Cassava (Manihot esculenta Crantz, Family: Euphorbiaceae) is a staple food in most of the tropical regions of Africa, Asia, and Latin America (Zinga et al., 2016). It originated in the northern Amazonian basin (Olsen and Schaal, 1999; Léotard et al., 2009) and was introduced by the Portuguese to the African continent during the sixteenth century (Fauquet and Fargette, 1990). The global production of cassava was estimated to be 302.7 million tons in 2020. West Africa’s production was 100.6 million tons, in the same year, representing more than 33.2% of the world’s production of cassava (FAOSTAT, 2022). Cassava is an allogamous species propagated predominantly by cuttings (Elias et al., 2001; Oliveira et al., 2014). The genetic diversity studies in the genus Manihot seemed to support a single event of domestication from the wild form M. esculenta ssp. flabellifolia (Roa et al., 1997; Olsen and Schaal, 1999, 2001). The cassava hybrids resulting from interspecific crosses involving cassava and wild Manihot species can be highly fertile (Second et al., 1997; Nassar, 1999). In addition, interbreeding between different cassava genotypes within and between fields is common, and the seeds produced may fall and germinate, leading to an increase in the genetic diversity of cassava (Elias et al., 2001). Some studies reported the presence of a high diversity of accessions in farmers’ fields due to in situ conservation and the exchange of planting materials between farmers (Park et al., 2005). This high genetic diversity could be used to develop new varieties with drought-tolerant, disease-resistant, high-quality, and high-yield attributes (Oliveira et al., 2014). However, through the exchange of planting materials between farmers, the same accessions are often given different names, or conversely, different accessions are given the same name, resulting in the existence of duplicate accessions collected in different localities (Salick and Cellinese, 1997; Rao et al., 2002). The costs of cassava germplasm maintenance are relatively high because the plants are kept in the field and/or in vitro (Albuquerque et al., 2019). Therefore, it is important to conduct studies aimed at identifying duplicate accessions in order to optimize the physical storage space both in the laboratory and in the field and to reduce the cost of maintaining the collection (Van Treuren and Van Hintum, 2003). Identification of genetic diversity in cassava germplasm has been already done using biochemical markers (Lefèvre and Charrier, 1993), quantitative and qualitative descriptors (Kawuki et al., 2011; Kamanda et al., 2020), and molecular markers such as microsatellites (Tiago et al., 2017; Adjebeng-Danquah et al., 2020) and the single nucleotide polymorphism (SNP; Oliveira et al., 2014; de Albuquerque et al., 2018; Prempeh et al., 2020). Most of the morphological descriptors, especially the quantitative ones, are not very reliable due to the strong influence of genotype-environment interaction (Al-Fares and Abu-Qaoud, 2012). Molecular markers are stable, easily detectable, and not influenced by the environment (Asare et al., 2011; Mezette et al., 2013). Among the molecular markers, SNPs seemed to have attracted research attention because of their genome abundance, chromosome-specific localization, low mutation rate, and ease of automation (Mammadov et al., 2012). SNP markers are used in genomic selection studies (Oliveira et al., 2012; Wolfe et al., 2016), the identification of sources of disease resistance by marker-assisted selection (Carmo et al., 2015), the identification of duplicate accessions (Albuquerque et al., 2019) and the characterization of germplasm (Oliveira et al., 2014; Mtunguja et al., 2015; Tiago et al., 2017; de Albuquerque et al., 2018; Prempeh et al., 2020).

Cassava was introduced to Burkina Faso from some neighboring countries (Ghana and Côte d’Ivoire) and has long been considered a neglected crop (Guira et al., 2017). In recent years, cassava production has increased through many government initiatives including the introduction of improved varieties in 2013 by the International Institute of Tropical Agriculture (IITA). However, despite these efforts, cassava production in Burkina Faso is relatively low compared to the demand. Indeed, the demand for cassava was estimated to be 124,917 tons in 2017 with the annual production estimated to be 22,104 tons (MAAH, 2019). This might be due to the use of susceptible varieties to pests and diseases, but also the high sensitivity to harsh environmental conditions (Akinwale et al., 2011). It may then be necessary to develop new varieties that are resistant to diseases and better adapted to these environments. Any progress made in breeding programs depends on a better understanding of the genetic variability present in the existing population (Adjebeng-Danquah et al., 2016). Unfortunately, since the introduction (formally or informally) of cassava accessions to Burkina Faso until today, no study on their genetic diversity has yet been conducted using molecular markers. The aim of this study was to assess the genetic diversity of cassava grown in Burkina Faso and to identify duplicate accessions in order to provide breeding programs with unique genotypes and reduce the conservation costs of cassava germplasm in the field and in vitro.

The minimum number of markers required to adequately determine genetic diversity in a population varies with the genetic diversity of the population, the scale of the study, and the type of marker used (Grünwald et al., 2017). Species that are highly inbred require a larger number of markers than those that are naturally heterogeneous, like cassava, for the detection of allelic variations. In addition, the number of alleles varies with the type of marker. SSR markers can have a large number of alleles at a locus while SNP markers have a fixed number of alleles (Adjebeng-Danquah et al., 2020). Thus, fewer microsatellite markers may be needed when compared with SNP markers to achieve the same degree of resolution (Schlötterer, 2004; Arnaud-Haond et al., 2007; Grünwald et al., 2017). Whatever the type of marker used or the diversity within the species, it is important to determine the appropriate number of markers for which the diversity observed in the population will not increase significantly if an additional marker is added (Arnaud-Haond et al., 2007). The genotype accumulation curve that was made to determine this appropriate number of markers showed that the selected markers were sufficient for the discrimination of the cassava accessions grown in Burkina Faso. Moreover, it revealed the existence of 79 unique multilocus genotypes among the 166 accessions.

Further analysis revealed that the retained 34 SNP markers were highly polymorphic. The high number of polymorphic SNP markers could be explained by the open-pollination mode of reproduction and the level of genetic variation in cassava (Oliveira et al., 2014). This finding could be a reflection of the diversified origin of cassava accessions grown in Burkina Faso through the informal introduction of accessions from other countries (Guira et al., 2017). Despite the high polymorphism of the SNP markers used in this study, the average PIC value was lower than those of other markers such as SSR (Asare et al., 2011; Pedri et al., 2019; Adjebeng-Danquah et al., 2020). This difference could be explained by the bi-allelic nature of SNP markers whose PIC values vary between 0.0 and 0.5, unlike SSR which are multi-allelic and can have PIC values up to 1 (Prempeh et al., 2020). Therefore, our results indicate that most of the SNPs used were sufficiently informative and can be used to study the genetic diversity of cassava accessions. The only exception we found was the Me_MEF_c_0566 marker which was moderately informative. Furthermore, the mean PIC value observed in this study was higher than those observed previously in some genetic diversity studies of cassava accessions (Oliveira et al., 2014; Kamanda et al., 2020; Karim et al., 2020; Prempeh et al., 2020). The average heterozygosity observed in this study was higher than the expected average heterozygosity suggesting a heterozygote excess within cassava accessions under the HWE. This heterozygote excess was confirmed by negative values of the F_IS and F_IT fixation indexes. Curiously, there are studies that have reported heterozygote excess within cassava accessions in relation to that expected under the HWE (Kamanda et al., 2020), and other studies reporting heterozygote deficit in relation to that expected under the HWE (de Albuquerque et al., 2018; Prempeh et al., 2020). Nevertheless, the differences observed between the mean values of PIC, H_o, and H_e in this study and those of other studies could be explained by the specificity of each cassava germplasm studied and the SNP markers used.

Molecular profiling of cassava accessions revealed a high rate of potential duplicates among the 166 accessions and a high variability in the proportions of potential duplicates across regions. The analysis also revealed that these potential duplicates belong to several genetic profiles and that the number of profiles varies between regions. This could be due to a high rate of exchange of cassava cuttings within regions. In addition, most of the potential duplicate profiles were found in several regions, indicating a high rate of exchange of cassava cuttings between regions. Similar results were obtained in Brazil in a study conducted on the identification of duplicates in the Embrapa cassava germplasm bank (Albuquerque et al., 2019). Accessions with profiles found in several regions probably have quite interesting characteristics for farmers (Rabbi et al., 2015). From the PCoA analysis, it was not possible to differentiate cassava accessions according to geographical origin. This lack of differentiation was confirmed by the low value of the genetic differentiation index (F_ST = 0.03) and by the low assignment probabilities of accessions to different regions. Furthermore, the AMOVA results revealed that more than 92%, for the 166 accessions, and more than 93%, for the 79 MLGs, of the total molecular variance were within accessions. This could be explained by the fact that some accessions are grown in several regions. The determination of the best number of clusters using the function best.cutree showed that the cassava accessions studied can be grouped into two large groups. The truncation was done at the top of the dendrogram and therefore does not accurately give the diversity of accessions. A lower truncation would more precisely give the real diversity of cassava accessions cultivated in Burkina Faso. The Bayesian analysis performed on the 166 accessions also grouped the accessions into two clusters, with more than 28% of the accessions not being assigned to a cluster at the 80% membership probability threshold. The result of the Bayesian analysis could be an underestimation of the diversity of cassava accessions. Indeed, as this analysis is based on the HWE model, it then assumes that the population is panmictic, that it is infinite, that there is no selection, mutation, or migration, and that successive generations are discrete (Oliveira et al., 2014; Rabbi et al., 2015). For populations with clonal and/or partially clonal reproduction such as cassava, these assumptions are violated because alleles are not always transmitted independently from one generation to another (Kamvar et al., 2014; Rabbi et al., 2015). The significant deviations from the HWE we detected during our analysis confirm this assertion as none of the subpopulations were at HWE. The DAPC performed on the 166 cassava accessions divided the accessions into 17 clusters with a higher individual assignment probability (100%) of accessions into clusters. The difference between the results of the Bayesian analysis (2 clusters) and the DAPC (17 clusters) could be explained by the multivariate approach used by the DAPC but also by the presence of many duplicates in the cassava germplasm. There are reports suggesting that the type of population structure influences the precision of the method (Jombart et al., 2010; Oliveira et al., 2014). In addition, the analysis of the population structure of the cassava germplasm involving samples from different genetic origins in different and unknown proportions leads to linkage disequilibrium between non-linked loci (Ersoz et al., 2007; Oliveira et al., 2014). In this case, DAPC would be more appropriate as it uses an approach that can identify genetic structures in the absence of any assumptions about the genetic model of the population (Jombart et al., 2010). Indeed, the results of the AMOVA of the 166 cassava accessions in relation to the clusters formed by the Bayesian analysis (STRUCTURE clusters) and by the DAPC (DAPC clusters) showed that the DAPC allows for better discrimination of cassava accessions. It was found that more than 80% of the total molecular variance was between the DAPC clusters, compared to only 16.05% within the accessions. In contrast to the DAPC, the molecular variance between the Bayesian analysis clusters represented 33.48% of the total molecular variance compared to 66.52% within the accessions. This moderate differentiation of the accessions by Bayesian analysis was confirmed by the low value of the genetic differentiation index (F_ST = 0.08). The criteria for classifying F_ST values showed that F_ST between 0 and 0.05 indicates low differentiation; F_ST between 0.05 and 0.15 reflects moderate differentiation; F_ST between 0.15 and 0.25 suggests high differentiation and above 0.25, the F_ST illustrates very high differentiation (Wright, 1978). These classification criteria showed that there was indeed moderate differentiation of the cassava accessions according to the clusters formed by the Bayesian analysis. Furthermore, the DAPC allowed a significant differentiation of the cassava accessions with an F_ST value higher than 0.25. The number of clusters proposed by the Bayesian analysis of the 79 MLGs (seven clusters) was very close to that proposed by the DAPC (eight clusters) which suggests that the elimination of duplicates improves the accuracy of the Bayesian analysis as mentioned in a previous study (Kamvar et al., 2014). Indeed, the results of the AMOVA in relation to the clusters formed by the Bayesian analysis (50.41% between clusters) and the F_ST value (0.23) showed a clear increase in the discriminatory power of the Bayesian analysis. The analysis also showed a weak differentiation of the accessions in relation to breeding patterns (improved varieties and landraces) with an F_ST of 0.03. This weak differentiation is probably due to the fact that most of the improved varieties were grown in cassava fields, but the difficulty in identifying them correctly led us to include them in the set of farmer accessions.

The names of the repository/repositories and accession number(s) can be found through this link: https://www.ebi.ac.uk/biostudies/studies/S-BSST1040?key=6eb67c56-96f6-44da-b772-14dc9d750616.

MS, FT, and KS: conceptualization. MS and JP: methodology. MS, DO, and EY: formal analysis. MS: data curation and writing—original draft preparation. FT, KS, JP, JM, JN, and DK: writing—review and editing. DK, JP, FT, and KS: supervision. FT and JP: project administration and funding acquisition. All authors have read and agreed to the submitted version of the manuscript.

This work was supported by the Bill and Melinda Gates Foundation and The United Kingdom Foreign, Commonwealth and Development Office (FCDO; grant no. OPP1082413) to the Central and West African Virus Epidemiology (WAVE) Program for root and tuber crops—through a subgrant from Université Félix Houphouet-Boigny (UFHB) to the Institut de l’Environnement et de Recherches Agricoles (INERA). Under the grant conditions of the Foundation, a Creative Commons Attribution 4.0 Generic License has already been assigned to the Author Accepted Manuscript version that might arise from this submission.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.