Influence of geographic provenance on phenotypic variation in seed and kernel traits of the African oil tree from southern Benin and implications for species breeding

Knowledge of oilseeds plants’ traits and features is fundamental to understanding the natural selection process and improving conservation programs through species selection. As a forest oilseed, Ricinodendron heudelotii is the object of growing interest among value chain actors, who are increasingly interested in its intrinsic characteristics. To date, there is a lack of information on plant morphological traits for the selection of the best genotype, especially as far as seed and kernel traits are concerned. The aim of this study was to make a typology and establish the relationship between seed and kernel traits of Ricinodendron heudelotii, depending on provenances represented by wild populations of trees in southern Benin. We sampled the provenances constituted equally and per tree, ten random samples each of fruits, seeds, kernels, and shells were collected for measurement. Hierarchical classification, common component analysis, Pearson’s, and Chisquare association tests were performed for statistical analysis. Substantial variation between seed and kernel traits was observed between and within provenances. High coefficients of variation for the quantitative descriptors (length, width, and mass) of seeds and kernels appear to be the essential traits that discriminate the plant into two morphotypes. Furthermore, seed and kernel characteristics show a significant positive correlation with these discrimination criteria. The upper morphotype contains accessions from three provenances with huge seeds and kernels but few seeds per fruit, while the lower morphotype includes the other provenances with tiny seeds and kernels. The results of the study provided essential information that could be an avenue for improvement with further biochemical and molecular characterization studies.

In Benin, Ricinodendron heudelotii is an important plant genetic resource, occupying the 2nd place among priority oilseed trees (Hounsou-Dindin et al., 2022). It is restricted to two phytogeographic zones, Pobè and Borgou-Nord (Akoègninou et al., 2006), and is the most abundant (77.6%) in the tree stratum of the semideciduous forest of Pobè (Sokpon 1995). It has been reported that between 2005 and 2015, the rate of disappearance of natural formations in Benin estimated at 2.2% each year (Mama et al., 2020). Currently, several studies have reported that the species is subject to serious threats linked to the massive exploitation of the trunk in the manufacture of Guèlèdè masks (Boko-Haya et al., 2017;Moussa et al., 2020). These constraints have meant that the species very rare in open woods, and its exploitation has already reached protected forests at the request of the royal authority (A. Badou, CRAPP, personal communication, September 2018). This human-induced habitat degradation and disturbance has a detrimental effect on the genetic diversity of species, with the possibility of extinction in small populations (Holsinger 2000;Helm et al., 2009;Schlaepfer et al., 2018;Dudley and Alexander, 2017;Jiang et al., 2018). To avoid the genetic extinction of R. heudelotii, knowledge of the phenotypic variability available within populations of this threatened species is a fundamental step in establishing effective breeding and conservation programs.
Previous work has revealed that the number of seeds per fruit varies from one to five (Fondoun et al., 1999;Mpeck et al., 2003;Eyog Matig et al., 2006), with significant variation in fruit and seed mass (Nehemiah et al., 2007). As a result, the relationship between phenotypic characteristics of fruits and seeds is tenuous (Mpeck et al., 2003;Tchoundjeu and Atangana 2006), which is a handicap in predicting seed production from fruit morphological traits in farmers (Ayuk et al., 1999;Franzel et al., 2007). Due to its woody endocarp being very hard and adherent to the kernel, and therefore very difficult to extract intact, information on the intrinsic quality characteristics of the kernels and on the seed-kernel relationship is lacking. Kernel quality is of paramount importance in the production and marketing of nut fruits. In a collection of R. heudelotii seeds, farmers and breeders are interested in morphometric characteristics and the seed-kernel relationship, while consumers are interested in kernel colour because of the importance of these traits in the value chain. For example, the size, colour, and shape of shea kernels are important criteria for purchasing, processing, and treatment equipment designs but also priority criteria in plant breeding programs (Honfo et al., 2012;Nde et al., 2016;Agúndez et al., 2020). Therefore, it is important for farmers, buyers, and processors to know the intrinsic characteristics of the seeds and kernels for better valorization of the species.
Knowledge of the morphological characteristics of the seeds and kernels is a priority for conservation and improvement of the resource; it can help understand phenotypic variation within populations of different provenances. Inter-and intra-provenance variation in edible fruit tree traits are prerequisites for cultivar development and plus tree identification. Variation in kernel colour is influenced by site (Sakar et al., 2021) and kernel mass can vary from tree to tree (Bai et al., 2021). Furthermore, between populations within the same agro-ecological zone, seed mass shows inter-and intra-specific variation (Tsobeng et al., 2020). Therefore, it is likely that the morphological characteristics of Ricinodendron heudelotii seeds and kernels are influenced by their provenances. To date, no studies have been undertaken to assess the seed and kernel characteristics of local populations of the African oil tree and how these traits can be used as indicators in the improvement of Ricinodendron heudelotii.
This study assessed the morphological characteristics of the 'African oil tree' populations in the southeastern region of Benin, described the natural variation traits of the seeds and kernels that can be used as tools for discriminating between populations by value chain actors, and investigated the links between these traits in order to guide growers and breeders in the choice of the best plant material for production and domestication strategies in a context of increasing demand for almonds. We hypothesized that (i) seed and kernel provenance influence the morphological characteristics of different populations of R. heudelotii and (ii) a strong correlation exists between the morphological variation of the seeds and their kernel.

Study site
The study was carried out in the phytogeography of southern Benin (6º25'E to 7º30'N), under a Guinean-Congolese climate with an annual rainfall varying from 900 to 1300 mm with hydromorphic, ferralitic soils without concretions (Adomou et al., 2006). Three boundaries of the Plateau phytodistrict characterized by dense semi-deciduous forests were considered: the western plateau west of the Ouémé valley with the Agrimey and Massi provenances located on hydromorphic soil and a rainfall of 1100-1300 mm; the southern border of the Eastern Plateau with Pobe with the provenances Akouho, Itchede, and CRAPP located on a ferralitic soil without concretions and a rainfall of 1100-1300 mm; and the northern border of the Eastern Plateau with Zou with the provenances Adaka, Ilagbe, Idena, Ilikimou, and Woroko located on a red ferralitic soil without concretions and a rainfall of 900-1300 mm. The majority of these provenances are located in southeastern Benin, near the Nigerian border ( Figure 1). The main geographical factors that influence the ecological conditions for plant growth are presented in Figure 2. The population around these phytogeographical zones is mainly made up of the Nago and Yoruba ethnic groups, heirs to endogenous knowledge of Guèlèdè mask making from R. heudelotii wood (Boko-Haya et al., 2017;Fagbohoun et al., 2019;Akpovo and Fandohan, 2021).  Sampling A preliminary survey with the help of key informants (carvers, forest reserve rangers, and endogenous cult leaders) identified through snowball sampling identified provenances and trees that produced fruit. Between July and October of the 2018 season, the period of maximum productivity and free-fall of fruit, ten newly fallen ripe fruits ( Figure 3a) were collected at random, below the total leaf area of each tree. Trees were at least 100 m apart to avoid collecting information on related individuals (Nevill et al., 2010;Amirchakhmaghi et al., 2018;Lloyd et al. 2020) and provenances were at least 10 km apart; therefore, trees from all provenances were considered as different populations (Te Beest et al., 2009). The total number of trees sampled was 56 equitably (n= 5 at least) to the provenances and the total number of fruits was 560.
To extract the seeds associated with each fruit and tree, an adapted protocol was tested. Fruits from each tree (Figure 4), were fermented separately in small labeled bags for about one month in a trench below ground to allow decomposition of the mesocarp and weakening of the sclerotized endocarp ( Figure 3b). The number of seeds per fruit was counted, and these freshly extracted seeds from each fruit were rinsed with tap water and the whole constitutes the seed collection for this tree. Then, in Akassato (2°20'58.382 "E, 6°30'25.207 "N), at an average temperature of 27 °C, these collections were dried alternately in the shade and in the sun for about six months in order to cause the natural detachment of the kernel from the shell wall and to facilitate the extraction of the intact kernels after a very light hammering of the extremely hard shell. From each specified seed collection, ten seeds were randomly selected and crushed to obtain the kernels. In the event that a seed was found to be kernelless, it was replaced by another seed from the corresponding collection. Thus, 560 seeds and 560 local kernels were obtained. Samples from Lobaye forest (3°52'4.11''E,17°59'21.22''N,Central African Republic) belong geographically to R. africanum subspecies, used as control.

Collection of morphological data
Quantitative data were collected based on the survey and collection guide for local crop varieties (Marchenay and Lagarde, 1986) and the descriptors selected by IPGRI for other fruit species (shea, citrus, and date palm). So, length, width and thickness were measure using a calliper with a precision level of 0.01 mm ( Figure 5). Seed mass were collected using a 0.01 g precision balance. In addition, the number of seeds per fruit, the number of striae and hilum-micropyle distance were recorded.
Qualitative data were also collected, seed colour and appearance ( Figure 6), seed apex shape ( Figure 7) and kernel colour ( Figure 8) and other characters and modalities detailed in Table 1. All these data were used to phenotypically differentiate seeds and kernels from different provenances.

Data analysis
We performed a common component analysis of variance to determine the source of variability in seed and kernel morphological traits using the lme4 (Bates et al., 2015) and lmerTest (Kuznetsova et al., 2017) packages. In this model, provenance and mother tree were respectively considered as fixed and nested factors. Besides, hierarchical principal component classification (HCPC) ward realized on the matrix of seed and kernel morphological characteristics by using Minitab 18 software. Also, due to the non-normality of the residuals shown by the Shapiro-Wilks test and the inequality of variances, generalized least squares models were used to test the difference of means for each fruit descriptor between the groups. Elsewhere, we performed Pearson's correlation test to describe the linearity relationship between morphological characteristics and the chi-square test to determine the relationship between the clusters and seed and kernel qualitative traits. All analyses were performed using R software, version 4.0.2 (R Core Team, 2022).

Results Results Results
Variation in phenotypic traits between and within provenances Significant variation (p<0.01) was observed between provenances for all morphological traits, with the exception of seed, kernel, and shell thickness (Table 2). Similarly, common component analysis of variation revealed considerable heterogeneity both between and within provenances (Table 3).
Overall, heterogeneity in organ functional characteristics was noted between provenances. This variance was greater than 50% for several of the characteristics assessed. Thus, between provenances, 80.41% and 61.88% of the variance in seed length (SdL) and width (SdW) was detected respectively, as well as 59% and 58.77% of the variation in ventral groove length (VGL) and width (VGW). Similarly, between provenances, 60.08% and 79.58% of the variance in the number of dorsal striae (SrN) and the distance between the micropyle and hilum (DHM) were detected. When kernels were included, the results were similar to those found with seeds, i.e., 70% and 74.38% of the variance in kernel length (AlL) and width (AlW), respectively, were observed between provenances. Provenance affected the seed and kernel length growth of R. heudelotii.
Within provenances, 52.00% and 87.30% of the differences in seed thickness (SdT) and kernel shell thickness (ShT), respectively, and 63.55% of the differences in kernel thickness (AlT), were identified. The mean seed, kernel, and shell thickness varied widely between mother trees of the same provenance.

Identification of morphotypes
Using principal component analysis, hierarchical clustering produced two large groups (Figure 9) that are statistically distinct (p<0.005). The first group encompasses 63.63% of the provenances (Akouho, Massi, Agrimey, Ilikimou, CRAPP, Woroko, and Adaka) and has over 37.5% similarity. Akouho forms one monophyletic group within this first group, while the other provenances form the other monophyletic group. Provenances of Woroko, Adaka, Agrimey and Massi, were quite close. Such, relationship can be understood by their shared ecological circumstances (or morphometric characters). On the other hand, the proximity of the Ilikimou and CRAPP provenances, which do not share similar ecological conditions, may be related to similarity in morphometric characteristics. Table 3 shows that this first group is characterized by seeds and kernels with means and coefficients of variation (CV) of length (1.46 cm, CV = 07.08 and 1.07, CV = 09.77), width (1.38 cm, CV = 07.15 and 0.93 cm, CV = 23.35), thickness (1.08 cm, CV = 10.98 and 0.72 cm, CV = 15.70), and distance between hilum and micropyle (1.24 cm, CV = 9.80). Similarly, at the fruit level, the respective means and coefficients of variation of the length (3.92 cm, CV = 10.73) and mass (26.38 g, CV = 21.53) of this first group are lower and significantly different (P<0.001) from those of the second group, except for the number of seeds per fruit (2.92, CV = 9.70), while the width of the fruits (3.31 cm) of the first group is not significantly different from that of the second group (3.37 cm). The second group includes 36.36% of provenances with less than 37.5% similarity (Lobaye, Itchede, Ilagbe, and Idena). Within this group, two distinct subgroups exist: Beninese and Central Africans. Illagbe, Idena, and Itchede constitute the Beninese sub-group. The seeds and kernels of Ilagbe and Idena are quite similar to those of Itchede in this Beninese subgroup. This similarity can be explained by their common agroecological zone which is Kétou, where the geographical gradient and ecological circumstances are more homogeneous than in the agro-ecological zone of Pobè, which includes Itchede. In contrast, the Central African subgroup is composed entirely of Lobaye, i.e., seeds and kernels indigenous to Central Africa, where the climatic and soil circumstances are different from those in Benin (West Africa). This second group is defined by seeds and kernels, with mean values and coefficients of variation for length (1.75 cm, CV = 06.25 and 1.29 cm, CV = 07.46), width (1.63 cm, CV = 08.29 and 1.12 cm, CV = 14.77), thickness (1.29 cm, CV = 07.97 and 0.90 cm, CV = 13.28), and number of dorsal striae (2.95, CV = 55.50) (Table 3). Similarly, the means and coefficients of variation for length (4.55 cm, CV = 6.24) and mass (30.36 g, CV = 14.62) of this second group are higher and significantly different (P< 0.001) from those of the first group, except for the number of seeds per fruit (1.86, CV = 21.75). The second group contains trees with large seeds and large kernels but few seeds per fruit, while the first group contains trees with more seeds and kernels but smaller sizes. Table  Table Table  Table 3 3 3 3. . . . Descriptive statistics (ME: mean, CV: coefficient of variation) of fruits, seed, and kernel morphological traits according to characteristics Comparing seed and kernel masses of R. heudelotii by groups, we observe that the generalized least squares results suggest that groups have a substantial influence on seed (F = 64.99, p<0.001) and kernel (F = 15.48, p<0.001) masses of R. heudelotii. The seed and kernel masses of group 1 (12.94 g ± 0.35) and 5.08 g ± 0.14) were significantly different from those of group 2 (21.43 g ± 0.58) and 6.71 g ± 0.35). Seeds and kernels in group 2 are also heavier than those in group 1. With the exception of the kernel, shell thickness (ShT), all seed and kernel characteristics are statistically extremely different (p<0.001) between the two groups or morphotypes (Table 4). The two morphotypes differed mainly in seed and kernel length, width, thickness, and mass, but also in the number of dorsal striae.
For the qualitative traits, the independent chi-square test (Table 4) reveals a significant difference (p<0.001) between the groups in terms of colour, seed appearance, dorsal aspect, and apex shape on the one hand, and seed appearance on the other. The relative frequency (%) of each of these characteristics at the group level shows that, seeds in group 2 are black (68.33%) or brown (20%), while those in group 1 are grey (32.15%) or khaki (27.25%); seeds in group 1 are rough (97.45%), while those in group 2 are rough (68. 33%) or hooked rough (26.66%); group 1 seeds are symmetrically pointed (54.70%) or round (28.62%), while group 2 seeds are mainly symmetrically pointed (80%). The shape of seeds in group 2 is obovate (83.33%) or elliptical (16.66%), compared to the shape of seeds in group 1 (62.74% and 18.23%, respectively), where the cordate shape also exists (17.84%); the dorsal aspect of seeds is grooved (97.05%) in cluster 1 compared to cluster 2 (80%), where 20% are also grooved and reticulated; the colour of the kernels was white (91.66%) in cluster 2, and white (74.70%) or yellowish (16.47%) in cluster 1; the appearance of the kernels was smooth (81.56%) in cluster 1 compared to rough (38.33%) or ashy rough (33.33%) in cluster 2.  Table  Table Table  Table 4 4 4 4. Results of chi-Deux independent test measuring the degree of association between clusters and organs traits R. heudelotii was identified as having two morphotypes: morphotype 2, located in Ilagbe, Itchede, and Idena, has black or brown seeds that are obovate or elliptical in shape, rough and/or hooked at the apical end, and have longitudinal striae (3)(4) or are reticulated on the dorsal side; morphotype 1 has grey or khaki (yellowbrown) seeds that are obovate with 4-7 longitudinal striae. In morphotype 2, the kernels are white, rough, and sometimes ashy, whereas in morphotype 1, they are white or yellowish to yellow with a smooth appearance. In general, the kernels have the same shape as the seed from which they are derived.

Relationships between descriptors of R. heudelotii
Pearson's correlation coefficient indicated that many of the attributes examined had substantial positive associations (p<0.001) ( Table 5). The correlations between seed length and width (r = 0.77), seed thickness (r = 0.57), distance between hilum and micropyle (r = 0.68), and kernel length (r = 0.62) were all positive.  Table  Table Table  Table 5  Kernel width was found to be favourably related to kernel thickness (r = 0.69) and distance from the hilum to the micropyle (r = 0.62). Similarly, the distance between the hilum and the micropyle was favourably related to the thickness of the seed (r = 0.51). In addition, there was a strong and positive correlation between shell thickness and kernel width (r = 0.79).

Discussion Discussion Discussion
Due to the significant correlation between seed thickness and length (r = 0.58), long seeds are also large. In contrast, the association between kernel thickness and length was very weak (r = 0.34); as long kernels do not necessarily have a large thickness, this discrepancy is due to kernels that are not filled in the shell. Furthermore, Pearson's correlation test revealed a substantial association between seed mass and kernel mass (Coef = 0.79, Pr = 0.001). Seeds with a high mass also have a high kernel mass.

Phenotypic traits show great variation
Planning sustainable domestication and plant breeding strategies requires knowledge of population variability. We examined the phenotypic variability of R. heudelotii seeds and kernels from wild populations to identify the most useful traits for breeding and conservation. In R. heudelotii, seed and kernel length, width, and mass, as well as the number of dorsal striae and the distance between the micropyle and the hilum, showed high inter-provenance variation, but seed, shell, and kernel thickness showed high intra-provenance variation. Cluster analysis confirmed the phenotypic diversity, grouping the provenances into two major groups (morphotypes).
This study emphasizes the hypothesis that provenances have a significant effect on seed and kernel dimensions. Previous studies showed that seed length and mass, seed width, colour, and texture of Jatropha curcas were found to be key characters for identification and discrimination of morphotypes between phytodistricts in Benin (Padonou et al., 2013). Many forest species have shown inter-provenance variation in seed length and width, including Faidherbia albida (Delile) A. Chev. (Fredrick et al., 2015), Pinus roxbrughii (Ghildiyal et al., 2009), Celtis australis (Singh et al., 2006), Argania spinosa (Belcadi Haloui et al., 2017), and Pterocarpus erinaceus Poir (Johnson et al., 2020), suggesting that seed length and width are strongly influenced by geographical origins (Harel et al., 2011;Singh and Thapliyal, 2012;Madjelia and Kadidiatou, 2015;Belcadi Haloui et al., 2017;Louati et al., 2019). Geographical origin is under the influence of climatic and soil factors, which are responsible for the overall variation in plant traits (Joswig et al., 2022). In the current study, all origins belong to the same climatic zone, the Guinean-Congolian zone, which suggests that the observed variations are not potentially climate-related. However, slight differences in rainfall, latitudinal, or physical and chemical soil properties are responsible for variations in morphological traits (Jongschaap et al., 2007;Abasse et al., 2011;Grozeva et al., 2016;Sama et al., 2022). In addition to soil conditions, the observed intra-provenance variation in thickness (seeds, hulls, and kernels) highlights a likely influence of the mother tree. Therefore, phenotypic variation in seeds and kernels may originate from soil conditions, maternal traits, or a combination of both (Kołodziejek, 2017). However, the cohabitation of both morphotypes at the CRAPP site, seems to support the hypothesis of a genetic difference between the two morphotypes.
Experiments in common gardens will be necessary to better elucidate the origin of this genetic difference. The high variation obtained means that the traits are heterogeneous and offer opportunities for selection of desirable traits to be maintained (Adler et al., 2013). In addition to these new traits, the two morphotypes were distinguished by variations in seed and kernel colour, dorsal aspect of the seeds, and number of striae. The colour of the kernels is an important quality characteristic that defines the visual acceptance of consumers and the preferences of agro-industrialists and is also a criterion for discriminating between cultivars of Prunus dulcis (Mill.) D.A. Webb. Also, according to Barthlott (1981), cuticular striations, inclusions, and thickenings of the visible outer walls of the seeds are among the surface characters used in taxonomy; for example, in most Leiothrix species, the striated seed coat is a distinctive ornamentation of the seeds that also increases the germination rate (Silva Mascarenhas and Scatena, 2021). For example, to avoid granivores, seeds take on the colour of the soil in which they grow to avoid detection (Porter, 2013).
The information on these new discriminating characters is valuable and complements the knowledge of the morphological description of the species. In the present study, the Central African accessions from Lobaye, phylogenetically closer to those from Ilagbe, Itchede, and Idena, and having an average of two seeds, would therefore be R. heudelotii var. africanum, whereas the accessions from Massi, Akouho, Woroko, and Agrimey with three perfectly mature seeds would be R. heudelotii var. heudelotii. Group 2, comprising Itchede, the best morphotype in our survey, could be considered for valorization in Benin given that the large seeds are largekernel and potentially have nutrient reserves responsible for good germination and seedling growth (Boko-Haya et al., 2021;Hounsou-Dindin et al., 2022). Future research should investigate this morphological differentiation using biochemical and molecular genetic markers to show the potential of the two morphotypes.

The extend of correlations between morphometric traits
The study shows a linear correlation between the productive traits of seeds and kernels. Intriguing correlations were found between most of the productive traits. The strong positive correlation (r = 0.58) between seed thickness and seed length, between seed thickness and kernel length (r = 0.34), and between seed length and kernel length (r = 0.62) indicates that large and long seeds also have large and long kernels. This newly discovered relationship is extremely interesting in the selection of productive traits; that is the same for mass, where seeds with high mass also have high kernel mass; this relationship is very beneficial for the productivity of the species. Similar observations have been made in Pentaclethra macrophylla Benth (Tsobeng et al., 2015), Phoenix dactylifera L. (Gros-Balthazard et al., 2016), Juglans regia (Orman et al., 2020) and Juglans mandshurica (Zhang et al., 2021).
These positive correlations between productive seed traits and kernels of R. heudelotii provide a useful indicator for predicting kernel yield of the species, a crucial challenge for farmers who may now consider large seeds as a key criterion for improving kernel yields. Thus, the study partly demonstrates a linear correlation between productive seed traits and kernels; these traits can predict each other (Talhouk et al., 2000) and an improvement in one quantitative trait can lead to changes in the other (Loha et al., 2006;Čolić et al., 2012). It is desirable that plus trees 'selection for yield could be based on qualities that show substantial correlations with the target attribute as criteria (Khadivi-Khub and Etemadi-Khah, 2015). So, do the differences in kernel colour and appearance clearly noted in this study reflect differences in the oil quality of R. heudelotii?

Conclusion Conclusion Conclusion Conclusion
Among the Benin accessions of R. heudelotii, two morphotypes were clearly distinguished based on the length, width, thickness, and mass of the seeds and kernels, as well as the number of dorsal striae on the seeds and kernels. Former descriptions of R. heudelotii used fruit to differentiate populations, but as seeds cannot be predicted from fruit, the results of the present study complement the former description by assessing seed and kernel morphology. Heavy seeds have heavy kernels, thus establishing a significant positive association between seed mass and kernel mass, which is crucial to the yield attribute of the species, which was a key concern of farmers. R. heudelotii has a high phenotypic diversity in nature and various correlated traits, which could be valuable for domestication and genetic improvement. Three provenances (Itchede, Illagbe, and Idena) were recommended for the selection of the best morphotype. The future application of genetic molecular markers to accessions from different provenances could help to elucidate this finding concerning phylogenetic relationship.

Authors' Contributions Authors' Contributions Authors' Contributions Authors' Contributions
Field work and drafting of the manuscript: BYY. Statistical analysis: AYY. Critical revision of the manuscript and supervising of the work: OAINC, and AC. Harel D, Holzapfel C, Sternberg M (2011