spe-

cific amino acid differences among isolates. Logistic

regression was used to explore associations between pro-

posed explanatory factors and phenotypic (amino acid)

distance of isolates from a proposed ancestor isolate, and

multinomial-response regression was used to explore

associations between proposed explanatory factors and

specific differences between amino acid sequences.

Sequence data

The Pan Asia strain of FMDV is thought to have arisen

as a strain unique from other circulating serotype O

FMDVs in the 1980s in India (Hemadri et al., 2002;

Knowles et al., 2005). Since that time, isolates of Pan Asia

have been collected and sequenced to track the transmis-

sion of the virus across Asia and to Europe and South

Africa (Knowles et al., 2001, 2005; Hemadri et al., 2002;

Mason et al., 2003b). Because this strain has a well-docu-

mented progression over a large area with diverse host

species and management practices in a relatively short

period of time, the genetic lineage of the published VP1

O Pan Asia sequences (Knowles et al., 2005) could be

assumed, mostly, to be described by the time and location

of isolation. The assumption that genetic lineage was

described by location and time of isolation greatly simpli-

fied the application of traditional epidemiologic methods

to the dataset.

A set of 147 aligned Pan Asia VP1 sequences (633 nu-

cleotides in length) – all of which were isolated from

1989 to 2004 – along with sequences from theorized

ancestor strains and outgroups were kindly provided by

Mr. N. Knowles of the Institute for Animal Health, Pir-

bright, UK and Dr J.-F. Valarcher currently of Interna-

tional Veterinary Investigations (iVi) Animal Health,

Uppsala, Sweden, and are listed in the appendix of Know-

les et al. (2005). The phylogenetic relationships among

these nucleotide sequences are graphed in one un-rooted

(Fig. 1a) and two rooted (Fig. 1b and c) phylogenetic

R. B. Garabed et al. Epidemiology of Pan Asia FMDV

ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156 143

(a)

(b) (c)

Epidemiology of Pan Asia FMDV R. B. Garabed et al.

144 ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156

trees that were constructed using MrBayes (Version 2)

(Huelsenbeck and Ronquist, 2001; Ronquist and Huelsen-

beck, 2003) and MrModeltest (Nylander, J. A. A. 2004.

Version 2. Evolutionary Biology Centre, Uppsala Univer-

sity). Figures were created using TreeView (Version 1.6.6)

(Page, 1996) and Adobe Illustrator CS3 (Adobe Systems

Incorporated, San Jose, CA, USA).

Information associated with these sequences (host-type,

location of isolation, and time of isolation) varied in

specificity, as did the number of sequences from a given

host-type, location, and/or time. For example, the time of

isolation for some sequences was recorded as a day,

month, and year whereas only a year was given for others;

and the location of isolation for some sequences was

recorded as a town or village name within a province and

country, whereas for others only the country was

recorded. Also, India and the United Kingdom each pro-

vided 20 sequences or more, whereas South Africa, Japan,

Pakistan, China, and Taiwan each provided one sequence.

The location of isolation was summarized to the

region. Regions for this study were assigned based on

consideration of general differences in management prac-

tices (including vaccination), relative prevalence of differ-

ent FMDV host species, perceived differences in sampling

of circulating FMDV strains, time of first Pan Asia FMD

isolate, and geographic proximity of the countries within

the region. The regions were (abbreviations and sample

size listed in parentheses): Middle East including Turkey

(ME, 32); South Asia including Pakistan, Afghanistan,

and the Himalayas (SA, 60); East Asia (EA, 6); Southeast

Asia (SEA, 24); Europe (EU, 24); and Africa (AF, 1).

In addition, the host-type categories used (with sample

size) were bovine (111), ovine (12), porcine (19), and

buffalo (5). Presumably, these names refer to members of

the Bos genus, the Ovis genus, the Sus genus, and the Bu-

balus bubalus species, respectively. Further identification

of host species, though hypothetically useful, was not pos-

sible. Isolates from a gazelle (Gazella dorcas) from Qatar

and an antelope (genus and species unknown) from the

United Arab Emirates were excluded because they were

the only isolates from these host genera.

Thus, the epidemiologic data available for analysis of

these 147 aligned nucleotide sequences were host-type,

region of isolation, and year of isolation.

Logistic regression

To allow for the examination of proposed explanatory

factors that were both continuous and categorical, the

associations between the proposed explanatory factors and

the phenotypic distance from an ancestor of Pan Asia VP1

were tested using a series of approximate Bayesian logistic

regression models. A Poisson approximation to the logistic

regression model, as discussed subsequently, was deemed

appropriate and necessary, and was implemented in

WinBUGSª (WinBUGS, Copyright 1996–2003, Imperial

College and MRC, U.K.). Nucleotide sequences were

translated to amino acid sequences using BioEdit (BioEdit,

Copyright 1997–2007; Tom Hall, Ibis Biosciences, Carls-

bad, CA, USA) and the universal genetic code.

The Bayesian logistic regression model (Expression 1)

quantified the associations of proposed explanatory fac-

tors (host type, isolation year, and region) with the prob-

ability of a difference between the amino acid sequence of

interest and the amino acid sequence of an ancestor iso-

late. The probability of a difference (p) was used here as

a measure of the phenotypic distance between the ‘typical’

site in the isolate of interest and the corresponding site in

the ancestor isolate and was allowed to vary for ‘atypical’

sites by use of a site-associated random effect (ci), as seen

below (Expression 1) and in the code in Appendix A. The

model is: Differenceij � Bernoulli(pij) where,

lnpij

1� pij

� �¼ hxj þ ci; and ci � Normal 0; sð Þ ð1Þ

where i is the site number in the VP1 amino acid

sequence, the response variable, (Differenceij), was an indi-

cator of an amino acid difference (yes or no) from the

proposed ancestor for a particular isolate (j) at each site

(i), and h were regression coefficients corresponding to the

proposed isolate-associated explanatory factor data, (xj).

The WinBUGSª code for the Poisson approximation

to the logistic regression model, is listed in Appendix A.

A Poisson approximation was used because the logistic

regression code ran very slowly. This was because all of

the logistic regression probabilities of a difference of one

(over various proposed explanatory factor combinations)

were small, and consequently the probabilities of no dif-

ference were generally close to one. Whereas the Poisson

distribution generally takes values 0,1,2…etc., using

the Poisson regression model in this instance results in

estimates of the probability of zero and one that add

nearly to one, leading to a reasonable approximation to

the logistic regression model. A Poisson regression model

running with appropriate rate (q), thus, approximates

very well the logistic regression model with probability

(p) and also runs quickly, resolving the computational

problem.

Fig. 1. Bayesian phylogenetic trees of the nucleotide sequences of Pan Asia isolates of FMDV examined in this study. (a) a tree with no outgroup

and no root, (b) a rooted tree using the ME-SA Branch B group of sequences reported by Knowles et al. (2005), and (c) a rooted tree using the

SEA group of sequences reported by Knowles et al. (2005).

R. B. Garabed et al. Epidemiology of Pan Asia FMDV

ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156 145

Amino acid sites that were conserved throughout the

dataset were not examined using this model. To exclude

these sites from computation without physically removing

them from the data, these sites were assigned a value of

0.00001 for the probability of a difference from the same

site in the ancestor sequence, as noted in the code

(Appendix A). Because sites that lacked variability in the

data set were excluded, the probability of a difference at

any ‘typical’ site from that in the ancestor sequence was

actually the probability for any ‘typical’ variable site –

referred to henceforth as simply ‘probability of a differ-

ence.’ Sites that lacked variability in the data set were

excluded because we were interested in finding factors

associated with variability, so sites with no variability pro-

vided no information pertinent to our exploration.

We fit models using as the approximate ancestor

sequences the sequence theorized by Hemadri et al.

(2002) and Knowles et al. (2005) to be closest to the true

ancestor sequence, O/IND/231/88 (Genbank AF186029),

and three other sequences close to the root of phyloge-

netic trees using different outgroups O/KUW/1/98 (Gen-

bank DQ164905, Fig. 1b), O/IND/77/00 (Hemadri et al.,

2002; Fig. 1c), and O/BHU/38/2002 (Genbank DQ165038,

Fig. 1b).

For each ancestor sequence, a regression model was fit

that controlled only for the anticipated variation over

time and over different sites along the VP1 sequence. This

model contained the minimum number of hypothesized

explanatory factors, one for year and one for the random

effect for site. In addition, models adding different com-

binations of the hypothesized explanatory factors were fit,

and the fits of the models to their respective ancestor data

were compared using the deviance information criterion

(DIC). The additional hypothesized explanatory factors

were the host type, the region, the interaction between

host type and region, and the interaction between region

and year. The interactions were fit to account for the

potential confounding effect of different relative host type

densities and different reporting rates for different species

in different regions, the confounding effect of different

genetic lineages that were present in a particular region at

a particular time, and the hypothesized effects of differ-

ences in host immunity and management over time and

region.

In addition to models using different hypothesized

explanatory factors, models containing different transfor-

mations of year were compared using the DIC. The dif-

ferent transformations of year that were compared were

as follows (i) the natural logarithm of the year, (ii) the

standardized year, where year was standardized by sub-

tracting the first year and dividing by the standard devia-

tion of the sample of years, and (iii) the natural

logarithm of the standardized value. The best model

according to the DIC statistic (smallest value) was used

to make inferences about the associations between region

and host type and FMDV amino acid distance.

Because these were Bayesian models, prior distributions

(priors) were used for the regression coefficients and vari-

ance components for random effects in the models. All

models used the same priors. The priors for the precision

(precision = 1/ variance = s) of the site-associated ran-

dom effect and of the intercept were ‘informed’ by use of

an independent data set retrieved from Genbank (http://

www.ncbi.nlm.nih.gov/Genbank/); a complete list of

strains can be found in Appendix B. These 552 aligned

VP1 amino acid sequences, henceforth referred to as the

‘Genbank data,’ were from FMDV type O strains, and

excluded the set of 147 Pan Asia sequences used to make

inferences in this study and any strains obviously manip-

ulated in a laboratory.

Using the Genbank data, the empirical average relative

frequency of differences from the consensus sequence,

where the consensus sequence is a crude approximation

of the ancestor sequence, was calculated over all variable

sites (0.073), as was the sample standard deviation of the

relative frequencies (0.126). After transformation, these

numbers were used to construct a Normal ()2.624,1)

prior distribution for the intercept (which was the first

component of the vector, h), and C(10.22285,10.11081)

for s, which was the precision of the random effects dis-

tribution, modelled as Normal(0,s). Priors for the

remaining components of h (the regression coefficients)

were relatively non-informative towards the effects of the

hypothetical descriptive factors, Normal(0,0.1) – where

0.1 is the precision.

The Genbank isolates spanned a wide range of years,

locations, and genetic lineages, so the variation in the

Genbank isolates would not be expected to have been

caused by the same selective pressures as in the Pan Asia

data. Therefore, the prior distribution was not intended

to be an exact match to the Pan Asia data, but rather a

rough estimate of values and frequencies that might be

observed in the Pan Asia data. The Genbank data were

adequate for this general estimate.

Multinomial-response regression

The association of site-specific amino acid differences with

host type and/or region was explored using a Bayesian

multinomial-response logistic regression in WinBUGSª

(code in Appendix C). This regression model was used to

estimate the probability of finding a particular amino acid

at a particular site. All sites were modelled independently

and separately, see Expression 2. Using this model, the

log-risk-ratio of the frequency of observing each amino

acid compared with the frequency of the reference amino

Epidemiology of Pan Asia FMDV R. B. Garabed et al.

146 ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156

acid was assumed to be a linear function of the regional

and host-type categories. The year of isolation was not

included as a predictor.

The regression model corresponding to each indepen-

dent amino acid site was as follows: Amino Acidj � Mul-

tinomial(n, a1j, a 2j, … a kj) where,

lnatj

a1j

� �¼ wtxj: ð2Þ

Here ‘Amino Acidj’ is the amino acid type present at

the site of interest in isolate, j (j = 1, 2,…, 147); n is the

number of times the amino acid site was observed for

isolate j (n = 1); atj is the probability of finding amino

acid type, t, at the site of interest in isolate, j, where

t = 1, 2,…, k. The log ratios of these probabilities, com-

paring the probability for the reference amino acid type,

a1j, were defined as a function of the proposed explana-

tory factors (number of factors = r), xj, and their associ-

ated vector of regression coefficients, wt (size is 1 by

r + 1). In this notation, wt includes an intercept for

t = 1,…, k)1.

The regression coefficients corresponding to the pro-

posed explanatory factors were modelled independently

with prior Normal(0,1) distributions. Using the Genbank

data, a partially informed Dirichlet prior was placed on

the probabilities (a1, a2, …, ak) of observing the most

common, second most common, …, and kth most com-

mon amino acid at an ambiguous amino acid site for a

typical serotype O FMDV isolate. This induced a partially

informed prior distribution on the intercepts. The conclu-

sions of the analysis were insensitive to minor modifica-

tions in these priors (sensitivity analysis).

To prevent computational errors, the number of amino

acids that could occur at each site was restricted in the

model to the number of amino acids found at the site in

the Pan Asia data.

Results and Discussion

Logistic regression

The logistic regression analyses provided some evidence

for host- and region-associated differences in phenotypic

distance from several approximate ancestor sequences.

The model with the lowest DIC (the ‘best model’ shown

in Appendix A and in Table 1) was one that used a log

transformation of the standardized year variable, the host

type variables, the region variables, and interactions of

region and host type and of region and year, which along

with the coefficient estimates in Table 1 suggest that all of

the proposed explanatory factors, in some way, describe

the process behind differences in the VP1 gene of Pan

Asia FMDV strains.

Because the estimated effects of the proposed explana-

tory factors on the probability of a difference involved

interactions, the results presented in Table 1 are difficult

to interpret directly. For example, the estimated probabil-

ity of a difference for pigs in Southeast Asia in 1990

would be the inverse logit of the sum of ‘Reference Cate-

gory,’ ‘pig,’ ‘year,’ ‘SEA,’ ‘SEA * pigs,’ and ‘SEA * year.’

Because of this difficulty in interpretation, the estimated

effects of the proposed explanatory factors are shown

graphically in Fig. 2 and described below. The 95% prob-

ability intervals for different regions (not shown in Fig. 2)

did overlap to some extent.

Small changes to the region designations (shifting Paki-

stani and Afghan isolates to the ME group and shifting

the South African isolate to the EU group) did not visibly

change the results presented in Fig. 2. However, differ-

ences in the choice of the ancestor sequence changed the

size of estimated associations between the probability of a

difference and the proposed explanatory factors and chan-

ged the width of the PIs in Table 1, though the directions

of the associations did not vary. The insensitivity of the

estimated associations to these slight changes suggests that

the results for this analysis were not dependent on the

methodological decisions supported by minimal objective

data (regional specification and choice of an ancestor

sequence).

Using O/IND/77/00 as the reference strain, the esti-

mated probabilities of phenotypic (amino acid) differ-

ences in the VP1 gene of hypothetical isolates with

different combinations of proposed explanatory factors

are shown in Fig. 2 to illustrate the estimated effects of

the different proposed explanatory factors. O/IND/77/00

was chosen as the reference strain for display pur-

poses because the strain produced the greatest magni-

tude of associations. Inferences about the African

region were not made because, with only one isolate,

the probability limits on estimates for that region were

extremely wide.

As seen in Fig. 2, the relationships among descriptive

factors and probability of a phenotypic difference from

O/IND/77/00 were complex and indicative of interactions

between region and time and region and host. In general,

the estimated probability of a difference decreased over

time or remained low for all host-types in SA, EU, and

EA but increased in ME and SEA, suggesting an interac-

tion between region and time. This regional association

might be explained by two biological mechanisms. One,

because ME and SEA neighbour the SA region where the

Pan Asia strain was thought to have originated, the trans-

mission to these areas (first isolates in 1998 and 1999,

respectively) may have proceeded through hosts in a

more ‘natural’ transmission cycle than the ‘imported’

transmissions that were thought to cause the EA and EU

R. B. Garabed et al. Epidemiology of Pan Asia FMDV

ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156 147

infections (first isolates in 1999 and 2001, respectively).

During a ‘natural’ transmission, one would expect amino

acid differences to occur in any gene because the virus

population would have been expanding as viruses were

passed among many hosts along the route of spread. An

expanding population would be more likely than a

restricted (smaller and less frequently reproducing) popu-

lation to have amino acid differences because of random

chance alone. Thus, the ‘importation’ of FMDV in one

host, in a small group of hosts or on material from a host

would not be expected to produce many amino acid

polymorphisms compared with progression of FMDV

through many hosts. Alternately, ME and SEA are regions

that had endemic circulation of FMDV type-O viruses

prior to Pan Asia’s introduction and were regions where

FMDV vaccination was practiced (Aidaros, 2002). Thus,

the second biological mechanism that might explain the

interaction between region and time is that the hosts in

these regions may have had some immunity to Pan Asia

creating a selective pressure for the amino acids of the

VP1 gene to vary.

However, the isolates from SA, a region that had ende-

mic FMDV type-O and practiced FMDV vaccination

prior to Pan Asia’s isolation (Bhattacharya et al., 2005),

did not appear to have the increasing probability of dif-

ference over time that was seen in ME and SEA. The Pan

Asia strain, apparently, was circulating in SA from before

1990 until after 2004, but increased phenotypic distance

was not seen over time. The lack of amino acid differ-

ences in Pan Asia isolates from SA, compared with ME

and SEA isolates, might be explained by some evolution-

ary advantage for strains similar to O/IND/77/00. How-

ever, a lower rate of replication (or transmission) of Pan

Asia strains in SA compared with ME and SEA because of

management conditions or because the strain was estab-

lished rather than advancing into a new region might also

explain the decreasing probability of difference for SA

over time.

The second interaction of interest, shown in Fig. 2, was

an interaction between host type and region. In SEA, buf-

falo, porcine and bovine hosts had similar changes in the

probability of a difference from O/IND/77/00 over time,

Table 1. Posterior mean coefficient values for

the different levels of the proposed explanatory

factors with 95% probability intervals (PIs)

and posterior probabilities (Pr) that the coeffi-

cient (coeff) Values were greater than and less

than zero

Mean 95% PI Pr(coeff > 0) Pr(coeff < 0)

Reference category

SA, bovine. 1989 )2.29 ()4.89–)1.07) 0.000 1.000

Sheep 0.87 ()0.40–1.97) 0.922 0.078

Buffalo )0.40 ()1.94–0.88) 0.282 0.719

Pig 1.09 (0.10–1.98) 0.987 0.013

year )1.26 ()2.56–0.28) 0.131 0.870

ME )1.64 ()3.46–0.46) 0.090 0.910

ME*sheep )0.78 ()2.08–0.64) 0.130 0.870

ME*buffalo 0.02 ()6.00–6.28) 0.492 0.509

ME*pigs 0.19 ()5.91–6.28) 0.525 0.476

ME*year 0.59 (0.04–1.07) 0.991 0.009

EA )2.81 ()4.86–)0.33) 0.007 0.993

EA*sheep 0.11 ()632–6.08) 0.515 0.485

EA*buffalo 0.05 ()6.34–6.29) 0.519 0.481

EA*pigs )0.43 ()1.65–0.99) 0.257 0.743

EA*year 0.75 (0.11–1.27) 1.000 0.000

SEA )2.00 ()4.14–0.42) 0.064 0.936

SEA*sheep 0.04 ()6.13–6.19) 0.516 0.484

SEA*buffalo 0.23 ()1.34–2.l0) 0.642 0.358

SEA*Pigs )1.40 ()2.72–)0.02) 0.023 0.977

SEA*year 0.69 (0.04–1.26) 0.987 0.013

EU )0.18 ()4.87–4.15) 0.474 0.526

EU*sheep )1.16 ()3.30–0.89) 0.132 0.868

EU*buffalo )0.01 ()5.96–6.20) 0.496 0.504

EU*Pigs )0.63 ()2.16–1.00) 0.208 0.792

EU*year 0.00 ()122–1.31) 0.486 0.514

AF 0.28 ()5.08–6.04) 0.539 0.461

AF*sheep )0.01 ()6.13–6.10) 0.497 0.503

AF*buffalo 0.02 ()6.26–6.24) 0.507 0.494

AF*pigs )2.34 ()5.31–0.33) 0.045 0.955

AF*year )0.22 ()2.21–1.61) 0.407 0.593

The estimates above used 0/IND/77/00 as the reference sequence.

Epidemiology of Pan Asia FMDV R. B. Garabed et al.

148 ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156

whereas, SA buffalo and SA and EU pigs had lower prob-

abilities of difference from O/IND/77/00, compared with

bovines in SA and EU. This host-type homogeneity in

SEA FMDV isolates might be explained by greater contact

among the different host types in SEA as opposed to a

separation of host types in SA and EU. If FMDV regularly

circulates between different host types, one would expect

isolates from the different host types to be similar. Con-

versely, viruses that are passed separately only to mem-

bers of the same host type would be expected to segregate

and, possibly, to mutate at different rates.

In addition to illustrating the associations above, the

logistic regression model identified amino acid sites that

were more likely than a typical variable site to differ

from the same site in the reference sequence. In Fig. 3,

one can see that sites 4, 9, 58, 83, 96, 138, 140, 141,

154, 172, 198 and 209 of the VP1 amino acid sequence

were more variable than the ‘typical’ variable site. This

hyper-variability was observed for sites 4, 9, 58, 83, 96,

138, 140, 141 and 198 using all of the different hypo-

thetical ancestor amino acid sequences. Whereas, sites

154, 172 and 209 appeared to be more variable than the

‘typical’ variable site, but posterior 95% PIs for the ran-

dom effects at these sites overlapped the ‘typical’ value

(zero) using one of the reference sequences. These extra-

variable sites might have indicated regions of the VP1

gene that were under greater pressure by the host

immune system, or were somehow influenced by other

sites under selective pressure.

Of the 12 amino-acid sites that appeared to be more

variable than the typical variable amino acid site, four

(138, 140, 141, and 154) were within the G-H loop (sites

135–160 in this dataset), which has been found to be one

of the major antigenic sites of VP1 (Strohmaier et al.,

1982; Fox et al., 1989) and was, thus, thought to be

highly variable (Belsham, 1993). Of the remaining

‘atypically variable’ sites found in this analysis, site 209

falls within the VP1 C-terminus antigenic site (site 1)

mentioned by Kitson et al. (1990) and DiMarchi et al.

(1986), and site 198 is very close to this area. However,

the sites in Kitson et al.’s (1990) antigenic site 3 (sites

43–45) were not found to be particularly variable except

for site 45 when compared with O/KUW/1/98. Recent

work by Tully and Fares (2006) suggests that site 83 (Tul-

ly and Fares site #86) is a structurally important site and

sites 4, 138, 140, 141, 198 and 209 (Tully and Fares sites

(a) (b)

(c) (d)

Fig. 2. Estimates of the probability of phenotypic difference from 0/IND/77/00 at a typical variable amino acid site in the VP1 sequence for Pan

Asia isolates over time from (a) bovine, (b) ovine, (c) buffalo, and (d) porcine isolates, comparing results for different regions of the world:

ME = Middle East, SA = South Asia, EA = East Asia, SEA = Southeast Asia, and EU = Europe. Host-region and year combinations that did not

occur in the data are not shown.

R. B. Garabed et al. Epidemiology of Pan Asia FMDV

ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156 149

#5, 146, 148, 149, 213, and 224) are functionally impor-

tant sites in the VP1 gene that have been under selective

pressure in serotype O FMDVs; though, Tully and Fares

(2006) make no suggestion about the specific function of

these sites. A literature search did not reveal any biologi-

cal cause for the variation in the other ‘atypically variable’

sites (9, 96 and 172), but future work in the molecular

biology of FMDV may provide some explanation.

Multinomial-response regression

The multinomial-response logistic regression analysis

compared the probability of observing each amino acid at

each site for the different host types and regions. As

shown in Table 2, several amino acids at sites identified

as variable in the logistic regression analysis were associ-

ated with certain geographical regions for comparisons

among bovine isolates. For example, bovine isolates from

ME were more likely than the bovine isolates from SA to

have a threonine (T) at site VP1#58 and an alanine (A) at

site VP1#96, whereas they were less likely than SA bovine

isolates to have a T at site VP1#141.

Inspection of the amino acid sequence data suggested

that some of the differences in the probability of observ-

ing different amino acids at each site as listed in Table 2

might have been related to the clustering of sampling in

the regions at specific times rather than overall differences

in amino acid sequences as a result of selection. For

example, the increased probability of finding a T at posi-

tion VP1#198 in isolates from SEA was due to nine simi-

lar samples from Laos in 2003 and one sample from

Vietnam in 2004 that all contained the same mutation.

This sample grouping can be seen as a branch labeled

‘Lao 2003 and Vit 2004’ on the phylogenetic trees in

Fig. 1. Similarly, all but one of the strains from the Uni-

ted Kingdom and Ireland in 2001 (the EU isolates) had

an A at position VP1#4 and a valine (V) at position

VP1#141. Thus, a clear distinction cannot be made

between amino acid differences that happened to be in

lineages that occurred in heavily sampled regions and

amino acid differences that had some selective advantage

in a certain region.

Though the year of isolation, biologically, should not

affect the likelihood of observing a specific amino acid at

a specific site, the confounding effect of year as it related

to sampling and phylogenetic lineage suggested that year

should have been included in the multinomial-response

regression. However, subsets of the data based on region,

host type and year were not sufficiently variable to permit

Fig. 3. Posterior mean estimates of site-associated random effects at

each amino acid site in the VP1 gene with 95% probability intervals.

Sites were numbered beginning with the first amino acid in VP1. Posi-

tive values indicate a site that was more likely to differ from the pro-

posed ancestor sequence, compared with the typical variable site with

similar proposed explanatory factors.

Table 2. Amino acids associated with geographica1 regions, based on regression coefficients and 95% PIs (interpreted as log risk ratios, com-

pared with the most common amino acid in South Asia for bovine isolates)

Amino acid site Middle East East Asia Southeast Asia Europe Africa

4 A: 3.06 (1.98, 4.26)

9 A: 2.56(1.24, 3.93)

58 T: 1.67 (0.21, 3.17)

83

96 A: 121 (0.03, 2.36)

138 A: 1.48 ()0.06, 3.03)

140

141 T: )1.64 ()3.04, )041)

A: )126 ()2.79, 0 06)

A: 1.28(0.15, 2.37) T: )1.31 ()2.80, 0.01)

154 R: 2.48 (1.17, 3.84)

172

198 T: 2.58 (1.96, 3.97)

209

The numbers listed are the regression coefficient value (and 95% PI) for the listed amino acid for the listed region. Amino acids that were more

common in the listed region have positive values and PIs that do not include zero. Conversely, less common amino acids have negative values and

PIs that do not include zero. Amino acids with PIs that include zero, but are heavily skewed to the negative or positive direction are also included.

A, alanine; T, threonine; R, arginine.

Epidemiology of Pan Asia FMDV R. B. Garabed et al.

150 ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156

estimation using a multinomial-response logistic regres-

sion model.

In this model ignoring the effect of year, no effects of

host type or the interactions of host type and geographi-

cal region were observed. One might conclude from the

lack of a host-type effect that no selective pressure for

any particular amino acid was associated with a particular

host type or a particular host type in a particular geo-

graphic region for FMDV Pan Asia VP1. However, the

effect of year as mentioned above might have masked or

confounded an effect of host type, so it would be inap-

propriate to make any conclusions regarding the host

type or region effects based on this analysis.

In addition, the VP1 region of the FMDV genome is

not known to have mutations affecting host specificity or

differences in virulence for different host species; there-

fore, a host-type effect, had one existed, would likely have

been related to the different management conditions, such

as differences in vaccination or natural immunity, present

for different hosts. In addition, use of isolates submitted

for diagnostic purposes, rather than systematically

collected isolates, prevented the separation of effects of

sampling and geographic region. However, the multino-

mial-response regression method presented above could

be used to screen sets of complete amino acid sequences

for sites where a given amino acid is favored in a specific

host type or geographic region.

General Remarks

The methods presented above demonstrate a novel appli-

cation of traditional epidemiologic methods to explore

the factors associated with genetic variation in the popu-

lation of FMD viruses.

Methods used in this study assume that the FMD

viruses do not recombine with one another. Though Hay-

don et al. (2004) concluded that FMDV sequences can

recombine, other analyses (Mason et al., 2003b) suggest

that recombination within the Pan Asia strain did not

occur during the relatively short period of time repre-

sented by this study and that recombination within the

capsid genes of FMDV is infrequent (Jackson et al., 2007).

A more important constraint to interpretation is that,

as with all previous observational studies of the epidemi-

ology of FMDV genetics, results of an analysis based on a

convenience sample of field isolates may not apply to

FMDV as a whole. For example, isolates with unusual

mutations causing altered virulence would be more likely

to be sequenced, because there is more potential use in

knowing their sequences than there is for an ordinary iso-

late. Therefore, if one were to calculate the mutation rate

for FMD viruses based on these abnormal sequences, one

would, presumably, get a higher rate than would be

observed for a random sample of FMD viruses. In this

study, the number of samples collected in a particular

region at a particular time did not seem to be directly

proportional to the number of infected animals in the

region at that time. Therefore, the results were weighted

to reflect the mechanisms underlying genetic differences

in frequently sampled lineages more than in frequently

circulating lineages.

One reason for the lack of previous studies exploring

risk factors for variation in the FMDV genome may be

the complexity and dependence of genetic sequence

data, which complicates many traditional epidemiology

methods. As mentioned above, genetic lineage and sam-

pling can confound the observed relationships between

factors hypothetically directing selection of amino acid

sequences. However, the Pan Asia strain of FMDV was

unique compared with other strains of FMDV because

its genetic lineage was thought to be described by time

and geographic location. This simplifying assumption

made the Pan Asia data set ideal to explore the molec-

ular epidemiology of FMDV because the lineage of Pan

Asia isolates could be replaced by simple factors

representing the time and location of a sequence’s

isolation.

Unfortunately, only the logistic regression method used

here could accommodate the effect of year of isolation

and neither of the methods could completely differentiate

the effects of sampling and lineage from the effects of

proposed factors explaining selection in FMDV VP1

sequences. Improved observational data that included

exact host species, management conditions (such as vacci-

nation, prior exposure to FMDV, and general health sta-

tus), the likely source of the virus, and use of methodical

sampling might help in the application of these methods.

In addition, other measures of sample similarity and line-

age possibly drawn from geostatistics or network analysis

might be applied to adapt the methods presented here to

situations where lineage is not described by time, location

and host type.

Whereas the methods presented here were not able to

accommodate completely the effects of lineage and

sampling, the phylogenetic trees and measures of general

variability that have been used to describe the epidemiol-

ogy of FMDV genetics have not incorporated factors pro-

posed to explain selection. Proposed explanatory factors

such as host type (bovine, porcine, …), host management

conditions (intensive, extensive, feedlot, dairy farm, …),

and host immunity could explain some of the selective

pressures influencing FMDV’s antigenic variation. How-

ever, because phylogenetic trees and measures of variation

do not quantitatively evaluate risk factor information,

only qualitative conclusions about the selective pressure

of host and management factors can be made from a

R. B. Garabed et al. Epidemiology of Pan Asia FMDV

ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156 151

phylogenetic tree. Whereas the regression methods above

ignore some of the effects of ancestry on the observed

amino acid sequence, other observational methods, with

the exception of Cottam et al. (2006, 2008), have ignored

the epidemiologic factors like time, location, host type

and vaccination status.

The methods presented above broadly tested the

hypothesis that host management (roughly represented

by host type and geographic region) could influence

selection in FMDV antigens by using available observa-

tional genetic and epidemiologic data. It is reasonable to

hypothesize that host management including intensive

versus extensive farming, vaccination protocol, nutrition,

and closed versus open herd replacement all could con-

tribute to host immunity to FMDV and might, thus,

influence evolution of FMDVs immunogenic amino acid

sequences. The results of the logistic regression implied

that factors related to host management were associated

with the phenotypic distance of an isolate from a pro-

posed ancestor sequence. In addition, the multinomial-

response logistic regression suggested that certain amino

acid polymorphisms were more common in isolates

from certain geographical regions. Though regional

amino-acid ‘fingerprints’ might be inferred to exist for

FMDV VP1 genes based on the multinomial regression

results, the confounding effect of lineage prevents a clear

conclusion. Ideally, with improved data, more hybrid

methods incorporating both ancestry and epidemiologic

data can be developed to help explain the selective pres-

sures on the immunogenic region of FMDV. Future sys-

tematic efforts to sample and sequence an endemic

population of FMDV and to collect more detailed infor-

mation relating to host management and immunity

would enable the methods presented above to be used

to understand more accurately the factors behind FMDV

evolution.

Acknowledgements

We thank Kara O’Keefe and Andres Perez for their help-

ful suggestions, Nick Knowles and Jean-Francois Valar-

cher for providing the sequence and epidemiologic data,

and Michelle Norris, Zachary Whedbee and Garry Kelley

for their technical assistance. This work was funded by

the National Center for Medical Intelligence and a Gradu-

ate Student Support Grant from the U.C. Davis School of

Veterinary Medicine.

References

Aidaros, H. A., 2002: Regional status and approaches to

control and eradication of foot and mouth disease in the

Middle East and North Africa. Rev. Sci. Tech. 21, 451–458.

Aylward, R. B., A. Acharya, S. England, M. Agocs, and J.

Linkins, 2003: Global health goals: lessons from the

worldwide effort to eradicate poliomyelitis. Lancet 362,

909–914.

Barnett, P. V., A. R. Samuel, and R. J. Statham, 2001: The suit-

ability of the ‘emergency’ foot-and-mouth disease antigens

held by the International Vaccine Bank within a global con-

text. Vaccine 19, 2107–2117.

Belsham, G. J., 1993: Distinctive features of foot-and-mouth

disease virus, a member of the picornavirus family; aspects

of virus protein synthesis, protein processing and structure.

Prog. Biophys. Mol. Biol. 60, 241–260.

Bhattacharya, S., R. Banerjee, R. Ghosh, A. P. Chattopadhayay,

and A. Chatterjee, 2005: Studies of the outbreaks of foot

and mouth disease in West Bengal, India, between 1985 and

2002. Rev. Sci. Tech. 24, 945–952.

Carrillo, C., E. R. Tulman, G. Delhon, Z. Lu, A. Carreno, A.

Vagnozzi, G. F. Kutish, and D. L. Rock, 2005: Comparative

genomics of foot-and-mouth disease virus. J. Virol. 79,

6487–6504.

Cottam, E., D. Haydon, D. Paton, J. Gloster, J. Wilesmith, N.

Ferris, G. Hutchings, and D. King, 2006: Molecular epidemi-

ology of the foot-and-mouth disease virus outbreak in the

United Kingdom in 2001. J. Virol. 80, 11274–11282.

Cottam, E., G. Thebaud, J. Wadsworth, J. Gloster, L. Mansley,

D. Paton, D. King, and D. Haydon, 2008: Integrating genetic

and epidemiologic data to determine transmission pathways

of foot-and-mouth disease virus. Proc. Biol. Sci. 275,

887–895.

Couacy-Hymann, E., S. C. Bodjo, M. Y. Koffi, and T. Dan-

ho, 2007: Observations on rinderpest and rinderpest-like

diseases throughout West and Central African countries

during rinderpest eradication projects. Res. Vet. Sci. 83,

282–285.

DiMarchi, R., G. Brooke, C. Gale, V. Cracknell, T. Doel,

and N. Mowat, 1986: Protection of cattle against foot-

and-mouth disease by a synthetic peptide. Science 232,

639–641.

Feigelstock, D. A., M. G. Mateu, M. L. Valero, D. Andreu, E.

Domingo, and E. L. Palma, 1996: Emerging foot-and-mouth

disease virus variants with antigenically critical amino acid

substitutions predicted by model studies using reference

viruses. Vaccine 14, 97–102.

Feng, Q., X. Chen, O. Ma, Y. Liu, M. Ding, R. A. Collins,

L.-S. Ko, J. Xing, L.-T. Lau, and others, 2003: Serotype

and VP1 gene sequence of a foot-and-mouth disease virus

from Hong Kong (2002). Biochem. Biophys. Res. Commun.

30, 715–721.

Fox, G., N. R. Parry, P. V. Barnett, B. McGinn, D. J. Row-

lands, and F. Brown, 1989: The cell attachment site on

foot-and-mouth disease virus includes the amino acid

sequence RGD (arginine-glycine-aspartic acid). J. Gen.

Virol. 70, 625–637.

Haydon, D. T., A. D. Bastos, N. J. Knowles, and A. R. Samuel,

2001a: Evidence for positive selection in foot-and-mouth

Epidemiology of Pan Asia FMDV R. B. Garabed et al.

152 ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156

disease virus capsid genes from field isolates. Genetics 157,

7–15.

Haydon, D. T., A. R. Samuel, and N. J. Knowles, 2001b: The

generation and persistence of genetic variation in foot-and-

mouth disease virus. Prev. Vet. Med. 51, 111–124.

Haydon, D. T., A. D. S. Bastos, and P. Awadalla, 2004: Low

linkage disequilibrium indicative of recombination in

foot-and-mouth disease virus gene sequence alignments.

J. Gen. Virol. 85, 1095–1100.

Hemadri, D., C. Tosh, A. Sanyal, and R. Venkataramanan,

2002: Emergence of a new strain of type O foot-and-

mouth disease virus: its phylogenetic and evolutionary

relationship with the PanAsia pandemic strain. Virus

Genes 25, 23–34.

Huelsenbeck, J. P., and F. Ronquist, 2001: MRBAYES: Bayesian

inference of phylogeny. Bioinformatics 17, 754–755.

Jackson, A. L., H. O’Neill, F. Maree, B. Blignaut, C. Carrillo,

L. Rodriguez, and D. T. Haydon, 2007: Mosaic structure of

foot-and-mouth disease virus genomes. J. Gen. Virol. 88,

487–492.

Kitching, R. P., 1998: A recent history of foot-and-mouth dis-

ease. J. Comp. Pathol. 118, 89–108.

Kitson, J. D. A., D. McCahon, and G. J. Belsham, 1990:

Sequence analysis of monoclonal antibody resistant mutants

of type O foot and mouth disease virus: evidence for the

involvement of the three surface exposed capsid proteins in

four antigenic sites. Virol 179, 26–34.

Klein, J., U. Parlak, F. Ozyoruk, and L. S. Christensen, 2006:

The molecular epidemiology of Foot-and-Mouth Disease

virus serotypes A and O from 1998 to 2004 in Turkey. BMC

Vet. Res. 2, doi:10.1186/1746-6148-2-35.

Knowles, N. J., and A. R. Samuel, 2003: Molecular epi-

demiology of foot-and-mouth disease virus. Virus Res. 91,

65–80.

Knowles, N. J., A. R. Samuel, P. R. Davies, R. P. Kitching, and

A. I. Donaldson, 2001: Outbreak of foot-and-mouth disease

virus serotype O in the UK caused by a pandemic strain.

Vet. Rec. 148, 258–259.

Knowles, N. J., A. R. Samuel, P. R. Davies, R. J. Midgley,

and J.-F. Valarcher, 2005: Pandemic strain of foot-and-

mouth disease virus serotype O. Emerg. Infect. Dis. 11,

1887–1893.

Martınez, M. A., J. Dopazo, J. Hernandez, M. G. Mateu, F.

Sobrino, E. Domingo, and N. J. Knowles, 1992: Evolution of

the capsid protein genes of foot-and-mouth disease virus:

antigenic variation without accumulation of amino acid

substitutions over six decades. J. Virol. 66, 3557–3565.

Mason, P. W., M. J. Grubman, and B. Baxt, 2003a:

Molecular basis of pathogenesis of FMDV. Virus Res. 91,

9–32.

Mason, P. W., J. M. Pacheco, Q.-Z. Zhao, and N. J. Knowles,

2003b: Comparisons of the complete genomes of Asian,

African and European isolates of a recent foot-and-mouth

disease virus type O pandemic strain (PanAsia). J. Gen.

Virol. 84, 1583–1593.

Nunez, J. I., P. Fusi, B. Borrego, E. Brocchi, M. L. Pacciarini,

and F. Sobrino, 2006: Genomic and antigenic characteriza-

tion of viruses from the 1993 Italian foot-and-mouth disease

outbreak. Arch. Virol. 151, 127–142.

Oem, J. K., K. N. Lee, I. S. Cho, S. J. Kye, J. Y. Park, J. H.

Park, Y. J. Kim, Y. S. Joo, and H. J. Song, 2005: Identifica-

tion and antigenic site analysis of foot-and-mouth disease

virus from pigs and cattle in Korea. J. Vet. Sci. 6, 117–124.

OIE 2006: Official Animal Health Status: Rinderpest. World

Organization for Animal Health, Paris. Available at http://

www.oie.int/eng/info/en_peste.htm.

Page, R. D. M., 1996: TREEVIEW: An application to display

phylogenetic trees on personal computers. Comput. Appl.

Biosci. 12, 357–358.

Paton, D. J., J.-F. Valarcher, I. Bergmann, O. G. Matlho, V. M.

Zakharov, E. L. Palma, and G. R. Thompson, 2005: Selec-

tion of foot and mouth disease vaccine strains – a review.

Rev. Sci. Tech. 24, 981–993.

Ronquist, F., and J. P. Huelsenbeck, 2003: MRBAYES 3:

Bayesian phylogenetic inference under mixed models.

Bioinformatics 19, 1572–1574.

Ruiz-Jarabo, C. M., N. Pariente, E. Baranowski, M. Davila, G.

Gomez-Mariano, and E. Domingo, 2004: Expansion of

host-cell tropism of foot-and-mouth disease virus despite

replication in a constant environment. J. Gen. Virol. 85,

2289–2297.

Samuel, A. R., and N. J. Knowles, 2001a: Foot-and-mouth

disease type O viruses exhibit genetically and geographically

distinct evolutionary lineages (topotypes). J. Gen. Virol. 82,

609–621.

Samuel, A. R., and N. J. Knowles, 2001b: Foot-and-mouth dis-

ease virus: cause of the recent crisis for the UK livestock

industry. Trends Genet. 17, 421–424.

Samuel, A. R., N. J. Knowles, R. P. Kitching, and S. M. Hafez,

1997: Molecular analysis of foot-and-mouth disease type O

viruses isolated in Saudi Arabia between 1983 and 1995.

Epidemiol. Infect. 119, 381–389.

Strohmaier, K., R. Franze, and K.-H. Adam, 1982: Location

and characterization of the antigenic portion of the FMDV

immunizing protein. J. Gen. Virol. 59, 295–306.

Sutmoller, P., S. S. Barteling, R. Casas-Olascoaga, and K. J.

Sumption, 2003: Control and eradication of foot-and-mouth

disease. Virus Res. 91, 101–144.

Tami, C., O. Taboga, A. Berinstein, J. I. Nunez, E. L.

Palma, E. Domingo, F. Sobrino, and E. Carrillo, 2003:

Evidence of the coevolution of antigenicity and host cell

tropism of foot-and-mouth disease virus in vivo. J. Virol.

77, 1219–1226.

Tully, D. C., and M. A. Fares, 2006: Unravelling selection shifts

among foot-and-mouth disease virus (FMDV) serotypes.

Evol Bioinform Online 2, 211–225.

Vosloo, W., A. D. S. Bastos, O. Sangare, S. K. Hargreaves, and

G. R. Thompson, 2002: Review of the status and control of

foot and mouth disease in sub-Saharan Africa. Rev. Sci.

Tech. 21, 437–449.

R. B. Garabed et al. Epidemiology of Pan Asia FMDV

ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156 153

Epidemiology of Pan Asia FMDV R. B. Garabed et al.

154 ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156

R. B. Garabed et al. Epidemiology of Pan Asia FMDV

ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156 155

Epidemiology of Pan Asia FMDV R. B. Garabed et al.

156 ª 2009 Blackwell Verlag GmbH • Transboundary and Emerging Diseases. 56 (2009) 142–156