American Journal of Botany 96(5): 1–8. 2009.
Agroforestry systems are the result of a long evolutionary
process during which an association between natural elements
such as trees and shrubs share the same stands with crops and
sometimes with households. They are characterized by the
dominance of several multipurpose tree species, which are con-
served and maintained in the ﬁ eld by farmers because of their
To date, the proportion and structure of variation maintained
on-farm in tree species during the development of tropical agro-
forestry systems is largely uncharacterized at an intraspeciﬁ c
level. Knowledge about population genetics is, however, of key
importance for understanding microevolutionary processes in
plant populations and supporting or developing appropriate use
and conservation strategies ( Lengkeek et al., 2006 ). The long-
term viability of tree species within agroforestry systems de-
pends upon a wide genetic base providing the capacity to adapt
to environmental ﬂ uctuations or changing farmer requirements,
such as change in species use or planting niche ( Lengkeek et al.,
To date, few studies have focused on genetic variation of
semidomesticated edible trees growing in parkland systems in
sub-Saharian Africa (Allaye Kelly et al, 2004 ; Sanou et al,
2005 ; Lengkeek et al, 2006 ). Generally, spatial genetic struc-
ture in tree species is inﬂ uenced by several biological factors
such as gene ﬂ ow (mediated by seed and pollen dispersal), den-
sity, fragmentation, colonization history, isolation into small
patches, differential mortality, and microenvironmental selec-
tion ( Wright, 1951 , Heywood, 1991 , Epperson, 1993 , Kang and
Chung, 2000 ; Vekemans and Hardy, 2004 ). More speciﬁ cally,
in agroforestry systems, all these factors may be inﬂ uenced by
human activity leading to many changes in ecosystem processes
with various impacts ( Young and Merriam, 1994 ; Aldrich et al.,
1998 ; Allaye Kelly et al., 2004 ; Sanou et al., 2005 ).
West African agroforestry systems are dominated by several
multipurpose tree species such as Adansonia digitata , Parkia
africana , Blighia sapida , Tamarindus indica , and Vitellaria
paradoxa . On the basis of threats (e.g., bush ﬁ re, overgrazing,
and overexploitation) and on the economic importance of Adan-
sonia digitata (baobab tree) to the rural poor in Africa, the In-
ternational Centre for Underutilised Crops (ICUC) has accorded
high priority to enhance research and development for Adanso-
nia digitata ( Sidibe and Williams, 2002 ). Also, Bioversity In-
ternational (previously called International Plant Genetic
Resources Institute) has classiﬁ ed the baobab tree among the 10
top agroforestry tree species to be conserved and domesticated
in West Africa ( Eyog Matig et al., 2002 ).
African baobab ( Adansonia digitata L., Malvaceae) is natu-
rally associated with the savannah, especially the drier parts
( Wickens, 1982 ). It is a multipurpose tree species used daily by
rural farmers for food and medicine and is economically and
culturally important to local people. However, ﬁ eld studies
1 Manuscript received 4 August 2009; revision accepted 14 January 2009.
This work was supported by Bioversity International and Pioneer Hi-
Bred International Inc., a Dupont Company, through a Vavilov-Frankel
Fellowship and by Rothamsted International through its African Fellowship
Program for the ﬁ nancial support in Europe. Additional funding for DNA
ﬁ ngerprinting and ﬁ eldwork was provided by the DADOBAT-Project (EU-
Funding) The King Leopold III Fund for Nature Conservation and
Exploration. T.K. received a postdoctoral grant from Ghent University
(BOF). O.J.H. is a Research Associate with the Belgian Fund for Scientiﬁ c
Research (FNRS). T.K. and A.E.A. equally contributed to this work.
6 Author for correspondence (e-mail: firstname.lastname@example.org)
S PATIAL GENETIC STRUCTURING OF BAOBAB ( ADANSONIA
DIGITATA , MALVACEAE) IN THE TRADITIONAL AGROFORESTRY
SYSTEMS OF WEST AFRICA 1
Tina Kyndt , 2,6
Achille E. Assogbadjo ,
Olivier J. Hardy ,
Romain Glele Kaka ï , 3
Brice Sinsin , 3
Patrick Van Damme,
and Godelieve Gheysen
2 Department of Molecular Biotechnology, Ghent University (UGent), Coupure Links 653, B-9000 Ghent, Belgium;
of Applied Ecology, Faculty of Agronomic Sciences, University of Abomey-Calavi,
05 BP 1752 Cotonou, Benin;
4 Behavioural and Evolutionary Ecology Unit, CP 160/12, Facult é des Sciences, Universit é Libre de
Bruxelles, 50 Av. F. Roosevelt, B-1050 Brussels, Belgium; and
5 Laboratory of Tropical and Subtropical Agriculture and
Ethnobotany, Department of Plant Production, Ghent University (UGent), Coupure links 653, B-9000 Ghent (Belgium)
This study evaluates the spatial genetic structure of baobab ( Adansonia digitata ) populations from West African agroforestry
systems at different geographical scales using AFLP ﬁ ngerprints. Eleven populations from four countries (Benin, Ghana, Burkina
Faso, and Senegal) had comparable levels of genetic diversity, although the two populations in the extreme west (Senegal) had less
diversity. Pairwise F ST ranged from 0.02 to 0.28 and increased with geographic distance, even at a regional scale. Gene pools de-
tected by Bayesian clustering seem to be a byproduct of the isolation-by-distance pattern rather than representing actual discrete
entities. The organization of genetic diversity appears to result essentially from spatially restricted gene ﬂ ow, with some inﬂ uences
of human seed exchange. Despite the potential for relatively long-distance pollen and seed dispersal by bats within populations,
statistically signiﬁ cant spatial genetic structuring within populations (SGS) was detected and gave a mean indirect estimate of
neighborhood size of ca. 45. This study demonstrated that relatively high levels of genetic structuring are present in baobab at both
large and within-population level, which was unexpected in regard to its dispersal by bats and the inﬂ uence of human exchange of
seeds. Implications of these results for the conservation of baobab populations are discussed.
Key words: Adansonia digitata ; agroforestry systems; Malvaceae; genetic structure; spatial autocorrelation; West Africa.
2American Journal of Botany [Vol. 96
2006 ). The DNA ﬁ ngerprints were scored by visual inspection for presence (1)
or absence (0) of speciﬁ c AFLP-bands. Only distinct, major bands were scored.
Data analyses — A model-based (Bayesian) clustering method was applied
on the presence/absence matrix to infer genetic structure in the data set, using
the software Structure version 2.0. ( Pritchard et al, 2000 ). Applying 250 000
iterations without using prior information of the number of populations (USE-
POPINFO = 0), different K -values (2 – 19) were evaluated to estimate the num-
ber of gene pools present in the data set. The most likely number of gene pools
was determined by Δ K as described in Evanno et al. (2005) . Individuals of the
11 populations were then assigned probabilistically to the inferred gene pools.
To estimate genetic diversity and differentiation among populations, we
analyzed population structure based on allele frequency using AFLPsurv ver-
sion 1.0. ( Vekemans, 2002 ) with the methods described by Lynch and Milligan
(1994) . Allelic frequencies at AFLP loci were estimated from the binary pres-
ence/absence matrix, under the assumption of Hardy – Weinberg equilibrium,
from the observed frequencies of fragments using the Bayesian approach pro-
posed by Zhivotovsky (1999) . A non-uniform prior distribution of allelic fre-
quencies was assumed with its parameters derived from the observed distribution
of fragment frequencies among loci (see note 4 in Zhivotovsky, 1999 ). Nei ’ s
(1973) gene diversity (also known as expected heterozygosity H e ) as well as
pairwise genetic differentiation ( F ST ) were computed. Signiﬁ cance of the ge-
netic differentiation between groups was tested by comparison of the observed
F ST with a distribution of F ST under a hypothesis of no genetic structure, ob-
tained by means of 1000 random permutations of individuals among groups.
Estimation of gene frequencies from dominant markers requires an a priori
assumption on the level of inbreeding within populations. Because ﬂ oral traits of
baobab suggest a predominantly outcrossing mating system, as observed for the
large majority of tropical trees, we assumed Hardy – Weinberg equilibrium to esti-
mate F ST and heterozygosity. However, because this assumption could not be
checked, we used an alternative phenetic method for partitioning genetic vari-
ability among and within populations and regions, an analysis of molecular vari-
ance (AMOVA) based on the presence/absence matrix using the program Arlequin
version 2.000 ( Schneider et al., 2000 ). This statistical analysis is recognized as an
effective tool to characterize population structure and degree of genetic differen-
tiation ( φ ST ) ( Excofﬁ er et al., 1992 ). It has also been shown to be effective in the
study of tetraploid species ( Jenczewski et al., 1999 ). This analysis is particularly
interesting in the case of A. digitata, for which cytological hypervariability and an
autotetraploid origin (resulting from aneuploid reduction from 4 × = 176) has been
suggested ( Baum and Oginuma, 1994 ) but not thoroughly established.
Under Wright ’ s isolation-by-distance model, the pairwise genetic diversity
between individuals and/or populations is expected to vary linearly with the
logarithm of their geographic distance on a two-dimensional scale ( Rousset
1997 , 2000 ; Hardy and Vekemans, 1999 , Hardy, 2003 ). To detect this kind of
spatial autocorrelation at large and regional scales, we calculated and tested the
correlation between F ST /(1 − F ST ) and the natural logarithm of geographical
distances between populations using the Mantel test ( Mantel, 1967 ) in the pro-
gram NTSYS-pc ( Rohlf, 2000 ). Statistical signiﬁ cance was evaluated with
have shown that the current semidomestication of baobab is al-
ready causing regeneration problems ( Assogbadjo et al, 2005 ),
and the species is threatened by overexploitation as well as by
bush ﬁ re, agriculture, and grazing in the parkland agroforestry
systems of West Africa ( Sidibe and Williams, 2002 ; Assogbadjo
et al, 2005 , 2006 ).
The only study on the population genetics of baobab, per-
formed by our research group in Benin ( Assogbadjo et al, 2006 ),
indicated some degree of physical isolation of the populations
collected in the three climatic zones of Benin and supposed a
certain impact of the environment and geographic distance on
the level of genetic structuring among the analyzed populations.
However, the study area was restricted in size.
The current study aims at studying the levels of spatial struc-
turing of baobab at different geographic scales. Speciﬁ cally,
this study involves a population genetic study of 11 baobab
populations from four West African countries where the spe-
cies is abundant and widely distributed in the parkland agrofor-
estry systems (Benin, Ghana, Burkina Faso, and Senegal). Our
main objectives were to characterize the level and organization
of genetic variation within and between A. digitata populations
at different geographical levels: large scale (the four West Afri-
can countries), regional scale (i.e., Benin, Ghana, and Burkina
Faso), and ﬁ ne scale (within populations).
This study addresses the following questions: (1) Is there a
geographical gradient of genetic diversity? (2) Does the organi-
zation of genetic variation at the large, country scale and at the
regional scale reveal discrete gene pools and/or a pattern of iso-
lation by distance? (3) Is there any evidence for spatial genetic
structuring within populations (SGS)?
This study provides insights into the level of gene ﬂ ow
among and within baobab populations and more speciﬁ cally on
the efﬁ ciency of humans and animals as seed dispersers of this
African tree species. The results of this study will be used to
build and enhance a database for species conservation and do-
mestication in the West African region.
MATERIALS AND METHODS
Sampling — In this study, baobab individuals from four countries were sam-
pled in the Sudanian and Sudano-Sahelian regions of West Africa: Benin,
Burkina Faso, Ghana, and Senegal. These countries are characterized by opti-
mal ecological conditions for the growth and development of this species ( FAO,
1981 ; Wickens, 1982 ).
Figure 1 shows the geographic distribution of the analyzed populations. Fif-
teen to 29 individuals were sampled within each population. A baobab popula-
tion was deﬁ ned as a group of baobab trees randomly and naturally distributed
in a traditional agroforestry system within a 30-km maximum radius. Two dif-
ferent populations are isolated from each other by a distance of at least 50 km.
Within a population, baobab individuals were randomly selected. In total, 11
populations of baobab represented by 251 individuals were sampled in the four
For Benin, sampling of a transect and megatransect within the studied eco-
logical zones and localities of Benin ( Assogbadjo et al., 2005 ) showed that the
natural mean population density is 5 baobabs/km
2 in the Sudanian zone. Per-
sonal observations in the ﬁ eld suggest that mean densities in the Sudano-Sahe-
lian zone of Ghana, Burkina Faso, and Senegal are 4, 6 and 7 baobabs/km
Molecular analyses — For each baobab, four or ﬁ ve leaves were harvested
and dried in silica gel for DNA extraction and AFLP analysis. Five primer pairs
(E-GT/M-ACGG, E-GT/M-ACGA, E-GA/M-ACGC, E-TC/M-GCGA, E-AT/
M-GCGG) were chosen based on an initial screening for polymorphism among
a limited number of samples and on band consistency and repeatability during
previous work on the genetic diversity of baobab in Benin ( Assogbadjo et al.,
Fig. 1. Map of West Africa, showing the geographical location of the
studied baobab populations.
May 2009] Kyndt et al. — Spatial genetic structure of ADANSONIA DIGITATA
assignment of individuals from the 11 sampled populations to
the inferred gene pools under the admixture model is shown in
Table 2 . The results revealed differences in gene pool represen-
tation among the populations, largely correlated with geogra-
phy. The populations from Benin (P1 to P3) mainly belonged to
gene pools 4, 5, and 8. In addition to these, population P3 also
included some members belonging to gene pool 1. All Ghana-
ian populations (P4 – P6) mostly contained members from gene
pool 6, while some individuals were attributed to gene pools 5
and 7. P7 from Senegal mainly contained members attributed to
gene pool 2. The second Senegalese population (P8) was con-
stituted by gene pools 1, 2, and 4. Populations P9 and P10 from
Burkina Faso had very similar gene pools, with individuals
from gene pools 3, 4, 7, and 8. Population P11 from this coun-
try, however, contained individuals attributed to gene pools 1,
4, 7, and 8, similar to the gene pool representation of population
P3 from Benin. Thus, despite the number of inferred gene pools
is close to the number of populations, most populations are rep-
resented by several gene pools, which may call into question
the actual signiﬁ cance of these inferred gene pools. The major-
ity of individuals were assigned to different gene pools, indicat-
ing that these inferred gene pools for the most part can be
attributed to variation in allelic frequencies caused by isolation-
by-distance rather than to actual genetically differentiated
Application of allele-frequency-based F -statistics revealed a
global F ST of 0.154 ± 0.080 ( P = 0.001). The total gene diver-
sity ( H t ) was estimated to be 0.354 ± 0.010, while the mean
gene diversity within populations ( H w ) and the average gene
diversity among populations ( H b ) were estimated at 0.299 ±
0.010 and 0.054 ± 0.004, respectively. Pairwise genetic dis-
tances between populations ( F ST ) were calculated (data not
shown), and all were statistically signiﬁ cant ( P < 0.001). While
the genetic distance between populations sampled in Senegal is
0.14, within-country genetic distance between populations from
Benin, Burkina Faso, and Ghana was generally lower or equal
to 0.07, except for P11 that again appeared to be more closely
related to P3 from Benin. Genetic distance between population
P7 of Senegal and those of other countries was very high, indi-
cating a low level of gene ﬂ ow between these populations and/
or substantial drift in P7, as suggested by the comparatively low
genetic diversity of this population. Maximum and minimum
F ST were 0.28 between P5 (Ghana) and P7 (Senegal), and 0.02
between P1-P2 (Benin) and P4-P5 (Ghana). Generally, these
results conﬁ rmed the observations from the gene pool cluster-
ing of the individuals ( Table 2 ).
Genetic variability distribution among and within popula-
tions was estimated by the analysis of molecular variance (AM-
OVA) ( Table 3 ). A two-level AMOVA of 251 individuals from
11 baobab populations showed that 79.32% of the total varia-
tion could be found at within-population level, while 20.68%
( φ ST = 0.2068, P < 0.001) of the genetic variation was present
among the populations. To investigate the impact of geography
on the level of variation, we performed a three-level AMOVA.
The data were investigated at large scale (between countries)
and at a smaller scale (within countries); 77.15% of the total
variation was observed within the populations, while the dif-
ferentiation at large scale was 14.43% and 8.42% at smaller
scale. All values were statistically signiﬁ cant ( P < 0.001).
Spatial autocorrelation at large and regional scales — F
(1 − F
ST ) matrices were compared with the natural logarithm of
geographical distance between pairs of populations from the
Spatial genetic structure within populations (SGS) refers to the decrease in
pairwise relatedness with distance within continuous populations ( Loiselle et
al., 1995 ). The strength of the genetic structuring can be estimated as the slope
of a kinship – distance curve, providing indirect information of gene dispersal
within populations ( Hardy, 2003 ; Vekemans and Hardy, 2004 ). Seven popula-
tions sampled with at least 22 individuals were used to assess their ﬁ ne-scale
genetic structuring. For each population, pairwise kinship coefﬁ cients ( F
Hardy, 2003 ) between the individuals were computed with the program SPA-
GeDi version 1.2 ( Hardy and Vekemans, 2002 ), and values were averaged for
four distance intervals, each holding at least 50 pairs of individuals. The slope
of the regression of F
ij with the spatial distance between individuals ( d
ij ) and it
natural logarithm [ln( d
ij )] was computed, providing the regression slopes b
Ld , respectively. Signiﬁ cance was tested by 10 000 permutations of the
spatial locations of individuals. Vekemans and Hardy (2004) developed a sta-
tistic to quantify the strength of spatial genetic structuring: Sp = − ( b
Ld )/(1 − F
1 is the coancestry between neighboring individuals [approximated by
F ( d ) for the ﬁ rst distance interval], to avoid bias from sampling effects.
AFLP ﬁ ngerprinting — For AFLP analysis, ﬁ ve primer pairs
were used based on the previous work on genetic diversity in
baobab populations from Benin ( Assogbadjo et al., 2006 ).
When bands from all individuals were considered, the ﬁ ve
primer combinations resulted in a total of 254 scored bands.
Only 53 bands were monomorphic across the complete germ-
plasm set, resulting in 79.13% of the scored bands being
Large- and regional-scale genetic structuring in bao-
bab — Nei ’ s (1973) gene diversity (expected heterozygosity)
within populations ( Table 1 ) ranged between 0.22 (P7; Senegal)
and 0.35 (P5; Ghana). Levels of polymorphism within popula-
tions varied between 41.7% (P7; Senegal) and 96.1% (P10;
Burkina-Faso) ( Table 1 ), reﬂ ecting a high level of polymor-
phism and variation within populations. Generally, populations
from Senegal had the lowest levels of gene diversity, with P7
also having an extremely low level of polymorphism in com-
parison with the other populations.
To detect genetic structuring in the analyzed sample set, we
used the model-based clustering method of Pritchard et al.
(2000) , without using prior information of the number of popu-
lations. This approach will identify any differentiated genetic
entities in the sample set. The results showed that clustering all
genotypes into eight “ gene pools ” correlates with a maximum
increase in the likelihood of the data ( Evanno et al., 2005 ). The
Table 1. Genetic diversity within 11 baobab populations.
Country Pop N d
max (m) H SD P p (%)
Benin P1 29 26 817 0.30 0.01 94.1
P2 26 25 008 0.30 0.01 94.5
P3 23 30 387 0.31 0.01 93.7
Ghana P4 28 46 511 0.32 0.01 94.1
P5 16 8 000 0.35 0.01 94.5
P6 15 50 000 0.29 0.01 95.3
Senegal P7 19 5 630 0.22 0.01 41.7
P8 26 4 876 0.27 0.01 95.3
Burkina Faso P9 22 1 084 0.32 0.01 94.0
P10 26 2 430 0.32 0.01 96.1
P11 21 6 894 0.28 0.01 94.5
Notes: Pop = Population; N = number of sampled individuals; d
maximum distance between sampled individuals; H e = Nei ’ s gene diversity
(expected heterozygosity); SD = standard deviation of H ; P p (%) = level of
polymorphism within population.
4American Journal of Botany [Vol. 96
In each population, the logarithmic regression slope b
statistically signiﬁ cant ( P < 0.05) or sometimes marginally sig-
niﬁ cant ( P < 0.1, Table 4 ). The linear regression slope b
also signiﬁ cant at P < 0.1 for all populations, except P2. The Sp
value for these populations varied from 0.009 (P2 and P3) to
0.049 (P10) with an average of 0.022 ( Table 4 ), suggesting a
statistically signiﬁ cant SGS within baobab populations.
Spatial genetic structuring of Adansonia digitata at large
and regional scales — Genetic variation among populations is
the result of various combinations of selection, mutation, mi-
gration, genetic drift, and mating behavior (VanderBank et al.,
1996). Generally, mating events have been shown to be the pri-
mary determinant of spatial genetic structuring in tree species
( Young and Merriam, 1994 ). Fruit bats are the major natural
dispersal agents for baobab pollen and seeds (Harris and Baker,
1959; Start, 1972 ; Baum, 1995 ), and although the breeding be-
havior of the species has not been extensively studied, baobab
is considered to be generally outbreeding ( Assogbadjo et al.,
2006 ; Ouedraogo, 2000 ). However, gene ﬂ ow between baobab
populations may also be inﬂ uenced by the status of baobab as a
semidomesticated species. Since the beginning of (semi)do-
mestication, rural populations, farmers, and traders have played
a role in gene ﬂ ow between geographically distant populations
by facilitating village-to-village transport of fruits ( Assogbadjo
et al., 2006 ).
Analysis of genetic variation among the West African popu-
lations of baobab conﬁ rms the general observations previously
reported at small scale within Benin ( Assogbadjo et al., 2006 ).
These populations appear to retain relatively high levels of ge-
netic variation within populations and modest variation among
populations, as conﬁ rmed by two independent estimates of ge-
netic differentiation: Bayesian ( F ST = 0.15), and AMOVA-
based ( φ ST = 0.21). This observation correlates with the fact that
long-lived tree species such as baobab show high levels of
within-population genetic diversity and low among-population
differentiation ( Hamrick et al., 1992 ; Austerlitz et al., 2000 ),
presumably because of their high levels of pollen-ﬂ ow and mul-
tigeneration populations (VanderBank et al, 1996; Austerlitz et
al., 2000 ).
Genetic structure analysis using a Bayesian clustering method
can potentially detect discontinuities in the pattern of variation
of genetic diversity and is thus useful to identify distinct “ gene
West African region ( Fig. 2 ) using the Mantel test, and a statis-
tically signiﬁ cant correlation was observed ( r= 0.598, P =
0.0001). A signiﬁ cant correlation also occurred at regional scale
( r = 0.531, P = 0.002), excluding Senegalese populations.
These results conﬁ rmed a clear isolation-by-distance pattern
between West African baobab populations at both large and re-
Spatial genetic structuring at population level — Spatial ge-
netic structure in A. digitata was estimated in seven populations
for which at least 22 individuals were sampled. Figure 3 shows
the average pairwise kinship coefﬁ cients, F
ij , on a logarithmic
distance scale. Distance scales varied from population to popu-
lation due to their sampling density (see the maximum distance
between individuals in this population, d max , in Table 1 and the
mean distance between individuals in the ﬁ rst distance interval,
d 1 , in Table 4 ) , which precluded more extensive SGS analyses
for some populations but which did not necessarily reﬂ ect natu-
ral baobab density in each region. The general trend seen in the
curves in Fig. 3 suggests a reduction of pairwise kinship with
increasing geographic distance.
In the ﬁ rst distance class, kinship coefﬁ cients were signiﬁ -
cantly larger than expected under random spatial distribution of
genotypes for P2, P4, P9, and P10 ( P < 0.05). While F
depended on the sampling schemes and the arbitrarily set dis-
tance intervals, the Sp statistic was robust to the sampling
scheme, at least as long as F
ij is approximately linear with the
distance or its natural logarithm ( Vekemans and Hardy 2004 ).
Table 2. Summary of the results of a model-based (Bayesian) clustering method using the software Structure version 2.0. ( Pritchard et al, 2000 ) with the
admixture model (250 000 iterations) for 11 populations of baobob. Individuals are assigned probabilistically to gene pools, based on the AFLP data.
All values > 0.080 are in boldface.
Inferred gene pool
Population 1 2 345678
P1 0.048 0.024 0.012 0.293 0.328 0.028 0.040 0.225
P2 0.019 0.020 0.009 0.346 0.327 0.005 0.015 0.258
P3 0.171 0.007 0.012 0.419 0.120 0.007 0.057 0.206
P4 0.008 0.021 0.022 0.056 0.088 0.744 0.042 0.019
P5 0.022 0.006 0.027 0.052 0.083 0.704 0.092 0.014
P6 0.007 0.043 0.031 0.056 0.020 0.819 0.013 0.012
P7 0.011 0.924 0.011 0.021 0.005 0.005 0.003 0.020
P8 0.484 0.266 0.014 0.152 0.044 0.007 0.003 0.030
P9 0.007 0.033 0.410 0.138 0.027 0.013 0.244 0.128
P10 0.048 0.014 0.494 0.153 0.018 0.020 0.139 0.115
P11 0.117 0.043 0.013 0.494 0.016 0.039 0.145 0.133
Table 3. AMOVA for 251 individuals of baobab from 11 populations
using 254 AFLP markers. The P -value is the probability of obtaining
an equal or more extreme value by chance alone, estimated from 1023
Source of variation DF SS Variance % of variation P
Among populations 10 616.592 2.32096 20.68 < 0.001
Within populations 240 2136.978 8.90407 79.32
3 976.312 4.02 14.43 < 0.001
7 518.851 2.34 8.42 < 0.001
Within populations 240 5158.271 21.49 77.15
Note: SS = sum of squares
May 2009] Kyndt et al. — Spatial genetic structure of ADANSONIA DIGITATA
geographic organization of genetic variation, continuous geo-
graphic gradients in allele frequencies due to isolation by dis-
tance cause the clustering algorithm to create discrete entities.
The spatial autocorrelation tests in this study conﬁ rmed that the
genetic differentiation between West African baobab popula-
tions is positively correlated to the logarithm of their geograph-
ical distance, as predicted under the isolation-by-distance
theory. This isolation by distance was observed at both large
scale (the whole West African zone) and medium scale (the
region covering Benin, Ghana, and Burkina Faso). Although
human inﬂ uence in the traditional agroforestry systems is ex-
pected to have a reducing effect on the genetic differentiation
between populations by seed exchange, this effect is apparently
only playing at narrow geographical scale, e.g., at local mar-
kets. Between-population seed exchange by humans is not able
to prevent some level of spatial isolation by distance between
baobab populations at larger scales.
Spatial genetic structuring at within-population level — The-
oretical models ( Wright, 1951 , 1978 ) predict an absence of
ﬁ ne-scale spatial structure when gene ﬂ ow is extensive and
there is local spatial genetic structure under locally restricted
gene ﬂ ow. The isolation-by-distance model, as proposed by
Wright (1946) predicts the formation of local pedigree struc-
tures as a result of limited gene dispersal and local random ge-
netic drift. Gene movement in plant populations involves both
pollen and seed, but the development of spatial genetic struc-
ture within populations is supposed to be inﬂ uenced more
strongly by seed dispersal ( Fenster, 1991 ; Chung et al., 2004 ).
Neighboring individuals will have an increased probability of
sharing at least one parent when seed dispersal is limited while
pollen dispersal is panmictic, leading to the formation of spatial
autocorrelative genetic structure ( Dyer, 2007 ).
In this study, a statistically signiﬁ cant SGS was detected
within seven populations from West Africa. Kinship between
individuals was shown to decrease following distance in all
populations and signiﬁ cant Sp -values were recorded despite the
small sample sizes. Although small sample sizes restrict the
available details regarding SGS and limit the statistical power
to detect signiﬁ cant SGS, our result show that a high number of
AFLP markers can partially compensate for small sample sizes
by providing more precise estimates of pairwise kinship coef-
ﬁ cients, in line with results from Hardy (2003) and Hardy et al.
On average, the Sp statistic equals 0.022, which would
correspond with a neighborhood size (Nb ≈ 1/ Sp ) of about 45
individuals (varying from 20 to 100 individuals depending
on the studied population) if we assume that drift-dispersal
equilibrium is reached. This observation implies that gene
dispersal, and hence seed dispersal, within populations is lim-
ited ( Loiselle et al., 1995 ), which is remarkable considering
the fact that baobab is known to be relying mainly on fruit bats
( E. helvum , Epomophorus gambiensis , and Rousettus aegyptiacus )
for pollination and seed dispersal ( Baum, 1995 ).
Fruit bats are among the most widely distributed gregarious
bats in Africa. Some populations are very localized, forming
vast colonies in areas where there is year long abundance of
fruit ( Kingdon, 1984 ), others (like E. helvum ) perform seasonal
migrations ranging up to thousands of kilometers looking for
food ( Richter and Cumming, 2005 ). Throughout the residence
period, a foraging range of up to 15 km has been reported for E.
helvum ( Richter and Cumming, 2005 ). This kind of movement
within the baobab population would suggest a substantial
pools ” that have diverged following an historical separation or
as a result of peculiar gene ﬂ ow barriers (e.g., Born et al., 2008 ).
Using the criterion designed by Evanno et al. (2005), eight such
“ gene pools ” were inferred in our population sample and popu-
lations from the same country are largely constituted of indi-
viduals belonging to the same gene pools. The only clear
exception to this rule is population P11 from Burkina Faso,
which shows a similar gene pool composition as P3 from Be-
nin. Nevertheless, this observation is not surprising considering
the fact that P11 is located close to the geographical border be-
tween Burkina Faso and Benin and is therefore very likely to be
inﬂ uenced by human seed exchange with populations from Be-
nin. Whether the inferred gene pools represent biologically rel-
evant entities can be questioned because there were nearly as
many inferred gene pools as populations, most populations
were represented by at least two gene pools, and the majority of
individuals were attributed to different gene pools. As men-
tioned in the documentation of the Structure 2.2 software, these
results suggest that instead of clear-cut discontinuities in the
Fig. 2. Scatter plot showing the relationship between F ST /(1 − F ST )
and the natural logarithm of geographical distances between populations.
Fig. 3. Mean pairwise kinship coefﬁ cients ( F
ij ) between individuals
according to the natural logarithm of geographical distance within seven
baobab populations from West Africa.
6American Journal of Botany [Vol. 96
The current high levels of genetic variation present within
populations of Benin, Ghana, and Burkina Faso imply that large
numbers of samples from a few populations will capture a suf-
ﬁ cient amount of the species ’ genetic variability for conserva-
tion programs. However, it has to be noted that such a practice
would increase the chance of missing rare alleles, particularly
in disjunct populations, which also express extreme phenotypes
for phenological traits related to climatic adaptation. Indeed,
this study and a previous study in Benin ( Assogbadjo et al,
2006 ) showed substantive levels of isolation by distance be-
tween and within baobab populations. Gapare et al. (2005) sug-
gested in these cases to sample populations from different
geographic areas to maximize genetic diversity for ex situ col-
lections. It has to be noted that AFLP provides neutral markers,
which are not correlated with morphological subclassiﬁ cation
of baobab ( Assogbadjo et al., 2008 ) across the West African
region. However, some morphometric variables were shown to
correlate with geographic distance and genetic differentiation
between baobab populations from Benin ( Assogbadjo et al.
2006 ). In addition, taking into account the isolation by distance
observed in the current study, we can see that sampling across
a wide ecogeographical range is essential.
Low genetic diversity and high levels of differentiation with
other populations imply the need for conservation strategies to
primarily focus on baobab populations from Senegal. Sampling
for conservation strategies should target more populations and
individuals in Senegal, and to a lesser extent in Burkina Faso,
where a substantial amount of genetic differentiation was ob-
served between populations. Although more populations should
be studied to conﬁ rm our observations about SGS within bao-
bab populations, conservation strategies should avoid sampling
neighboring trees, which are likely to be closely related.
To obtain a view of the long-term genetic variation of baobab
in agroforestry systems, we need to monitor the allelic richness
and inbreeding effects in the populations across generations. A
comparison of SGS between populations with different levels
of human inﬂ uence would provide interesting insights in the
long-term effects of agroforestry systems on the population ge-
netics of savannah trees.
Aldrich , P. R. , J. L. Hamrick , P. Chavarriaga , and G. Kochert .
1998 . Microsatellite analysis of demographic genetic structure in
fragmented populations of the tropical tree Symphonia globulifera.
Molecular Ecology 7 : 933 – 944 .
Allaye Kelly , B. , O. J. Hardy , and J.-M. Bouvet . 2004 . Temporal and
spatial genetic structure in Vitellaria paradoxa (shea tree) in an agro-
forestry system in southern Mali. Molecular Ecology 13 : 1231 – 1240 .
Assogbadjo , A. E. , R. Gl è l è Kaka ï , F. J. Chadare , L. Thomson , T.
Kyndt , B. Sinsin , and P. Van Damme . 2008 . Genetic ﬁ ngerprint-
ing using AFLP cannot distinguish traditionally classiﬁ ed baobab
morphotypes. Agroforestry Systems .
Assogbadjo , A. E. , T. Kyndt , B. Sinsin , G. Gheysen , and P. Van
Damme . 2006 . Patterns of genetic and morphometric diversity in
baobab ( Adansonia digitata L.) populations across different climatic
zones of Benin (West Africa). Annals of Botany 97 : 819 – 830 .
Assogbadjo , A. E. , B. Sinsin , J. T. C. Codjia , and P. Van Damme .
2005 . Ecological diversity and pulp, seed and kernel production of the
baobab ( Adansonia digitata ) in Benin. Belgian Journal of Botany 138 :
7 – 56 .
Austerlitz , F. , S. Mariette , N. Machon , P. H. Gouyon , and B.
Godelle . 2000 . Effects of colonization processes on genetic diver-
sity: Differences between annual plants and tree species. Genetics 154 :
1309 – 1321 .
amount of gene dispersal (up to 15 km), which is expected to
erase SGS within populations ( Hamrick and Loveless, 1986 ;
Kalisz et al., 2001 ).
Based on the SGS found in this study, it might be hypothe-
sized that bats are not so effective to promote long-distance dis-
persal of seeds within baobab populations. Although the lack of
natural pollinators around the sampled baobab populations
might account for the observed phenomenon, more plausible ex-
planations lie in baobab ’ s possession of heavy diaspores and its
low density populations. Heavy diaspores have been suggested
to result in short-distance seed dispersal, leading to short gene-
dispersal distances in neotropical tree species ( Hardy et al.,
2006 ). At small distances, localized seed fall from related indi-
viduals will enhance the spatial genetic structure within popula-
tions. In addition, baobab populations in agroforestry systems
reveal very low densities: 1 – 5 individuals per km
2 have been
reported in Benin ( Assogbadjo et al., 2005 ). The Sp statistic was
found to be higher in low density plant populations, a pattern
regularly found in other species ( Vekemans and Hardy 2004 ). It
can be explained by the reduction of seed shadow overlap at low
density so that nearby individuals are more related and kinship
coefﬁ cients decrease faster with geographic distance.
Consequences for conservation of baobab genetic variation
in West Africa — Baobab populations in West African agrofor-
estry systems are under the constant inﬂ uence of human activi-
ties, which may affect the genetic structure of tree species
through their effect on seed and pollen dispersal, density, frag-
mentation, and selection ( Young and Merriam, 1994 ; Aldrich et
al, 1998 ; Allaye Kelly et al., 2004 ). These perturbations of natu-
ral populations can have a major impact on breeding structure
and hence reduce genetic diversity and ﬁ tness by promoting in-
breeding ( Aldrich et al., 1998 ; Chung et al., 2004 ). The current
study of SGS suggests a spatial aggregation of related geno-
types and therefore risks for future inbreeding depression. How-
ever, a substantial amount of genetic variability was observed
within the 11 analyzed baobab populations from West Africa
and baobab populations with strong human inﬂ uence (e.g., in
Benin) even revealed slightly higher levels of variation than
less-inﬂ uenced populations (e.g., from Senegal). Most proba-
bly, active seed exchange on local markets will enhance gene
ﬂ ow between populations, maintaining genetic variation within
populations and reducing interpopulation genetic differentia-
tion and inbreeding levels ( Loiselle et al., 1995 ; Chung et al.,
2000 ). Nevertheless, this local seed exchange does not obstruct
the development of some signiﬁ cant level of genetic differentia-
tion between and within populations, correlated with their geo-
graphic distance, following the isolation-by-distance theory.
Table 4. Sp statistic α α ( Vekemans and Hardy, 2004 ) and 95% conﬁ dence
intervals (CI) revealing strength of spatial genetic structuring in seven
baobab populations. d 1 : mean distance (m) between individuals in the
ﬁ rst distance interval. P -values refer to Mantel tests.
1 (m) Sp CI ( Sp ) P -value
P1 627 0.012 (0.007 – 0.017) 0.024
P2 635 0.009 (0.002 – 0.016) 0.073
P3 627 0.009 (0.001 – 0.015) 0.061
P4 764 0.019 (0.012 – 0.026) 0.002
P8 300 0.011 (0.002 – 0.021) 0.038
P9 153 0.046 (0.028 – 0.064) 0.001
P10 320 0.049 (0.030 – 0.066) 0.001
May 2009] Kyndt et al. — Spatial genetic structure of ADANSONIA DIGITATA
phic DNA (RAPD) markers and comparison to allozymes. Molecular
Ecology 8 : 1317 – 1330 .
Kalisz , S. , J. D. Nason , F. M. Hanzawa , and S. J. Tonsor . 2001 . Spatial
population genetic structure in Trillium grandiﬂ orum : The roles of
dispersal, mating, history, and selection. Evolution 55 : 1560 – 1568 .
Kang , S. K. , and M. G. Chung . 2000 . High levels of allozyme variation
within populations and low allozyme divergence within and among
species of Hemerocallis (Liliaceae). American Journal of Botany 87 :
1634 – 1646 .
Kingdon , J. 1984 . East African mammals: An atlas of evolution in
Africa, vol. 2, part A, Insectivores and bats. University of Chicago
Press, Chicago, Illinois, USA.
Lengkeek , A. G. , A. M. Mwangi , C. A. C. Agufa , J. O. Ahenda , and
I. K. Dawson . 2006 . Comparing genetic diversity in agroforestry
systems with natural forest: A case study of the important timber tree
Vitex ﬁ scheri in central Kenya. Agroforestry Systems 67 : 293 – 300 .
Loiselle , B. A. , V. L. Sork , J. Nason , and C. Graham . 1995 . Spatial
genetic structure of a tropical understory shrub, Psychotria ofﬁ cinalis
(Rubiaceae). American Journal of Botany 82 : 1420 – 1425 .
Lynch , M. , and B. G. Milligan . 1994 . Analysis of population genetic
structure with RAPD markers. Molecular Ecology 3 : 91 – 99 .
Mantel , N. A. 1967 . The detection of disease clustering and a general-
ized regression approach. Cancer Research 27 : 209 – 220 .
Nei , M. 1973 . Analysis of gene diversity in subdivided populations.
Proceedings of the National Academy of Sciences, USA 70 : 3321 – 3323 .
Ouedraogo , L. G. 2000 . Rapport du Burkina-Faso. In O. Eyog Matig,
O.G. Gaou é , and B. Dossou [eds.], R é seau Esp è ces Ligneuses
Alimentaires: Compte rendu de la premi è re r é union du R é seau tenue
11 – 13 d é cembre 2000 au CNSF Ouagadougou, Burkina Faso. Institut
International des Ressources Phytog é n é tiques, Rome, Italy.
Pritchard , J. K. , M. Stephens , and P. Donnelly . 2000 . Inference of
population structure using multilocus genotype data. Genetics 155 :
945 – 959 .
Richter , H. V. , and G. S. Cumming . 2005 . Food availability and annual
migration of the straw-colored fruit bat ( Eidolon helvum ). Journal of
Zoology 268 : 35 – 44 .
Rohlf , F. J. 2000 . NTSYS-pc, numerical taxonomy and multivariate
analysis system, version 2.1. Exeter Software, Setauket, New York,
Rousset , F. 1997 . Genetic differentiation and estimation of gene ﬂ ow from
F -statistics under isolation by distance. Genetics 145 : 1219 – 1228 .
Rousset , F. 2000 . Genetic differentiation between individuals. Journal of
Evolutionary Biology 13 : 58 – 62 .
Sanou , H. , P. N. Lovett , and J. M. Bouvet . 2005 . Comparison of
quantitative and molecular variation in agroforestry populations of the
shea tree ( Vitellaria paradoxa CF Gaertn) in Mali. Molecular Ecology
14 : 2601 – 2610 .
Schneider , S. , D. Roessli , and L. Excoffier . 2000 . Arlequin: Software
for population genetic data. Genetics and Biometry Laboratory,
University of Geneva, Switzerland.
Sidibe , M. , and J. T. Williams . 2002 . Baobab: Adansonia digitata L..
International Centre for Underutilised Crops, Southampton, UK.
Start , A. N. 1972 . Pollination of the baobab ( Adansonia digitata ) by the
fruit bat ( Rousethus agyptiacus ). East African Wildlife Journal 10 :
71 – 72 .
VanderBank , H., B.-E. Van Wykt, and M. VanderBank . 1996 . Genetic
variation within and geographical relationships between four natu-
ral populations of Virgilia oroboides (tribe Podalyrieae: Fabaceae).
Biochemical Systematics and Ecology 24 : 135 – 143 .
Vekemans , X. 2002 . AFLP-surv, version 1.0. Computer program distrib-
uted by the author, Laboratoire de G é n é tique et Ecologie V é g é tale,
Universit é Libre de Bruxelles, Brussels, Belgium.
Vekemans , X. , and O. J. Hardy . 2004 . New insights from ﬁ ne-scale
spatial genetic structure analyses in plant populations. Molecular
Ecology 13 : 921 – 935 .
Wickens , G. E. 1982 . The baobab: Africa ’ s upside-down tree. Kew
Bulletin 37 : 173 – 209 .
Wright , S. 1946 . Isolation by distance under diverse systems of mating.
Genetics 31 : 39 – 59 .
Baum , D. A. 1995 . The comparative pollination and ﬂ oral biology of bao-
babs ( Adansonia – Bombacaceae). Annals of the Missouri Botanical
Garden 82 : 322 – 348 .
Baum , D. A. , and K. Oginuma . 1994 . A review of chromosome numbers
in Bombacaceae with new counts for Adansonia. Taxon 43 : 11 – 20 .
Born , C. , O. J. Hardy , M. H. Chevallier , S. Ossari , C. Atteke , J.
Wickings , and M. Hossaert-McKey . 2008 . Small-scale spatial ge-
netic structure in the Central African rainforest tree species Aucoumea
klaineana : A stepwise approach to infer the impact of limited gene
dispersal, population history and habitat fragmentation. Molecular
Ecology 17 : 2041 – 2050 .
Chung , M. G. , M. Y. Chung , G. S. Oh , and B. K. Epperson . 2000 . Spatial
genetic structure in a Neolitsea sericea population (Lauraceae).
Heredity 85 : 490 – 497 .
Chung , M. Y. , J. D. Nason , and M. G. Chung . 2004 . Spatial genetic
structure in populations of the terrestrial orchid Cephelanthera longi-
bracteata (Orchidaceae). American Journal of Botany 91 : 52 – 57 .
Dyer , R. J. 2007 . Powers of discerning: Challenges to understanding
dispersal processes in natural populations. Molecular Ecology 1 6 :
4881 – 4882 .
Epperson , B. K. 1993 . Recent advances in correlation studies of spatial
patterns of genetic variation. Evolutionary Biology 27 : 95 – 155 .
Evanno , G. , S. Regnaut , and J. Goudet . 2005 . Detecting the number of
clusters of individuals using the software STRUCTURE: A simulation
study. Molecular Ecology 14 : 2611 – 2620 .
Eyog Matig , O. , O. G. Gaou é , and B. Dossou . 2002 . R é seau « Esp è ces
ligneuses alimentaires » . Compte rendu de la premi è re r é union du r é seau
tenue 11 – 13 d é cembre 2000 au CNSF Ouagadougou, Burkina Faso.
Institut International des Ressources Phytog é n é tiques , Rome, Italy.
Excoffier , L. , P. E. Smouse , and J. M. Quattro . 1992 . Analysis of
molecular variance inferred from metric distances among DNA hap-
lotypes: Application to human mitochondrial DNA restriction data.
Genetics 131 : 479 – 491 .
FAO . 1981 . An eco-climatic classiﬁ cation of inter-tropical Africa. Plant
production and protection paper 31. FAO, Rome, Italy.
Fenster , C. B. 1991 . Gene ﬂ ow in Chamaecrista fasciculata (Leguminosae).
I. Gene dispersal. Evolution 45 : 410 – 422 .
Gapare , W. J. , S. N. Aitken , and C. E. Ritland . 2005 . Genetic diver-
sity of core and peripheral Sitka spruce ( Picea sitchensis (Bong.) Carr.)
populations: Implications for gene conservation of widespread species.
Biological Conservation 123 : 113 – 123 .
Hamrick , J. L. , M. J. O. W. Godt , and S. L. Sherman-Broyles .
1992 . Factors inﬂ uencing levels of genetic diversity in woody plant
species. New Forests 6 : 95 – 124 .
Hamrick , J. L. , and M. D. Loveless . 1986 . The inﬂ uence of seed dispersal
mechanisms on the genetic structure of plant populations. In A. Estrada
and T. H. Fleming [eds.], Frugivores and seed dispersal, 211 – 223, W.
Junk, The Hague, Netherlands.
Hardy , O. J. 2003 . Estimation of pair wise relatedness between individuals
and characterization of isolation-by-distance processes using dominant
genetic markers. Molecular Ecology 12 : 1577 – 1588 .
Hardy , O. J. , L. Maggia , E. Bandou , P. Breyne , H. Caron , M.
Chevallier , A. Doligez , C. Dutech , A. Kremer , C. Latouche-
Halle , V. Troispoux , V. Veron , and B. Degen . 2006 . Fine-scale
genetic structure and gene dispersal inferences in 10 neotropical tree
species. Molecular Ecology 15 : 559 – 571 .
Hardy , O. J. , and X. Vekemans . 1999 . Isolation by distance in a con-
tinuous population: Reconciliation between spatial autocorrelation
analysis and population genetics models. Heredity 83 : 145 – 154 .
Hardy , O. J. , and X. Vekemans . 2002 . SPAGEDI: A versatile computer
program to analyse spatial genetic structure at the individual or popu-
lation levels. Molecular Ecology Notes 2 : 618 – 620 .
Harris , B. J. , and H. G. Baker . 1959 . Pollination of ﬂ owers by bats in
Ghana. Nigerian Field 24 : 151 – 159 .
Heywood , J. S. 1991 . Spatial analysis of genetic variation in plant popu-
lations. Annual Review of Ecology and Systematics 22 : 335 – 355 .
Jenczewski , E. , J. M. Prosperi , and J. Ronfort . 1999 . Differentiation
between natural and cultivated populations of Medicago sativa
(Leguminosae) from Spain: Analysis with random ampliﬁ ed polymor-
8American Journal of Botany [Vol. 96
Wright , S. 1951 . The genetic structure of populations. Annals of Eugenics
15 : 323 – 354 .
Wright , S. 1978 . Evolution and the genetics of populations, vol. 4,
Variability within and among populations. University of Chicago
Press, Chicago, Illinois, USA.
Young , A. C. , and H. G. Merriam . 1994 . Effect of forest fragmenta-
tion on the spatial genetic structure of Acer saccharum Marsh. (sugar
maple) populations. Heredity 72 : 201 – 208 .
Zhivotovsky , L. A. 1999 . Estimating population structure in diploids
with multilocus dominant markers. Molecular Ecology 8 : 907 – 913 .