13
Phylogeography in continuous space: coupling species distribution models and circuit theory to assess the effect of contiguous migration at different climatic periods on genetic differentiation in Busseola fusca (Lepidoptera: Noctuidae) S. DUPAS,* B. LE RU,* †‡ A. BRANCA, § N. FAURE,* G. GIGOT,* P. CAMPAGNE,* M. SEZONLIN, R. NDEMAH,** G. ONG’AMO, ‡†† P.-A. CALATAYUD* †‡ and J.-F. SILVAIN* *Laboratoire Evolution, Ge ´nomes et Spe ´ciation, UPR 9034, Centre National de la Recherche Scientifique, Institut de Recherche pour le De ´veloppement, UR 072, 91198 Gif sur Yvette, France, Universite ´ Paris-Sud 11, 91405 Orsay, France, Icipe - African Insect Science for Food and Health, PO Box 30772-00100, Nairobi, Kenya, §Ecologie, Syste ´matique et Evolution, B^ atiment 360, Universite ´ Paris-Sud, F-91405 Orsay, France, De ´partement de Zoologie et de Ge ´ne ´tique, Faculte ´ des Sciences et Techniques, Universite ´ d’Abomey - Calavi, 01 BP 526 Cotonou, Be ´nin, **International Institute of Tropical Agriculture, PO Box 2008, Messa, Yaounde ´, Cameroon, ††School of Biological Sciences, University of Nairobi, PO Box 30197 Nairobi, Kenya Abstract Current population genetic models fail to cope with genetic differentiation for species with large, contiguous and heterogeneous distribution. We show that in such a case, genetic differentiation can be predicted at equilibrium by circuit theory, where con- ductance corresponds to abundance in species distribution models (SDMs). Circuit- SDM approach was used for the phylogeographic study of the lepidopteran cereal stemborer Busseola fusca F uller (Noctuidae) across sub-Saharan Africa. Species abun- dance was surveyed across its distribution range. SDMs were optimized and selected by cross-validation. Relationship between observed matrices of genetic differentiation between individuals, and between matrices of resistance distance was assessed through Mantel tests and redundancy discriminant analyses (RDAs). A total of 628 individuals from 130 localities in 17 countries were genotyped at seven microsatellite loci. Six population clusters were found based on a Bayesian analysis. The eastern margin of Dahomey gap between East and West Africa was the main factor of genetic differentiation. The SDM projections at present, last interglacial and last glacial maxi- mum periods were used for the estimation of circuit resistance between locations of genotyped individuals. For all periods of time, when using either all individuals or only East African individuals, partial Mantel r and RDA conditioning on geographic distance were found significant. Under future projections (year 2080), partial r and RDA significance were different. From this study, it is concluded that analytical solu- tions provided by circuit theory are useful for the evolutionary management of popula- tions and for phylogeographic analysis when coalescence times are not accessible by approximate Bayesian simulations. Keywords: circuit theory, dispersal, landscape genetics, phylogeography, species distribution modelling Received 3 October 2012; revision received 5 March 2014; accepted 13 March 2014 Correspondence: Ste ´phane Dupas, Fax: 33 (0)1 69 82 37 38; E-mail: [email protected] © 2014 John Wiley & Sons Ltd Molecular Ecology (2014) doi: 10.1111/mec.12730

Phylogeography in continuous space: coupling species distribution models and circuit theory to assess the effect of contiguous migration at different climatic periods on genetic differentiation

Embed Size (px)

Citation preview

Phylogeography in continuous space: coupling speciesdistribution models and circuit theory to assess the effectof contiguous migration at different climatic periods ongenetic differentiation in Busseola fusca (Lepidoptera:Noctuidae)

S . DUPAS,*† B. LE RU,*† ‡ A. BRANCA,§ N. FAURE,* G. GIGOT,* P . CAMPAGNE,*

M. SEZONLIN,¶ R. NDEMAH,** G. ONG’ AMO,‡ † † P. -A. CALATAYUD*† ‡ and J . -F . SILVAIN*

*Laboratoire Evolution, Genomes et Speciation, UPR 9034, Centre National de la Recherche Scientifique, Institut de Recherche

pour le Developpement, UR 072, 91198 Gif sur Yvette, France, †Universite Paris-Sud 11, 91405 Orsay, France, ‡Icipe - African

Insect Science for Food and Health, PO Box 30772-00100, Nairobi, Kenya, §Ecologie, Systematique et Evolution, Batiment 360,

Universite Paris-Sud, F-91405 Orsay, France, ¶Departement de Zoologie et de Genetique, Faculte des Sciences et Techniques,

Universite d’Abomey - Calavi, 01 BP 526 Cotonou, Benin, **International Institute of Tropical Agriculture, PO Box 2008,

Messa, Yaounde, Cameroon, ††School of Biological Sciences, University of Nairobi, PO Box 30197 Nairobi, Kenya

Abstract

Current population genetic models fail to cope with genetic differentiation for species

with large, contiguous and heterogeneous distribution. We show that in such a case,

genetic differentiation can be predicted at equilibrium by circuit theory, where con-

ductance corresponds to abundance in species distribution models (SDMs). Circuit-

SDM approach was used for the phylogeographic study of the lepidopteran cereal

stemborer Busseola fusca F€uller (Noctuidae) across sub-Saharan Africa. Species abun-

dance was surveyed across its distribution range. SDMs were optimized and selected

by cross-validation. Relationship between observed matrices of genetic differentiation

between individuals, and between matrices of resistance distance was assessed

through Mantel tests and redundancy discriminant analyses (RDAs). A total of 628

individuals from 130 localities in 17 countries were genotyped at seven microsatellite

loci. Six population clusters were found based on a Bayesian analysis. The eastern

margin of Dahomey gap between East and West Africa was the main factor of genetic

differentiation. The SDM projections at present, last interglacial and last glacial maxi-

mum periods were used for the estimation of circuit resistance between locations of

genotyped individuals. For all periods of time, when using either all individuals or

only East African individuals, partial Mantel r and RDA conditioning on geographic

distance were found significant. Under future projections (year 2080), partial r and

RDA significance were different. From this study, it is concluded that analytical solu-

tions provided by circuit theory are useful for the evolutionary management of popula-

tions and for phylogeographic analysis when coalescence times are not accessible by

approximate Bayesian simulations.

Keywords: circuit theory, dispersal, landscape genetics, phylogeography, species distribution

modelling

Received 3 October 2012; revision received 5 March 2014; accepted 13 March 2014

Correspondence: Stephane Dupas, Fax: 33 (0)1 69 82 37 38;

E-mail: [email protected]

© 2014 John Wiley & Sons Ltd

Molecular Ecology (2014) doi: 10.1111/mec.12730

Introduction

Phylogeographic processes generate patterns in biodi-

versity. Their study has evolved from descriptive

approaches, using phylogenetics and population struc-

ture analyses, to inference methods with nested clade

analysis (Strasburg et al. 2007) and, more recently,

model-based statistical analyses (Knowles & Carstens

2007; Knowles et al. 2007; Lozier & Mills 2009). Species

distribution models (SDM) generate phylogeographic

hypotheses that can be further evaluated by several sta-

tistical methods (Alvarado-Serrano & Knowles 2013).

Coalescent models of isolation or coalescent models of

isolation with migration are used to test hypotheses on

the role of refuge populations in genetic structure (Hey

2010). Until very recently, such statistical approaches

were, however, limited to the study of discrete popula-

tions (with or without migration) while much of the

biodiversity on earth may have been generated in a

nondiscrete spatial continuum. To cope with spatial

continuity, He et al. (2013) simulated coalescence of

genes from present and past occurrence SDM projec-

tions (SPLATCHE, Ray et al. 2010) and compared phyloge-

ographic scenarios in an approximate Bayesian

computation framework (iDDC method). In this

approach, coalescent events occur in a spatially explicit

framework, although ultimate coalescent steps occur in

the source populations before expansion (source popu-

lations are not explicitly spatially structured). Another

pertinent and efficient approach based on extinction

recolonization processes has been proposed (Barton

et al. 2010) and implemented (Kelleher et al. 2013) to

explain large-scale patterns. This approach is still lim-

ited to a landscape of homogenous density, and to our

knowledge, it has not been applied to any empirical

systems.

Correlative approaches provide an alternative to coa-

lescence models for the study of differentiation in contin-

uous and heterogeneously distributed species with large

distribution. Wang et al. (2008) proposed to calculate

least cost path connectivity from habitat suitability

models and test their relationships with genetic differen-

tiation through Mantel tests. To really account for pro-

cesses of differentiation across multiple paths in

complex landscape, the study reported here focused on

circuit theory. Circuit theory is based on the equivalence

between graph theory representation of random walks

of electrons in a circuit and movements of genes in a

landscape across generations. The equivalence relies on

the hypothesis that migration between contiguous demes

is isotropic (same number of migrants in both direc-

tions). The backward time required for two genes pres-

ent in different demes to occur in the same deme is

related to commute time in an electric circuit and can be

calculated from circuit resistance distance. This enables

us to express the Slatkin (1991) FST/(1�FST) relation to

coalescence times within and between demes in terms of

circuit resistance distance (McRae 2006). Therefore,

genetic differentiation can be predicted by friction maps

representing the number of migrants between neigh-

bouring cells. Although tested in a correlative manner,

the model is based on a mechanistic formulation of

genetic differentiation at equilibrium derived from dem-

ogenetic parameters (McRae 2006). Such an approach

produced significant results in the detection of gene flow

barriers (McRae 2006; McRae & Beier 2007; Lee-Yaw

et al. 2009). As the number of migrants depends as much

on population size as on migration rate, we propose here

to calculate circuit resistance distance from species

abundance maps predicted by niche models (assuming a

constant migration rate). The effects of present and

palaeoclimatic periods on genetic differentiation in a

geographic continuum of heterogeneous density may

thus be deciphered by correlations linking genetic differ-

entiation to SDM-circuit resistance (from present projec-

tions or palaeoclimatic reconstructions).

The possibility of testing phylogeographic hypotheses

in continuous space is of particular interest for tropical

taxa. In tropical regions, ice ages may not have discret-

ized populations as much as in temperate regions. In

this context, the number and location of refuge popula-

tions are not necessarily relevant while required for

coalescent analyses. Nonetheless, strong phylogeo-

graphic structure, probably related to forest contrac-

tions, has been observed in forest trees (Dauby et al.

2010; Koffi et al. 2011) and rodents (Nicolas et al. 2008).

Fragmentation is also held responsible for strong geo-

graphic patterns in savannah trees (Allal et al. 2011)

and cheetahs (Charruau et al. 2011). In contrast, the

phylogeographic patterns of species associated with for-

est–savannah transition biotopes are still poorly docu-

mented. In Lepidoptera species, research has also been

limited by the slow development of reliable microsatel-

lite markers (Zhang 2004). To our knowledge, there is

no phylogeograhic study in Africa using these markers

on Lepidoptera species, and only a few have been pub-

lished in other continents [Vandewoestijne & Van Dyck

(2010) for Nymphalidae in Europe and Fuentes-Contre-

ras et al. (2008) for Tortricidae in Chile]. In Busseola fusca

(F€uller) (Lepidoptera: Noctuidae), a major maize pest

species with a wide distribution, three major well-sup-

ported mitochondrial clades have been observed across

sub-Saharan Africa (Sezonlin et al. 2006). One is specific

to the West of Africa (W, from Benin to Mali), and two

are observed only in the east of Africa (KI, distributed

between the two East African Rift Valleys, and KII, with

a broad distribution from Cameroon to eastern and

southern African countries). Eastern and western

© 2014 John Wiley & Sons Ltd

2 S . DUPAS ET AL.

populations did not appear to exchange mitochondrial

migrants (Sezonlin et al. 2006) and were structured

according to their altitudinal range. Western popula-

tions are observed at lower altitudes than eastern popu-

lations (Kfir et al. 2002). The present study aims at

assessing the existence of such a structure in microsatel-

lite markers and assessing the main factors of differenti-

ation in continuous space, by combining circuit theory

and niche modelling.

To have a population density to estimate circuit-SDM

resistance, a variety of SDM modelling method exist.

GAMs (Hastie & Tibshirani 1990) are of specific interest

to produce abundance maps to link with circuit theory

as opposed to presence-only models such as MaxEnt

(Phillips et al. 2006). In a comparative study, Hijmans &

Graham (2006) observed that GAMs were less prone to

false positives as compared to MaxEnt models. Identify-

ing population units corresponding to different niches

is challenging in a species such as B. fusca because it

may exhibit climatic niche differentiation across their

distribution range (Schulthess et al. 1997). Moreover,

genotypes responsible for niche differentiation are

unknown (Banta et al. 2012). In this study, we tested

both the single-niche and the two-niche hypotheses

using GAM and MaxEnt models with training test sta-

tistics. The connectivity was then calculated from pre-

dicted distribution maps based on circuit theory

(McRae 2006). Present connectivity and past connectiv-

ity were assessed from the corresponding distribution

maps, and their relative role in the genetic differentia-

tion was evaluated using partial Mantel tests (Landguth

et al. 2010).

Materials and methods

Insect DNA sampling and microsatellite sequencing

Busseola fusca larvae and pupae were collected on maize

and sorghum between 2001 and 2004 in 130 localities of

17 sub-Saharan African countries (latitudes between S

16° and N 8.6°, longitude between W 14° and E 40.6°).All larvae collected from plants were reared on an arti-

ficial diet under laboratory conditions, following Ony-

ango & Ochieng’-Odero (1994), to obtain pupae. Moths

were preserved in absolute ethanol immediately after

emergence. A total of 628 individuals collected were

analysed at seven microsatellite loci, amplified using

the method described in Faure & Silvain (2005) and

sequenced on an ABI 3130 sequencer.

Genetic analyses

Population clustering was analysed with complemen-

tary Bayesian inference programs. GENELAND 2.012

(Guillot et al. 2008) was used because it handles null

alleles that are common in Lepidoptera microsatellite

data. BAPS 4.14 (Corander et al. 2007) and TESS 1.2 (Chen

et al. 2007) were used for comparison purposes. These

different softwares rely on different rationale in their

definition of population units (Hardy–Weinberg equilib-

rium in Geneland, genetic homogeneity in BAPS, spatial

autocorrelation in TESS) (Franc�ois & Durand 2010).

Geneland was run using null allele and spatial depen-

dence options for 106 generations, with 20 repetitions.

The burn-in period was determined by plotting log pos-

terior probabilities on sampled generations. Additional

runs were performed using the nonspatial model option

to detect migrants. An admixture analysis of Geneland

cluster partitioning was performed in BAPS. The num-

bers of iterations, reference individuals and iterations

for reference individuals were set at 1000, 30 and 100,

respectively. BAPS was run using the spatial clustering

option. TESS was run using several values for the spa-

tial dependence parameter (w = 0, 1, 1.5 and 2), without

admixture. The maximum number of clusters was esti-

mated based on the minimum deviance information cri-

terion (DIC). Shared allele distances (DAS; Chakraborty

& Jin 1993) between Geneland clusters were obtained

using Populations software (Langella 2006) and were

used to reconstruct a population tree among the clusters

inferred by Geneland (using 100 bootstrap replicates on

individuals). Tests of population expansion (within-

locus k-test and between-locus g-test; Reich et al. 1998)

were performed on the Geneland clusters. The g interlo-

cus test statistic is the ratio of observed between-locus

allele length variance to the expected variance, assum-

ing a single-step mutation rate at all loci. The k statistics

is based on moment statistics for population at equilib-

rium, assuming a stepwise mutation model; k < 0 for

populations in expansion. Data from Kenya were suit-

able for analysis with the wombling method (wombling

analysis requires regularly spaced sampling coverage)

to detect areas of genetic homogeneity, areas of genetic

heterogeneity and the orientation of heterogeneity gra-

dients. The wombsoft R package was used (Crida &

Manel 2007).

Distribution models

The density of insects per stem was evaluated across

sub-Saharan Africa over a total of 612 visits in 140 local-

ities (nine countries). At each visit, an average of 100

randomly selected maize stems were checked for stem-

borer infestation. Infested plants were dissected in the

laboratory and the stemborer larvae reared on an artifi-

cial diet until emergence, whereupon they were identi-

fied. Additional data were also obtained from the

literature (Schulthess et al. 1997; Moyal 1998; Ndemah

© 2014 John Wiley & Sons Ltd

PHYLOGEOGRAPHY IN CONTINUOUS SPACE 3

2001) (152 samples from 44 localities in six additional

countries). Thirty absence points were added in areas of

the Sahara desert where maize is not grown and

B. fusca is known to be absent (Kfir et al. 2002). Maxi-

mum entropy models (MaxEnt) (Phillips et al. 2006) and

maximum-likelihood models – polynomial generalized

linear models (GLM) of the first, second and third

order, and generalized additive models (GAMs) – were

optimized on WorldClim bioclimatic (Hijmans et al.

2005) and elevation layers (100 resolution). To compare

likelihood-based and entropy-based models, the indi-

vidual predictions were converted into presence/

absence using different thresholds for sensitivity analy-

sis of model performance. Due to the ecological differ-

entiation between East and West African B. fusca

populations (Kfir et al. 2002) and to the absence of mito-

chondrial exchange between the two regions (Sezonlin

et al. 2006), we also tested whether SDMs separating the

West and East African data had a better predictive

value than the joint SDM (the border between East and

West was set at longitude 4 corresponding to the east-

ern margin of Dahomey gap, east of Benin, which is

recognized as an important biogeographic barrier for

many taxa). All optimized models were compared for

their predictive value using three training test proce-

dures: area under the curve (AUC), nonparametric cor-

relation between training and test data and generalized

cross-validation (GCV). The optimal model selected on

present data was used to predict present, past and

future climatic distributions. Details about bioclimatic

variables and model selection are given in the Appen-

dix S1 (Supporting information). Figure S1 (Supporting

information) presents a workflow diagram describing

the procedure for model selection, connectivity analysis

and Mantel correlation analysis (see below) (the corre-

sponding R script is given in Appendix S1, Supporting

information). Data were analysed in R2.10 (R Develop-

ment Core Team 2010). Figures were produced in QGIS

(Quantum GIS Development Team 2011).

Connectivity analyses

Species distribution models projected from best-selected

model were used to estimate connectivity measures.

Mantel correlations and redundancy discriminant analy-

ses (RDAs) (Legendre & Fortin 2010) were performed at

the individual level between the matrix of genetic dif-

ferentiation and the matrices of resistance distance

between individuals assuming different connectivity

models, taking into account or not geographic distance

(Geffen et al. 2004). Genetic differentiation between

individuals was estimated by a values, which corre-

sponds to FST/(1�FST) at the individual level (Rousset

2000), using GENEPOP v3.1c (Raymond & Rousset 1995;

Rousset 2008). Individuals with missing data at more

than three loci were removed. a values below 0.0001

were all set at 0.0001 to permit log transformations. The

first connectivity model assumed the eastern margin of

Dahomey gap to be the only barrier to migration. This

was represented by a resistance distance matrix

between individuals calculated from a friction map with

a contiguous connectivity of one everywhere except for

a reduced migration strip at longitude 4. The second

connectivity model between individuals was the classic

Mantel test stepping-stone model assuming homoge-

nous migration and population size. Geographic dis-

tance and log geographic distance matrices were used

for the one- and the two-dimensional cases, respectively

(Rousset 1997). The next connectivity models were the

circuit-SDM model presented above. Circuit theory

resistance distances between focal points corresponding

to the location of genotyped individuals were calculated

with the software circuitscape (McRae 2006). Habitat

maps were SDM projections [at present, last interglacial

(LIG, �130 000 yBP) and last glacial maximum (LGM,

�12 000 yBP) periods]. The effect of SDM-circuit resis-

tance distance on genetic differentiation was tested

using partial Mantel tests, with geographic distance as

a covariable (Landguth et al. 2010). McRae (2006) dem-

onstrated equivalence between the number of migrants

between demes in a landscape and the effective conduc-

tance between nodes in a circuit. Cell resistance maps

are built based on physical or topographical maps of

potential barriers. McRae (2006) based his demonstra-

tion on demes with variable migration rates and uni-

form population sizes but observed that the method

was robust to heterogeneity in population size. We

therefore based the cell conductance map directly on

predictions from the SDM. The assumption is made that

migration distance is less than the resolution of the

map and that migration rate is constant. In this situa-

tion, migration occurs only between adjacent cells, and

absolute number of migrants entering to a cell is pro-

portional to adjacent cell density. Once these assump-

tions are met, provided that conductance assigned to

resistors reflects the number of migrants, the number of

migrants between distant cells in continuous space can

be directly estimated from circuit theory on distribution

maps corresponding to the conductance. We set the

connection scheme at four neighbours and used average

conductance to calculate resistance between neighbours.

To estimate maximal dispersal distance, we used data

on the geographic expansion of resistance of B. fusca to

genetically engineered maize expressing Bt toxin intro-

duced in South Africa (Kruger et al. 2011). Bt resistance

expanded 50 km within 1 year (Kruger et al. 2011). As

Bt resistance is positively selected and genetically domi-

nant (Campagne et al. 2013), this may be close to the

© 2014 John Wiley & Sons Ltd

4 S . DUPAS ET AL.

maximal dispersion distance. To ensure that migration

occurs only among neighbouring cells, we therefore

aggregated SDM grids to a resolution of more than

three times this dispersion distance (1.66° by 1.66°, cor-responding to squares of about 185 km side in tropical

region). Distances and connectivity measures are calcu-

lated on individuals. Mantel tests and RDA are con-

ducted on individual measures or on averages of

individual measures within or between Bayesian

genetic clusters (R script in Appendix S1, Supporting

information). Simple and partial Mantel r values were

calculated. As significance of Mantel r cannot be

assessed by classical permutations (Guillot & Rousset

2013), RDAs of genetic differentiation were performed

on principal component vectors of connectivity matri-

ces, conditioning on geographic distance (Geffen et al.

2004).

Results

Genetic analyses

The Geneland analyses converged to yield six clusters,

West Africa (W), Cameroon (C), central to southern

Africa (S), Tanzania to eastern Kenya (KE), western

Kenya to Uganda (KW) and Horn of Africa (H) (Fig. 1).

Geneland and BAPS results were comparable, although

BAPS merged some Geneland clusters (C with S and

KW with H) and separated KE and KW at a slightly

different place: closer to the centre of the Gregory Rift

Valley in BAPS and at the eastern border of the Greg-

ory Rift Valley in Geneland. Note that Gregory Rift Val-

ley corresponds to a zone of lower stemborer density

when altitude is low and to the limit between the

wombsoft zones in Fig. 1. Some of the Geneland runs

also grouped together Cameroon and central to south-

ern Africa, yielding a total of five clusters. The FIS and

null allele proportions are listed in Table S1 (Supporting

information). The geographic ranges of each cluster con-

tained small areas where B. fusca was predicted to be at

medium to high density across present (Fig. 1) and past

climates (Fig. 2). These high-density areas were located

in the Ethiopian highlands for cluster H, in the north of

Victoria lake for cluster KW, on the Mount Kenya and

regions at the east of Gregory Rift Valley in Tanzania

for cluster KE, in the southern African highlands and

western borders of Western Rift Valley for cluster S, in

Cameroon’s volcanic chain for cluster C and in Ghana

and Liberia-Sierra Leone for cluster W. This relatively

sustained high population density across geological

times is consistent with the absence of signature of pop-

ulation expansion in k or g tests in any of the Geneland

clusters nor in the entire species (Table S2, Supporting

information). Among the six Geneland clusters, the

West African cluster was the most genetically distant

from the others (level plot of differentiation in Fig. 3A,

DAS distance neighbour joining tree in Fig. S3, Support-

ing information). The admixture analysis in BAPS

detected 5.7% introgression in the clusters produced by

Geneland (5.7% of the individuals have a probability

below 0.05 of belonging to their own Geneland modal

cluster). There was zero per cent admixture in all West

African individuals. The nonspatial model in Geneland

detected 27 wrongly assigned individuals (4.3%), which

were located in clusters H (five from cluster KW; six

from cluster S), KW (nine from cluster KE; all individu-

als from the locality of Nyahururu), KE (two from clus-

ter KW) and C (four from cluster S; one from cluster

KW). No wrongly assigned individuals were observed

in cluster S or cluster W. TESS led to a less repeatable

clustering. The five best runs showed different cluster

partitioning, independent of interaction parameter val-

ues and of the model chosen (F-model or admixture

models). The wombsoft analysis in Kenya (Fig. 1B)

detected two areas: one was genetically homogeneous,

at the east of the Gregory Rift Valley, and another

was considered as a suture zone because it was geneti-

cally heterogeneous, at the West of the Gregory Rift

Valley. Gradients of heterogeneity ran from southeast to

northwest.

Distribution models

According to three training test criteria (AUC, nonpara-

metric correlation and GCV criteria), the GAM had a

better predictive performance than the GLM or MaxEnt

models, and models grouping the East and West popu-

lations had better performance than models separating

them (Fig. S2, Table S3, Supporting information). Sensi-

tivity analysis was performed on the threshold used to

convert abundance data to presence/absence data. All

the thresholds gave the same result (we tested between

the 40% and the 70% quantiles with a 5% quantile step;

note: 35% quantile and below correspond to zero insects

per stem). Distribution maps across palaeoclimatic

extremes (Fig. 2) suggest that the eastern margin of

Dahomey gap (longitude 4, east of Benin) has always

been unsuitable for West African populations. A gen-

eral increase in population densities by the time horizon

of 2080 is predicted for the four climatic models:

CCCMA, CSIRO, HCCPR and NIES (Fig. 2).

Mantel tests, RDA and connectivity analyses

Mantel r values were calculated between individual

genetic differentiation and connectivity measures. The

strongest r value was observed for east–west Dahomey

gap separation matrix [r = 0.501, CI = (0.487, 0.513)],

© 2014 John Wiley & Sons Ltd

PHYLOGEOGRAPHY IN CONTINUOUS SPACE 5

(A)

(B)

Fig. 1 Population density distribution and Bayesian genetic clustering of Busseola fusca in Africa. Population density was estimated

from GAM grouping East and West populations. Full black background corresponds to stemborer density above two per maize stem.

Cluster names: Horn: Horn of Africa, Kenya West: western Kenya to Uganda. Kenya East: Tanzania to Eastern Kenya, South: central

to southern Africa, Centre: North Cameroon, West: West Africa. (A): Africa, (B): Kenya, with wombsoft genetic homogeneity regions,

boundary regions and gradients.

Fig. 2 Insect densities calculated at different periods from the best predictive distribution model (GAM). LIG: last interglacial

(�130 000 yBp), LGM: last glacial maximum (�12 000 yBp), CCCMA, CSIRO, HCCPR and NIES scenarios: 2080 predictions in their

A2a (normal development) and B2a (ecologically oriented development) versions. Full black background corresponds to stemborer

densities above two insects per maize stem. Arrows indicate regions containing significant population densities across climatic cycles

for each cluster, represented by its colour.

© 2014 John Wiley & Sons Ltd

6 S . DUPAS ET AL.

© 2014 John Wiley & Sons Ltd

PHYLOGEOGRAPHY IN CONTINUOUS SPACE 7

followed by geographic distance matrix [r = 0.424,

CI = (0.409, 0.438)] and log geographic distance matrix

[r = 0.340, CI = (0.327, 0.325)] (Table 1). Circuit theory

resistance distances were calculated from distribution

map projections obtained from the best-selected model

(GAM joining East and West African data, as described

above). Partial Mantel correlation between circuit theory

resistance distance calculated from present projection

and genetic data, with geographic distance as a covari-

able, was r = 0.043, CI = (0.025, 0.061), and correspond-

ing RDA conditioning on geographic distance was

significant (P < 0.01) (Table 1). Partial Mantel correla-

tion between circuit theory resistance distance calcu-

lated from past projection and genetic data, with

geographic distance as a covariable, were r = 0.003,

CI = (�0.008, 0.004), and r = 0.016, CI = (0.003, 0.028),

for LIG and LGM periods, respectively. The corre-

sponding RDA conditioning on geographic distance

was significant (P < 0.05) (Table 1).

As no genetic exchange was observed between East

and West Africa, the connectivity measures, Mantel

tests and RDA were further performed within each

region. RDAs were significant in East Africa only

(P-value < 0.005; Table 1 and Table S4, Supporting

(A)

Row

Col

umn

W

H

KE

KW

S

Cam

W H KE KW S Cam

0.2

0.4

0.6

0.8

1.0

1.2(B)

Row

Col

umn

W

H

KE

KW

S

Cam

W H KE KW S Cam

051015202530354045

Fig. 3 Level plots of (A) average Fst/(1�Fst) and (B) average connectivity between individuals of the six Geneland clusters. Cluster

names: W: West Africa, C: North Cameroon, S: central to southern Africa, KE: Tanzania to eastern Kenya, KW: western Kenya to

Uganda, H: Horn of Africa.

Table 1 Mantel r values between genetic distance and SDM-circuit resistance distances. Significance of the relationship was assessed

by redundancy discriminant analyses (see Table S4, Supporting information for details). Simple: simple Mantel tests. Partial: partial

Mantel test with geographic distance as covariable. CI: 95% confidence intervals assessed by 10 000 permutations. East/West: differ-

entiation between east and west. Geo: geographic distance. LogGeo: log geographic distance. Simple: simple Mantel test. Partial: par-

tial Mantel tests with log geographic distance as covariable

Simple East/West CI Geo CI LogGeo CI

All 0.501*** 0.488, 0.515 0.424*** 0.409, 0.439 0.340*** 0.329, 0.352

East 0.078*** 0.038, 0.111 0.040*** 0.026, 0.054

West 0.039 NS 0.025, 0.054 0.113 NS 0.102, 0.122

Partial Present CI LIG CI LGM CI

All 0.043** 0.025, 0.061 0.003*** �0.008, 0.004 0.016* 0.003, 0.028

East 0.121*** 0.101, 0.136 0.033*** �0.002, 0.037 0.115*** 0.095, 0.128

West �0.106 NS �0.133, �0.076 0.041*** 0.024, 0.060 �0.124*** �0.155, �0.093

Partial CCCMAA2A CI CCCMAB2A CI CSIROA2a CI CSIROB2a CI

All �0.030 NS �0.043, �0.007 0.046** 0.031, 0.078 0.096*** 0.079, 0.124 0.107** 0.089, 0.126

East 0.132 NS 0.111, 0.150 0.143** 0.124, 0.160 0.103*** 0.084, 0.116 0.104*** 0.081, 0.118

West �0.060 NS �0.085, �0.035 �0.058 NS �0.081, �0.036 �0.068 NS �0.089, �0.041 �0.039 NS �0.065, �0.014

Partial HCCPRA2A CI HCCPRB2A CI NIESA2a CI NIESB2a CI

All 0.036 NS 0.020, 0.054 0.075 NS 0.057, 0.094 �0.035 NS �0.049, �0.015 �0.014 NS �0.025, 0.011

East �0.056 NS �0.074, �0.034 0.057 NS 0.033, 0.074 �0.030 NS �0.043, �0.009 0.002 NS �0.015, 0.027

West �0.086 NS �0.110, �0.064 �0.023*** �0.053, 0.003 0.064* 0.029, 0.109 �0.003 NS �0.026, 0.033

*P < 0.05, **P < 0.01, ***P < 0.005.

© 2014 John Wiley & Sons Ltd

8 S . DUPAS ET AL.

information). Mantel r values in East Africa were 0.078

[CI = (0.038, 0.111)] for geographic distance and 0.040

[CI = (0.026, 0.054)] for log geographic distance. In East

Africa only as well, connectivity measures based on

electrical resistance distance calculated from SDM pro-

jections added significant information to the geographic

distance predictors. Partial r values were 0.121,

CI = (0.101, 0.136), 0.033, CI = (�0.002, 0.037), and

0.115, CI = (0.095, 0.128), for present, LIG and LGM

projections, respectively. All were significant (P < 0.005)

in RDAs. In West Africa, RDAs were not significant for

both geographic distances and circuit resistance dis-

tance calculated from SDM at the different periods, con-

ditioning on geographic distance.

The circuit theory resistance distance within and

between Geneland populations was obtained by averag-

ing the resistance distances between individuals belong-

ing to different clusters (Fig. 3B). The populations KE,

KW, S and C were connected to each other with rela-

tively small resistance distance (<0.3), whereas W and

H exhibited higher resistance, both to one another

(resistance distance 0.6) and to KE, KW, S and C (resis-

tance distance >0.2). Mantel correlation between average

a values (Fig. 3A) and average connectivity estimated

from present distribution was very high, 0.72 (P < 0.05).

The RDA and partial mantel r calculated from circuit

theory resistance distance using year 2080 scenarios

SDM projections (Table 1, Fig. 2) remained significant

for CCCMAB2a, CSIROA2a and CSIROB2a, but not for

the other scenarios (CCCMAA2a, HCCPRA2a,

HCCPRB2a, NIESA2a and NIESB2a), both for East Afri-

can and whole African data sets.

Discussion

Phylogeographic methods for species occupying largecontiguous range

Important advances in the study of phylogeography

have been obtained by the coupling of SDM past cli-

matic reconstructions with genetic analyses (Carstens &

Richards 2007; Knowles et al. 2007; Richards et al. 2007;

Orsini et al. 2008; Knowles & Alvarado-Serrano 2010;

Alvarado-Serrano & Knowles 2013). Coalescent simula-

tions enable assessing drift and migration hypotheses

among discrete refuge populations inferred from SDM

reconstructions (Richards et al. 2007). When populations

are contiguous, coalescent simulations from SDM recon-

structions are also possible using spatial expansion sim-

ulation software SPLATCHE (Knowles & Alvarado-Serrano

2010), which can be analysed statistically by approxi-

mate Bayesian computation techniques (He et al. 2013).

However, in species characterized by large population

size and large geographic ranges, computation of

coalescent events in contiguous landscape may not be

possible and many coalescence events may have to

occur in noncontiguous source populations before

expansion. Computational burden requires that these

last coalescent events occur in a manner not consistent

with the model. The present study focuses on applying

faster analytical solution to the problem of differentia-

tion in continuous space. McRae (2006) deduced a rela-

tionship between effective resistance distance calculated

from circuit theory on friction maps and equilibrium

values of genetic differentiation. The relationship dem-

onstrated for heterogeneous migration rate and homog-

enous density was robust to variation in density

(McRae 2006). We suggest here it can therefore provide

an analytical solution to estimate patterns of genetic dif-

ferentiation at equilibrium associated with a particular

SDM. The equilibrium value may not be reached within

a climatic period and may only represent a nonattained

limit (Barton et al. 2010). The differentiation matrix

between genetic samples would therefore result from

the combined influence of these trends towards differ-

ent equilibriums at present and palaeoclimatic periods.

Phylogeographic patterns in Busseola fusca

In the present study, we estimated geographic connec-

tivity based on circuit theory (McRae 2006) for past dis-

tributions predicted from a niche model, and we tested

their effect on genetic differentiation. We used partial

Mantel tests for quantification (Landguth et al. 2010)

and RDA for significance testing (Geffen et al. 2004).

Most of the genetic differentiation was associated with

the east margin of Dahomey gap, between East and

West Africa. Previous results showing complete isola-

tion of mitochondrial markers are therefore supported.

In addition, no admixed individual between East and

West African individuals was observed in BAPS admix-

ture analysis of Geneland clusters. In East Africa,

genetic differentiation was significantly correlated with

circuit-SDM resistance distance, with geographic dis-

tance as a covariable. The correlation was significant for

all the periods (present, LIG and LGM). In West Africa,

the RDAs were not significant. Considering that only

RDAs are relevant for significance testing, not Mantel

test (Guillot & Rousset 2013), we can conclude that no

significant isolation by distance or by resistance was

observed in this study in West Africa. The West African

insects may have a long-distance dispersal behaviour

not affected by habitat suitability, which blurs patterns

of isolation by distance or by resistance. Interestingly,

in West Africa, although not significant, partial mantel r

values with SDM-circuit model were negative for most

periods. He et al. (2013) also observed the same phe-

nomenon in an Australian lizard. Partial correlations

© 2014 John Wiley & Sons Ltd

PHYLOGEOGRAPHY IN CONTINUOUS SPACE 9

between genetic differentiation and circuit theory resis-

tance distance calculated on MaxEnt suitability maps

with geographic distance as covariable were highly neg-

ative. The authors considered that this negative correla-

tion pattern was generated by colonization processes

associated with climate induces habitat shifts, not caus-

ally linked to circuit resistance distance (Knowles &

Alvarado-Serrano 2010). We propose here, that negative

r value is due to higher migration rates in areas with

low habitat suitability (there could be more migrants

going through demes of smaller population size).

Correlation and causation

Such correlative approach must certainly be interpreted

with caution because correlation patterns do not neces-

sarily reflect causal effects (He et al. 2013). A correlation

between past climate and present genetic differentiation

can be a by-product of actual causal differentiation

force, for instance present climate SDM-circuit model

distance. Conversely, the absence of correlation with

past climate can occur even if the differentiation is actu-

ally influenced by the period but has been modified by

posterior processes of recolonization (Knowles & Alva-

rado-Serrano 2010). Coalescent-based models can be

considered superior in this respect because they can

account for these processes of directional recolonization,

whereas circuit-theory-resistance-distance-based models

assume that migration patterns are symmetrical. Never-

theless, both coalescent and circuit-based approaches

rely on specific hypotheses of dispersal behaviour.

Therefore, both approaches might lead to spurious

results due to unrealistic assumptions about dispersal.

In our study, we assumed that migration rate was con-

stant and differentiation was due to variation in densi-

ties. The same assumption was made by He et al. (2013)

in their coalescent simulations.

Tropical Africa phylogeographic patterns

The circuit theory resistance model may be useful to

document phylogeography in the tropics where the

restrictions of distribution ranges could have been less

drastic across climatic cycles. The glacial periods in Eur-

ope correspond to arid periods in Africa, and the pres-

ence of refuge and range restrictions during these arid

periods, or during wetter, periods depends on the taxa

considered. Forest species may have diversified by

range restrictions during arid periods (Couvreur et al.

2008). Conversely, purely savannah species may have

diversified through fragmentation or refuge isolation

during wetter periods (Barnett et al. 2006; Zinner et al.

2009). Busseola fusca distribution (Fig. 1) corresponds to

transition areas and deciduous woodlands in African

vegetation maps (White 1983; Mayaux et al. 2004). Buss-

eola fusca may therefore represent another model of a

species mostly associated with forest/savannah transi-

tion areas, which is likely to maintain large and contig-

uous populations across climatic extremes. It remains

uncertain whether the distinct populations result from

past refuge differentiation or current drift–migration

equilibrium or both. The refuge hypothesis is not sup-

ported by the observation that the geographic range of

all six genetic clusters contained relatively large popula-

tions across past and present climatic conditions that

may have prevented strong genetic drift over time.

Likewise, the absence of signature of population expan-

sion in k-g tests is not in agreement with the refuge

hypothesis. The continuous space circuit theory method

may be useful for understanding genetic differentiation

in the absence of a clear refuge scenario. In this study,

we observed that genetic differentiation was correlated

with circuit resistance distance calculated from the pro-

jected present, LGM and LIG distribution in East Africa.

Yet, r values of contiguous circuit resistance models

remained limited. The 5.7% introgression observed

among the population clusters suggests that long-dis-

tance migrations could have partially masked contigu-

ous migration patterns.

Effects of climatic changes

The impact of climatic change on species diversity is

still largely unknown. However, it has benefited from

an increasing attention (B�alint et al. 2011). Some meth-

ods were proposed to predict the impact of global

changes on the loss of genetic diversity (Pfenninger

et al. 2012). In our study, a general increase in popula-

tion size is forecasted in the scenarios referring to 2080.

As a consequence, the correlations between genetic dif-

ferentiation and isolation by resistance may change

accordingly. In some scenarios, correlation seems to dis-

appear, suggesting that the genetic patterns may evolve.

SDM-connectivity analyses may therefore provide new

cues to assess the impact of global changes, not only on

evolutionary significant units as usually addressed but

also on the forces that are structuring genetic diversity.

Circuit-SDM model and the management of Btresistance in insect pests

Knowledge about the demographic history and the

level of gene flow is important to adapt the strategies of

pest resistance management (Porretta et al. 2007; Timm

et al. 2008). Bt resistance has been reported in B. fusca.

Genetically engineered maize expressing Bt toxins has

been planted at large scale in South Africa, and new

strains are to be released in Kenya. Resistance to Bt

© 2014 John Wiley & Sons Ltd

10 S . DUPAS ET AL.

appeared in B. fusca after 8 year of plantation in South

Africa and spread rapidly (Van Rensburg 2007). The

analysis of connectivity showed little resistance to

migration within and between Kenyan and South Afri-

can clusters (Fig. 3), which could promote the invasion

of Bt resistance alleles (Van Rensburg 2007). In the

crambid Ostrinia nubilalis (H€ubner), high connectivity

between populations is also expected to favour the dis-

persal of Bt resistance (Krumm et al. 2008). In the noc-

tuid Trichoplusia ni (H€ubner), genetic analyses revealed

long-distance dispersion from west to east among North

American populations, but geographic areas with large

greenhouse populations showed high isolation by

distance, correlated with Bt resistance differentiation

(Franklin et al. 2010). Our results suggest that B. fusca is

a good disperser: (i) the same microsatellite cluster is

present from southern Africa to southern Cameroon,

confirmed by the extension of East African mitochon-

drial clades in Cameroon (Sezonlin et al. 2006, 2012); (ii)

apart from the east–west separation (0% introgression),

the clusters exchanged migrants (9.0% introgression

between the eastern and southern clusters). Locally, Bt-

free refuge strategy or the plantation of a certain area

with non-Bt plants may help to delay the apparition of

resistance. In Kenya, new local strain of Bt is being

introduced in maize and its large-scale production is

planned. This study showed that genetic structure in

eastern clusters of B. fusca was correlated with SDM-cir-

cuit resistance. Circuit theory may therefore be useful

in designing optimal refuge areas to insure mixing

between Bt-resistant and Bt-susceptible insects for the

management of Bt resistance.

Acknowledgements

We are very grateful to anonymous reviewers and to the editor

that helped us to improve, by their high expertise and careful

consideration, the relevance of this work to the state of the art.

This work was supported by IRD. The authors are grateful to

Dominique Vautrin for her technical support for microsatellite

genotyping. The study also benefited from the technical assis-

tance of several persons in the ICIPE stemborer team.

References

Allal F, Sanou H, Millet L et al. (2011) Past climate changes

explain the phylogeography of Vitellaria paradoxa over Africa.

Heredity, 107, 174–186.Alvarado-Serrano DF, Knowles LL (2013) Ecological niche

models in phylogeographic studies: applications, advances

and precautions. Molecular Ecology Resources, 14, 233–248.B�alint M, Domisch S, Engelhardt CHM et al. (2011) Cryptic bio-

diversity loss linked to global climate change. Nature Climate

Change, 1, 313–318.

Banta JA, Ehrenreich IM, Gerard S et al. (2012) Climate enve-

lope modelling reveals intraspecific relationships among

flowering phenology, niche breadth and potential range size

in Arabidopsis thaliana. Ecology Letters, 15, 769–777.Barnett R, Yamaguchi N, Barnes I, Cooper A (2006) The origin,

current diversity and future conservation of the modern lion

(Panthera leo). Proceedings of the Royal Society B: Biological Sci-

ences, 273, 2119–2125.Barton NH, Kelleher J, Etheridge AM (2010) A new model for

extinction and recolonization in two dimensions: quantifying

phylogeography. Evolution, 64, 2701–2715.

Campagne P, Kruger M, Pasquet R, Le Ru B, Van Den Berg J

(2013) Dominant inheritance of field-evolved resistance to Bt

corn in Busseola fusca. PLoS One, 8, 1–7.Carstens BC, Richards CL (2007) Integrating coalescent and

ecological niche modelling in comparative phylogeography.

Evolution, 61, 1439–1454.

Chakraborty R, Jin L (1993) A unified approach to study hy-

pervariable polymorphisms: statistical considerations of

determining relatedness and population distances. In: DNA

Fingerprinting: State of the Science (eds Pena S, Epplen J &

Jeffreys A), pp. 153–175. Birkhauser Verlag, Basel.Charruau P, Fernandes C, Orozco-Terwengel P et al. (2011)

Phylogeography, genetic structure and population diver-

gence time of cheetahs in Africa and Asia: evidence for long-

term geographic isolates. Molecular Ecology, 20, 706–724.Chen C, Durand E, Forbes F, Franc�ois O (2007) Bayesian clus-

tering algorithms ascertaining spatial population structure: a

new computer program and a comparison study. Molecular

Ecology Notes, 7, 747–756.

Corander J, Siren J, Arjas E (2007) Bayesian spatial modelling

of genetic population structure. Computational Statistics, 23,

111–129.Couvreur TLP, Chatrou LW, Sosef MSM, Richardson JE (2008)

Molecular phylogenetics reveal multiple tertiary vicariance

origins of the African rain forest trees. BMC Biology, 6, 54.

Crida A, Manel S (2007) Wombsoft: an r package that imple-

ments the Wombling method to identify genetic boundary.

Molecular Ecology Notes, 7, 588–591.Dauby G, Duminil J, Heuertz M, Hardy OJ (2010) Chloroplast

DNA polymorphism and phylogeography of a Central Afri-

can tree species widespread in mature rainforests: Green-

wayodendron suaveolens (Annonaceae). Tropical Plant Biology,

3, 4–13.

Faure N, Silvain J-F (2005) Characterization of eight microsatel-

lite loci in the maize stalk borer Busseola fusca (Fuller, 1901).

Molecular Ecology Notes, 5, 846–848.Franc�ois O, Durand E (2010) Spatially explicit Bayesian cluster-

ing models in population genetics. Molecular Ecology

Resources, 10, 773–784.

Franklin MT, Ritland CE, Myers JH (2010) Spatial and temporal

changes in genetic structure of greenhouse and field popula-

tions of cabbage looper, Trichoplusia ni. Molecular Ecology, 19,

1122–1133.

Fuentes-Contreras E, Espinoza JL, Lavandero B, Ram�ırez CC

(2008) Population genetic structure of codling moth (Lepi-

doptera: Tortricidae) from apple orchards in Central Chile.

Journal of Economic Entomology, 101, 190–198.

Geffen E, Anderson M, Wayne R (2004) Climate and habitat

barriers to dispersal in the highly mobile grey wolf. Molecu-

lar Ecology, 13, 2481–2490.Guillot G, Rousset F (2013) Dismantling the Mantel tests. Meth-

ods in Ecology and Evolution, 4, 336–344.

© 2014 John Wiley & Sons Ltd

PHYLOGEOGRAPHY IN CONTINUOUS SPACE 11

Guillot G, Santos F, Estoup A (2008) Analysing georeferenced

population genetics data with Geneland: a new algorithm to

deal with null alleles and a friendly graphical user interface.

Bioinformatics, 24, 1406–1407.Hastie TJ, Tibshirani RJ (1990) Generalized Additive Models.

Chapman & Hall, New York.

He Q, Edwards DL, Knowles LL (2013) Integrative testing of

how environments from the past to the present shape genetic

structure across landscapes. Evolution, 67, 3386–3402.

Hey J (2010) The divergence of chimpanzee species and sub-

species as revealed in multipopulation isolation-with-migra-

tion analyses. Molecular Biology and Evolution, 27, 921–933.Hijmans RJ, Graham CH (2006) The ability of climate envelope

models to predict the effect of climate change on species dis-

tributions. Global Change Biology, 12, 2272–2281.

Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A (2005)

Very high resolution interpolated climate surfaces for global

land areas. International Journal of Climatology, 25, 1965–1978.Kelleher J, Barton NH, Etheridge AM (2013) Coalescent simula-

tion in continuous space. Bioinformatics, 29, 955–956.Kfir R, Overholt AW, Khan RZ, Polaszek A (2002) Biology and

management of economically important lepidopteran cereal

stem borers in Africa. Annual Review of Entomology, 47, 701–

731.

Knowles LL, Alvarado-Serrano DF (2010) Exploring the popu-

lation genetic consequences of the colonization process with

spatio-temporally explicit models: insights from coupled eco-

logical, demographic and genetic models in montane grass-

hoppers. Molecular Ecology, 19, 3727–3745.Knowles LL, Carstens BC (2007) Estimating a geographically

explicit model of population divergence. Evolution, 61, 477–493.

Knowles L, Carstens BC, Keat ML (2007) Coupling genetic and

ecological-niche models to examine how past population dis-

tributions contribute to divergence. Current Biology, 17, 940–946.

Koffi KG, Hardy OJ, Doumenge C, Cruaud C, Heuertz M

(2011) Diversity gradients and phylogeographic patterns in

Santiria trimera (Burseraceae), a widespread African tree typi-

cal of mature rainforests. American Journal of Botany, 98, 254–

264.

Kruger M, van Rensburg JRJ, Berg JVD (2011) Resistance to Bt

Maize in Busseola fusca (Lepidoptera: Noctuidae) from Vaal-

harts, South Africa. Environmental Entomology, 40, 477–483.

Krumm JT, Hunt TE, Skoda SR et al. (2008) Genetic variability

of the European corn borer, Ostrinia nubilalis, suggests gene

flow between populations in the Midwestern United States.

Journal of Insect Science, 8, 1–12.

Landguth EL, Cushman SA, Schwartz MK, McKelvey KS, Mur-

phy M, Luikart G (2010) Quantifying the lag time to detect

barriers in landscape genetics. Molecular Ecology, 19, 4179–4191.

Langella O (2006) Populations 1230: population genetic software.

Available at: http://bioinformaticsorg/~tryphon/populations/.

Lee-Yaw JA, Davidson A, McRae BH, Green DM (2009) Do

landscape processes predict phylogeographic patterns in the

wood frog? Molecular Ecology, 18, 1863–1874.Legendre P, Fortin M-J (2010) Comparison of the Mantel test

and alternative approaches for detecting complex multivari-

ate relationships in the spatial analysis of genetic data.

Molecular Ecology Resources, 10, 831–844.

Lozier JD, Mills NJ (2009) Ecological niche models and coales-

cent analysis of gene flow support recent allopatric isolation

of parasitoid wasp populations in the Mediterranean. PLoS

One, 4, e5901.

Mayaux P, Batholome E, Fritz S, Behward A (2004) A new

land-cover map of Africa for the year 2000. Journal of Biogeog-

raphy, 31, 861–877.

McRae BH (2006) Isolation by resistance. Evolution, 60, 1551–1561.

McRae BH, Beier P (2007) Circuit theory predicts gene flow in

plant and animal populations. Proceedings of the National

Academy of Sciences, USA, 104, 19885–19890.Moyal P (1998) Infestation patterns and parasitism of the maize

stalk borer, Busseola fusca (Fuller) (Lepidoptera: Noctuidae),

in the Ivory Coast. African Entomology, 6, 289–296.

Ndemah R (2001) Distribution, relative importance and effect

of lepidopterous borers on maize yields in the forest zone

and mid-altitude of Cameroon. Journal of Economic Entomol-

ogy, 94, 1434–1444.

Nicolas V, Bryja J, Akpatou B et al. (2008) Comparative phylog-

eography of two sibling species of forest-dwelling rodent

(Praomys rostratus and P. tullbergi) in West Africa: different

reactions to past forest fragmentation. Molecular Ecology, 17,

5118–5134.Onyango FO, Ochieng’-Odero JPR (1994) Continuous rearing

of the maize stem borer Busseola fusca on an artificial diet.

Entomologia Experimentalis et Applicata, 73, 139–144.

Orsini L, Corander J, Alasentie A, Hanski I (2008) Genetic spa-

tial structure in a butterfly metapopulation correlates better

with past than present demographic structure. Molecular

Ecology, 17, 2629–2642.Pfenninger M, B�alint M, Pauls SU (2012) Methodological

framework for projecting the potential loss of intraspecific

genetic diversity due to global climate change. BMC Evolu-

tionary Biology, 12, 224.

Phillips SJ, Anderson RP, Schapire RE (2006) Maximum

entropy modelling of species geographic distributions. Eco-

logical Modelling, 190, 231–259.

Porretta D, Canestrelli D, Bellini R, Celli G, Urbanelli S (2007)

Improving insect pest management through population

genetic data: a case study of the mosquito Ochlerotatus caspi-

us (Pallas). Journal of Applied Ecology, 44, 682–691.

Quantum GIS Development Team (2011) Quantum GIS Geo-

graphic Information System Open Source Geospatial Founda-

tion Project. http://qgisosgeoorg

R Development Core Team (2010) R: A Language and Environ-

ment for Statistical Computing. R Foundation for Statistical

Computing, Vienna.

Ray N, Currat M, Foll M, Excoffier L (2010) SPLATCHE2: a

spatially-explicit simulation framework for complex demog-

raphy, genetic admixture and recombination. Bioinformatics,

26, 2993–2994.

Raymond M, Rousset F (1995) GENEPOP (version 12): popula-

tion genetics software for exact tests and ecumenicism. Jour-

nal of Heredity, 86, 248–249.Reich DE, Feldman MW, Goldstein DB (1998) Statistical prop-

erties of two tests that use multilocus data sets to detect

population expansions. Molecular Biology and Evolution, 4,

453–466.Richards CL, Carstens BC, Knowles L (2007) Distribution

modelling and statistical phylogeography: an integrative

© 2014 John Wiley & Sons Ltd

12 S . DUPAS ET AL.

framework for generating and testing alternative biogeo-

graphical hypotheses. Journal of Biogeography, 34, 1833–1845.Rousset F (1997) Genetic differentiation and estimation of gene

flow 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.

Rousset F (2008) GENEPOP’007: a complete reimplementation

of the Genepop software for Windows and Linux. Molecular

Ecology Resources, 8, 103–106.Schulthess F, Bosque-Perez A, Chabi-Olaye NA, Gounou S,

Ndemah R, Georgen G (1997) Exchange of natural enemies

of lepidopteran cereal stemborer between African regions.

International Journal of Tropical Insect Science, 17, 97–108.Sezonlin M, Dupas S, Le Ru B et al. (2006) Phylogeography

and population genetics of the maize stalk borer Busseola

fusca (Lepidoptera, Noctuidae) in sub-Saharan Africa. Molec-

ular Ecology, 15, 407–420.Sezonlin M, Ndemah R, Goergen G et al. (2012) Genetic struc-

ture and origin of Busseola fusca populations in Cameroon.

Entomologia Experimentalis et Applicata, 145, 143–152.

Slatkin M (1991) Inbreeding coefficients and coalescence times.

Genetics Research, 58, 165–175.

Strasburg JL, Kearney M, Moritz C, Templeton AR (2007) Com-

bining phylogeography with distribution modeling: multiple

Pleistocene range expansions in a parthenogenetic gecko

from the Australian arid zone. PLoS One, 2, e760.

Timm AE, Geertsema H, Warnich L (2008) Population genetic

structure of the oriental fruit moth Grapholita molesta (Lepi-

doptera: Tortricidae) in South Africa, inferred by AFLP

analysis. Annals of the Entomological Society of America, 11,

197–203.

Van Rensburg JBJ (2007) First report of field resistance by the

stem borer, Busseola fusca (Fuller) to Bt-transgenic maize.

South African Journal of Plant Soil, 24, 147–151.Vandewoestijne S, Van Dyck H (2010) Population genetic dif-

ferences along a latitudinal cline between original and

recently colonized habitat in a butterfly. PLoS One, 5, e13810.

Wang Y-H, Yang K-C, Bridgman CL, Lin LK (2008) Habitat

suitability modelling to correlate gene flow with landscape

connectivity. Landscape Ecology, 23, 989–1000.White F (1983) The vegetation of Africa, a descriptive memoir

to accompany the UNESCO/AETFAT/UNSO vegetation

map of Africa. UNESCO, Natural Resources Research, 20,

1–356.Zhang D-X (2004) Lepidopteran microsatellite DNA: redundant

but promising. Trends in Ecology and Evolution, 19, 507–509.

Zinner D, Groeneveld LF, Keller C, Roos C (2009) Mitochondrial

phylogeography of baboons (Papio spp.): indication for intro-

gressive hybridization? BMC Evolutionary Biology, 9, 83.

B.L., M.S., R.N., and G.O. sampled the insects across

Africa for genetic analyses and distribution modelling.

N.F. and G.G. genotyped microsatellite markers. A.B.

and P.C. did bayesian cluster analyses. S.D. performed

the circuit-SDM and other analyses and wrote the paper

as main author. J.-F.S. initiated and co-ordinated the

IRD program of research on Busseola fusca population

genetics between Africa and Gif-sur-Yvette, France.

Data accessibility

Phylogenetic data: TreeBASE submission ID 7854914. R

scripts for circuit-SMD modeling and analysis, climatic,

and genetic data: Dryad entry doi:10.5061/dryad.081q1.

Supporting information

Additional supporting information may be found in the online ver-

sion of this article.

Fig. S1 Workflow diagram for the connectivity analysis based

on species distribution modelling (Circuit-SDM analysis).

Fig. S2 Population tree of Busseola fusca obtained in Geneland

Bayesian analysis.

Table S1 Statistics of the six Geneland genetic clusters detected

in Busseola fusca.

Table S2 Population expansion tests on Geneland populations.

Table S3 Species distribution model comparison on test data.

Table S4 Association between genetic distance and circuit-

SDM model of genetic differentiation assessed by redundancy

discriminant analysis, conditioning on geographic distance.

Appendix S1 Materials and methods.

© 2014 John Wiley & Sons Ltd

PHYLOGEOGRAPHY IN CONTINUOUS SPACE 13