**4. Discussion**

A well-characterized and diverse germplasm is an essential requisite for genetic enhancement of crops. In this study, we applied GBS technology to explore the genetic diversity and population structure of a maize panel comprising landrace gene pools from Burkina Faso, Ghana, and Togo, plus a reference population from IITA-MIP. The results of the estimated diversity indices revealed ample genetic diversity within the maize panel indicated by average He (0.36) and Ho (0.5). The He obtained in this study was comparable to the 0.36 reported for provitamin A (PVA) quality protein maize (QPM) germplasm from IITA-MIP [21] but was higher than that reported for maize landraces from Eastern Africa (He = 0.25), Western Africa (He = 0.18), and Sahel Africa (He = 0.24) [9] as well as tropical maize breeding populations (He = 0.22) [27] including IITA early-maturing white inbred lines [20]. Characterization of the Burkinabe, Ghanaian, and Togolese maize pools showed different values for the estimated diversity indices. The results indicated that the Togolese gene pool (He = 0.36, Ho = 0.50) contained slightly higher diversity than the Burkinabe (He = 0.30, Ho = 0.41) and Ghanaian (He= 0.32, Ho = 0.34) landrace pools. Further, the low variation in the genetic indices identified between the landraces as a group, and the reference populations showed that the two germplasm sets possessed similar genetic diversity (Table 1). These results agreed with previous findings that tropical maize germplasm is highly diverse with He > 0.3 [37–39]. The mean PIC obtained in the present study, 0.29 using 5974 DArTseq SNPs for the 208 maize accessions was higher than the 0.19 and 0.26 reported for tropical early-maturing maize inbred lines using 15,047 [30] and 7224 SNPs for a sample size of 94 and 134, respectively [20,27]. The discrepancies between the results of our study and those of earlier researchers may be due to the use of different genetic materials, the sample sizes, and the number of SNPs used. Nonetheless, the mean PIC value in this study was like the 0.29 recently reported for tropical PVA-QPM maize germplasm using 8171 DArTseq SNP markers [21].

The Evanno criterion employed for the model-based simulation of population structure identified the peak level of ΔK at K = 3 (Figure 1a), which depicted the presence of three genetically distinct subpopulations (Figure 1b). The proportion of admixed accessions (47%) in the maize panel, based on a membership probability threshold of 80%, suggested moderate genetic differentiation and gene flow. The DAPC, PCoA, and NJ phylogenetic analyses results all illustrated the existence of three subpopulations in the whole set of 208 maize accessions. Comparison of the results of the four complementary clustering methods (STRUCTURE, DAPC, NJ tree, and PCoA) revealed high consistency in the individuals assigned to each group, which reinforced the findings that the identified groups were indeed genetically distinct. The close proximity between Togolese and Ghanaian gene pools suggested high genetic relatedness of the two gene pools. This result was expected due to the geographical proximity of the two countries and the similarity of the climatic conditions. The Burkinabe gene pool largely diverged from all others, suggesting its adaptation to Sahel conditions, which is in agreemen<sup>t</sup> with its pattern of phenotypic diversity [12]. Multivariate analyses revealed high a ffinity of Ghanaian and Togolese landraces with the reference populations (Figures 1 and 3–5). It is likely that some of these accessions are not true landraces but, rather, old improved cultivars that were either recollected or wrongly classified, as farmers usually consider improved varieties cultivated over longer periods in a given area as landraces [40]. The grouping together of some landraces with the reference populations also suggested a pedigree relationship. Hence, it is possible that some of the landraces analyzed in this study were local varieties that were selected by earlier maize breeders in IITA, based on high grain yield, earliness, and resistance to the maize streak virus (MSV), and adaptation to the drought and heat stress as starting materials for the development of inbred lines that were later involved in cross-breeding (see http://r4dreview.iita.org/index.php/tag/maize-improvement/). The additional smaller peaks observed at K = 4 (276.33) and K = 2 (273.07) implied the presence of subgroups within the three major groups (Figure 1). Therefore, an independent STRUCTURE run was performed for each gene pool (Figure 2). The high degree of genetic admixtures within each landrace gene pool observed with ancestry share of <80% probably reflects considerable levels of gene flow or germplasm exchange. Results of previous studies have shown that such an admixture is not unusual in landraces from restricted geographical backgrounds [40].

According to Frankham et al. [41], an *FST* value greater than 0.15 can be considered as significantly di fferentiating populations. Thus, in the present study, the overall *FST* value (0.21) supported the presence of significant genetic divergence within the maize panel. Wright [36] reported that an Nm value less than 1 indicated limited gene exchange among populations. In the present study, the overall Nm value of 1.58 (Table 2) indicated that moderate genetic exchange or gene flow may have occurred, leading to the moderate genetic di fferentiation between gene pools. This observation was consistent with the AMOVA results (Table 2), which indicated that 14% of the total variation was accounted for by gene pool variations. This result is consistent with the findings of previous studies [42]. According to the *FST* values, the Burkinabe gene pool was the most di fferentiated (Table 3), in agreemen<sup>t</sup> with its divergence as revealed by the clustering methods (STRUCTURE, PCoA, DAPC, and NJ analyses). The divergence between the reference populations and landraces varied among the di fferent gene pools. In particular, the low a ffinity of the Burkinabe gene pool with the reference population (*FST* = 0.31, Nm = 0.98) suggested little involvement of the original Sahelian gene pool in the development of the modern maize varieties presently grown in the sub region. This observation is biologically and historically meaningful since in West Africa, the reference maize gene pool called Composite Y [43], which was developed through recombination of 145 flint landraces of West Africa savannah zone, contained only 2% each of the genetic materials from Burkina Faso and Niger, as well as 1% of those from Senegal [44]. In the analyses of the isozyme variability in West African maize cultivars, Sanou et al. [45] showed that Burkinabe landraces were distinct, even though some levels of gene flow between them and an elite open pollinated variety (SR 22) developed by IITA in 1984 from CIMMYT Pop 22 and widely adopted in Burkina Faso [46] was observed. Therefore, the Burkinabe gene pool, having been grown and selected by farmers over many generations under warmer and drier conditions, could harbor novel and favorable alleles for improving maize for tolerance to drought and heat stresses. It is notable that in our earlier work on this maize panel, the high degree of tolerance of the Burkinabe landraces to drought, heat, and the combined heat and drought stresses was unrivalled [47,48]. The high genetic similarity observed between the Ghanaian and Togolese landraces was supported by their low *FST* (0.14) and high Nm (2.63) values. This result further reflected

the gene flow via seed exchanges and local preferences towards a given agrotype owing to similar climatic conditions. These may have significantly shaped the distribution of the genetic diversity within Ghanaian and Togolese maize landraces, as was previously suggested [12]. The *FST* and Nm values (Table 3) suggested that the Ghanaian gene pool was closer to the reference populations, in agreemen<sup>t</sup> with PCoA stratification (Figure 4). Indeed, the reference set analyzed in this study included two popular cultivars that are commonly cultivated in Ghana (Aburoheema and Obatanpa GH, coded IM1 and IM6, respectively). The deep knowledge of the genetic diversity and structure of Sahel and coastal West African maize landraces revealed in the present study provides an essential platform for efficient use of these valuable maize gene pools.
