doi:10.1111/j.1365-2052.2004.01157.

Microsatellite diversity, population subdivision and gene flow

in the Lipizzan horse
R. Achmann*, I. Curik†, P. Dovc‡, T. Kavar‡, I. Bodo§, F. Habe‡, E. Marti¶, J. Sölkner** and G. Brem*
*Ludwig Boltzmann-Institut für immuno-, zyto- and molekulargenetische Forschung, Veterinärplatz 1, A-1210 Wien, Austria. †Animal
Science Department, Faculty of Agriculture, University of Zagreb, Svetosimunska 25, 10000 Zagreb, Croatia. ‡Department of Animal
Science, Biotechnical Faculty, University of Ljubljana, Groblje 3, SI-1230 Domzale, Slovenia. §Department of Animal Husbandry, University
of Veterinary Science Budapest, H-1078 Budapest, Istvan u.2, Hungary. ¶Division of Immunogenetics, Institute of Animal Breeding,
Bremgartenstrasse 109A, CH-3012 Bern, Switzerland. **Department of Livestock Sciences, University of Agricultural Sciences Vienna,
Gregor Mendel Strasse 33, A-1180 Wien, Austria

Summary Blood samples of 561 Lipizzan horses from subpopulations (studs) of seven European
countries representing a large fraction of the breed’s population were used to examine the
genetic diversity, population subdivision and gene flow in the breed. DNA analysis based on
18 microsatellite loci revealed that genetic diversity (observed heterozygosity ¼ 0.663,
gene diversity ¼ 0.675 and the mean number of alleles ¼ 7.056) in the Lipizzan horse is
similar to other horse breeds as well as to other domestic animal species. The genetic
differentiation between Lipizzan horses from different studs, although moderate, was
apparent (pairwise FST coefficients ranged from 0.021 to 0.080). Complementary findings
explaining the genetic relationship among studs were revealed by genetic distance and
principal component analysis. One genetic cluster consisted of the subpopulations of Aus-
tria, Italy and Slovenia, which represent the classical pool of Lipizzan horse breeding.
A second cluster was formed by the Croatian, Hungarian and Slovakian subpopulations.
The Romanian subpopulation formed a separate unit. The largest genetic differentiation
was found between the Romanian and Italian subpopulation. Genetic results are consistent
with the known breeding history of the Lipizzan horse. Correct stud assignment was
obtained for 80.9% and 92.1% of Lipizzan horses depending on the inclusion or exclusion of
migrant horses, respectively. The results of the present study will be useful for the devel-
opment of breeding strategies, which consider classical horse breeding as well as recent
achievements of population and conservation genetics.
Keywords conservation genetics, gene flow, genetic structure, Lipizzan horse, microsat-
ellite diversity.

Introduction
The Lipizzan horse, established in 1580 in Lipica by the
Habsburg Archduke Charles II, is one of the most famous
horse breeds in the world. The breed arose by crossing local
horses from the Slovenian Karst region with Andalusian and
Neapolitanian horses. Later, Arab, Frederiksborg and Kla-
drub horses were also used. Although it was threatened by
the Wars of Napoleon as well as by the First and Second
Address for correspondence
GmbH, Falkenried 88, 20251 Hamburg, Germany
E-mail: achmann@genteQ.de
Accepted for publication 1 May 2004
World War the breed has survived. The main breeding pur-
pose of the Ôold Lipica studÕ was to breed exclusive noble
horses for show and parade at the Imperial Court in Vienna.
Therefore, breeding of Lipizzan horses was restricted for
nearly four centuries to a few studs preventing a strong
expansion of the breed. Since the middle of the 20th century
Lipizzan horses have also been owned and bred privately.
Today the worldwide number of purebred horses ranges
between 2000 and 3000 animals. The breeding goals for the
Lipizzan horse are mainly defined by a few state-owned stud
farms in Austria, Croatia, Hungary, Italy, Romania, Slovakia
and Slovenia that were once part of the Austro-Hungarian
Empire. The breeding objectives are not identical in all
countries and are sometimes changing. For example,
the main breeding goal for the Austrian Lipizzan horse is

Ó 2004 International Society for Animal Genetics, Animal Genetics, 35, 285–292 285
286 Achmann et al.
classical dressage, the Hungarian breeding goal is coach
driving while the Italian Lipizzan horse is kept as a genetic
reserve (Zechner et al. 2002). The breeding books are closed
and migration of horses between studs was impeded until
recently by geographical and socio-political barriers. In a
population with such characteristics (closed population,
small census number, presence of selection and limited
gene flow) the loss of genetic variation through inbreeding
might lead to the extinction of the population (Amos &
Balmford 2001). Being among the oldest horse breeds in
Europe and because of its historical connection to the Austro-
Hungarian Empire, the Lipizzan horse is a living part of the
European cultural heritage. Thus, in addition to biological
reasons (see Kavar et al. 1999), there are historical and
cultural needs for monitoring the conservation status of the
Lipizzan horse. Within the past decade microsatellites have
been proven to be a reliable and frequently used tool for
the estimation of population parameters useful for conser-
vation management of wild and domestic animal populations
(Beaumont & Bruford 1999; Canon et al. 2000; Maudet et al.
2002).
The main objective of this study was to determine levels of
genetic variability within and between Lipizzan horse sub-
populations (studs). Secondly, we estimated the degree of
population subdivision and gene flow among Lipizzan
horses from different studs. Thirdly, we analysed and
graphically illustrated molecular genetic relationships
among subpopulations as well as among individual horses.
Finally, we tested the potential of microsatellite genotypes
for the assignment of individual horses with respect to their
stud of origin.
Materials and methods
Population sampling
A large fraction of the world Lipizzan breed population
(20–28% of the living purebred horses) was sampled
from state owned studs from seven countries: Austria –
Piber (153 horses including stallions from the Spanish
Riding School in Vienna); Croatia – Ðakovo (50 horses);
Hungary – Szilvesvarad (76 horses); Italy – Monterotondo
(63 horses); Romania – Fagaras (118 horses including
horses from the sister stud Beclean), Slovakia – TopolÔcianky
(42 horses) and Slovenia – Lipica (59 horses). Horses from
each country were considered as subpopulations. Being a
part of the Lipizzan breeding history, a sample of 47 Kladrub
horses (here considered as a single population) was taken as
a reference population. We reassessed Lipizzan stud samples
with the help of pedigree records to account for recent gene
flow due to the exchange of horses and due to crossbreeding
between studs. Horses with one or both parents originating
from a foreign stud were termed ÔmigrantÕ horses and were
removed from the data set for some analyses. This adjusted
data set consisted of 407 Lipizzan horses.
Ó 2004 International Society for Animal Genetics, Animal Genetics, 35, 285–292
DNA isolation and characterization of microsatellites
Genotyping was performed for 561 Lipizzan and 47 Kladrub
horses and included 22 dinucleotide repeat microsatellite
loci (AHT4, AHT5, AHT21, ASB2, HMS1, HMS2, HMS3,
HMS6, HMS7, HMS8, HTG4, HTG6, HTG7, HTG10,
LEX053, LEX054, MPZ002, NVHEQ18, UCDEQ405,
UCDEQ437, UCDEQ505 and VHL20) dispersed over 16
different chromosomes (Table 1). Blood samples were taken
from the jugular vein with ethylenediaminetetraacetic acid
(EDTA)-containing vacutainers (VacuetteÒ Greiner, Huber
u. Co., Reinach, Switzerland). The DNA was isolated from
whole blood using a Nucleon BACC3 DNA Extraction Kit
(Amersham Biosciences, Vienna, Austria) following the
manufacturer’s instruction. The DNA microsatellites were
amplified by polymerase chain reaction (PCR) and fluores-
cently labelled PCR products were separated on a capillary
electrophoresis instrument (ABI PRISM 310 Genetic Ana-
lyzer; Applied Biosystems, Vienna, Austria) applying
standard loading and electrophoresis conditions (Wenz et al.
1998). Allele sizes were determined after processing of raw
data with software packages GENESCAN 2.0 and GENOTYPER 2.1
(Applied Biosystems). Allele designations refer to the centre
value (calculated as the weighted average) of similar sized
alleles (within the range of 1 bp) rounded to the next
integer as suggested by the Ôadd categoryÕ submenu of
GENOTYPER software. Therefore, alleles did not necessarily
differ in size by multiples of 2 bp, which would be expected if
the number of dinucleotide repeats is the variable factor.
Supplementary information on primers and PCR conditions
are available on request.
Statistical analysis
Allele frequency, the number of alleles, observed hetero-
zygosity and gene diversity [heterozygosity expected from
Hardy–Weinberg equilibrium (HWE) assumptions] were
calculated across different loci and/or subpopulations using
the SAS package (SAS Institute 1999–2001). Tests for
deviations from HWE were performed across all loci and
subpopulations by GENEPOP v.3.3 (March 2001) software
package (Raymond & Rousset 1995; available at ftp://
ftp.cefe.cnrs-mop.fr/genepop/) applying the exact test with
default settings of the Markov Chain Monte Carlo meth-
odology (Guo & Thompson 1992). Where appropriate,
P-values were adjusted according to Hochberg’s step-up
Bonferroni method (Hochberg 1988) using the SAS package
(SAS Institute 1999–2001). Genetic structuring of the
Lipizzan horse breed was analysed by Wright’s F statistics
(Wright 1951). Fixation coefficients (FST, FIS and FIT) were
calculated according to Weir & Cockerham (1984) using
the GENETIX 4.04 (January 2003) software package (Belkhir
et al. 2002; available at http://www.univ-montp2.fr/

genetix/genetix/intro.htm). The number of effective
migrants per generation (Nm) was based on FST estimates
 Lipizzan horse population subdivision 287

Table 1 Characteristics and summary statistics for the microsatellite loci analysed in the Lipizzan population (561 horses); number of successfully
typed horses (N), size range (bp), observed heterozygosity, gene diversity and chromosome location.

Number Observed Gene Chromosome

Locus N of alleles Size range heterozygosity diversity location

AHT4 559 7 144–159 0.716 0.732 24

AHT5 545 6 127–138 0.758 0.746 8
AHT21 559 7 195–211 0.481 0.511 1
ASB21 501 7 235–251 0.725 0.773 15
HMS1 561 5 174–184 0.538 0.527 15
HMS2 560 8 217–236 0.720 0.736 10
HMS31 557 7 149–169 0.591 0.778 9
HMS6 556 6 157–168 0.665 0.637 4
HMS7 557 7 171–186 0.732 0.753 1
HMS8 561 5 205–221 0.635 0.673 19
HTG4 502 6 127–137 0.647 0.660 9
HTG6 555 6 78–97 0.544 0.568 15
HTG7 558 3 118–126 0.643 0.645 4
HTG10 557 10 86–108 0.682 0.685 21
LEX053 537 5 122–132 0.730 0.715 23
LEX0541 554 7 164–182 0.509 0.576 18
MPZ0021 513 5 72–89 0.374 0.590 16
NVHEQ18 557 11 112–150 0.785 0.821 10
UCDEQ405 532 7 247–274 0.579 0.607 25
UCDEQ437 554 9 165–191 0.590 0.646 3
UCDEQ505 561 10 175–195 0.724 0.729 16
VHL20 558 9 86–105 0.765 0.764 30

1
Microsatellite loci that were excluded from analysis (see Materials and Methods).

(Weir & Cockerham 1984) and was calculated by the for-
mula described in Wright (1969). The relationship between
inter-stud Nm estimates (Table 3) obtained from genetic
analysis (561 horses) and the proportion of migrating
horses between pairs of studs (sum of emigrating and
immigrating horses divided by the total sample size for pairs
of studs, Table 4) was analysed by linear regression analy-
sis.
Molecular genetic relationships among subpopulations
were derived using the allele sharing distance DAS
(Chakraborty & Jin 1993). DAS is not based on an evolu-
tionary model and it has been shown to be appropriate for
phylogenetic studies of closely related populations where
drift is the main force of differentiation (Goldstein & Pollock
1997). Distance matrices were calculated using POPULATION
v.1.2.28 software package (Olivier Langella; available at
http://www.pge.cnrs-gif.fr/bioinfo/populations/). Phyloge-
netic dendrograms were constructed from the DAS matrix
applying the unweighted pair group method with arithmetic
mean algorithm (UPGMA, Sneath & Sokal 1973). The
UPGMA algorithm was chosen because it is considered to be
superior to the Neighbor-Joining algorithm when sub-
sequent migrations between populations are present.
Robustness of the dendrogram was tested by 1000 boot-
strap samplings. The dendrogram was drawn with the
TREEVIEW v.1.6.6. (2001) software package (Page 1996;
available at http://www.taxonomy.zoology.gla.ac.uk/rod/
rod.html). DAS distances between subpopulations as well as
allele sharing between individuals were used for principal
component analyses (PCA). The PCA and related graphics
(bubble plot) were prepared with the SAS software (SAS
Institute 1999–2001). The contour plot was created using
algorithms written for the SAS/IML module (Friendly
1991).
Assignment of individual horses to their most likely
subpopulation was performed by GENECLASS v.1.0.02
(February 1999) software package (Cornuet et al.
1999; available at http://www.montpellier.inra.fr/URLB/
geneclass/geneclass.html). Among several methods (ÔDASÕ,
ÔNei standardÕ, ÔNei minimumÕ, ÔNei DAÕ, ÔCavalli-SforzaÕ,
ÔGoldsteinÕ and ÔBayesianÕ) applied, the best assignment was
obtained by the Bayesian method. Self-classification of indi-
viduals in reference populations was applied using the Ôleave
one outÕ procedure.
Results
Genetic diversity
After Hochberg’s step-up Bonferroni correction for multiple
comparisons (Hochberg 1988), a significant deviation from
HWE was observed for loci HMS3, LEX054 and MPZ002
(see Table 1). At locus HMS3 as well as at ASB2 a point
mutation in the primer-binding site is responsible for allele

Ó 2004 International Society for Animal Genetics, Animal Genetics, 35, 285–292
288 Achmann et al.

calling errors because of weak amplification of some alleles
(Achmann et al. 2001). As a consequence we found a sig-
nificant excess of homozygotes at HMS3. For ASB2 the
deviation was not significant. Although we cannot prove
their existence it is likely that low- or non-amplifying alleles
(null alleles) account for the observed homozygote excess at
LEX054 and MPZ002. To avoid potential bias on popula-
tion genetic estimates, loci ASB2, HMS3, LEX054 and
MPZ002 were not further considered.
The total number of alleles found for the 18 microsatellite
loci in the Lipizzan horse was 127. The mean number of
alleles per locus was 7.06, with a range of 3–11. Observed
heterozygosity ranged from 0.481 for microsatellite AHT21
to 0.785 for microsatellite NVHEQ18. The same markers,
AHT21 and NVHEQ18, have the lowest (0.511) and the
highest (0.821) gene diversity, respectively. More informa-
tion related to diversity and summary statistics across
microsatellite loci is given in Table 1.
Microsatellite diversity statistics across Lipizzan sub-
populations are presented in Table 2. Diversity statistics
(observed heterozygosity, gene diversity and the mean
number of alleles) were very similar for all subpopula-
tions. The observed heterozygosity (0.635), gene diversity
(0.607) and the mean number of alleles (4.722) were
lowest in Italy. The highest observed heterozygosity
(0.707) and gene diversity (0.675) was observed in
Hungary while the highest mean number of alleles
(6.222) was found in Austria. Significant negative FIS
coefficients were estimated for Austrian, Croatian, Hun-
garian and Italian subpopulations suggesting an excess of
heterozygotes because of selective breeding that
avoids matings between closely related individuals
(inbreeding).
Genetic structure and gene flow
Population differentiation of Lipizzan subpopulations and
the Kladrub population is presented by pairwise FST coeffi-
cients (see Table 3). For the current Lipizzan subpopulations
the FST coefficients ranged from 0.021 (between Austria
and Slovenia) to 0.080 (between Italy and Romania).
Thus, from 2.1 to 8.0% of the microsatellite variability is
explained by the subdivision of populations while the
remaining variability is explained by the variation within

Table 2 Microsatellite diversity of the Lipizzan horse across subpopulations based on 18 loci; observed heterozygosity and gene diversity with SD in
parenthesis, the mean number of alleles (MNA), and heterozygote deficiency coefficient (FIS).

Sample N Observed heterozygosity Gene diversity MNA FIS (95% CI)1

Austria 153 0.656 (0.112) 0.636 (0.106) 6.222 )0.027 ()0.053 to )0.009)
Croatia 50 0.674 (0.120) 0.645 (0.094) 5.167 )0.035 ()0.083 to )0.003)
Hungary 76 0.707 (0.097) 0.675 (0.067) 5.778 )0.041 ()0.083 to )0.009)
Italy 63 0.635 (0.092) 0.607 (0.103) 4.722 )0.038 ()0.086 to )0.003)
Romania 118 0.671 (0.101) 0.665 (0.086) 5.611 )0.005 ()0.033 to 0.016)
Slovakia 42 0.657 (0.144) 0.632 (0.125) 5.278 )0.027 ()0.088 to 0.007)
Slovenia 59 0.645 (0.154) 0.631 (0.130) 5.222 )0.012 ()0.066 to 0.015)
Total 561 0.663 (0.088) 0.675 (0.084) 7.056

1
Wright’s heterozygote deficiency coefficient (FIS) is presented with empirical confidence interval (95%) obtained from 1000 bootstrap replicates.

Table 3 Population differentiation among Lipizzan subpopulations and Kladrub horse, based on 18 microsatellite loci, is presented by FST estimates1
as well as by the number of effective migrants per generation2 (Nm) that are in balance with genetic drift (in parenthesis). The estimates related to the
current origin of horses (561 Lipizzans, 47 Kladrub horses) are presented below diagonal, while estimates for the adjusted data set3 (407 Lipizzans, 47
Kladrub horses) are presented above diagonal.

(Sub)population Austria Croatia Hungary Italy Romania Slovakia Slovenia Kladrub

Austria 0.073 (3.16) 0.067 (3.37) 0.051 (4.65) 0.061 (3.83) 0.063 (3.69) 0.038 (6.24) 0.103 (2.17)
Croatia 0.049 (4.81) 0.058 (4.09) 0.091 (2.50) 0.068 (3.40) 0.066 (3.52) 0.069 (3.39) 0.110 (2.02)
Hungary 0.045 (5.27) 0.040 (6.04) 0.077 (3.00) 0.053 (4.46) 0.072 (3.21) 0.047 (5.03) 0.086 (2.64)
Italy 0.031 (7.71) 0.076 (3.05) 0.067 (3.46) 0.087 (2.63) 0.101 (2.22) 0.048 (4.95) 0.109 (2.03)
Romania 0.052 (4.59) 0.059 (3.94) 0.044 (5.40) 0.080 (2.89) 0.081 (2.84) 0.060 (3.94) 0.084 (2.72)
Slovakia 0.037 (6.50) 0.034 (7.05) 0.046 (5.16) 0.074 (3.11) 0.066 (3.55) 0.082 (2.79) 0.128 (1.70)
Slovenia 0.021 (11.69) 0.059 (3.98) 0.048 (5.01) 0.040 (6.03) 0.059 (4.01) 0.054 (4.38) 0.102 (2.20)
Kladrub 0.097 (2.34) 0.110 (2.03) 0.083 (2.77) 0.106 (2.11) 0.084 (2.74) 0.115 (1.91) 0.102 (2.20)

1
FST estimates were calculated as described by Weir & Cockerham (1984).
2
The number of effective migrants per generation (Nm) was estimated using the formula, Nm ¼ (1 ) FST)/4 FST, derived by Wright (1969).
3
See Materials and Methods.

Ó 2004 International Society for Animal Genetics, Animal Genetics, 35, 285–292
 Lipizzan horse population subdivision 289

Table 4 Gene flow between Lipizzan subpopulations deduced from stud book records of 561 horses.

From/To Austria Croatia Hungary Italy Romania Slovakia Slovenia Emigrants (total)

Austria 0 4.0 4.0 6.0 4.5 12.0 30.5

Croatia 4.5 0.5 0 0 9.0 0.5 14.5
Hungary 7.0 3.0 0 1.0 0.5 0 11.5
Italy 8.0 0 0 0 0 0.5 8.5
Romania 0.5 0 1.0 0 0 0 1.5
Slovakia 3.0 0 3.0 0 0.5 0 6.5
Slovenia 11.0 3.0 0 0 0 0 14.0
Immigrants (total) 34.0 6.0 8.5 4.0 7.5 14.0 13.0 87.0

Gene flow is presented as number of migrant horses per generation from a subpopulation (rows) and to a subpopulation (columns). Horses for which
both parents originate from a foreign stud were counted as Ô1Õ. Horses for which only one parent originate from a foreign stud were counted as Ô0.5Õ.

subpopulations. As expected from the breeding history

pairwise FST values between Lipizzan subpopulations and
the Kladrub population were higher (ranging from 0.083 to
–
0.115). When migrant horses were excluded from the
analysis estimated pairwise FST values were to some extent
higher (ranging from 0.038 to 0.101 between Lipizzan –S
subpopulations and from 0.084 to 0.128 between Lipizzan
subpopulations and Kladrub horse). Gene flow between
–
subpopulations is shown by the estimates of the number of
migrants per generation (Nm, where N is the total effective
number of horses and m is the migration rate; see Table 3). –
The highest migration rate was estimated between Austria
and Slovenia (11.69) and the lowest migration rate between
–
Italy and Romania (2.86).
The regression analysis demonstrated that a large pro-
portion of the Nm variation is explained by the observed –
proportion of migrating horses between studs (R2 ¼ 0.713,
P < 0.0001; see Tables 3 and 4). –

Figure 1 Unweighted pair group method with arithmetic mean algo-

Molecular genetic relationship among Lipizzan
subpopulations
The UPGMA dendrogram and the PCA-based bubble plot
further clarify the genetic relationships between subpop-
ulations (see Figs 1 and 2). The UPGMA dendrogram
shows the relationship between subpopulations while
bootstrap values support the reliability of the dendrogram.
Low bootstrap values do not support well the clusters
(interior branches) for Croatia, Hungary and Slovakia. In
the bubble plot, the placements of populations corres-
ponds approximately with the geographic origin of studs.
A total of 91.9% of the variance was accounted by first
three PCA variables.
Molecular genetic relationship among individual horses
The PCA performed on individual genotypes of Lipizzan and
Kladrub horses revealed additional information not dedu-
cible from the phylogenetic dendrogram and the bubble
plot. The molecular genetic relationship among individual
horses is presented by the contour plot (Fig. 3). The contour
rithm (UPGMA) dendrogram of the molecular genetic relationship
among Lipizzan subpopulations (561 horses) and a Kladrub population
(47 horses). The UPGMA dendrogram is based on allele sharing
distance (DAS, Chakraborty & Jin 1993) derived from 18 microsatellite
loci. The number for each interior branch is the bootstrap value from
1000 replications.
plot shows the genetic overlap between studs and illustrates
that, although genetic differences between Lipizzan horse
subpopulations are present, the degree of genetic similarity
is high.
Assignment of the Lipizzan and Kladrub horses
The results of the assignment tests are shown in Table 5.
Of 608 analysed Lipizzan and Kladrub horses 492 (80.9%)
were assigned properly. The assignment precision for
Lipizzans ranged from 61.9 (Slovakia) to 95.8% (Romania)
properly assigned horses. The assignment precision was
influenced by migrant horses. Of 154 Lipizzan migrants,
having at least one parent originating from a foreign

Ó 2004 International Society for Animal Genetics, Animal Genetics, 35, 285–292
290 Achmann et al.
Figure 2 Bubble-plot of the molecular genetic relationship among
Lipizzan subpopulations (561 horses) and a Kladrub population (47
horses). The bubble-plot is based on the allele sharing distance (DAS,
Chakraborty & Jin 1993) derived from 18 microsatellite loci and is
presented by the first three components from a principal component
analysis (PCA, the third component is proportional to the diameter of
the bubble). The first component explains 66.4%, the second 16.8%
and the third 8.7% of the total variation. Lipizzan horse: 1, Italy; 2,
Slovenia; 3, Austria; 4, Slovakia; 5, Croatia; 6, Hungary; 7, Romania;
8, Kladrub horse.
Figure 3 Contour plot with 75% confidence region of the molecular
genetic relationship among individual horses (561 Lipizzan and
47 Kladrub horses, respectively). The contour plot is based on the allele
sharing distance (DAS, Chakraborty & Jin 1993) derived from
18 microsatellite loci and is presented by the first two components from
a principal component analysis (PCA). The first component explains
25.9% and the second 6.9% of the total variation. Lipizzan horse:
1, Italy; 2, Slovenia; 3, Austria; 4, Slovakia; 5, Croatia; 6, Hungary;
7, Romania; 8, Kladrub horse.
subpopulation, 74 (48.2%) horses were not assigned
properly. When assignment testing was carried out with
the adjusted data set of 454 horses (407 Lipizzans and
47 Kladrub horses) the assignment precision improved
substantially and 418 (92.1%) horses were assigned cor-
rectly (Table 5). Neither a Kladrub horse was misclassified
(see Table 5) as Lipizzan or a Lipizzan horse misclassified
as a Kladrub horse.
Ó 2004 International Society for Animal Genetics, Animal Genetics, 35, 285–292
Discussion
On a population level, the Lipizzan horse has maintained a
level of genetic (microsatellite) diversity comparable with
that of other horse breeds (see Wimmers et al. 1998; Canon
et al. 2000; Bjørnstad et al. 2000; Cunningham et al. 2001;
Vilà et al. 2001) as well as to that of other domestic animal
species (see Buchanan et al. 1994; MacHugh et al. 1998;
Saitbekova et al. 1999; Fan et al. 2002). The differences
among the subpopulations are only minor and subpopula-
tions show no obvious signs of genetic impoverishment. The
results from microsatellite analyses are concordant with
those obtained from mtDNA analysis (Kavar et al. 1999),
genetic analysis of the equine MHC gene region (Eder et al.
2001; Curik et al. 2003a) and pedigree analysis (Zechner
et al. 2002).
The genetic differentiation between Lipizzan subpopula-
tions, although moderate, is apparent. A cluster is obvious,
which consists of the subpopulations of Austria, Italy and
Slovenia. These studs represent the classical pool of Lipiz-
zan horse breeding. Historically these studs arose by
splitting the Lipica population after the First and Second
World War. Moreover, all three subpopulations focus on
breeding the classical Lipizzan lines, whereas in other
subpopulations additional mare lines are much more fre-
quently used for breeding. Another cluster is formed by the
Croatian, Hungarian and Slovakian subpopulations. While
Croatian and Hungarian subpopulations are genetically
more similar and form a separate cluster, the Slovakian
subpopulation is positioned closer to the classical Lipizzan
subpopulations. The origin of the Croatian and Hungarian
subpopulations goes back to the Wars of Napoleon, they
have similar breeding goals and they are geographically
closely located. Therefore, their higher genetic relationship
is not surprising. The Slovakian subpopulation was formed
after the First World War from the horses from Lipica.
Later, because of the breeding goal, geographical and
socio-political barriers, the Slovakian subpopulation
exchanged relatively more horses with Croatia, Hungary
and Austria than with other subpopulations. The Roma-
nian subpopulation forms a separate unit with the largest
differentiation from the Italian subpopulation. Among
other subpopulations, the Romanian population is histor-
ically and geographically closest to the Hungarian sub-
population. Although there is a close historical
relationship, there was a clear genetic differentiation
between the Kladrub and the Lipizzan horse breed.
The data available allowed comparisons between gene
flow estimates obtained from microsatellites and pedigree
records. In all analyses, conclusions drawn from microsat-
ellite data were substantiated by information from pedi-
grees. First, the comparison between FST values estimated
from the complete data set vs. without migrant horses
(higher FST values for the complete data set; see Table 3)
indicates that exchange of horses between studs is an
 Lipizzan horse population subdivision 291

Table 5 Population assignment1 (%) between seven Lipizzan subpopulations (561 horses) and a Kladrub population (47 horses) by means of 18
microsatellite loci.

Subpopulation Austria Croatia Hungary Italy Romania Slovakia Slovenia Kladrub Percentage of errors

Austria 69.3 (89.3) – 5.8 (1.9) 4.6 (1.0) 3.3 (1.9) 5.2 (1.0) 11.8 (4.9) – 30.7 (10.7)
Croatia 2.0 (–) 86.0 (95.2) 4.0 (–) – 2.0 (4.8) – 6.0 (–) – 14.0 (4.8)
Hungary 8.0 (–) 2.6 (3.7) 85.5 (92.6) – (1.9) – 1.3 (–) 2.6 (1.9) – 14.5 (7.4)
Italy 9.5 (2.0) – – 87.3 (94.1) – – 3.2 (3.9) – 12.7 (5.9)
Romania – (0.9) 1.7 (0.9) 1.7 (–) 0.9 (–) 95.8 (98.3) – – – 4.2 (1.7)
Slovakia 19.5 (6.4) 4.8 (–) 14.3 (6.4) – – 61.9 (83.9) – (3.2) – 38.1 (16.1)
Slovenia 20.3 (9.4) 1.7 (3.1) 3.4 (6.4) 6.8 (9.4) – 5.1 (3.1) 62.7 (68.8) – 37.3 (31.2)
Kladrub – – – – – – – 100.0 (100.0) 0.0 (0.0)

Values in parenthesis refer to the adjusted Lipizzan horse data set (407 horses, see Materials and Methods). Dashes (–) indicate none assignment.
1
The Bayesian method and the Ôleave one outÕ procedure was applied for self-classification.

important force counteracting the genetic differentiation of
Lipizzan horse subpopulations. Secondly, a high percentage
(71.3%) of the variability among pairwise FST values is
explained by actual migrations (positive linear regression of
Nm values on the proportion of observed migrations
between subpopulations) demonstrating, further, strong
relationship between population structure and observed
gene flow. Finally, the assignment precision was substan-
tially improved by exclusion of migrant horses suggesting
that pedigree information, when available (for example,
available migrant status of at least one parent), might be
combined efficiently with genetic assignment analysis.
In conclusion, the results of the present study are in
accordance with the historical description of the subpopu-
lation (stud) formation, with the observed migration pattern
as well as with pedigree information (see Zechner et al.
2002). Together with information obtained from pedigree
analyses (Zechner et al. 2002), from morphological meas-
ures (Zechner et al. 2001; Curik et al. 2003b), from mit-
ochondrial DNA analyses (Kavar et al. 1999) and
immunogenetics (Eder et al. 2001; Curik et al. 2003a), they
could provide a basis for the development of breeding
strategies, which consider classical horse breeding as well as
recent achievements in the field of population and conser-
vation genetics. Further, the results from assignment tests
show that a relatively accurate classification precision
might be reached even at the subpopulation level, offering
the possibility for the molecular genetic control of the
purebred status of the Lipizzan horse breed.
Acknowledgements
Authors are particularly grateful to the Austrian Federal
Ministry of Agriculture and Forestry, to the stud of Piber
and to the Spanish riding school in Vienna for their support.
Authors greatly acknowledge the hospitality and the sup-
port of the stud directors and their staff. Finally, authors are
grateful to Marion Deutschmann for help with DNA
extraction, to Dragan Tupajić for the graphical design of the
figures and to three anonymous reviewers for comments on
an earlier version of the manuscript. Data analysed in this
study are part of the joint research project ÔBiotechnical
methods in the maintenance of the genetic diversity in the
Lipizzan horse breedÕ (EC-INCO-Copernicus project No.
IC15CT96-0904).
References
Achmann R., Huber T., Wallner B., Dovc P., Müller M. & Brem G.
(2001) Base substitutions in the sequences flanking microsatellite
markers HMS3 and ASB2 interfere with parentage testing in the
Lipizzan horse. Animal Genetics 32, 52.
Amos W. & Balmford A. (2001) When does conservation genetics
matters? Heredity 87, 257–65.
Beaumont M.A. & Bruford M.W. (1999) Microsatellites in con-
servation genetics. In: Microsatellites: Evolution and Applications
(Ed. by D.B. Goldstein & C. Schlötterer), pp. 165–82. Oxford
University Press, Oxford, UK.
Belkhir K., Borsa P., Chikhi L., Raufaste N. & Bonhomme F.
(2002) GENETIX, Version 4.04 logiciel sous Windows TM pour la
Genetique des populations. Laboratoire Génome, Populations,
Interactions CNRS UMR 5000, Université Montpellier II,
Montpellier, France.
Bjørnstad G., Gunby E. & Røed K.H. (2000) Genetic structure of
Norwegian horse breeds. Journal of Animal Breeding and Genetics
117, 307–17.
Buchanan F.C., Adams L.J., Littlejohn R.P., Maddox J.F. & Crawford
A.M. (1994) Determination of evolutionary relationship among
sheep breeds using microsatellites. Genomics 22, 401–5.
Canon J., Checa M.L., Carleos C., Vega-Pla J.L., Vallejo M. & Dunner
S. (2000) The genetic structure of Spanish Celtic horse breeds
inferred from microsatellite data. Animal Genetics 31, 39–48.
Chakraborty R. & Jin L. (1993) Determination of relatedness
between individuals by DNA fingerprinting. Human Biology 65,
875–95.
Cornuet J.-M., Piry S., Luikart G., Estroup A. & Solignac M. (1999)
New methods employing multilocus genotypes to select or
exclude populations as origins of individuals. Genetics 153,
1989–2000.
Cunningham E.P., Dooley J.J., Splan R.K. & Bradley D.G. (2001)
Microsatellite diversity, pedigree relatedness and the contribu-
tions of founder lineages to thoroughbred horses. Animal Genetics
32, 360–4.

Ó 2004 International Society for Animal Genetics, Animal Genetics, 35, 285–292
292 Achmann et al.
Curik I., Fraser D., Eder C., Achmann R., Swinburne J., Binns M.,
Crameri R., Brem G., Sölkner J. & Marti E. (2003a) Association
between the MHC gene region and variation of the serum IgE
levels against specific mould allergens in the horse. Genetics
Selection Evolution 35 (Suppl. 1), S177–90.
Curik I., Zechner P., Sölkner J., Achmann R., Bodo I., Dovc P.,
Kavar T., Marti E. & Brem G. (2003b) Inbreeding, microsatellite
heterozygosity, and morphological traits in Lipizzan horses.
Journal of Heredity 94, 125–32.
Eder C., Curik I., Brem G., Crameri R., Bodo I., Habe F., Lazary
S., Sölkner J. & Marti E. (2001) Influence of environmental
and genetic factors on allergen-specific immunoglobulin-E
levels in sera from Lipizzan horses. Equine Veterinary Journal
33, 714–20.
Fan B., Wang Z.-G., Li Y.-J. et al. (2002) Genetic variation analysis
within and among Chinese indigenous swine populations using
microsatellite markers. Animal Genetics 33, 422–7.
Friendly M. (1991) SASÒ System for Statistical Graphics, 1 edn.
SAS Institute Inc., Cary, NC, USA.
Goldstein D.B. & Pollock D.D. (1997) Launching microsatellites: a
review of mutation processes and methods of phylogenetic
inference. Journal of Heredity 88, 335–42.
Guo S.W. & Thompson E.A. (1992) Performing the exact test of
Hardy-Weinberg proportion for multiple alleles. Biometrics 48,
361–72.
Hochberg Y. (1988) A sharper Bonferroni procedure for multiple
significance testing. Biometrika 75, 800–3.
Kavar T., Habe F., Brem G. & Dovč P. (1999) Mitochondrial D-loop
sequence variation among the 16 maternal lines of the Lipizzan
horse breed. Animal Genetics 30, 423–30.
MacHugh D.E., Loftus R.T., Cunningham P. & Bradley D.G.
(1998) Genetic structure of seven European cattle breeds
assessed using 20 microsatellite markers. Animal Genetics 29,
333–40.
Maudet C., Miller C., Bassano B., Breitenmoser-Würsten C., Gautier
D., Obexer-Ruff G., Michallet J., Taberlet P. & Luikart G. (2002)
Microsatellite DNA and recent statistical methods in wildlife
conservation management: applications in Alpine ibex [Capra ibex
(ibex)]. Molecular Ecology 11, 421–36.
Ó 2004 International Society for Animal Genetics, Animal Genetics, 35, 285–292
Page R.D.M. (1996) TREEVIEW: an application to display phylo-
genetic trees on personal computers. Computer Applications in the
Biosciences 12, 357–8.
Raymond M. & Rousset F. (1995) GENEPOP, Version 1.2 popula-
tion genetics software for exact tests and ecumenicsm. Journal of
Heredity 86, 248–9.
Saitbekova N., Gailard C., Obexer-Ruff G. & Dolf G. (1999) Genetic
diversity in Swiss goat breeds based on microsatellite analysis.
Animal Genetics 30, 36–41.
SAS Institute Inc. (1999–2001) SAS/STAT Software: Changes and
Enhancements Through Release 8.02. Cary, NC, USA.
Sneath P.H.A. & Sokal R.R. (1973) Numerical Taxonomy. W.H.
Freeman, San Francisco, CA, USA.
Vilà C., Leonard J.A., Götherström A., Marklund S., Sandberg K.,
Lindén K., Wayne R.K. & Ellegren H. (2001) Widespread origins
of domestic horse lineages. Science 291, 474–7.
Weir B.S. & Cockerham C.C. (1984) Estimating F-statistics for the
analysis of population structure. Evolution 38, 1358–70.
Wenz H., Robertson J.M., Menchen F., Oaks F., Demorest D.M.,
Scheibler D., Rosenblum B.B., Wike C., Gilbert D.A. & Efcavitch
J.W. (1998) High precision genotyping by denaturating capillary
electrophoresis. Genome Research 8, 69–80.
Wimmers K., Ponksuksili S., Schmoll F., Hardge T., Hatzipanagio-
tou A., Weber J., Wostmann S., Olek K. & Schellander K. (1998)
Efficiency of microsatellite markers of the international standard
panel for parentage control in German horse population. Zuch-
tungskunde 70, 233–41.
Wright S. (1951) The genetical structure of populations. Annual
Eugenetics 15, 323–54.
Wright S. (1969) The Theory of Gene Frequencies: Evolution and the
Genetics of Populations, Vol. 2. Chicago University press, Chicago,
USA.
Zechner P., Zohman F., Sölkner J., Bodo I., Habe F., Marti E. & Brem
G. (2001) Morphological description of the Lipizzan horse popu-
lation. Livestock Production Science 69, 163–77.
Zechner P., Sölkner J., Druml T., Baumung R., Achmann R., Bodo
I., Marti E., Habe F. & Brem G. (2002) Analysis of diversity and
population structure in the Lipizzan horse breed based on pedi-
gree information. Livestock Production Science 77, 137–46.