Genetic diversity of bitter and sweet African bush mango trees ( Irvingia spp . , Irvingiaceae ) in West and Central Africa

1 Biosystematic Group, Radix Building, Droevendaalsesteeg, 6708 PB Wageningen University, P.O. Box 647, 6700 AP, Wageningen, The Netherlands. 2 Faculty of Agronomic Sciences, University of Abomey-Calavi, 01 BP 526 Cotonou, Benin. 3 Laboratoire de Génétique et des Biotechnologies (LGB/UAC), University of Abomey-Calavi, 01 BP 526 Cotonou, Benin. 4 Botanic Garden Meise, Nieuwelaan 38, 1860 Meise, Belgium.


INTRODUCTION
The food tree species in African tropical forests are important sources of food in sub Saharan Africa (Hladik et al., 1996;Malaisse, 1997;FAO, 2008;Augustino et al., 2011).Despite the increased food deficits, ecosystems face very high destruction (Laurance, 1999;Archard et al., 2002).Thus, domestication of the most important food tree species used by local communities in their daily diets remains a logical policy.This will strengthen traditional and regional strategies for agrobiodiversity maintenance while enhancing the global production of agro forestry systems.
The International Centre for Research in Agroforestry (ICRAF, now called the World Agroforestry Centre) has become a leading institution in traditional food tree species domestication in West and Central Africa.Since decades, the African bush mango trees (ABMTs) are systematically preserved and integrated in various traditional agroforestry systems in humid sub-Saharan Africa (Okafor and Fernandes, 1987;Franzel et al., 1996;Tabuna, 2001;Okunomo and Egho, 2010).ABMTs are widely distributed and taxonomically ambiguous taxa within the family of Irvingiaceae thus the bitter fruited and sweet fruited forms are difficult to be distinguished.The only way to easily differentiate them is by assessing the bitterness versus sweetness of the mesocarp because there are limited morphological differences between the ABMTs (Harris, 1996).
However, these mangoes exhibit a high morphological and phenological diversity, which is vital for domestication and selection programs (Harris, 1996;Atangana et al., 2002).There is also no clear relation between biochemical properties and the type of ABMT (Tchoundjeu and Atangana, 2007).Due to this overlap of morphological, phenological and biochemical properties, the correct taxonomy of ABMTs needs to be re-visited.Okafor (1975) proposed the variety level for sweet and bitter ABMTs, respectively, Irvingia gabonensis (Aubry-LeComte ex O'Rorke) Baill.var.gabonensis and Irvingia gabonensis var.excelsa (Mildbr.)Okafor.Although, based on a thorough taxonomic revision (Harris 1996) and a random amplified polymorphic DNA (RAPD) analysis (Lowe et al., 2000), a distinction at species level was suggested: I. gabonensis and I. wombolu Vermoesen, for sweet and bitter trees, respectively.The latter taxonomic grouping was not supported due to the low reproducibility of the RAPD analysis.
Moreover, a sound quantitative morphological comparison between bitter and sweet ABMTs is still lacking.Still, in order to ensure proper in-and ex situ conservation of the genetically diverse material and to support its development and genetic improvement (Grace et al., 2008), a clear taxonomic identity of the material is essential.Furthermore, the geographic origin of ABMTs that occur in the Dahomey gap (which is the wide savannah area separating the West African forest into the Upper and Lower Guinean forest blocks in Benin and Togo) remains an important topic.Like RAPDs, the amplified fragment length polymorphisms (AFLPs) are dominant markers.However, the power of AFLPs to reveal genetic diversity and difference even for closely related species are demonstrated better than that of the RAPDs.Thus, AFLPs were successfully used by Ude et al. (2006) to separate population of the Lower Guinean and Congolian forest block provenances of the sweet ABMTs.Simple sequence repeat (SSR) are PCR basedco-dominant markers, consist of repeats of short nucleotide sequences and are proven to have great importance in the studies of genome (Stafne and Clark, 2005).The chloroplast simple sequence repeats (cpSSRs) are genetic markers with uniparental inheritance used particularly to investigate phenomena of evolution of species and to assess the phylogenetic relations between populations of species (Wills et al., 2005).
This study intends to revisit the pattern of genetic diversity of ABMTs in answering two main questions: 1) What is the genetic diversity and differentiation within and between bitter and sweet ABMTs, and what do the patterns tell us about the geographical origin of the Dahomey Gap material?2) Are our markers (AFLPs and cpSSRs) suitable to consistently discriminate populations and types of ABMTs?

Sampling
Sweet and bitter ABMTs were sampled throughout the Dahomey Gap and in Cameroon.Materials were also sampled in four gene banks: IITA-Ibadan (International Institute for Tropical Agriculture) and NAGRAB (National Centre for Genetic Resources and Biotechnology) both in Nigeria and Kolbison and Mbalmayo both established by the World Agroforestry Centre in Cameroon (Table 1).For each sampled tree, young leaves were collected and stored in silica gel.DNA was extracted from each sample following the protocol described in Fulton et al. (1995).An AFLP analysis was first carried out on 33 samples (10 bitter and 23 sweet trees).This was followed by an independent cpSSR analysis carried out with 47 samples (14 bitter and 33 sweet trees), including more individuals from the Dahomey Gap and the regions postulated as ABMTs genetic diversity centres by Lowe et al. (2000) and Ude et al. (2006) in the Lower Guinean forest and the Congolian forest blocks.Thus, in total, 59 accessions (39 in the Dahomey Gap and 20 from Nigeria and Cameroon) were used in this study with 21 samples common to both analyses (Table 1).

Genetic diversity
The AFLPs are more reliable genetic markers that generate large number of bands and have high reproducibility.Therefore, they are *Corresponding author.E-mail: rlvihotogbe@gmail.com.Tel: +22995451295.Fax: +22921303084.
Author(s) agree that this article remains permanently open access under the terms of the Creative Commons Attribution License 4.0 International License widely used in genetic analysis (Berchowitz et al., 2001;Assogbadjo et al., 2010) as opposed to the RAPDs with low reproducibility (Powell et al., 1996).Microsatellites (or simple sequence repeats, SSRs) and especially chloroplast microsatellites (cpSSRs), have the power to reveal genetic diversity as well as phylogenetic relationships and hybridization between plant species (Wills et al., 2005;Panwar et al., 2010).Thus AFLPs and cpSSRs were used for these analyses.

AFLP data
The AFLP analysis was carried out at the Biosystematics Group, University of Wageningen, The Netherlands, following the procedure of Vos et al. (1995) with minor modifications.Three primer combinations previously identified to amplify sweet mango tree material (Ude et al., 2006) were used to analyze all the 33 samples.These included: E38M59 (Eco ACT/ se CTA), E40M62 (Eco AGC/Mse CTT) and E33M48 (Eco AAG/Mse CAC).PCR reactions were performed using a MJ PTC200 thermocycler.Prior to the selective amplification, the EcoR1 primer was fluorescently labelled with IRD700.AFLP fragments were separated on a LYCOR 4300 (Westburg, The Netherlands), and the resulting profiles were scored using the Quantar software (Key Gene Products, Wageningen, The Netherlands 2000) to produce the presence/absence data matrix.

cpSSR data
The cpSSR analysis was carried out in the Laboratory of Genetics and Biotechnology of the University of Abomey-Calavi, Benin.Eighteen (18) chloroplast SSR primers were tested on independent samples (bitter and sweet mango tree accessions

Genetic diversity and structure
Three datasets were considered in this study: the AFLP set, the cpSSRs set and the one containing the accessions that showed a result for both AFLP and cpSSR.An analysis of the genetic diversity and population structure based on allele frequency using AFLP-SURV version 1.0 (Vekemans, 2002) was performed on each dataset.The type of ABMT (sweet versus bitter) was considered as well as the geographical origin of the sample.Four geographic 'populations' were considered: (i) bitter mango trees from the Dahomey Gap (Benin and Togo = population BDG), (ii) bitter mango trees from Tropical Africa (population BTA), (iii) sweet mango trees from the Dahomey Gap (Benin and Togo = population SDG) and (iv) sweet mango trees from Tropical Africa (population STA; Table 1).For each dataset, we computed the mean Nei genetic diversity (Nei, 1973) for each population, the global genetic differentiation (Fst statistics) and the pair wise genetic distance among populations and between sweet and bitter mango trees.
Assuming there was no genetic structure among populations under Hardy-Weinberg equilibrium (Vekemans, 2002), the significance of the genetic differentiation was assessed by comparing the observed Fst with the distribution of obtained Fst using 100 random individual permutations.

Identification of populations and distinction between sweet and bitter ABMTs
To assess the effectiveness of the genetic markers used in the discrimination of the four geographically recognized populations as well as the two types of ABMT, a cluster analysis was carried out on each dataset and a dendrogram was produced using the well as the two types of ABMT, a cluster analysis was carried out on each dataset and a dendrogram was produced using the unweighted pair group method with arithmetic mean (UPGMA) method based on Jaccard similarity index (Jaccard, 1908) in Past software (Hammer et al., 2001): (1) Where, for a random pair of individuals, a = number of totally loci scored present for the two individuals, b = number of loci scored present exclusively present for only one individual and c = number of loci exclusively present for only the second individual.When classifying individuals using the principal component analysis (PCA), there is unclear grouping pattern due to the abundance of factors of low contribution to an existing pattern in the dataset.Therefore, the PCA axes that explained a high percentage of the total variance within the dataset or the factors correlated with those PCA often used in a subsequent multivariate analysis to get a better signal from the dataset (Mohammadi and Prasanna, 2003;Bidogeza et al., 2009).Because the separate use of the AFLPs and cpSSRs data generated more confusing patterns, only the combined AFLPs + cpSSRs dataset was used in the rest of the analysis.First, all the alleles with no variability (totally shared presence and absence) in the AFLPs + cpSSRs dataset were excluded.A principal coordinate analysis (PCoA) was performed in order to highlight the main groups yielded in the cluster analysis.The AFLPs + cpSSRs datasets without the totally shared presence and absence alleles were used to detect relationship among our sample.First, a PCA was performed and the axes that accumulated at least 70% of the total variation within the dataset were retained.The loci that were highly correlated (at least 70%) with those axis were used to produce a Neighbor Joining (NJ) tree using Kulczynski similarity index.Like Jaccard index, the Kulczynski similarity index (Kulczynski, 1927) does not integrate totally shared absent alleles and is one of the most consistent similarity index used in systematic and taxonomy (Boyce and Ellison, 2001).
Especially, this index is influenced by the total number of loci that make the different individuals between two randomly chosen individuals (Kronberg, 1987). (2)

RESULTS
For the AFLPs analysis, a total of 141 polymorphic alleles were scored for all the 33 individuals.The cpSSR analysis yielded 20 polymorphic alleles (one to six per locus).Thus, a total of 161 polymorphic alleles were available for the set of 21 samples with both AFLP and cpSSR results.

Genetic diversity and differentiation of ABMTs
The AFLP analysis indicated that the mean number of fragments scored as present for an individual tree was 47.The number of segregating fragments in the dataset was high (96.5%).The polymorphism was higher within the Dahomey Gap sweet mango tree population and lowest in the Dahomey Gap bitter mango trees population.The same tendency was observed regarding the within population Nei genetic diversity.The sweet mango tree population in the Dahomey Gap showed significantly highest genetic diversity, while the bitter ones in this eco-region displayed the lowest diversity (Table 2a).The test for genetic differentiation among populations indicated a global F st value of 0.108 (P = 0.024).The pairwise F st values among populations was higher between bitter and sweet mango tree populations within the Dahomey Gap and no genetic difference was found between bitter and sweet tree populations outside of the Dahomey Gap (Table 2b).Sweet mango trees showed a higher polymorphism and Nei genetic diversity than bitter mango trees (Table 2a).However, the difference based on this AFLP data was not significantly different (global F st = 0.034; P = 0.0639).The F st between bitter and sweet mango tree populations were very low (0.011).
The cpSSR results showed that the mean number of fragments scored as present at individual tree level was 8 and all the 20 scored loci have segregation power.The polymorphism was highest in the sweet trees population outside the Dahomey Gap and lowest for the Dahomey Gap bitter trees.The sweet tree populations in the Dahomey Gap and the bitter one in tropical Africa presented the highest genetic diversity, while the bitter tree population in the Dahomey Gap presented the lowest diversity again (Table 2a).No clearly significant genetic differentiation was found among populations (global F st = 0.105; P = 0.048  0.05) even though the highest pairwise F st was found between bitter tree population of the Dahomey Gap and the sweet trees of Tropical Africa (Table 2b).
Considering bitter versus sweet trees, the genetic diversity was highest in the sweet mango trees (Table 2a).However, based on the cpSSRs there was no significant genetic differentiation amongst the two types (F st = 0.0537; P = 0.077), and the F st between bitter and sweet tree populations was low (0.0540).
The combined AFLPs + cpSSRs data also indicated a high number of alleles with segregating power (83.9%), with 24 alleles presenting no variability.The within population proportion of polymorphism and the Nei genetic diversity showed the same tendency as the separate AFLPs and cpSSRs results: the highest value was calculated for the sweet trees population from the Dahomey Gap and the lowest in the bitter trees population from this eco-region (Table 2a).A significant genetic differentiation was detected among populations (global F st = 0.0176; P = 0.016).The bitter tree population in the Dahomey Gap and the sweet one in Tropical Africa had the highest pairwise F st while the lowest distance was shown between the sweet tree population in the Dahomey Gap and that in Tropical Africa (Table 2b).Considering sweet and bitter mango trees, 85.1% of the combined AFLP and SSR alleles had segregation power.The proportion of polymorphic loci was higher within sweet trees than within bitter trees (Table 2b).
The Nei genetic diversity was 0.2453 and 0.21 for sweet and bitter trees, respectively.The genetic differentiation between sweet and bitter trees was low (F st = 0.0335) and not significantly different (global F st =

Cluster analysis
The dendrogram based on the AFLP results (Figure 1a) showed no clear pattern among populations and no clear distinction between bitter and sweet mango trees.The majority of individuals from each considered population were spread across many clusters.However, apart from few accessions, there is a tendency for the Togo (bitter and sweet) and Benin (sweet) materials to cluster together (Figure 1a and Table 1).
The dendrogram based on the cpSSR results (Figure 1b and Table 1) showed an even less clear pattern with higher similarity among individuals and completely failed to discriminate between the two types of ABMTs or geographically defined populations.
The dendrogram resulting from the cluster analysis of the combined APLP + cpSSR data (Figure 1c and Table The first two axes of the PCoA on the combined dataset with 137 alleles (Figure 2 and Table 1) accounted for 62.5% of the variance (46.13 and 16.38% for coordinate 1 and 2, respectively).The PCoA tended to separate the bitter from sweet ABMTs, with the two sweet trees from Nigeria (IGIBGB1 and IGIBGB2) falling within the bitter tree group.Within each of the two groups, the populations were not clearly distinguishable apart from the sweet trees from Benin as shown in Figures 1a and c.
Forty-eight (48) alleles were highly correlated (at least 70%) with the first 8 PCA axes accounting for 72.8% of the total variation.The clustering obtained with NJ based on the Kulczynski similarity index of these alleles (Figure 3) confirmed the pattern in the PCA.The inability of the combined AFLP + cpSSR data to accurately separate populations and the clustering of the sweet trees from Nigeria within the bitter tree cluster was also confirmed.

ABMTs genetic diversity: Failure of AFLPs and cpSSRs or influence of domestication
For all the three datasets considered, the lowest polymorphism and genetic diversity was observed in the bitter mango tree population in the Dahomey Gap, while the highest values for these parameters was noted among the cultivated sweet tree population from the same eco-region.These results are not consistent with those of Lowe et al. (2000) and Ude et al. (2006) who indicated a higher genetic diversity for ABMTs in Cameroon and Nigeria.Environmental transformation through logging, extension of agricultural productive space through yearly bush fires, and urbanisation are the main causes of biodiversity loss in Tropical Africa (FRIG, 2003;Sodhi, 2007;Jose, 2012).Fragmentation and decreased population size have changed the climatic characteristics of the Volta forest region; a unique ecosystem in which bitter trees are found in the wild in the Dahomey Gap (Vihotogbé et al., 2014a).Additionally, the economic potential of bush mangoes' seed highly threatens the ABMTs, since the market of this non timber forest product relies mostly on natural populations (Agbor, 1994;Lowe et al., 2000).Thus, the population size of wild bitter ABMTs has decreased in the Volta forest region in their entire distribution range due to a lack of sufficient natural regeneration (Agbor, 1994;Zapfack and Ngobo-Nkongo, 2002;Vihotogbé et al., 2014a).Consequently, the reduction of their ecological variability has narrowed their morphological and genetic diversity.Nevertheless, the domestication and cultivation of sweet trees in various climatic zones in the Dahomey Gap, has helped to preserve or increase the existing diversity (Casas et al., 2005;Jose, 2012).Although, the provenance of sweet ABMTs in the Dahomey Gap is unknown (Harris, 1996;Asaah et al., 2003;Lesley and Brown, 2004;Ude et al., 2006, Vihotogbe et al., 2014a), their higher diversity in this eco-region may well be due to the fact that the ongoing traditional domestication process in this region includes material from geographically different origins: the Upper and Lower Guinean forest blocks as well as the Congolian forest region.We conclude that this is a consequence of the random genetic material transfer between and within local communities, not only for ABMTs but related to any economically important food tree species in agroforestry systems (Jose, 2012).In general, for the three datasets, no significant genetic differentiation was found between bitter and sweet ABMTs.Similarly, no genetic differentiation among populations was observed with the AFLP dataset.Genetic differentiation was detected within the cpSSR and AFLP + cpSSR datasets but the pattern was not the same for the two datasets.The dominance of cultivated material in our samples (collected in the field as well as in gene banks) resulted in the expression of artificially generated variation.
Comparatively few sharp bands could be scored with the AFLP and the cpSSR products.This weakness was noted during an AFLP study by Ude et al. (2006) who used 12 pairs of primers.The genetic diversity among populations and most importantly between bitter and sweet mango trees posed important questions concerning the suitability of the genetic markers used in our study.Thus, including wild materials from every ecoregion in the entire distribution range of ABMTs and using sound genetic analyses' methods would be of great importance in the evaluation of the genetic diversity, the influence of domestication and the genetic adaptability of ABMTs.

Suitability of the markers
Apart from the sweet mango tree population from Benin, which formed the most consistent and distinct clusters throughout our analyses, none of the methods used in this study clearly separated the ABMTs into geographically distinct populations (Figures 1 to 3).The PCA and NJ (Figures 2 and 3) showed a distinction between the two types, with the exception of the sweet tree samples from Nigeria which clustered with the bitter trees (Figures 2 and 3).This was also observed in the study of Lowe et al. (2000), but was explained in terms of inaccuracy of sampling.In our study, the fact that neither the PCA plots nor the NJ dendrogram were able to discriminate either geographical populations or fruit types implied that the markers used to achieve this goal were probably unsuitable.This idea corroborated with the fact that we know that in the area where bitter and sweet ABMTs co-occur (in the Volta forest region) successful gene flow between bitter and sweet trees is hardly expected for phenological reasons: (i) very short coflowering time, (ii) flowers abortion on all sweet trees after this co-flowering time and (iii) consistent overall difference in phenology between both types (Vihotogbé et al., 2014b).Further support comes from the presence of ecological differences between both types (Ainge and Brown, 2004;Vihotogbé et al., 2014a).Therefore, the unclear species distinction from the AFLP and cpSSR data might be attributed to the high level of sweet ABMTs diversity (Kelleher et al., 2005).So, in conclusion, we attributed the failure of AFLPs, cpSSRs and AFLPs + cpSSRs to distinguish populations to the effect of domestication and large scale transfer of genetic material via seeds of the economically and nutritionally appreciated ABMT morphotypes.

Figure 1a .
Figure 1a.UPGMA dendrogram for AFLP data of the 33 accessions based on Jaccard's similarity index.

Figure 1b .
Figure 1b.UPGMA dendrogram for cpSSR data of the 47 accessions based on Jaccard's similarity index.

Figure 2 .
Figure 2. Plot of first two principal coordinates based on Jaccard's similarity index with the 48 AFLPs + cpSSRs for the 21 common accessions.DG, Dahomey Gap; NC, Nigeria and Cameroon.

Table 1 .
ABMT accessions and types of analysis applied.

Table 2a .
Results of the genetic diversity analysis with AFLP-SURV.