Abandonment of pearl millet cropping and homogenization of its diversity over a 40 year period in Senegal

Cultivated diversity is considered an insurance against major climatic variability. However, since the 1980s, several studies have shown that climate variability and agricultural changes may already have locally eroded crop genetic diversity. We studied pearl millet diversity in Senegal through a comparison of pearl millet landraces collected 40 years apart. We found that more than 20% of villages visited in 1976 had stopped growing pearl millet. Despite this, its overall genetic diversity has been maintained but differentiation between early- and late-flowering accessions has been reduced. We also found stronger crop-to-wild gene flow than wild-to-crop gene flow and that wild-to-crop gene flow was weaker in 2016 than in 1976. In conclusion, our results highlight genetic homogenization in Senegal. This homogenization within cultivated pearl millet and between wild and cultivated forms is a key factor in genetic erosion and it is often overlooked. Improved assessment and conservation strategies are needed to promote and conserve both wild and cultivated pearl millet diversity.


Introduction
Since the 20 th century, major agricultural transformations have been taking place alongside social, technical and climatic changes. Agriculture has been striving to keep step with these changes and population growth via specialization in a few major crop species and a handful of varieties [1,2]. Meanwhile, the first impacts of climate change were felt in the 1980s when the Green Revolution took hold in developing countries [2]. In West Africa, climate change had already led to a 10-20% and 5-15% decrease in pearl millet and sorghum yields, respectively [3]. those collected in 2016, thus including the pivotal period of major climatic and socio-agricultural changes. Moreover, as Senegal is viewed as a pearl millet diversity hotspot, partly due to the high extent of gene flow with local wild populations [37], we also investigated how wildcrop gene flow evolved over this 40-year period.

Plant materials and sampling strategy
This study was conducted jointly with the Senegalese Institute of Agricultural Research (ISRA), which has a national mandate to collect and conserve all plant genetic resources for food and agriculture in Senegal. During field surveys in 2016, regional and departmental services were informed and, subsequently, they participated in some of the collections. Once we arrived in villages, we first explained the project objectives to local authorities, who then authorized the survey in the village. We then obtained the "prior and informed consent" of each farmer participating in this study, in accordance with the Code of Ethics of the International Society of Ethnobiology. Cultivated pearl millet sampling was first conducted in 1974-1976 and then in 2016 in Senegal. The 1976 collection [38] was conserved at National Research Institute for Sustainable Development (IRD) in France. For the 1976 collection, villages' geographical coordinates had been manually estimated on maps by first collection agents according to villages' names. We conducted new surveys from February to April 2016 and visited 69 sites where samples had been collected in 1976 (Fig 1). We were only able to survey 50 sites with the initial coordinates because some villages had moved and/or disappeared. For each of the remaining 19 sites, we visited a village within a 10 km radius of the initially surveyed village.
We first interviewed farmers to ensure that they had not changed their landraces since 1974-1976. This meant that farmers had always self-produced their seed without any voluntary introduction of seed from outside (purchase, donation). We found that farmers only fulfilled this condition in 43 villages out of 69 total sites. We sampled a maximum of five farmers from these villages, identical to the 1974-1976 sampling campaign. Two spikes or a handful of seeds were sampled from each farmer's granaries, and these samples were subsequently combined to constitute the final village pooled sample. Out of these 43 villages surveyed in 2016, 13 of their respective samples collected in 1974-1976 had good germination rates (> 80%) (S1 Fig). Among these 13 villages, 11 were growing early-flowering (EF) landraces (i.e Souna type) and two villages were growing late-flowering (LF) landraces (i.e Sanio type).
In addition to this cultivated pearl millet collection over a 40-year timespan, wild pearl millet populations (Pennisetum glaucum monodii (L.) R. Br.) were sampled at different sites across Senegal from September to October 2015 (S1 Fig). Wild pearl millet can be easily distinguished from cultivated pearl millet by their smaller spike (~10 cm) and small plant size with numerous tillers. A total of 35 wild accessions were sampled based on their wild phenotype and distance from cultivated fields. No wild populations were found at longitudes above -14.91 (S1 Fig).

DNA extraction and microsatellite amplification
DNA extraction was carried out according to an MTAB protocol, as previously described in [40], from wild pearl millet inflorescences and cultivated pearl millet leaves after plants were grown from seeds in a greenhouse. Twelve polymorphic SSRs were used for genotyping, as described in [41]. A total of 271 individuals from the 1976 pearl millet collection (i.e. 21 individuals/accession on average; range 12 to 29), 309 individuals from the 2016 pearl millet collection (i.e. 24 individuals/accession on average; range 6 to 31), and 642 individuals from the wild pearl millet collection (i.e. 18 individuals/accession on average; range 3 to 36) were kept for subsequent analysis. The average percentage of missing data per individual was 1.75% (S1 Table).

Diversity statistics estimation
We used FSTAT v.2.9.3.2 [42] to calculate the observed (H Obs ) and unbiased expected (H Exp ) heterozygosity, rarefied allelic richness (A R ), inbreeding coefficient (F IS ) and differentiation index (F ST ). We also tested for significant differences in the genetic diversity parameters (H Exp , H Obs, A R, F IS ): i) between wild and cultivated forms, and ii) over sampling years. F ST significance was assessed after 10000 permutations using FSTAT v.2.9.3.2 [42]. Finally, Pearson coefficient correlations were used to assess relationships between geographic coordinates, ancestry and genetic diversity parameters.

Genetic structure
A panel of complementary methods were used to assess genetic structure within and between cultivated and wild forms. We performed scaled and centred principal component analyses (PCA) using the FactomineR package v.1.4.2 [43] and factoextra package v.1.0.5 [44] in the R environment [45] v.3.6.0. Then, we conducted Bayesian clustering analyses in STRUCTURE v.2.3.4 [46] using the admixture model [47] and correlated allele frequency option, with a burn-in period of 100000 steps and 500000 MCMC replicates. Ten independent runs were performed for each K, ranging from K = 1 to K = 10. The most probable K value was determined by the D.ΔK criterion [48] and the log-likelihood (Ln P (D | K)) plot with STRUCTURE HAR-VESTER v0.6.94 [49]. To investigate further the wild pearl millet genetic structure, we incorporated spatial information by performing a spatial principal component analysis (sPCA) implemented in the adegenet package v.2.1.1 [50] in the R environment [45] v.3.6.0. We used a Gabriel graph as spatial connection network.

Pearl millet growing area
In our survey, we found that 78% (54/69) of the initially surveyed villages were still growing pearl millet in 2016, out of which 16% (11/69) had introduced new varieties. Farmers from over 20% of the villages visited in 1976 had stopped growing pearl millet. Those villages were located above 14˚N and below 13˚N (Fig 1). Most reasons put forward by the farmers to explain this choice were: insufficient rainfall, pest and bird attacks, and poor soil fertility (pers. observations).

Wild and cultivated pearl millet differentiation
Marked differentiation was observed between wild and cultivated pearl millet samples. The first two PCA axes explained 7.1% and 1.7% of the inertia, respectively (Fig 2A). Cultivated and wild accessions were clearly separated on the first axis, while the second reflected genetic differences within cultivated pearl millet. A handful of wild accessions with intermediate positions on the first axis suggested potential hybrids samples. Those accessions were spatially distributed throughout Senegal. The structure analysis findings fit those of the PCA analysis, with the most likely number of clusters being K = 2 (S3 Fig), representing cultivated and wild pearl millet forms ( Fig 2B). Ten wild samples fell within the cultivated cluster, with q-values > 0.5. Those individuals were removed from subsequent analysis. Had the individuals not been removed, the study's main results would remain unaltered. The genetic differentiation F ST between wild and cultivated pearl millet was 0.227 (permutation test, P = 0.001). The cultivated and wild pearl millet samples were not significantly different with reference to the expected heterozygosity (Table 1, Wilcoxon paired test, V = 31, P > 0.05), but they were significantly different for allelic richness A R (Table 1, Wilcoxon paired test, V = 1, P = 0.003). The average population crop-to-wild ancestry (q cw ) was significantly higher than the wild-to-crop ancestry (q cw = 0.033 vs. q wc = 0.019, Mann-Whitney-Wilcoxon test, W = 315, P = 0.041).

Wild diversity and structure
Wild pearl millet accessions showed a weak structure and only slight geographical structuring. We found that PC1 and PC2 summarized 2.3% and 2.1% of the inertia, respectively (Fig 3A). A significant correlation was nevertheless obtained between PC1 and latitude (S2 Table), (r = -0.26, P = 2.27 10 −11 ) or longitude (r = -0.24, P = 8.29 10 −10 ), thus reflecting a weak spatial effect on wild pearl millet diversity. The STRUCTURE analysis revealed the highest ΔK for K = 2 and K = 4 (S3 Fig). However, no clear groups were identified ( Fig 3B).
Adding spatial information as a variable could help in identifying groups when there is very low genetic structuring. However, the spatial PCA did not improve the structuration analysis (S4 Fig). The expected heterozygosity (H Exp ) averaged 0.607 per site (range 0.463 to 0.669) and the mean allelic richness (A R ) was 2.3 (range 1.9 to 2.5; Table 2). We found that crop-to-wild

PLOS ONE
Temporal evolution of pearl millet diversity  ancestry (q cw ) was negatively correlated with latitude (r = -0.41; P = 0.015) and positively correlated with allelic richness (r = 0.34, P = 0.044) and the expected heterozygosity (r = 0.40; P = 0.018). We noted a nearly significant negative correlation between wild expected heterozygosity and longitude (r = -0.32, P = 0.057; S3  Table 3). The cultivated pearl millet genetic structure changed over time. The PCA showed that PC1 and PC2 jointly accounted for 7.1% of the total variance (Fig 4).  Table 4).

Abandonment of pearl millet cropping in Senegal
A staggering proportion of villages (20%) were found to have abandoned pearl millet cropping mainly at latitudes above 14˚N. Several studies have reported evidence of the adoption and southward displacement of short-cycle cultivars in response to the Sahelian drought of 1980 [26,27] associated with a 0.9˚C temperature increase since 1975 [27] and declining rainfall in West Africa [6,[51][52][53], yet ours is the only study that highlights the complete abandonment of pearl millet cropping. We noted this pattern in regions at latitudes above 14˚N, i.e. in the Saint-Louis, Louga and Matam regions. The need for varieties with very short cycles (< 60-75 days; [54]) had been stressed since the 1970s for these regions. Except for Saint-Louis where the cultivated area has increased due to the introduction of intensive Asian rice cultivation policies, Louga and Matam lost 12% to 18% of their overall cultivated area between 2001 and 2017 [55], including 50% of the pearl millet cropping area.
Similarly, we observed pearl millet cropping abandonment in southern villages under suitable rainfall conditions. The rice, maize, cassava and cowpea cultivated area has increased by more than fourfold in this region [55]. Most of these crops are mainly used as cash crops and sold on markets. Pearl millet yields never exceed 0.9 t/ha while rice yields may reach 2 t/ha and cassava yields are sometimes over 10 t/ha [55]. Pearl millet has hence been replaced by more productive crops, in line with the trend in other West African countries [56].
Overall, this suggests that, in recent years, there has been a shift in agricultural strategy from food to cash crops, with a marked reduction in cultivated areas in northern Senegal. Despite the observed changes in the pearl millet growing area and the magnitude of drought events, we did not find a significant decrease in terms of genetic diversity between the two periods. Two non-exclusive explanations can be put forward: 1) pearl millet is a very diversified allogamous species, and 2) despite some seed loss events, which were probably very localized, diversity has likely been maintained through the reacquisition of landraces via traditional seed exchange systems [57][58][59][60][61].

Homogenization of pearl millet diversity across Senegal
Neutral diversity has been maintained but its structure has changed. Our results revealed homogenization within and between early-and late-flowering landraces. We acknowledge that our LF sample size was small due to the germination issues discussed in the Materials and Methods section; nonetheless, homogenization within and between EF and LF landraces was congruent with: 1) the southward displacement of EF landraces, thus increasing the overlap of EF and LF growing areas [53], and 2) the over-representation of EF at the national scale, thus increasing the likelihood of EF to LF gene flow. Genetic homogenization within EF landraces was also revealed by the loss of genetic diversity structuring over a latitudinal gradient. Genetic homogenization could be the result of an intensification of seeds' exchange between farmers and along larger scales. Another possibility is that homogenization results from the diffusion and release at large scale of improved varieties such as Souna 3 [32,62]. Improved varieties are not always adopted [36], nevertheless, because farmers test them over a few years, they contribute to local pollen and seed mediated gene flow in local varieties. This could explain the pattern we observed, a large-scale homogenization of diversity.
This genetic homogenization may have major impacts pertaining to agricultural adaptation and resilience. From an agroecological standpoint, there is now a growing body of evidence on the role varietal mixtures could play in adaptation to climate change, notably in terms of resilience, i.e. ensuring minimal production to climate uncertainty [63][64][65]. While LF landraces are preferred by farmers (higher crop and fodder yields, etc.), these crops are riskier with respect to the uncertainty regarding the rainy season. Farmers may grow EF and LF landraces on the same plot-EF landraces ensure a minimal yield under poor rainy season conditions, whereas LF landraces can make profitable use of late rainfall of the rainy season and produce sufficient fodder [18].

PLOS ONE
Temporal evolution of pearl millet diversity

Wild and cultivated pearl millet gene flow
Weedy pearl millet and sorghum plants can sometimes be used by farmers for their earliness in harsh conditions [66][67][68], therefore facilitating gene flow between wild and cultivated gene pools. We found that admixture rates were higher in 1976 than in 2016; this phenomenon could possibly be explained by the fact that EF landraces were cultivated further south, which could have reduced the overlap of the wild and cultivated pearl millet distribution area (Fig 1).
Our study also showed that the diversity of wild relatives is at risk. Indeed, cultivated-towild pearl millet gene flow was significantly higher, leading to a change in terms of A R and H Exp in the wild diversity, thus increasing the risk of genetic swamping. Similar results have also been obtained in the Sahelian region [69]. When sampling, we sought avoiding weedy phenotypes and removed miss-assigned samples (potential weeds) from analyses. Estimates were therefore conservative, and could be considered as the lower limit of the estimate of contamination of the wild pool by the cultivated pearl millet pool.
Wild relatives are seen as a reservoir for future adaptation [24,70], their importance in climate change mitigation strategies is increasingly recognized [71]. Until recently, cultivated alleles were thought to be maladaptive and counter-selected very quickly. It has now been demonstrated that cultivated diversity could be beneficial and thus positively selected in wild conditions [72,73]. Despite their value, crop wild relatives (CWR) are generally poorly represented in ex situ collections [74,75] and efforts ought to be encouraged to conserve more wild diversity in gene banks. In Senegal, the low pairwise F ST noted between wild populations (S5 Table) confirmed the lack of structuring, which means that a limited number of accessions would be enough to conserve wild diversity. Ex-situ collection could also enable time-course studies and help gain insights into wild pearl millet evolution [76,77].
Genetic homogenization might negatively impact long-term adaptation [78]. If improved varieties contribute to the local maintenance or increase of genetic diversity [9], their large dispersal is likely to lead to a strong spatial homogenisation of diversity [78,79]. Recent methodological developments make it possible to measure the adequacy of the diversity currently available to survive in the future climatic conditions, i.e. the genomic vulnerability [80,81]. These approaches developed in trees [80] and birds [81] allow identifying diversity crucial for the adaptation to future climate conditions. Conservation strategies ought to focus on enhancing the functional diversity from wild and cultivated populations in current and future gene banks. It is essential to improve quantification of not only diversity but also erosion of the functional diversity. Characterizing erosion of functional diversity will make possible to determine the real impact of the large-scale dissemination of few improved varieties and thus influence the rationale underlying agricultural development policymaking.

Conclusion
First, this study is one of the only studies documenting a decrease of the pearl millet growing area. With an even drier future climate, pearl millet cultivation might continue to decrease in the northern part of Senegal. A strong homogenisation of diversity is now observed across Senegal, likely hampering abilities of pearl millet to adapt to future climate. Two major strategies for increasing diversity in pearl millet and favouring its adaptation to climate changes ought to be implemented: 1) better define plant breeding/variety development programs and appropriate seed dissemination and 2) development of strategies on assisted migration. Assisted migration, i.e. introducing crops/varieties from areas where current ecological niches could correspond to those of the target region in the future, can be a major adaptive strategy to cope with future climatic conditions. However, this approach may not be suitable in a system where large-scale diversity homogenization is underway, thus highlighting the need to focus on the conservation of functional and local diversity for crops and their wild relatives.