Phenotypic Diversity of Doum Palm (Hyphaene compressa), a Semi‐Domesticated Palm in the Arid and Semi‐Arid Regions of Kenya

Hyphaene compressa is an economically important palm in Africa. Despite its significant role in the livelihoods of rural communities, the diversity of doum palm is poorly documented and studied. In addition, it has no model descriptor that can aid such studies. Ninety H. compressa accessions collected from Northern, Eastern, and Coastal regions of Kenya were examined to determine the morphological variability of the vegetative and fruit traits of H. compressa and to identify its morphotypes for improvement. A total of 19 morphological characters including seven quantitative and 12 qualitative traits of fruit and vegetative traits were selected. Linear mixed-effects models, principal component analysis, and linear discriminant analyses were used to assess the variation in the morphological traits of doum palm based on the regions. Hierarchical clustering was performed to identify the morphotypes of H. compressa. There was variability in H. compressa morphological traits, particularly at the Kenyan Coast. All seven quantitative traits were able to effectively discriminate doum palm phenotypically (p ≤ 0.001). The 90 accessions clustered into five morphotypes designated as 1, 2, 3, 4, and 5. Morphotype 4 was specific only to the Coastal region. Morphotype 5 had the tallest trees with the biggest fruits and included palms from Eastern and Coastal regions making it the best morphotype for fruit traits. This study will inform the domestication, improvement, and conservation of H. compressa by selecting elite accessions.


Introduction
Hyphaene compressa (doum palm) H. Wendl. is a common palm in East Africa [1,2]. It belongs to the Coryphoideae subfamily of the Arecaceae family [3]. e genus Hyphaene also known as the "doum palms" is predominant in Africa and has eight species, namely, H. compressa [2,4]. In Africa, the genus Hyphaene has a wide range of uses that include but are not limited to the source of non-timber products for construction materials, food, medicine, and woven products as documented by several studies [1,5,6].
Despite the important economic role and contributions the genus makes to the palm family diversity in Africa, the genus is still poorly understood and evaluated [4,7]. Of concern is the steady decline of doum palm populations in Africa due to destruction of their cradle habitat, drought, and overharvesting, thereby exacerbating pressure on the remaining African doum palm accessions which could inevitably lead to loss of their gene pool [4].
In Kenya, H. compressa plays a significant role in the livelihoods of people especially the pastoralist communities who rely on it for food, construction materials, medicine, and income through the sale of woven products [6]. e most important use of H. compressa in Kenya is food. ere is increasing interest in doum palm domestication in some of the arid and semi-arid regions of Kenya. e drive for this is the decline in doum palm germplasm resources in these areas due to human interference and biotic stress [8]. H. compressa in situ conservation is found in six protected areas and five ex situ conservation areas globally [9,10]. In Africa, in situ conservation status of doum palms is limited and difficult to ascertain [11]. Moreover, the IUCN red list has categorized doum palm as a least concern species and is not yet in the category of near-threatened species. is could be the reason for the limited conservation efforts in the region. However, increased anthropogenic activities might lead to loss of doum palm biodiversity and ultimately more conservation efforts will be advocated for in the future. erefore, conservation and diversity studies are needed to hasten the process of domestication and genetic improvement.
Diversity can be assessed using morphological variations which are informative enough for evaluation and description [12,13]. Phenotypic characterization is the basic step for the classification, conservation, and utilization of genetic resources [14].
e present lack of knowledge on H. compressa limits access to its important traits and hence a hindrance to its improvement. Besides, it has no model descriptors which can aid in diversity studies. It, therefore, has no reference values at the International Plant Genetic Research Institute (IPGRI). It is important to determine the unique phenotypic descriptors for H. compressa which can be relied upon to distinguish members of this group [15]. A descriptor is a collection of standardized features used to provide information for describing and classifying a specific group of genetic resources [16]. According to IPGRI (https:// www.bioversityinternational.org/e-library/publications/ descriptors/), to facilitate international exchange and use of genetic resources uniformly, it is important to standardize these descriptors. Other palms like coconut, sago palm, peach palm, and date palm have descriptors that can be assessed at the IPGRI website. Morphological diversity study is the initial step for plant breeding. erefore, to enhance doum palm, the diversity of its morphology is important. e objectives of this study were to determine the morphological variability of the vegetative and fruit traits of H. compressa and to identify the morphotypes of doum that are important for its improvement. It is assumed that the vegetative and fruit traits of doum palm are important in doum palm phenotypic diversity. It is also assumed that there are different morphotypes of doum palm.

Study Area.
is study was done in three regions of Kenya: Northern (Turkana County), Eastern ( araka Nithi County), and Coastal (Tana River and Kwale County) as shown in Figure 1. ese regions are characterized by high temperatures ranging from 20°C to 41°C and erratic rainfall of 280 mm to 2200 mm. e attributes of each of these study areas are summarized in Table 1.

Sampling.
Sampling was done between January and July 2018 when doum palm trees were fruiting. Identification of doum palm was done with the aid of a taxonomist from the National Museums of Kenya. e selection criteria included the gender of the plant, the maturity of the tree, and the general good health of the palm and fruits. Only fruiting palms were selected for morphological diversity study. is is because distinguishing the nonflowering males from nonfruiting females is difficult in the wild populations. Moreover, doum palm has limited descriptors that can aid in diversity studies; therefore, fruit traits are important which are lacking in the male. Purposive sampling was used to select 30 female trees from each region. Doum palm trees sampled were separated from each other by at least 200 meters to reduce the probability of sampling close relatives [22]. From each sampled female tree, 10 fruits were randomly collected. e collected fruits were labelled, placed in bags, and transported to the laboratory for morphological assessment. e fruits collected from each tree were pooled and stored in one bag [22]. Some of the descriptors used for morphology were adapted from a descriptor list available for date palm [16]. All collected fruits were cleaned in running sterile water and left in the open to dry in the sun [22]. is was followed by the assessment of fruit morphological descriptors ( Table 2). Fruit length and width were measured using vernier calipers [23]. e fruit weight was measured using an electronic weighing scale (Sartorius Entris 64-1S).
Some of the morphological parameters used in this study included those used by Rizk and El Sharabasy [16]. e morphology of the leaves and stem was assessed in the field during sampling. Leaf morphological characters were assessed as an average of five well-developed doum palm leaves [24]. Quantitative and qualitative vegetative traits were recorded ( Table 2). Photographs of the plant leaves, stem, and fruits were taken to document their morphology and any differences were noted.

Data Analysis.
e mean, range, and coefficient of variation were calculated for quantitative traits per sampled region.
e frequencies for qualitative data were also recorded. e analysis of variance (ANOVA) was performed to determine the difference in the mean among categories and sites [25]. e Games-Howell Post Hoc Test was used to determine specifically which two treatments differed significantly from each other for the different phenotypic traits. Standardization of data was done because different scales of measurement were used for the different quantitative parameters assessed [14,26]. Linear mixed-effects model using Ime4 package in R was used to assess the morphological diversity of doum palm according to the geographical regions of collection. e principal component analysis (PCA) using prcomp package in R was done to identify the most discriminating traits among the sampled sites. Discriminant analysis was done to estimate and describe each population using the MASS package in R. All the quantitative data were standardized prior to discriminant analysis. Clustering was done using Gower distance with the PAM (Partitioning around Medoids) algorithm using the daisy package in R. Both the numeric and categorical data were used for the cluster analysis. e silhouette coefficient was used to determine the number of clusters. All the statistical analyses were done in R version 4.0.2.

Morphological Diversity of Fruit and Vegetative Traits.
e frequencies of the quantitative traits are summarized in Table 3. ere was high variability for doum palm height (cv � 38.3%). e fruit sizes ranged from 48.2 g to 148.8 g (cv � 21.5).
ere was low variability in the fruit length (cv � 11.8). ere was variability in the fruit and vegetative  quantitative traits of H. compressa per region (Table 4). All the seven quantitative traits were able to effectively discriminate doum palm phenotypically (p ≤ 0.001; Table 4).
ere was no significant difference in the quantitative traits of doum palm between Kwale and Turkana for leaf length, leaf breadth, fruit length, and fruit weight. araka Nithi had the highest mean height (13.5 m) with the least being Kwale (5.65 m). e leaf breadth was significantly smaller (p ≤ 0.001) in Tana River (55.87 cm) than the other sampling sites. Tana River had the highest mean leaf length (120.2 cm) and fruit length (7.64 cm) with a p value of 0.000473 and <6.32e -12, respectively. ere was a positive correlation between doum palm height and leaf length (p ≤ 0.001), leaf breadth (p � 0.006), fruit breadth (p � 0.029), fruit weight (p ≤ 0.001), and fruit length (p ≤ 0.001). ere was a negative correlation between petiole length and all the quantitative fruit traits, fruit length (p ≤ 0.001), fruit breadth (p � 0.004), and fruit weight (p ≤ 0.001) as shown in Table 5.
A linear mixed-effects model was fitted to predict H. compressa fruit weight with height, leaf length, leaf breadth, petiole length, fruit length, and fruit breadth. e model included the four sampling regions as random effects. e model's total explanatory power was substantial (conditional R 2 � 0.80), and the part related to the fixed effects alone (marginal R2) was 0.66. e model's intercept was at -70.25. Within this model, the effect of fruit length on fruit weight was significant (beta � 19.01, std. beta � 0.68, p ≤ 0.001) and the effect of fruit breadth on fruit weight was significant (beta � 7.35, std. beta � 0.16, p < 0.05). e effects of height, leaf length, leaf breadth, and petiole length were not significant.

Scientifica
Qualitative fruit and vegetative traits showed greater variability in trunk branching, mature fruit colour, trunk colour, leaf colour, and trunk diameter whereas little diversity was seen in terms of fruit shape, fruit apex shape, fruit base shape, mid-rib colour, unripe fruit colour, and petiole colour. All the doum palm fruits sampled had shiny skin which was fused with the flesh. e mesocarp was orange in colour and fibrous in texture with a characteristic strong aroma (Figures 2(a), 2(g), 2(h), and 2(i)). All the fruits sampled from araka Nithi and Turkana were oblong shaped with truncate bases and apices (Table 6). e fruits from Kwale showed the most diverse traits with differing shapes, bases, and apices ( Figure 2(d)). e colour of unripe doum palm fruits was green in Kwale (Figure 2(c)), Tana River, araka Nithi, and partly in Turkana. A total of 43.3% of the fruits sampled from Turkana were maroon when unripe (Figure 2 Table 6). e colour of mature doum palm fruits differed across the four sampling sites with the majority of the fruits being reddish-brown. All the fruits sampled from Tana River were reddish-brown when ripe while the fruits sampled from araka Nithi were either brown (30%), orange-brown (63.3%), or orange (6.7%) as shown in Table 6.
All leaf petioles were stouter at the base than at the top with varying petiole colours and curved costa ( Table 6, Figure 3) e branching pattern observed in doum palm differed with some palms not branching at all. However, the majority of the palms had dichotomizing trunks. In Kwale, 46.7% of the sampled palms did not have any trunk branching. Twotrunk branching was common in all of the study sites with Tana River having the highest number of palm trees with twotrunk branching (80%) as shown in Table 6. On the other hand, Turkana and araka had 10% and 33.3%, respectively, of the sampled doum palm trees with more than 2-trunk branching ( Table 6). Trunk branching was either at the base (Figure 4(d)) or mid-section (Figures 4(b) and 4(c)).

Relationships between Discriminant Morphological
Descriptors.
e following discriminant models were derived: where LD1, LD2, and LD3 are discriminant functions, Ht is the height, LL is the leaf length, LB is the leaf breadth, PL is the petiole length, FL is the fruit length, FB is the fruit breadth, and FWGT is the fruit weight. LD1 explained 76.2% of the variation while LD2 and LD3 explained 15.03% and 8.8%, respectively. e second and third factors do not contribute much to discriminating between the groups. ere were samples within Kwale that did not show any overlap with any of the groups from araka, Turkana, and Tana River. ere was an overlap of samples between Turkana and Kwale and between Tana River and araka ( Figure 5).

Principal Component Analysis.
e first, second, and third components explained up to 59% of the variability in doum palm qualitative traits (Table 7). Component 1 explained 25% of the variability which was positively correlated with fruit shape, fruit apex, fruit base, trunk diameter, and pinnae density and negatively correlated with fruit colour when unripe and trunk branching (Figure 6(a)). e second component explained 18% of the variability which was positively correlated with leaf colour while the third component explained 15% of the variability correlated with trunk colour and trunk branching. e first, second, and third components explained 75% of the variability in the quantitative traits (Table 8). e first 7.59 ± 0.34a 7.64 ± 0.3a 6.33 ± 1.36b 6.56 ± 0.25b 6.32e − 12 * * * Fruit breadth (FB) 6.031 ± 0.227a 6.27 ± 0.228a 5.6 ± 0.798b 6.32 ± 0.40a 6.36e − 06 * * * Fruit weight (FWGT) 127.6 ± 11.24a 111.53 ± 9.3b 91.73 ± 34.94c 92.37 ± 9.11c 2.09e − 12 * * * Same letters within the row indicate no significant difference between the means while different letters indicate a significant difference between the means at α � 5% significance codes * � 0.01 and * * * � 0.000. e Games-Howell post hoc test was used for multiple comparison.  (Table 8). Component 1 was negatively correlated with petiole length and positively with all the fruit traits, leaf length, width, and tree height; that is, the bigger the fruit, the shorter the petiole. Component 2, on the other hand, was negatively correlated with fruit characteristics and positively correlated with vegetative data (Figure 6(b)).
Individual PCA based on qualitative and quantitative traits clustered the doum palm into three and two major clusters, respectively (Figures 7(a) and 7(b)). Five samples from Kwale clustered on their own using both qualitative and quantitative traits. e same samples also formed their own cluster after hierarchical clustering and are represented as morphotype 4 (Table 9, Figure 8).

Cluster Analysis.
e hierarchical clustering of quantitative traits of the 90 doum palm samples clustered the samples into 5 morphotypes (Table 9, Figure 8). Morphotype 1 had 77.3% of doum palm from Turkana. All the doum palms belonging to morphotype 4 were from Kwale. A total of 90.5% of the palms belonging to morphotype 5 were from araka. Morphotype 3 had representative palms from the four sampled regions of Kenya (Table 9). Some of the sampled palms from Kwale clustered with morphotypes 1, 3, and 5 indicating that these palms in Kwale are heterogeneous.

Identification of Elite Doum Palm.
e minimum, maximum, and mean of the morphological traits are shown in Table 9. Morphotype 5 had the tallest trees (mean � 14) with the biggest fruits (mean � 129.4). Members of this cluster include palms from araka (90.5%) and Kwale (9.5%).
araka samples that clustered together showed close homogeneity. Morphotypes 2 and 3 showed intermediate fruit sizes and traits. In addition, morphotype 2 had the longest leaves. Morphotype 4 had the shortest palms (mean � 3.96) with the smallest fruits (mean � 53.62) and the longest petioles (mean � 141.6). Morphotype 5 should be selected for improvement due to its fruit traits.

Discussion
High diversity was observed in quantitative traits within the individual doum palm trees sampled as well as among the different geographical sites sampled. ere was also a high variability in the fruit qualitative traits with Kwale having the most diverse fruits. e mature fruits varied from reddish-brown, brown, to orange. Doum palm fruits are mostly green when unripe but later mature to orange, brown, red, or yellow [7]. However, other studies have reported that the colour of mature fruits tends to be orange-brown in colour [27]. In this study, the fruit weight varied from 48.2 g to 148.8 g. Another study on the update of the African palms noted that fruit size seemed to be greater in areas with no water stress [7]. e smallsized fruits in Turkana could possibly be explained by phenotypic plasticity due to resource limitation [28]. However, fruit sizes in Kwale varied from very large (morphotype 5) to very small (morphotype 4), which could be a result of belonging to different varieties. According to Stauffer et al. [2]. Hyphaene compressa fruits are extremely polymorphic with green immature fruits which turn to orange-brown at maturity. is indicates that the fruits in Kwale are heterogeneous. e analysis of variance of all the quantitative traits evaluated in this study was significant. is is similar to a study done to assess the phenotypic and molecular diversity in H. thebaica, where the authors reported significant differences in all the phenotypic traits evaluated. at study further indicated that phenotypic and molecular analyses were complementary to each other in evaluating H. thebaica even though they gave different relationships among the samples tested [26]. e PCA clustering of the samples using both quantitative and qualitative traits indicates that two major clusters are  8 Scientifica formed with a subset of samples from Kwale clearly forming their own cluster from the rest of the accessions. ese five accessions seem to be distantly related to the others. ese samples also seem not to show any overlap with any samples from the other regions based on linear discriminant analysis. However, this cannot be used to delineate this group since advanced markers would be required to genotype it [13]. Most palm species have cylindrical, elongated, and unbranched stems.
H. compressa, on the other hand, has dichotomizing trunks which is a unique feature for Hyphaene where the basal stem is overbuilt to handle the later dichotomous branching [29].
However, 46.7% of the accessions from Kwale did not have dichotomizing trunks. is suggests variability at the Kenyan Coast especially in Kwale compared to the rest of the regions. e representation of the different samples in different clusters indicated that doum palm is genetically diverse. erefore, the different morphotypes identified in this study might not be directly influenced by their environment.
is is supported by Gower cluster analysis and projection of the samples on PCA which indicated a high level of heterogeneity. Cluster analysis in this study revealed phenotypic diversity and heterogeneity within  samples from the same region. For instance, accessions from Kwale were clustered in morphotypes 1, 3, and 5. Kwale also had some accessions forming a lone cluster, the morphotype 4. ey were, therefore, the most diverse with some accessions having very tall trunks with very large fruits while others were very short with small fruits. is heterogeneity was also observed among accessions from araka and Turkana with fruits from each region clustering into three different morphotypes. is heterogeneity within samples from the same region has been previously reported [13].
Identification of the different morphotypes in existence will help farmers and stakeholders to identify specific accessions for their own use, improvement, and conservation.
ere are no known improvement strategies for doum palm. Farmers in araka prefer a specific doum palm for weaving because the leaves are longer and wider than the other trees. Such information can help breeders select traits for improvement and mass production. e present study noted that the longest and widest leaves used for weaving were found in araka (90.5%) and Kwale (morphotype 5). e fact that Turkana which is the most arid region of all the  sampled areas had accessions that form long and wide leaves just like in araka, which receives a slightly higher amount of rainfall than Turkana, suggests that the difference in leaf lengths and breadths might not be influenced by the environment in H. compressa. It is indeed in these regions ( araka and Turkana) where massive weaving is done using doum palm leaves. If the morphotypes are superior for a specific trait that is desired by farmers, then it is only prudent that they are selected for improvement/breeding. In this study, morphotype 5 had the biggest fruits and can be selected for improvement.
H. compressa has costapalmate, fan-shaped leaves with entire margins, curved costa, and curved thorns on the leaf stalk [30]. e petiole length seemed to be an important trait in discriminating this palm. Petiole length was significantly longer in morphotype 4. ere was a negative correlation between the petiole length and fruit traits. at is, the bigger the fruits, the shorter the petiole and vice versa. Petioles are important resources for the local communities especially the nomadic-pastoralists of Kenya who use them for furniture and construction of houses [6]. Significantly, longer petioles on shorter trees are important for construction. Local communities would then prefer this accession for this purpose. e present study also reports that morphotype 4, in spite of having longer petioles, is a short accession.
is advantage will benefit the users by making the petioles easily accessible compared to taller palms. Morphological diversity is of benefit to preliminary doum palm genetic resource evaluation. However, the superior traits identified cannot adequately resolve the differences in diversity and should be confirmed if indeed they are genetically determined. e diversity of H.  compressa could be further investigated by genome-wide associations or other next-generation approaches. e main limitation of this study is the exclusion of male doum palms. Additionally, few descriptors were used in the present study as doum palm has no known standard descriptors. Future studies should evaluate additional descriptors so that the male doum palm diversity can also be determined.

Conclusion
is study assessed the variability in morphological traits of H. compressa and identified its morphotypes. e results show that there was variability in the fruit and vegetative traits of H. compressa per region. is study identified five morphotypes of H. compressa from the Northern, Coastal, and Eastern regions of Kenya. Different morphotypes showed superior traits for fruits, leaves, and petioles making it possible to select superior accessions for domestication and genetic improvement. Morphotype 5 should be considered for the improvement of leaf and fruit traits.

Data Availability
All data generated or analysed during this study are included in this published article.

Conflicts of Interest
e authors declare that there are no conflicts of interest.  12 Scientifica