Diversity and conservation of traditional African vegetables: Priorities for action

Traditional African vegetables have high potential to contribute to healthy diets and climate resilience in sub‐Saharan African food systems. However, their genetic resources are likely at threat because they are underutilized and under the radar of agricultural research. This paper aims to contribute to a conservation agenda for traditional African vegetables by examining the geographical diversity and conservation status of these species.


| INTRODUC TI ON
Sub-Saharan Africa (SSA) has some of the areas of highest hidden hunger in the world, with these acute nutritional deficiencies exacerbated by climate change (von Grebmer et al., 2014).
Governments have highlighted the important role of food production diversification to promote a wider range of healthier foods and support more sustainable production systems (Covic & Hendriks, 2016;von Grebmer et al., 2014). In a revised research and development agenda, traditional African vegetables have high potential to contribute to food production diversification and healthier diets in SSA. These vegetables are naturalized (introduced long ago and now accepted as "traditional") or indigenous to SSA and are adapted to local food and farm systems after generations of interactions with humans and the environment. Many are highly nutritious (Odhav et al., 2007;Yang & Keding, 2009) and are easy to incorporate into farm systems because they require limited space and fit within short rotations (Schreinemachers et al., 2018). Traditional African vegetables therefore potentially help to diversify farm systems and diets with nutritious foods and generate more climate-resilient food production (van Zonneveld, Turmel, et al., 2020). So far, however, the majority of these species have not been considered in climate smart agriculture strategies (e.g. Pironon et al., 2019;Rippke et al., 2016).
Current rural-to-urban migrations in African countries provide opportunities for new urban markets and peri-urban processing of these vegetables, and this may particularly benefit the livelihoods of women (Dinssa et al., 2016;Weinberger & Pichop, 2009). On the other hand, these migrations result in reduced human populations in rural areas and drive the loss of traditional knowledge about these species, as well as the loss of local landraces and populations because of reduced use (Keller et al., 2005;Pilling et al., 2020). These losses are exacerbated by limited research. This means relatively little is known about the origins and distributions of many of these vegetables, and how genetic resources-that can be exploited in crop promotion-are structured within their distribution ranges. It is clear therefore that conserving and documenting the diversity and traditional knowledge of these vegetables is important, before these resources are lost. This is especially relevant within biodiversity-rich food production systems that still exist in Africa but are now threatened by trends to food production homogenization (Dawson, Park, et al., 2019;Khoury et al., 2014).
In this study, we provide insights into the geographical patterns in the diversity and origin of 126 traditional African vegetables in SSA. These vegetables were selected after a review of five renowned species lists of important traditional African vegetables (African Orphan Crops Consortium, 2019; Dinssa et al., 2016;Grubben & Denton, 2004;Guarino, 1995;Maundu et al., 2009).
Second, we prioritize actions to safeguard traditional African vegetables following six conservation principles (Game et al., 2013): (a) Delineation of priorities: we aim to support decision-making in resource allocation to safeguard the genetic resources of traditional African vegetables as a basis for food and nutrition security in SSA; (b) Clear objective functions: our analysis contributes to a conservation agenda to safeguard SSA genetic resources of the 126 selected traditional African vegetables by 2030. We propose ex situ and in situ conservation of a minimum of 50 populations per species (following Brown & Marshall, 1995), while at the same time maintaining diverse ex situ collections of at least 200 genebank accessions and ideally 1,000 or more accessions for those species with high cultivation and breeding potential; (c) Prioritized actions: countries and conservation actions will be prioritized with a focus on multiple-crop germplasm collecting missions for ex situ conservation; (d) Scoring rules: a clear rational is provided for the selection of the 126 species, and we apply a set of indicators to assess and compare the ex situ and in situ conservation status of each species and to prioritize countries for conservation actions; (e) Transparency: scoring is explained in detail and the R coding we use can be requested for verification of our results or to apply to other geographical regions or crop groups; and (f) Risks need to be managed by establishing standard operating procedures to ensure the safety of people when they implement conservation actions, and by establishing germplasm backups, such as at the Global Seed Vault in Svalbard.

| Selection of traditional African vegetables with high potential
We consulted five key species lists on African vegetables to identify which vegetables, according to experts and other stakeholders, have high potential to support food and nutrition in SSA: • The vegetable volume of the Plant Resources of Tropical Africa (PROTA) (Grubben & Denton, 2004) is the standard reference on vegetables in Africa. We retreived a list of 337 plants from PROTA that were identified as species that are principally used as a vege-  (Guarino, 1995). Our 126 selected species include 111 of these species (65% of this list).
• Three renowned scientists in traditional African vegetables identified 64 important vegetables according to country species lists, ethnobotanical surveys in specific communities and the authors' own experiences (Maundu et al., 2009). Our 126 selected species include 60 of these species (94% of this list).
• The African Orphan Crops Consortium (AOCC) (African Orphan Crops Consortium, 2019) species list includes 43 vegetables. These vegetables were selected by a diverse group of stakeholders including experts in the early 2010s as part of a wider set of annual and perennial plants to be considered in regional breeding research on new and orphan crops. Our 126 selected species include 38 of these species (88% of this list).
• Breeding priorities of the World Vegetable Center (WorldVeg) for traditional African vegetables based on Dinssa et al. (2016) help to define which vegetable genetic resources are used at a regional level. Farmers in Cameroon, Mali, Madagascar and Tanzania selected 15 traditional vegetables during 2008 as promising for crop diversification. Our 126 selected species include all 15 of these species (100% of this list).
In total, these five lists returned an extensive initial list for potential conservation action of 422 species (https://dx.doi.org/10.6084/ m9.figsh are.11954001). However, limited resources exist to safeguard traditional African vegetables both ex situ and in situ for a sufficient number of populations for conservation and breeding. To avoid the dilution of effort, and to be targeted and pragmatic, we decided to focus only on those species on this initial list that have the most potential for food and nutrition. As most of the species lack suitable metrics on production and consumption that could be used for prioritization purposes, we used a simple consensus-based approach for ranking. This was based on the frequency of each species inclusion in our five species lists. Such an approach, though imperfect, is recognized as sound when detailed information is not available (Romney et al., 1986). In our case, we selected any vegetable mentioned in at least two assessments for more detailed study at regional scale, resulting in a final list of 126 species. The remaining 296 of the 422 species were only mentioned by one of the species lists and were excluded accordingly from this regional assessment. • Records from herbaria and inventories stored in the Global Biodiversity Information Facility (GBIF, 2019). These were collected with the "rgbif" package (Chamberlain et al., 2016). The contributing organizations are listed in Text S1.

| Species' presence records
To detect geographical patterns of species richness, the georeferenced records from WIEWS and GBIF were selected and checked for quality. Presence records with inconsistencies (outside a border buffer zone of 1 arc minute) between geographical coordinates and the given country, as reported in associated passport data, were removed following the procedure of van Zonneveld et al. (2018).
Coordinates of presence records located in coastal waters within a 1-arc minute buffer zone to the coastline were relocated to the nearest land point. Presence records with coordinates of country middle points were removed using Coordinate Cleaner (Zizka et al., 2019), as these records likely reflect low georeferenced precision. For each species, duplicate records (i.e. with the same coordinates) were removed to reduce sample bias. Synonyms were checked using the "Taxonstand" package (Cayuela et al., 2012), and presence records were removed when synonyms were rejected by the Plant List.

| Conservation status
For each vegetable, the number of genebank accessions originating from SSA was recorded from WIEWS to determine to what extent its genetic variation is safeguarded ex situ in genebanks and therefore to what extent it is possible to use their genetic resources for crop improvement programmes. Our threshold number for conservation is 50 accessions: this is the minimum number of populations recommended to be sampled in a region with no prior knowledge of genetic structure (Brown & Marshall, 1995). Our threshold number to start a crop improvement programme is 200 accessions: this is the proximate size of mini-core collections that are used for screening germplasm for crop improvement (Schafleitner et al., 2015;Upadhyaya et al., 2006). Our threshold number to sustain long-term breeding programmes is 1,000 accessions: this follows audit recommendations for the WorldVeg genebank for the minimum size of vegetable crop collections, as part of its Genebank Quality Management System. While these numbers are somewhat arbitrary, they provide an expert-based and practical framework to review, monitor and improve the conservation and coverage of existing genebank collections for conservation and breeding.
To strengthen this framework with details about the geographical and ecological representativeness of the genetic variation safeguarded ex situ, three standard indicators for ex situ conservation were calculated for each species: (a) the amount of all genebank and herbarium records (non-georeferenced and georeferenced) were compared, resulting in the Sampling Representativeness Score (SRS); (b) the Geographic Representativeness of georeferenced genebank records (GRSex); and (c) the Ecological Representativeness of these records across terrestrial ecoregions (Olson et al., 2001) (ERSex). The final conservation score for ex situ conservation (FCSex) is the average of SRS, GRSex and ERSex. All indicators were calculated with the "GapAnalysis" package (Carver et al., 2020) following the approach of Khoury et al. (2019).
For each species, standard indicators for in situ conservation status were also calculated following the approach of Khoury et al. (2019) with a focus on conservation in protected areas. First, the Geographical Representativeness Score for in situ conservation (GRSin) was calculated as the proportion of the overlap of the modelled distribution range with the World Database of Protected Areas (Bingham et al., 2019). To support decision-making on prioritizing a type of conservation, a t test was used to check for differences between the ex situ and in situ conservation scores. An ANOVA was used to assess whether there was a relationship between the potential of speciesaccording to the frequency of their inclusion in the species lists-and their ex situ and in situ conservation status.
Finally, as a measure of sampling representativeness at country level, we compared the number of species and accessions recorded as safeguarded ex situ with the number of species and the number of observations of these species from presence records. All presence records, including non-georeferenced records, were used from the WIEWS and GBIF databases. The larger the number of observed species that is not safeguarded ex situ and the greater the disparity between observations and accessions collected, the greater the urgency to collect germplasm for ex situ conservation in that country.

| Observed species richness
As a basic initial analysis of geographical patterns to confirm species presence, we mapped the observed richness of traditional African vegetables using our extracted location data. We used the "raster" package and the circular neighbourhood approach to assess these diversity patterns (Hijmans & Etten, 2012;van Zonneveld et al., 2012). In this approach, each cell receives the number of species found within a circle with a specified diameter centred on the cell. Geographical coordinates were transformed to the Mollweide equal area projection to optimize comparison between geographical areas. Cell resolution was set at 50 km and the circular neighbourhood diameter at 300 km; these are scales suitable for a cross-continent analysis.

| Observed species richness corrected by resampling without replacement
While the observed richness is a good baseline indicator to plan conservation actions, it is limited in detecting biogeographical patterns of plant diversity because the values are biased by the varying densities of observations that occur in most geographical datasets. To account for this, we used a second approach where observed richness estimates were corrected by a common resampling without replacement procedure (Thomas et al., 2012). In our case, a 100-times bootstrap was used to reduce the sampling bias and the minimum sample number was set as the median number of species observed per grid cell.

| Modelled species richness
We then used a third approach, based on species distribution modelling (SDM), to identify in which agroecological zones people may be growing the selected vegetables or harvesting them from the wild. This approach helps to detect potential areas of high species richness in locations with low sampling density and to confirm areas of observed high species richness. We modelled the distributions of species with the "BiodiversityR" package (version 2.11-2) (Kindt, 2018). Details on the selection of environmental variables, on environmental and spatial thinning, and on model evaluation, are provided in Text S2. After spatial and environmental thinning of presence records, we undertook SDM only on 110 of our 126 species that had sufficient records to qualify for our modelling approach (Text S2). Species suitability maps were created for these species

| Vegetable composition structure
As a measure of spatial structure in vegetable diversity, we arranged grid cells in geographical clusters according to their composition of observed species. Parameters for cell resolution and circular neighbourhood diameter were as set out above.
Hierarchical clustering was used with Bray-Curtis, Jaccard and Kulczynski dissimilarity indices, to develop distance matrices for dendrograms using the "vegan" package (Oksanen et al., 2019).
The Ward linkage method was used to ensure even groups of grid cells to detect geographical patterns of vegetable composition.
The final number of clusters was set at five following the Kelley-Gardner-Sutcliffe penalty function, with a maximum number of possible clusters of 25, using the "maptree" package (Grum & Atieno, 2007;White & Gramacy, 2015). Then, each grid cell was assigned to its corresponding cluster. Consensus indicator species were determined per cluster following Dufrêne and Legendre (1997). Indicator values were determined with frequency-only data and calculated with the "indicspecies" package (De Cáceres et al., 2012). Consensus indicator species of a specific cluster returned an average indicator value of higher than 0.3 across the three applied clustering methods.

| Uses and domestication level
For each selected species, its food uses were recorded from the available literature. Then, for all species, their domestication level was determined from the literature using three categories: wild, semi-domesticated and domesticated. Broadly, the term "wild" is applied when a species is predominantly harvested from natural stands; "semi-domesticated" when a species is both widely harvested from natural stands and is in significant "cultivation"; and "domesticated" when a species is principally cultivated. These are loose definitions because "domesticated" has a specific biological meaning distinct from to be "cultivated," but our applied terms suffice for the current analysis (for further discussion, see .

| Prioritization of countries to implement conservation actions
We developed eight indicators to score the priority of countries for the implementation of conservation actions. For each cluster of vegetable composition, and for the primary regions of crop diversity, the countries with the highest scores were prioritized for conservation actions.

| Species summary
The 126 Table 1). The resampling exercise also revealed another hotspot that was not at first evident when considering the uncor-

| Vegetable composition structure
The three distance methods (Kulczynski, Bray-Curtis and Jaccard) revealed similar cluster patterns: clustering with the Kulczynski distance method can therefore be considered broadly representative Clusters 1 (Sahel) and 5 contained consensus species following our criterion, while the other clusters did not (Table S1). Cluster 1 included one consensus species, the indigenous wild jute mallow (Corchorus fascicularis). Cluster 5 included two consensus species, both naturalized and originating from the Americas, winter squash (Cucurbita maxima) and crookneck squash (C. moschata).

| Origin and cultivation status
In total, 79 of the 126 selected vegetables were indigenous to SSA.
Of these, we were able to identify for 34 species their primary re- Wild vegetables were most prevalent in Northeast and East Tropical Africa, and Southern Africa ( Figure 5).

| Conservation status
On average, the final conservation scores for in situ conservation (FCSin) were significantly higher than those for ex situ conservation (FCSex) (t = −23.54, p < .0001; Figure 6; Species that were included in many species lists had a better ex situ conservation status than those included less frequently  (Table S3). In addition, Kew's Millennium Seed Bank maintains 311 accessions from a total of 72 of the selected species (4.3 accessions per species and a Shannon index of 3.9); here, species coverage is therefore good, though the depth of coverage is less than the three aforementioned genebanks (Table S3).
The five countries with the most observed species missing in genebank collections were DR Congo, Angola, Cameroon, Benin and Togo ( Figure 7; Table 1). The five countries with the biggest difference in the total number of specimens recorded against the accessions noted in genebanks included DR Congo and Benin that were in common with the nations with the most observed missing species, but with South Africa, Burkina Faso and Ethiopia also featuring ( Figure 7; Table 1).

| Both human and phytogeographical history explain geographical distribution of traditional African vegetables
Our analysis identifies two hotspots of species diversity in primary regions of diversity of traditional African vegetables: the Dahomey gap covering South Benin, Togo and Ghana; and the Ethiopian highlands. These hotspots overlap with centres of crop domestication in both cases (Larson et al., 2014;Scarcelli et al., 2019). We hypothesize that the relatively dry conditions in the Dahomey gap compared to surrounding upper and lower Guinean rain forests (Salzmann & Hoelzmann, 2005) could have allowed the cultivation of a wide range of vegetables after their initial domestication in the West Tropical African region. The Ethiopian highlands in Northeast Tropical Africa were already a recognized Vavilov centre of diversity for cereals and coffee (Larson et al., 2014;Vavilov, 1992). Our analysis suggests this holds for a wider group of crops that includes vegetables.
Our analysis identifies South Cameroon as another hotspot of species diversity that is especially rich in semi-domesticated vegetables. South Cameroon does not overlap with an historic centre of crop domestication, but does so with a hotspot of cultural diversity (Loh & Harmon, 2005), providing support that the area could be a secondary region of vegetable crop diversity and domestication.
While the Sahel is recognized as a centre of domestication of several vegetables, including roselle, baobab, Bambara groundnut, hyacinth bean (Lablab purpureus) and kenaf (Hibiscus cannabinus) (Larson et al., 2014), our findings show low levels of species richness in this area. This finding is in contrast with our observed overlap between high species diversity and centres of crop domestication in the coastal areas of West Tropical Africa and the Ethiopian highlands.
One possible reason for this contrast is that, for a large number of vegetables, rain-fed production is possible in humid and seasonally dry regions, while their production is constrained in the semi-arid

| Conservation of traditional African vegetables requires an integrated approach
The conservation indicators show that, in general, the ex situ conservation status of traditional African vegetables is poor compared to their in situ conservation status. Even for most species with high potential, the in situ conservation status is still higher compared to ex situ, and most of these species do not have sufficiently large collections to sustain regional breeding programmes.
Our analysis therefore indicates that urgent efforts are needed to strengthen the ex situ conservation status of traditional African vegetables.
Even so, the conservation of traditional African vegetables requires an integrated conservation approach considering both ex situ and in situ. Ex situ conservation refers to storage or planting in external locations such as genebanks and botanic gardens, while in situ conservation refers to safeguarding wild populations in their original habitat or to dynamic conservation with local communities of local varieties or populations in their original areas of cultivation (Frankel et al., 1995). A third form of conservation, circa situm, is often used for perennial species in agroforestry systems. This refers to safeguarding planted and/or remnant trees in farmland where natural forest or woodland containing the same trees was once found, but this wild habitat has been lost or modified significantly through agricultural expansion (Dawson et al., 2013).
Our analysis suggests that people collect a high percentage of these vegetables at least in part from the wild, as 66% of the selected vegetables are wild or semi-domesticated. In addition, many of these species play an important role in the diets of wild herbivores, including baobab, Hibiscus, Solanum and Cleome spp. (Barnes et al., 1994;Kartzinel et al., 2015). In situ conservation in protected areas seems therefore to be a suitable measure to safeguard populations of these Genetic resources outside protected areas are priority for ex situ conservation, especially local varieties and populations that are vulnerable to extirpation. At the original locations of many herbarium records, people may have abandoned the production and consumption of these species or will do so in the near future because of trends to food production-and consumption-homogenization, and rural-to-urban migrations (Dawson, Park, et al., 2019;Pilling et al., 2020). The high scores for in situ conservation in protected areas of our 126 selected species suggest that this conservation approach could be promising too for many of the remaining 296 of the 422 species that were initially identified in our study. Although these 296 species were not considered further by us in the current study (see Section 2), they could be important to consider in local and/or national conservation plans. Assuming that many of these species are wild or semi-domesticated and that they have high in situ conservation scores, a relatively low investment in resources might safeguard the populations of these wild or semi-domesticated species in protected areas.
Botanic gardens in African countries could play an important role in ex situ conservation of especially wild and semi-domesticated species including many of the remaining 296 species. While genebanks may play a key role in sustainable agricultural development with a focus on safeguarding the genetic variation of the genepools of species with high potential for food and nutrition, botanic gardens especially may play a societal role in safeguarding biocultural heritage with a focus on conserving species diversity (Engelmann & Engels, 2002). In this way, genebanks and botanic gardens can develop complementary ex situ conservation approaches, engaging with different stakeholders (Pearce et al., 2020). Stakeholders for genebanks include researchers and breeders, while for botanic gardens they include the general public as well as researchers.

| West Tropical Africa and South Cameroon are priority areas for conservation actions
Of the main areas of high vegetable diversity identified above, veg-  (Larson et al., 2014;Paris, 2015). Only a few herbarium samples and genebank records for South Sudan were reported in the current study, and it is therefore a priority for germplasm exploration. DR Congo and Angola encompass high levels of non-georeferenced herbarium records and merit further exploration of the vegetable diversity occurring in them.

F I G U R E 6
Conservation indicators for the status of the genetic resources of the 126 selected traditional African vegetables in sub-Saharan Africa (SSA). For each species, the following three indicators are shown: (i) number of accessions with SSA origin safeguarded; (ii) final conservation score for ex situ conservation (FCSex) for the SSA genetic resources; and (iii) final conservation score for in situ conservation (FCSin) for the SSA genetic resources. The species are categorized according to the frequency of their inclusion in the five species lists consulted for this study, as a proxy for their potential for food and nutrition. For example, category five refers to the species that were included in all five species lists: accordingly, these species are considered to have most potential.
For each category, the species are sorted from low to high numbers of SSA accessions safeguarded, and then from low to high values of the two conservation scores combined. In this way, the species with most urgent conservation needs are listed at the top of each category The fact that in Africa it is still used primarily as a leafy vegetable indicates a different domestication trajectory that will have impacted on genetic diversity in the crop over the last 500 years or so since introduction to Africa. Clearly, in such cases, understanding processes of local African adaptation is crucial.

| CON CLUS IONS
Even though traditional African vegetables have a high potential for food and nutrition security, and climate change adaptation, F I G U R E 7 Gap analysis of traditional African vegetables at country level in sub-Saharan Africa. Panel a shows per country the number of species observed; panel b shows per country the number of species for which germplasm has been collected and safeguarded in one or more genebanks inside and/or outside the country; panel c shows the number of species for which germplasm has not yet been collected in the country for ex situ conservation. Panel d shows per country the number observations of species; panel e shows per country the number of accessions collected in the country that is safeguarded in one or more genebanks inside and/or outside the country; panel f shows, for each country, the differences between the number of observations of the selected species in that nation and the number of genebank accessions collected they have been relatively little invested in for food production, with limited germplasm conservation to support breeding and cul- Overall, the ex situ conservation status of vegetable diversity in West and West-Central Tropical Africa is poor compared to Northeast and East Tropical Africa, and Southern Africa, arguing for particular action in the first two of these regions. These efforts should include the collection of germplasm as well as corresponding traditional knowledge on use and management. Without these efforts, these genetic resources will be lost, as traditional food production systems in SSA are outcompeted by conventional ones.

ACK N OWLED G EM ENTS
Funding for WorldVeg general research activities is provided by core donors: Taiwan

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13188.