On the necessity of combining ethnobotany and genetics to assess agrobiodiversity and its evolution in crops: A case study on date palms (Phoenix dactylifera L.) in Siwa Oasis, Egypt

Abstract Crop diversity is shaped by biological and social processes interacting at different spatiotemporal scales. Here, we combined population genetics and ethnobotany to investigate date palm (Phoenix dactylifera L.) diversity in Siwa Oasis, Egypt. Based on interviews with farmers and observation of practices in the field, we collected 149 date palms from Siwa Oasis and 27 uncultivated date palms from abandoned oases in the surrounding desert. Using genotyping data from 18 nuclear and plastid microsatellite loci, we confirmed that some named types each constitute a clonal line, that is, a true‐to‐type cultivar. We also found that others are collections of clonal lines, that is, ethnovarieties, or even unrelated samples, that is, local categories. This alters current assessments of agrobiodiversity, which are visibly underestimated, and uncovers the impact of low‐intensity, but highly effective, farming practices on biodiversity. These hardly observable practices, hypothesized by ethnographic survey and confirmed by genetic analysis, are enabled by the way Isiwans conceive and classify living beings in their oasis, which do not quite match the way biologists do: a classic disparity of etic versus. emic categorizations. In addition, we established that Siwa date palms represent a unique and highly diverse genetic cluster, rather than a subset of North African and Middle Eastern palm diversity. As previously shown, North African date palms display evidence of introgression by the wild relative Phoenix theophrasti, and we found that the uncultivated date palms from the abandoned oases share even more alleles with this species than cultivated palms in this region. The study of Siwa date palms could hence be a key to the understanding of date palm diversification in North Africa. Integration of ethnography and population genetics promoted the understanding of the interplay between diversity management in the oasis (short‐time scale), and the origins and dynamic of diversity through domestication and diversification (long‐time scale).


| INTRODUC TI ON
The date palm, Phoenix dactylifera L., is a major perennial crop of the hot and arid regions in the Middle East and North Africa (Barrow, 1998). Its sugar-rich fruit, the date, has been consumed for millennia (Tengberg, 2012) and has long been rooted in Berber/Amazigh and Arabic cultures. Phoenix dactylifera belongs to the Arecaceae family and, along with 12 or 13 other interfertile species, composes the genus Phoenix (Barrow, 1998). Date palms exist mostly as either cultivated or feral (i.e., uncultivated but derived from cultivated palms) domesticated forms (for review, Gros-Balthazard, Hazzouri, & Flowers, 2018). Only a few relictual populations of its wild progenitor are known today in Oman , even though Tuaregs of the Tassili n'Ajjer (Algeria) consider it to be wild in their gardens (Battesti, 2004). Phoenix dactylifera is dioecious (either male or female, and only females bear fruit). Today, thousands of female cultivars are reported, varying in fruit shape, color, texture, or taste, but also in their vegetative aerial architecture (Chao & Krueger, 2007). Without addressing the existence of male "cultivars" (there is little or no vegetative reproduction of identified and named males outside of research stations), male varieties are also locally identified, but are less studied. Nevertheless, local categorization of this diversity by farmers requires clarification, starting with the adequacy of the notion of "cultivar" or what it is supposed to be.
Phoeniciculture (date palm cultivation) involves a mix of clonal and sexual propagation. In palm gardens, palm trees bear a name.
The unquestioned and often implicit assumption is that each female date palm named type (for instance the famed Medjool or Khalas) is a cultivar, that is a clone, multiplied only through vegetative propagation. In order to obtain specifically female trees and to ascertain that fruits will be of the desired predictable quality, farmers mostly make use of the asexual reproduction abilities of this plant through offshoot multiplication. Indeed, sexual reproduction of date palms only leads in rare cases to progenies having equivalent or superior fruit qualities (4‰ according to Peyron, 2000), although these assessments remain subjective. For the oasis system to be efficient, despite the scarcity of water, irrigable lands, and manure, oasis communities plant and maintain 95%-99% of female palms (Battesti, 2005) instead of a natural 50:50 sex ratio (Chao & Krueger, 2007).
Opting for such an artificial sex ratio requires hand pollination, a practice already used in southern Mesopotamia during the 3rd millennium BC (Landsberger, 1967), as a substitute to natural wind pollination (Henderson, 1986). Hence, farmers almost entirely restrict sexual reproduction by seeds, with a few exceptions (e.g., in India, Newton et al., 2013). Accidental seedlings are, however, sometimes spared. A resulting male may be later used for pollination. If female, its fruits are sometimes harvested and, although rare, it can lead to a new selected and named line of clones, that is, a cultivar.
Another possibility, highlighted in our previous work in the oasis of Siwa, is to integrate this new genotype into an existing named type, because, from a local perspective, it is the very same variety, the same "form," or phenotype (Battesti, 2013;Battesti et al., 2018). This farming practice challenges the presumption of a named type being a true-to-type cultivar, that is, aggregating solely vegetatively propagated individuals. This may, in turn, lead to a misinterpretation and an inaccurate estimation of local agrobiodiversity.
Siwa is a desert oasis located in the Libyan Desert 300 km south of the Mediterranean coast and the closest city Marsa Matruh, and about 30 km east of the Libyan border ( Figure 1). The territory is occupied by salt lakes, the "sabkha," and by the cultivated area in the form of palm groves, highly concentrated around settlement areas ( Figure 1; Figure S1A). With roughly 200,000 to 250,000 palms (Battesti, 2013), dates are the main commercial crop in Siwa Oasis, closely followed by olive (Olea europaea L.). As in most oasis systems, date palm is used to feed the oasis inhabitants, but also for drinks, fodder, building materials (beams, hedges), crafts (baskets, various utensils), and daily uses (ropes, ties, brooms, furniture) (Battesti, 2005). The date palm is of primary importance, more generally, as the keystone of the oasis ecological system (microclimate effect, Riou, 1990). It is also the cornerstone of the local economy of exportation to large urban centers. The oases have a very relative autarchy and self-sufficiency, and their inhabitants export what they have in abundance: dates (Battesti, 2013;Battesti et al., 2018).
In Siwa Oasis, we previously described a complex system where named types are not necessarily cloned genotypes, but ethnovarieties or local categories (Battesti, 2013;Battesti et al., 2018). To synthesize our "objectification" of the local organization of the date palm agrobiodiversity system, we proposed these definitions: • Cultivar or true-to-type cultivar: a set of clonal individuals, that is, the association of a name and a single genotype reproduced vegetatively (asexually, by offshoot) by humans. Genotypes of a same cultivar are genetically identical, except in case of somatic mutations (McKey et al., 2010).
• Ethnovariety: a set of similar (according to local standards) lines of clonal individuals reproduced vegetatively (asexually, by offshoot) by humans deliberately under a single local name.
• Local category: a set of individuals sharing some characteristics (according to local standards, i.e., fruit color, harvesting season, usage, rusticity), typically found as seedlings, sometimes reproduced vegetatively (asexually, by offshoot) by humans but identified under a single local name. The two largest local categories are óṭem, the male date palms, and úšik, which includes all date palms resulting from sexual reproduction (not to be confused with the named types with the form úšik xxx, such as úšik n gubel).
The origins of this date palm heritage and Siwa Oasis altogether are lost in the mists of time. Siwa Oasis was well known during antiquity, in its Egyptian dynastic period, under the name of "Sexet-ȧm," the land of date palms (Duemichen, 1877), that included the current oasis but also the oases now abandoned in its periphery (Kuhlmann, 2013). In the classical period, the oasis was famous throughout the Mediterranean Basin for its oracle built in the 6th century BCE (Kuhlmann, 1988(Kuhlmann, , 2013Leclant, 1950), under the name of Amon Oasis (later Hellenized in Ammon). Alexander the Great was one of the most famous to consult this desert F I G U R E 1 Localization of Siwa Oasis and sampling strategy in Siwa region oracle, for him to confirm his divine ascendancy before his campaign of conquest in Persia. The dates of Siwa were mentioned as early as the 5th century BCE by Hellanicus of  in his Journey to the Oracle of Ammon (cited by Leclant, 1950, p. 248). The date palms of Siwa were then mentioned or even celebrated by Theophrastus (c. 371-c. 288 BCE) and later Pliny the Elder (23-79 CE) and , before a long silence and a return under the famous Arabic authors al-Bakrī (1040-1094), al-Idrīsī (1100-1165), and al-Maqrīzī (1364-1442. A few genetic studies involved date palms from Siwa (Abd El-Azeem et al., 2011;Abou Gabal et al., 2006;Gros-Balthazard et al., 2017;Hemeid et al., 2007). Nevertheless, not all named types of the oasis were studied, levels of categorization of date palm names were confused (sometimes mistaking úšik, seedlings, for a cultivar), farming practices were neglected, and only date palms cultivated in the current oasis were considered, ignoring the abandoned oases scattered in the desert. A proper assessment of date palm agrobiodiversity in Siwa region, and in a broader sense an understanding of its origin, hence, is still lacking.
Beyond the peculiar history of Siwa Oasis, the scenario of the beginning of date palm cultivation in Egypt in particular, and in North Africa in general, remains incomplete. Older evidence of exploitation is found around the Persian Gulf, while it seems that in North Africa, cultivation is more recent (Flowers et al., 2019;Tengberg, 2012). In Egypt, the date palm seems exploited or cultivated sporadically since the Old Kingdom (about 2,700-2,200 BCE), but phoeniciculture is only established since the New Kingdom, about 1,600-1,100 BCE (Tengberg & Newton, 2016).
Genetic analyses of the current date palm germplasm identified two differentiated genetic clusters in North Africa and the Middle East, with evidence of gene flows, especially in Egypt (Flowers et al., 2019;Gros-Balthazard et al., 2017;Hazzouri et al., 2015;Mathew et al., 2015;Zehdi-Azouzi et al., 2015). A recent study showed that the cultivated North African pool has mixed ancestry from Middle Eastern date palms and the Aegean endemic wild relative Phoenix theophrasti Greuter, a.k.a. the Cretan date palm (Flowers et al., 2019). Nevertheless, the geographic, chronological, and historical contexts of this introgression remain enigmatic.
The oasis of Siwa is located at the crossroads between Greek, Libyan, and Egyptian influences. It is on one of the rare passage points, a rare node in the network , between the east and west of the distribution area of the cultivated date palm (Figure 1). A deep understanding of date palm diversity in the region could therefore enlighten the diversification history of Phoenix dactylifera in North Africa.
In this paper, we are taking our previous work on Siwa date palms  to the next level, using a joint ethnographic study and genetic analysis. First, we greatly expanded the number of named types sampled in Siwa to further test whether named types are true-to-type cultivars or ethnovarieties or local categories. Secondly, we also increased our sampling of uncultivated individuals from the abandoned oases in the desert nearby Siwa (also known as "feral" in Battesti et al., 2018). This enabled a full assessment of local biodiversity and potential connections between the currently cultivated pool of the oasis and the uncultivated abandoned date palms. Lastly, we used genotyping data from more than 200 cultivated date palms sampled across the entire historical range of the species in order to locate Siwa diversity located within a wider germplasm diversity. By including Phoenix theophrasti, we can assess a possible gene flow in this particular population of date palms. From a broader perspective, our work aims at documenting the origins of phoeniciculture in Egypt and in North Africa in general.

| Ethnobotanical study and sample collection
We sampled 176 cultivated and uncultivated date palms in Siwa Oasis (Egypt) and surrounding desert ( Figure 1; Table 1; Table S1), of which 52 were included in our previous study . Those samples were collected in situ with the essential cooperation of the local farmers while conducting an ethnobotanical study ( Figure S1B). VB conducted social anthropological fieldwork between 2002 and 2017, including about six months dedicated to date palm categorization and naming (Battesti, 2013). While he occasionally conducted structured interviews or held focus group discussions and free listings of date palm given names, most of his data are derived from participant observation . This collection is composed of 109 accessions of date palms growing in about 46 private gardens that were deliberately chosen scattered throughout the current oasis. For each named type, we collected more than one date palm, when possible (some are rare), in order to test, with genetic data, whether they represent cloned accessions or not. We also sampled accidental seedlings growing in gardens (referred to as úšik #1) and on the border of gardens or palm groves (referred to as úšik #2). Further, 27 uncultivated date palms growing in abandoned oases in the desert were sampled (Figure 1). Those oases probably already existed during the Roman/Ptolemaic period, or at least some of them, and have been presumably abandoned since the 9th or 10th century CE (Battesti, 2013).
We additionally sampled nine Phoenix theophrasti Greuter in their native habitat in Crete and Turkey. Two accessions of Phoenix reclinata Jacq., collected in the botanical garden of the Villa Thuret in Antibes, France, and originally from sub-Saharan Africa, were included as outgroup population. For each accession, a few leaflets were collected and dried in the shade, with or without silica gel.

| DNA extraction and microsatellite genotyping
The collection was genotyped using 17 nuclear microsatellites and one chloroplastic minisatellite (Table 2), following the protocol of Zehdi-Azouzi et al. (2015). We crushed 40 mg of dried leaves in a fine powder using bead-mill homogenizer TissueLyser (Qiagen, Courtaboeuf, France). Total genomic DNA was extracted from leaf powder using DNeasy Plant MINI Kit (Qiagen, Courtaboeuf, France).
In addition to this newly generated genotyping dataset, we utilized P. dactylifera genotyping data from previous studies (Moussouni, Pintaud, Vigouroux, & Bouguedoura, 2017;Zango et al., 2017;Zehdi-Azouzi et al., 2015) (Table S1). Both these data and our new data relied on the same set of microsatellite markers and were generated by the same company (ADNid, Montpellier, France) with the same protocol, enabling a meta-analysis. These additional accessions are a good representation of the cultivated germplasm as their origin spans the historic date palm distribution, stretching from North Africa to the Middle East and Pakistan (Barrow, 1998).

| Genotyping data analysis
Statistical analysis was conducted with the R Statistical Programming Language (R Core Team, 2015), unless otherwise stated. To identify duplicated genotypes among the whole dataset, we performed an identity analysis using Cervus v3.0.7 (Kalinowski, Taper, & Marshall, 2007). For each pair of accessions, Cervus calculates the number of matching genotypes across the 17 nuclear loci. We used the same software for calculation of the polymorphic information content (PIC) for each locus. We estimated null allele frequencies using null.all function in the R package PopGenReport (Adamack & Gruber, 2014;Gruber & Adamack, 2015). For each locus, the number of alleles N A , and the observed (H O ) and expected heterozygosity (H S ) were estimated using the R package pegas (Paradis, 2010). The deviation from Hardy-Weinberg equilibrium was estimated using the function hw.test from the same package.

| Date palm agrobiodiversity in Siwa
For each named type of Siwa, we checked whether they are actual true-to-type cultivars or rather represent a group of more or less Phoenix reclinata Sub-Saharan Africa Wild relative 2 Note: The species and the country/region of origin are provided along with the status. For Siwa current oasis, the status is set as follows: "named types," for cultivated accessions for which we want to check the status; "úšik #1" and "úšik #2" which refer to accidental seedlings growing respectively in gardens (#1) or on the border of gardens or palm grove (#2). Were also analyzed uncultivated date palm accessions sampled in abandoned oases scattered in the desert surrounding the current Siwa Oasis (Figure 1). To investigate the extent of the diversity in the whole region, and not only in the oasis, we included uncultivated date palms from the abandoned oases in the following analyses. We generated a neighbor-joining tree based on Nei's genetic distance (Nei, 1972) using aboot function in poppr package (Kamvar, Tabima, & Grünwald, 2014) with 100 bootstrap replicates.

TA B L E
For each line of clones identified with these analyses, we kept a single accession for downstream analyses and performed a principal component analysis using function dudi.pca in the ade4 package (Dray & Dufour, 2007). Missing data were replaced by the mean allele frequencies using the function scaleGen from the R package adegenet (Jombart & Ahmed, 2011).

| Extent and partitioning of worldwide date palm diversity
To draw up a picture of the overall Phoenix population structure, Genotyped individuals were allocated to one to eight clusters K. All runs were performed using a model allowing admixture and correlated allele frequencies among populations (Falush, Stephens, & Pritchard, 2003). A 100,000-iteration burn-in period was followed by 1,000,000 MCMC steps. Ten independent runs were performed for each specified K, and the convergence of likelihood values was checked for each K. The optimal value of K was estimated using both the approach of Pritchard et al. (2000) based on the maximization of the log likelihood, and the approach of Evanno, Regnaut, and Goudet (2005) based on the rate of change in the log likelihood between successive K values (delta K).
For each population pair, we estimated a measure of differentiation (F ST ) using the Genepop software v. 4.7 (Rousset, 2008).

| RE SULTS
A total of 406 Phoenix spp. samples (Table 1; Table S1) genotyped across 17 nuclear microsatellites and one chloroplastic minisatellite (Table 2) were analyzed in the present study. We report new genotyping data for 176 cultivated and uncultivated date palms from the Egyptian oasis of Siwa and the surrounding abandoned oases ( Figure 1; Figure S1C), nine Phoenix theophrasti, and two Phoenix reclinata (Table S2). Additional genotyping data for 98 Middle Eastern/ Asian and 121 North African date palms were retrieved from previous studies (Moussouni et al., 2017;Zango et al., 2017;Zehdi-Azouzi et al., 2015).
Missing data across the full dataset were very limited with an average of 1.1%, and the mean null allele frequency was 6.8%, on average, across all 18 chloroplastic and nuclear loci ( Table 2). All loci were polymorphic, with four to 31 alleles and an average polymorphic information content of 0.625. All loci deviated significantly (p < .05) from Hardy-Weinberg equilibrium except Cuc3-ssr1. Except for some Siwa samples (see below), all accessions were unique, as no pair of accessions displayed 18 matches across the 18 chloroplastic and nuclear loci (Table S3).

| On the local named types of Siwa Oasis
The ethnobotanical field survey identified the existence of 18 named types in Siwa Oasis (Table 3). Based on ethnographic work, some were hypothesized as true-to-type cultivars, as they are supposed to only arise by planting offshoots, according to farmers' accounts. Meanwhile, the fieldwork already allowed us to suppose that some of those named types are not true-to-type cultivars, as farmers recognized the possibility that they partially or totally arise from seeds.
To investigate this question, we calculated genetic identity among accessions of each given named type, based on 17 nuclear microsatellites and one chloroplastic minisatellite (Table S3). The named types ṣaɛidi and alkak, the main cultivated and exported dates of Siwa, show proportion of identities of 97.6% and 100% at the nuclear level, respectively, and of 100% at the chloroplastic locus (Figures 2 and 3; Table 3). The named type alkak is locally known to come only from an offshoot, but different "qualities" can be distinguished, depending on growth conditions and age. The ṣaɛidi is the emblematic "palm reproduced by offshoot," which Isiwans always oppose to úšik, the seedlings. Although slightly different (on a single locus; Table S2), the 13 ṣaɛidi accessions cluster together (Figure 3), and we hypothesize that this subtle genetic variation is due to somatic mutations or genotyping error.
Hence, the two Siwa elite named types are indeed true-to-type cultivars.
TA B L E 3 Named type of Siwa, sampling effort, and genetic identity Note: For each named type, the local name of the dates is given, followed by the possible name of the palm [in brackets]. The identity is the average proportion of matching genotypes among accessions of a given named type and across the 17 nuclear microsatellites (Nuc Identity) and the chloroplastic minisatellite (Cp Identity).
*one accession (3,377) was identified as an identification error of the informer. When included, average identity across this set of accession is 79.41%.
Have been also analyzed as true-to-type cultivars three other named types: aɣzāl, taṭṭagt, and ḥalu en ɣanem. The famous aɣzālrare but valued by the Isiwan as a tonic for lack of energy (Fakhry, 1990, p. 27) and for its aphrodisiac properties (for men)-is here represented by four accessions, three of which being identical and one showing a single allelic difference that we thus hypothesized to be a somatic mutation or genotyping error ( Figure 3; Table 3; Tables   S2-S3). The named type taṭṭagt refers to a very soft date rich in water, whose name is taken from a maturation stage in the Siwa language: One half of the fruit is brown and ripe, and the other half is yellow and immature. Because it keeps very poorly, it is always eaten on the spot, but is highly appreciated locally. It was studied using eight accessions, of which a single one (3377) appears to be very different from the other genetically uniform accessions (Figures 2,3; Table 3; Tables S2-S3). The informant for this accession was a young farmer, and we presume that he was wrong about his identification.
Hence, we believe that taṭṭagt is a true-to-type cultivar, even if different qualities of taṭṭagt are reported by our ethnography. The fifth confirmed cultivar is ḥalu en ɣanem (100% identity over the three studied accessions in both nuclear and chloroplastic loci, Figure 3; Table 3; Tables S2-S3): relatively few farmers in Siwa know of its existence, but its long dates are highly appreciated by connoisseurs, so soft and sweet that the seed remains attached to the bunch when the fruit is plucked.
Further, we can confirm some named types as ethnovarieties: ɣrom aɣzāl and úšik niqbel indeed group some individuals having the same genotypes (clones) but also some that are isolated in different clades ( Figure 3; Table 3). The first is said to resemble aɣzāl and the second ṣaɛidi, but in both cases of lower quality.
An intermediate case is alkak wen žemb: The six samples are unrelated except two which are from the same clone, but coming from the same garden a few meters away. Etymologically, alkak wen žemb is a "relegated," "put aside," that is, a second class alkak, but it has always been, without much ambiguity, referred to as a cultivar and cannot come from a seed, according to farmers. Under the names lekrawmet, amenzu (etymologically "early" dates, Laoust, 1932), or ɣrom ṣaɛid, we only identified unrelated individuals (Figures 2 and 3; Table 3). These are clearly not true-to-type cultivars despite them being explicitly thought by local farmers as never coming from a seedling; apparently, they often have been, although the occurrences may be too distant for local memory. They are thus interpreted as ethnovarieties.
Decisions are more difficult to reach in the following cases: úšik azzugaɣ, úšik amaɣzuz, tažubart, and tazuwaɣt. The last two are only F I G U R E 2 Intra-named type variation and structure in Siwa date palms. (a) Relatedness among the 149 cultivated date palms sampled in the current Siwa Oasis and genotyped across 17 nuclear microsatellite loci. Intra-named type variation is expected to be zero or near zero (red), in case of somatic mutations, while for ethnovarieties and local categories, we expect a higher intra-named type variation (yellow known to a handful of farmers, and we have only one sample of each, preventing an intra-type identity analysis. Tažubart is said to look like ṣaɛidi and tazuwaɣt like úšik ezzuwaɣ (zuwaɣ). The first two are more widely known, but two amaɣzuz (etymologically "late" dates, Laoust, 1932) samples do not allow us to reach a conclusion. Regarding úšik azzugaɣ, we have two closely related palms from unrelated gardens and farmers, but also two isolated samples. Local farmers say úšik azzugaɣ are not (necessarily) reproduced by offshoot and can be from a seedling (úšik). They are sometimes described as "all úšik (seedlings) that yield red dates," "azzugaɣ" meaning clearly "red" in jlan en Isiwan, the local Amazigh language. By combining genetic information and ethnographic data, these four types thus all seem to be possible "local categories." We are quite certain that the following named types are neither cultivars nor ethnovarieties: obviously, and as expected, the inclusive local category of úšik in general (all female seedlings), but also úšik ezzuwaɣ (úšik selected/cultivated with good reddish/dark dates), kaɛibī (úšik selected/cultivated with good dry dates), and úšik nekwayes (all úšik selected/cultivated with dates suitable for human consumption). Indeed, the field survey showed they refer to individuals sharing fruit characteristics and/or originate from seedlings, and the genetic analysis confirmed that samples are unrelated (Figures 2 and 3; Table 3). Four úšik nevertheless present the same microsatellite profiles as the Siwa widespread cultivar ṣaɛidi ( Figure 3) meaning that these were actually ṣaɛidi but thought wrongly, by the sampler, to be accidental seedlings (because they were found in abandoned areas of the palm grove).
Following this identity analysis, we removed 58 genotypes from the initial Siwa dataset in order to only keep 128 unique genotypes in downstream analyses.

| Structure of date palm diversity within Siwa region
To study the structure of the diversity of date palms in the region of Siwa, we studied named types and seedlings (úšik) from the oasis, but also uncultivated date palms from abandoned oases scattered across the desert surrounding Siwa (Figure 1). We found that the genetic diversity is mainly distributed between oasis samples versus samples from outside the oasis (i.e., the ancient abandoned oases in the desert), as the principal component (PC) 1 mostly draws apart those two types of date palms ( Figure 2b). Identically, the NJ tree shows that the uncultivated date palms from the desert roughly form a distinct clade (Figure 3). Some accessions from the current oasis can nevertheless be found in the cluster formed by the desert uncultivated palms, mostly seedlings (úšik #1 and #2), but also named types, such as taṭṭagt. The second PC opposes the two main cultivars, namely alkak and ṣaɛidi. Alkak, and a few other accessions, appears isolated from the other accessions of the region, including its relegated counterpart, alkak wen žemb (Figure 2b). The seedlings (úšik) do not form a distinct population. Here, we differentiated the úšik from the gardens that may be tended, depending on their use, irrigated, as other date palms of the garden, and pollinated (úšik #1) from those that are not (úšik #2). For farmers, there is no difference and we could not separate them based on their microsatellite profiles ( Figure 2b).

| Comparing the diversity of date palm in Siwa and worldwide
We

| Partitioning and extent of the diversity in P. dactylifera and P. theophrasti
The principal components (PCs) 1 and 2 both separate Phoenix theophrasti from Phoenix dactylifera accessions (Figure 4a), accounting for the large differentiation observed between these two species (F ST = 0.32; Table S4). Additional PCs do not provide notable results  (Table 4).
Interestingly, Siwa date palms overlap only slightly with the western and eastern clusters in the PCA (Figure 4a,b). Similarly, in the Bayesian clustering analysis, Siwa accessions form a distinct cluster from K = 3 ( Figure 4c). With an additional K, four clusters, corresponding mostly to P. theophrasti, eastern cultivars, western cultivars (excluding Siwa samples), and Siwa date palms, can be identified (Figure 4c; Table S5).
Nevertheless, over-representation of related accessions may lead to spurious clustering in both PCA and structure analyses. In this study, we included a very large number of accessions from Siwa compared to the number of date palms from the Middle East and North Africa.
Hence, we ran two other PCAs (with and without P. theophrasti) in which we randomly sampled ten accessions from Siwa Oasis and five accessions from the surrounding desert ( Figure S5). We found that the accessions from the current oasis are found within the diversity  Figure S5).  Table S5). The only Siwa accession with mostly eastern ancestry is an úšik ezzuwaɣ (3363), collected in the heart of the old palm grove (in Jubba annēzi). Noteworthy, Siwa accessions are as diverse as eastern accessions in terms of allelic richness (calculated using equal sample size; Figure 4d) and even more diverse in terms of expected heterozygosity (Table 4).
On the one hand, the inferred ancestry constituting the Siwa cluster ( Figure 4c, in black) is not only found in Siwa. Although it is almost absent in eastern cultivars (0.93% on average), it is substantial in western date palms (11.85% on average). It is prominent in the two Libyan accessions (Figure 4c) Table S5). They are also closer, and especially the uncultivated ones, to this wild Cretan relative on PC1 than western date palms are (Figure 4c, from K = 4) and share more alleles (35.58%) with this species than western do (32.49%).

| Shared ancestry between P. dactylifera and P. theophrasti
Finally, Structure results show that there is also potential gene flow in the other direction, from Phoenix dactylifera to P. theophrasti.
Indeed, two samples of the Cretan date palm display ancestry attributed to the date palm (Figure 4c).

| Partitioning and extent of the diversity within Siwa region
Siwa accessions can be further split into two distinct clusters, as seen previously (Figure 2b), and highlighted in both the Bayesian clustering (Figure 4c, from K = 5) and the PCAs (Figure 4a,b). Diversity estimates were calculated on western (or North African date palms) including or not Siwa accessions. Siwa population was split into two populations to oppose accessions sampled in the oasis (both named types and seedlings) with those sampled in abandoned oases in the surrounding desert. Further, accessions sampled in the oasis were further split into three populations: the named types and the seedlings collected in or nearby the gardens (úšik #1 and #2, respectively).

TA B L E 4 Diversity estimates calculated in
These two clusters mostly fit the predefined Siwa populations, with a cluster comprising mostly accessions from the current oasis ( Figure 4c, in black at K = 5), while the second comprises mostly uncultivated desert date palms from the abandoned oases (in orange at K = 5  (Table S4).
Remarkably, the desert uncultivated accessions appear even less differentiated from P. theophrasti (F ST = 0.28) than the accessions sampled in the oasis (F ST = 0.35, Figure 4a,c,  and 47.57% shared with the named types cluster, respectively (in black at K = 5, Figure 4c and Figure S8).

| Diversity at the chloroplast minisatellite in Siwa and worldwide
We identified five alleles at the chloroplastic locus psbZ-trnfM (Table S2). Two of them were restricted to Phoenix reclinata and one of them to Phoenix theophrasti ( Figure S9). In Phoenix dactylifera, we Two accessions of Phoenix theophrasti display the date palm western chlorotype ( Figure S9). These accessions also have ancestry that is mostly attributed to date palm in the Structure analysis (Figure 4c).
This could indicate potential gene flow from the date palm to this wild relative, as previously reported (Flowers et al., 2019).
In Siwa, occidental and oriental chlorotypes are found in almost equal proportion (47.65% and 52.34%, respectively). Among the named types of the oasis though, the oriental chlorotype is slightly predominant (59.70%, Figure S9), while we found the opposite pattern in the Nilotic Egyptian cultivars (42.86%). As for the uncultivated date palms from the abandoned oases of Siwa desert region, we found that they display mostly the occidental chlorotype (74.1%).

| D ISCUSS I ON
In this paper, we studied date palms from Siwa using a combined molecular population genetic and ethnographic approach in order to (a) better understand folk categorization in conjunction with local agrobiodiversity and (b) infer the origins and the dynamic of the diversity found in this oasis and around, by comparing it to the worldwide date palm germplasm.

| Differentiating cultivar, ethnovariety, and local category
Our genetic analysis of intra-named type variability confirmed the existence of true-to-type cultivars as we found, for some named types such as the elite alkak, a 100% genetic identity across the 18 nuclear and chloroplastic loci. Nevertheless, we also pointed out the lack of genetic uniformity within other named types. Previous stud- is the result of a cultivation practice that we proposed before, that is, the incorporation of new clonal lines of seedlings under an existing name . In this study, we further substantiate this statement, using a larger number of named types and, for each, more samples. We also highlight that a significant number of named types, beyond the very inclusive úšik and males, are neither true-to-type cultivars nor even ethnovarieties, but of the "local category" order.
This peculiar cultivation practice we observe in Siwa could also exist in other palm groves and explain previously described cases of intra-varietal genetic variation, for instance, for Medjool/majhūl, the famous Moroccan variety, which is not a "genetically uniform" clone (Elhoumaizi et al., 2006, p. 403

| Determining the number of named types
Surveying the precise number of named types that refer to distinct local date palms in an oasis is an already complex operation, as in Siwa (Battesti, 2013) where given lists of named types can refer to varying degrees of inclusiveness (in no hierarchical order of exclusive taxa, unlike what Brent Berlin's ethnobiological theory of taxonomic categories implies: Berlin, Breedlove, & Raven, 1973;Berlin, Breedlove, & Raven, 1974). However, in a system with cultivation methods based on massive vegetative reproduction, the identification of the accurate number of named types should have been sufficient to correctly estimate the number of genotypes. Until now, therefore, and at best, the overestimation of the agrobiodiversity of the date palm at the regional level had been considered, taking into account the phenomenon of synonymy (a cultivar takes another name by changing oasis), somewhat offset by homonymy (the same name is used in different oases to designate a different cultivar) (Battesti, 2013;Battesti et al., 2018). In a single oasis, Siwa, one of the difficulties of the field survey resides in massive local synonymy, including for plant names and in particular for date palm varieties, as already mentioned (Battesti, 2013;Battesti et al., 2018). The first hypothesis addressing such a synonymy is that a landlocked Berberspeaking community should promote its export products by adding Arabic trade names. Another explanation is the co-presence on the same territory of Arabic-speaking minorities (sedentary Bedouins, especially Awlad 'Alī, Battesti et al., 2018) who use these Arabic names; for example, rather than the names tasutet/ alkak/ úšik, they will systematically use (or even only know) the names ṣaɛidi/ freḥī/ azzawī to denote the same palms. Besides, we observed that local practices of categorization and integration of seedlings can lead, as explained above, to a massive phenomenon of underestimation of agrobiodiversity for an uninitiated external observer and all assessments have stumbled over this obstacle (even the recent Atlas of date palm in Egypt, El-Sharabasy & Rizk, 2019).
In Siwa, our ethnobotanical survey revealed the existence of 15 to 20 named types (Battesti, 2013;Battesti et al., 2018). The exact number is difficult to pinpoint as the field survey reveals that farmers offer local, more or less shared, quality distinctions, even for dates that we depicted as true-to-type cultivars (alkak and taṭṭagt, for example). Therein, a higher alkak is called "alkak n amles," meaning smooth or wrinkle-free alkak dates, and a lower alkak with smaller dates, and three times cheaper than the upper one, is called "alkak nifuɣen." Fifteen to 20 named types in Siwa is not a lot compared to the number of named types described in other oases. Assessing the agrobiodiversity of date palm is a difficult exercise and carried out using noncomparable competing methodologies (Jaradat, 2016).
As a result, it is difficult to establish the terms of comparison of agrobiodiversity between oases. We can provide some comparative data. For instance, in the Jerid region of Tunisia (about twice the area of old palm groves compared to Siwa), there is a collective collection of more than 220 varieties (Battesti, 2015;Rhouma, 2005Rhouma, , 1994; some of them may very well be ethnovarieties or local categories, but this hypothesis still has to be checked in situ.
Hence, we could believe that the agrobiodiversity is relatively low in Siwa, if considering a conventional system where named types are thought to be genotypes. Nevertheless, with this combined ethnobotanic/genetic approach, we showed that this assumption is wrong, and that, despite a low number of named types compared to other oases, the number of genotypes in Siwa is high and so is the overall diversity.

| Etic versus. emic categorizations
Although we confirm the validity of the notions "cultivar," "ethnovariety," and "local category," we also underline their limitations:  (Battesti, 2013). This difficulty is a classic conflict for anthropologists, also known as a disparity of etic versus. emic categorizations (Olivier de Sardan, 1998).
Deeply characteristic of human societies, categorization processes are also at the heart of both mundane and scientific thoughts and practices. Naturally, our aim is not to use our paradigm (genetics) to try to evaluate the paradigm of indigenous local knowledge (Roué & Nakashima, 2018), but to translate the latter: For social sciences, there is no such thing as one science but several incommensurable sciences, and multiple modes of existence coexist responding to various forms of veridiction (Latour, 2013). In the emic version (the local point of view), the differentiation between cultivars/ethnovarieties (which farmers assimilate) and local categories is clearly thought out.
However, it is not so clearly expressed: Farmers can list all named types at the same level even if they do not refer to the same object classes. It is worth bearing in mind that for local farmers, reproducing by offshoot is the rule. Naming/identifying a seedling date palm after a known cultivar (ethnovariety process) is a possible (and appropriate) practice but of an exceptional occurrence on a human life scale. Hence, the difficulty for local farmers to know whether, for instance, all the úšik niqbel are cultivars (which they tend to present as such) or ethnovarieties. Although they do not remember seeing an úšik n gubel coming from a seed, this has apparently been the case, perhaps several decades or generations ago (a date palm easily outlives a human being). The question is of little relevance to them, since the palm trees in question behave and have/produce the same form. But, the other process, categorizing a date palm from a seedling, with reddish/dark dates for instance, as having a valuable production and naming/qualifying it, as úšik ezzuwaɣ for instance, is a more common experience (local category process). Hence, for local categories, we have to reverse the point of view. For example, some date palms do not bear reddish/dark dates because they belong to the variety úšik ezzuwaɣ or zuwaɣ, but the fact that their dates are reddish/dark qualifies them as úšik ezzuwaɣ or zuwaɣ. As scientists, It is therefore preferable to consider the distance between the ethnovariety and the local category as a continuum ( Figure 5).
Indeed, among the 18 named types we analyzed here, we found a gradient of intra-cultivar relatedness: Some named types corresponded actually to unique genotypes while some did not, as evidenced in our previous study . It is worth noting that an expanded sampling could show that what we believe today to be a true-to-type cultivar is in fact an ethnovariety, by discovering a new line of clone under the same name. Further, it could allow to identify lines of clones for the inconclusive alkak wen žemb, lekrawmet, amenzu, or ɣrom ṣaɛid to confirm them as ethnovarieties. To formally confirm that a given name is either a true-to-type cultivar, an ethnovariety, or a local category, it would require that the 200,000 to 250,000 date palms of Siwa to be all genotyped. This is obviously not conceivable.

| Usefulness of the folk categorization
What are the explanations for such a complex local system for categorizing date palms in Siwa? They might be both cognitive and agricultural. First, all "classifications are functionally linked to the effective storage, retrieval and communication of large quantities of information relating to the animal and plant worlds" (Meilleur, 1987, p. 9-10) and must be easily mobilized to guide action and communicate (knowledge, experience, etc.). This system must therefore be shared. The utilitarian nature of folk biological classifications had already been discussed by Eugene Hunn (1982).
Date palm cultivation in Siwa is largely dominated by a few "elite" types (probably for centuries, an integration into the Saharan trading network, Battesti, 2018Battesti, , 2013. Here, we found that they are apparently true-to-type cultivars, despite their prevalence and therefore the mechanical possibility of becoming an ethnovariety.
The ethnovariety and local category system make it possible to "put in order" the profusion of all the other date palms, less commercially valued, while not multiplying the denominations for the same characteristics.
Secondly, a peculiar system allows for a peculiar action on the world, in this case, a fairly flexible management of agrobiodiversity. To date, cassava is the only clonal crop for which in-depth information exists on how mixed clonal/sexual systems work (ibid.). We propose here another crop model, quite different as the date palm is a perennial plant that lives decades or even over a century (Chao & Krueger, 2007). This critically impacts farming practices, especially those inherent to propagation (by seed or offshoot, sexual or clonal). However, it is very likely that maintenance of a mixed clonal/sexual system is a local key strategy for managing and preserving an agrobiodiversity of date palms in Siwa. And this is made possible for Siwa farmers because they developed a classification system that enables them to do so.

| A unique and high genetic diversity in Siwa date palms
Siwa date palms form a partially distinct cluster from western accessions, rather than a subset of the diversity found in North Africa  (Terral et al., 2012). Nevertheless, both a limited sampling in Siwa and an incomprehension/ignorance of the cultivation practice had hampered those previous researches to pinpoint the singularity and high diversity of Siwa date palms. We note that although there are existing studies based on microsatellite data that provide diversity estimates for date palm populations from various regions (i.e., Chaluvadi et al., 2019;Zehdi-Azouzi et al., 2015), the use of distinct loci and/or different sample size prevent comparisons between our results and that of these studies. in turn account for a higher diversity than expected at first sight.
It is thus possible that in Siwa, more than in other oases, sexual reproduction being more common, there is a more limited loss of diversity through time. In addition, the position of Siwa, at the crossroads between western Africa and the Middle East, promotes the creation of novel hybrids, and a particularly high outcrossing rate, where many individuals contributing to the next generations may have allowed the retention of this peculiar diversity. These hypothesis remains to be tested by ethnographic survey in other oases in order to test whether sexual reproduction is more prevalent in Siwa than elsewhere. This leads us to the third hypothesis: The high diversity found in Siwa could reflect, rather than a reality, a sampling bias. Indeed, our sampling, considering practices and knowledge of farmers, may have led us to sample an appropriate representation of the existing diversity of Siwa. On the opposite, the non-Siwa date palms included here have been sampled without such an ethnobotanical survey and may in turn only be a poor representation of the actual diversity. Hence, if such a sampling methodology were applied everywhere, we may very well discover more diversity elsewhere too.
We identified that date palms in the current oasis of Siwa and in the ancient abandoned palm groves in the surrounding desert constitute two subpopulations. Changes in allele frequencies in the uncultivated palms may have been driven by a relaxation of human-induced selective constraints, natural selection, or genetic drift, following the abandonment of these groves. Additionally, while we identified potential gene flow from eastern accessions at both chloroplastic and nuclear level in the date palms from the current oasis, accessions from the abandoned palm groves may have received less diversity from this population. Uncultivated palms from the desert share more ancestry with P. theophrasti than both western and Siwa Oasis date palms. Locally, these desert date palms are considered as úšik but are also specifically called igizzã (sing. agzzu).
Our ethnographic survey revealed that they do not seem to be used by the farmers as a reservoir of diversity. Some of these abandoned oases could nevertheless constitute casual date harvest sites, and hence, seeds from the harvested fruits could potentially end up as úšik in the cultivated palm grove. This could explain the intermediate genetic profiles of úšik #1 and #2: Their diversity seems in between that found in the current oasis and the abandoned oases.
Siwa date palm origins are unknown, but their presence in Siwa dates back at least to the 5th century (mentioned by Hellanicus of Mytilene, see above); the oasis being then an independent state related to the Libyan world, well established as an essential trade and religious hub with Libyan, Egyptian, and Hellenistic influences (Kuhlmann, 1999). Declining or abandoned at the end of the first millennium, the oasis was probably recolonized in the 11th or 12th century (a priori by Amazighs from Libya, then Arabs, see Battesti, 2013 (Flowers et al., 2019). This has been interpreted as an asymmetry in the direction of the interspecific cross, where gene flows from P. theophrasti were pollen-mediated (ibid.), but the possibility of a plastome-genome incompatibility (Greiner, Sobanski, & Bock, 2015) cannot be ruled out. Going beyond these two occidental and oriental chlorotypes, Mohamoud et al. (2019) identified four chlorotypes, using genome-wide genotyping data.
Our microsatellite data do not allow us to know which of these four chlorotypes are displayed by our samples, but this represents an exciting direction for future genomic studies.
Although a major event in the diffusion and diversification of cultivated date palms to North Africa, the introgressive event by P. theophrasti remains puzzling, especially in terms of localization (Flowers et al., 2019). Indeed, today, this species is not distributed in North Africa, but is found in the Aegean region, mostly in Crete (Barrow, 1998), but also in coastal Turkey (for review, Boydak, 2019). One hypothesis explaining the introgression by P. theophrasti is that it had a much wider distribution in the past, encompassing North Africa and/or the Levantine region (Flowers et al., 2019 (Colin, 1997;Kuhlmann, 2013), and Greek or even Cretan workers employed in the construction of the temples have left engraved inscriptions of their passage (Aldumairy, 2005;Kuhlmann, 2013).
Those persistent connections could have set the opportunity of plant material exchanges, including of Phoenix.
We note that date palms from the abandoned oases of Siwa show an even closer affinity to P. theophrasti than do the palms from the current oasis. One speculative hypothesis for this pattern is that those palm groves were abandoned shortly after introgres-

| The Siwa region at the diffusion crossroads of the domestic date palm?
Siwa was in early contact with Egypt and the Libyan Sahara, a position that was essential for the diffusion of the date palm. In Egypt, unequivocal evidence for the economic and cultural importance of dates and for the local cultivation of the date palm for its fruits begins with the New Kingdom in the mid-2nd millennium BCE (Tengberg & Newton, 2016). However, there is now earlier evidence for the presence of date palm products in Egypt, but whether the dates were produced locally, in the Nile delta, or imported remains unknown. For instance, administrative texts document that two regions (nomes) located on the Mediterranean coast provided dates, of unknown origin, to workers under the reign of Khufu (around 2,600 BCE, Tallet, 2017), while a very small amount of date palm remains were found at Giza (2,700-2,100 BCE) (Malleson, 2016;Malleson & Miracle, 2018) and at the site where the texts were found (Wadi al-Jarf, a Red Sea harbor site, Newton, unpublished data).
The extension of the route leading west from Siwa, via Jaghbūb, Awjila, to Fezzan, and plausibly onwards, appears to date to the late second or early first millennium BCE (Mattingly, 2017, p. 8 (Pelling, 2005, p. 401). These are some of the earliest evidence for oasis agriculture in North Africa ( Van der, Veen, & Westley, 2010). Siwa, the realm of the "two deserts" (Kuhlmann, 2013), could have been in the first millennium an important, even unavoidable node in the date palm exchange network between Egypt and Libya.

| A missing contributor to the modern date palm genetic makeup?
North African date palms display a higher diversity than Middle Eastern ones Hazzouri et al., 2015). Flowers et al. (2019) showed that the more the genomic regions are introgressed by P. theophrasti or a P. theophrasti-like population, the higher their diversity is, concluding that the excess of diversity found in North African date palms can be explained by the introgressive event(s) by this wild relative. Our results corroborate these findings, as we found that the fraction of alleles uniquely found in North Africa is reduced when P. theophrasti is included in the analysis ( Figure S4).
Nevertheless, both Flowers et al. (2019) and our data indicate that there could still be an unknown population that contributed to the modern date palm germplasm. Indeed, in genomic regions with no or limited P. theophrasti introgression, the diversity in North African date palms remains higher than that in Middle Eastern date palms (Flowers et al., 2019). Further, of the two main chlorotypes described in date palms , we ignore the origins of the so-called occidental one, prevalent in western date palms.
Indeed, the few relictual wild date palm populations (in Oman) bear the so-called oriental chlorotype (Gros-Balthazard et al., 2017), and P. theophrasti an even more divergent type .
Here, we could also identify diversity that is unique to the North theophrasti, but also from a yet unidentified North African Phoenix population.
So far, there is no evidence of date palms or other Phoenix populations in North Africa before the establishment of oasis agriculture.
In contemporary times, wild date palm populations are exclusively known in the mountainous regions of Oman . The natural distribution of P. dactylifera (before its domestication and diffusion) is unknown (Barrow, 1998). It likely covered at least the Middle East and may have shrunk, in connection to climate change (Collins et al., 2017). In Egypt, it is possible that the date palm was growing wild or that it was grown for ornamental purposes (more precisely, not mainly for its fruit), perhaps from the 4th until the mid-2nd millennium BCE.
Siwa Oasis and the Libyan region in general could be a region of prime importance for the understanding of date palm origins in North Africa and the history of gene flows with the Cretan date palm Phoenix theophrasti. More precisely, not only date palms from the current oasis could shed light on the history of this species, but also the uncultivated palms from the abandoned oases. So far, research has focused mainly on cultivated germplasm, neglecting uncultivated date palms, even if local oasis communities have a use for them. Here, we demonstrate that they can be of interest as, not only they can inform on past history, but they may represent untapped reservoir of diversity for future breeding programs.

| Articulating the scales of ethnography and domestication over the long term
This study demonstrates that the combination of observation angles and methodologies on the cultivation of a crop, in particular perennial with a clonal/sexual reproduction, is essential. Finally, this interdisciplinary framework could be useful not only to date palm scientists, but also to scholars studying agrobiodiversity and its evolution in many different plant contexts.

ACK N OWLED G EM ENTS
The present study was funded by BioDivMeX program (Working-