An analytically derived delineation of the West African Coastal Province based on bivalves

To assess the pattern of biotic regions (BR) and their boundaries, to detect chorotypes and to relate these patterns to key environmental factors.


| INTRODUC TI ON
The existence of differentiated biogeographical regions in the sea has been long recognized since the pioneer works of Forbes (1856) and McAndrew (1857). However, the placement of boundaries has been persistently based on authoritative decisions or on an empirical knowledge (Briggs, 1974;Ekman, 1935Ekman, , 1953Schilder, 1956), and rarely based on a formal analysis of the occurrence data of taxa (e.g. the pioneer work of Valentine, 1966Valentine, , 1973. With a wealth of occurrence data being increasingly available, it becomes feasible and necessary to address those empirical units from the very basic data, namely the occurrence patterns of individual species (e.g. Belanger et al., 2012;Costello et al., 2017;Kulbicki et al., 2013). Tropical West African is unanimously recognized as one of the four major provinces for tropical coastal and shelf marine biota, together with the Caribbean, Panamanian and Indo-West Pacific (Briggs, 1974;Briggs & Bowen, 2012;Ekman, 1953;Kulbicki et al., 2013). Yet, the exact location of the boundary is unclear and the within-province biogeography is much less known than in the other tropical provinces.
A sound classification of biogeographical areas is an essential tool for analysing basic questions in historical and ecological biogeography, evolutionary biology and systematics underlying global biodiversity patterns (Lomolino et al., 2006;Valentine, 1973;Witman & Roy, 2009), and for assessing priorities for conservation planning (Spalding et al., 2007;Whittaker et al., 2005). The assessment of representativeness for the conservation evaluation also requires a previous biotic regionalization to define biogeographical contexts (Austin & Margules, 1986).
Another perspective of biogeography is to analyse species distributions by looking for shared overall distribution patterns (chorotypes; see Fattorini, 2015 for a detailed discussion of that term).
Both biotic regions and chorotypes reflect species distribution, which in turn respond to environmental and historical attractors (Kreft & Jetz, 2010;Real et al., 2008). The species composition of a biotic region can be viewed as the regional superposition of several chorotypes, considered as biogeographic features with properties beyond the simple sum of their parts (Olivero et al., 2013).
Molluscs are one of the major phyla in the marine benthos, accounting for as much as one-quarter of all species (Appeltans et al., 2012) and occupying a large variety of habitats. Among these, the class Bivalvia is not only the second largest with ca. 8500 living species (Huber, 2015) but also that one with the most economically and ecologically important species, commonly outnumbering gastropods in number of individuals, if not of species. Therefore, bivalves have attracted the attention of biogeographers as an appropriate group for large-scale analysis, for example Krug et al. (2009), Belanger et al. (2012, Jablonski et al. (2013).
The Bivalvia of tropical West Africa have been treated in a comprehensive identification guide (Cosel & Gofas, 2019). The wealth of original distribution data examined for this publication is such that it could provide a new, independent insight on the regional biogeography. Here, we aim to analyse the occurrence data therein in order to (1) test the placement of a major boundary between the Lusitanian or west European province and the West African one, (2) characterize the tropical West African biotic region and its subdivisions in terms of correlated chorotypes, species richness and levels of endemism, (3) attempt to relate the faunal break between Europe and West Africa, and the subordinate faunal breaks within the West African Province, to key environmental factors, and (4) to draw conclusions from this study regarding conservation priorities. 1.6 for details), being most thorough in Mauritania, Senegal, Cape Verde Islands, Guinea and Guinea Bissau, Côte d'Ivoire, Gabon, Congo and Angola. Mauritania and Guinea were covered by a comprehensive sampling using grabs, aimed at a sedimentological characterization of the entire continental shelf; other countries such as Côte d'Ivoire, Gabon and Angola were sampled repeatedly in particular places and not at all in others. For the remainder of West Africa, the data rely on sampling efforts that spanned the whole coast, or large parts of it, among which the most noteworthy are the mostly shore-based Mission Gruvel 1909-1910(Dautzenberg, 1910, 1912, the expedition of R/V Atlantide in 1949 (Nicklès, 1955), the expeditions of R/V Calypso to the Cape Verde Islands in 1956and Gulf of Guinea in 1959, and the Guinean Trawling Survey in 1964 unique endemic component is insufficiently represented in current marine protected areas.

K E Y W O R D S
biogeographic regions, bivalves, chorotypes, Lusitanian, marine biogeography, marine regionalization, marine regions, Mauritanian, Mediterranean., West Africa Altogether, the sampling considered in this work is representative of the second half of the 20th century as a baseline. The material figured in Cosel & Gofas (2019) is detailed in the book's supplementary material, but the distributions scored here take into account all the examined material, including the unfigured lots.
Data for distributions outside tropical West Africa were taken from the official Spanish list of marine organisms (Gofas et al., 2017) in which five areas are considered separately; from Pasteur-Humbert (1962) and Salas (1996) for Morocco; from Fauna d'Italia (Schiaparelli, 2008) in which nine regions are scored; from the list of Roscoff area (Cornet & Marche-Marchad, 1951, with online updates at <http://www.sb-rosco ff.fr/fr/obser vatio n/biodi versi te/espec es/ inven taire s/inven taire s-de-la-faune -et-de-la-flore -marines>) and Seaward (1990)  summarizes occurrence data for 595 species (Appendix S1), among which 65 were marked as data-deficient and so some analyses were repeated without including them, as a test for the robustness of the results. These species were treated as data-deficient because of either taxonomic difficulty (more than one-third are Galeommatoidea, which for the same reason were excluded by Jablonski et al., 2013) or rarity/difficulty to collect (e.g. species known only or mostly from their type specimens), which make records less predictable.  Figure 3) of OGUs, with the drawbacks above mentioned.
A 1°grid was not feasible taking into account the disparity in sample density and precision of labelling. The MEOW Ecoregions, conversely, would have provided only 20 OGUs within the same scope, with some of them spanning several countries, and we rejected them as too coarse. We have considered, following Phipps (1975), that the size of the OGUs must seek the best trade-off between too many units with poor data for most of them, and too coarse units with a poor resolution.

| Regionalization
The method used to objectively identify biogeographic regions (BR) and boundaries was based on Olivero et al. (2013). OGUs were classified hierarchically according to the presence/absence of bivalve species, using the Baroni-Urbani & Buser (1976) similarity index and the unweighted pair group method with arithmetic mean agglomerative algorithm (Sneath & Sokal, 1973). All clusters in the resulting classification dendrogram were assessed for statistical significance using G tests of independence (Sokal & Rohlf, 1981) and the 'RMacoqui 1.0' software (http://rmaco qui.r-forge.r-proje ct.org/) (see more details in Olivero et al., 2013). Two types of biotic boundaries are shown in the resulting classification tree: 'weak boundaries' defined by significant similarities within the bounded region, and 'strong boundaries' defined by significant between-region dissimilarities. Thus, a coastline section that was bounded by strong borders was considered to be a 'strong BR', within which other biotic units bounded by weak borders, called 'weak BR', could nest. Consistent weak BR were characterized by a significant value of internal homogeneity (Olivero et al., 2013). Conversely, a set of clustered OGUs delimited by a significant boundary could be identified as a fuzzy biotic region if its value of internal homogeneity was not significant, thus denoting the presence of a biogeographic transition zone.
There is no consensus regarding which method is best for delimiting biogeographic regions. Quantitative techniques based on cluster analysis and similarity indices have been recently used for marine regionalization (Costello et al., 2017;Freitas et al., 2019;Kulbicki et al., 2013). Our preference went to taking into account that some biogeographic boundaries can be gradient zones or transition zones (Jacquez et al., 2000;Leung, 1987), that is that some of the areas may belong to a certain degree to more than one biogeographical region.

| Characterization of the biogeographical patterns with respect to spatial/environmental factors
We characterized the biogeographical patterns according to 17 variables related to spatial trends and to geophysical and hydrological factors ( Table 1).
Forward-backward stepwise logistic regression analyses were used to environmentally characterize the biotic regions, using IBM SPSS v. 24. The analyses were performed on each BR, using the binary membership/non-membership of the OGUs in the BR (possible values: 1 or 0) as dependent variables, and the set of spatial/environmental descriptors in Table 1 as independent variables. Variable selection and parameterization along the stepwise procedure were based on score-test significance and iterative loglikelihood maximization, respectively; the goodness-of-fit regression was assessed using a chi-squared test; and the significance of variables in the regression was tested using Wald's tests (Hosmer & Lemeshow, 2000).
Spatial factors (latitude and longitude) were used, because they could point to historical reasons behind the biogeographic pattern (Legendre, 1993). The extent of the continental shelf was calcu- Distance to river mouths was considered, in order to take into account the influence of river run-off in the sea (Mann & Lazier, 1991). A classification was made according to each river flow in two categories: mighty rivers (Q ≥ 100 km 3 /year) and low-flow rivers (Q < 100 km 3 /year). River flow data were obtained from Milliman and Farnsworth (2011). The distance to river mouths was calculated by Zonal Statistics ArcGIS 10.3.
The presence of permanent or seasonal upwellings was considered as a variable: Alboran Sea upwellings and the patterns of upwellings described by Le Loeuff & Cosel (1998).

| Biotic characterization of regions based on chorotypes
The chorotypes of bivalve species were analysed following the method outlined by Olivero et al. (2011). Chorotype analysis is similar to the above-described regionalization, but now, we analyse the similarity between species distributions (based on shared OGUs) instead of between OGUs (based on shared taxa). This analysis does not predefine the number of resulting chorotypes. All groups of distributions meeting the requirements for forming chorotypes to the highest level (i.e. maximizing within-group similarity) were considered chorotypes provided that they were significantly clustered.
Similarity values, clustering and statistical significance based on G tests of independence were calculated using the 'RMacoqui 1.0' TA B L E 1 Variables used in the analysis for the spatial/environmental characterization of biogeographic regions software (http://rmaco qui.r-forge.r-proje ct.org/) (see more methodological details in Olivero et al., 2011). When a chorotype showed disaggregated diversity cores (i.e. the chorotype species were geographically grouped in separated areas), these areas were considered 'subchorotypes' (Olivero et al., 2011) if they corresponded to different clusters of species in the hierarchical classification and these clusters were statistically significant.
The correlation between biotic regions and chorotypes was tested using Spearman's coefficient. This involves the correlation between the membership of each OGU in a region but not in the rest of the study area, on the one hand, and the number of species of a particular chorotype in those OGUs, on the other hand (see Olivero et al., 2013).
The degree of membership of an OGU in a region but not in the rest of the study area was calculated using the fuzzy difference tool: where μ is 'degree of membership', BR is 'biotic region', and RSA is 'rest of the study area', being μ BR (OGU i ) and μ RSA (OGU i ) obtained with RMacoqui 1.0.

| Levels of endemism
The percentage of endemic species has been retained as an essential feature for the characterization of marine provinces (Briggs & Bowen, 2012). Species were scored as endemic to the West African biotic region determined by the analysis, when they were present in one or more of the OGUs comprised therein and absent elsewhere. The north-western African region, delimited by a weak biotic boundary, shows a positive but not significant internal homogeneity index, and so it is identified as a fuzzy BR or biogeographic transition zone. The same applies to Baía dos Tigres and Namibia. Although the southern part of Angola, São Tomé and Principe-Annobón, and the Cape Verde Islands are geographically distant, they are recovered as a consistent BR, but with a much lower internal homogeneity index and significance. The remaining BR, delimited by strong or weak boundaries, are consistent with a high internal homogeneity value.

| Regionalization
Within the western Africa tropical region, there is no strong signal for significant grouping subsets of OGUs. However, some localities in the cluster show a greater similarity: Senegal (Casamance), Guinea Bissau and Guinea share the same node with more than 90% similarity; Sierra Leone and Liberia are more similar to Nigeria and Cameroon (90%) than to Ivory Coast and Ghana-Togo-Benin (85%), although they are not contiguous locations; and Gabon, Congo and the northern part Angola are more similar to each other (80%) than they are to the rest of the West African OGUs.
The same analysis, performed without 65 data-deficient species, yields a very similar topology. The strong biotic boundaries and the topology of the West African cluster are not altered, the only difference in the topology being that the northern and middle Adriatic OGUs cluster with the remaining Mediterranean OGUs rather than with southern Adriatic.

| Biotic characterization of regions based on chorotypes
Sixteen chorotypes and five gradual patterns were identified for bivalve species, the latter not considered significant chorotypes ( Table 3). The Appendix S1 indicates for each species the pertinence F I G U R E 2 Biotic regions in European and West African coastal areas as determined in the present analysis   Note: DS > 0 and a significant G(S) indicate the presence of a strong biotic boundary; DW > 0 and a significant G(W) indicate a weak biotic boundary between clusters; IH is the internal homogeneity index, and IH > 0 and a significant G(IH) indicate that the cluster can be considered a consistent BR (otherwise, the cluster is a fuzzy BR or biogeographic transition zone). Statistical significance associated with the G tests (degrees of freedom = 1): ns = p ≥ .05; *p < .05; **p < .005; and -: not applicable because the cluster is composed by a single OGU. BR are named as in Figure 1.
to the chorotypes. The correlation of the BR with chorotypes is also represented in Table 3. The most representative chorotypes are shown in Figure 3a- (Figure 4a-c), including species that are not present in Sierra Leone and Liberia and/or Nigeria and Cameroon.

| Species richness and endemism
Of the 429 species of our matrix (Appendix S1) present in the West and 2 in the Gulf of Guinea islands ( Figure 5), but there is no species endemic to the cluster and present in more than one of its compo-  Table 4 with the appropriate sources.
Considering individual OGUs, species richness ranges from 78 to 258, with very few species endemic to a particular OGU ( Figure 5).

F I G U R E 3
A selection of the most representative chorotypes, showing a high positive correlation with one or more biogeographic regions, and comprising a higher number of species (see Table 3 for species numbers). The colour scale and corresponding numbers represent species richness in each OGU

| Gaps of knowledge
Baía dos Tigres and Namibia cluster outside the West African supraregion but the occurrence data totalize only 18 species. Whereas

Baía dos Tigres (sampled from shore and by dredging in August 1985
by SG) is genuinely species-poor because of the homogeneity of its sandy habitats, Namibia is one of the deepest gaps of knowledge in World faunistic records.

| Regionalization support
There is general agreement, since Ekman's (1935Ekman's ( , 1953  Taking into account the boundaries detected in the analysis and the levels of endemism within these boundaries, the biogeographic units resulting from the analysis should be given a rank in a hierarchical system. Level of endemism (see below) is the main basis for doing so, but Kulbicki et al. (2013) wrote that 'There is currently no universally accepted terminology for a biogeographical hierarchy'. Briggs (1974) and Briggs & Bowen (2012)

| Characterization of the biogeographic regions
The West African BR is mainly characterized by the mean sea surface temperature and the range of chlorophyll. Characterization of tropical West Africa without the islands also involved substrate with the negative correlation of rocky shore reflecting the extensive sandy coast (Bird & Schwarz, 1985). As a difference with the islands, the extension of the shelf also characterized the tropical West Africa proper.
The north-western African transition region from west Sahara to Senegal is characterized by an arid-desert climate, lowest rainfall and few important rivers. The hydrological conditions that are influenced by seasonal upwelling present during the Northern Hemisphere winter-spring (Pelegrí & Benazzouz, 2015) are strongly influenced by important seasonal variation of chlorophylls.
The cluster that comprises Angola, Sao Tomé and Principe-Annobón and the Cape Verde Islands does not hold any species endemic to the cluster and present in two or in all three OGUs. These three OGUs contrast with the remainder of tropical West Africa in having a predominantly rocky shore, and this rather than a shared history likely drives the occurrence of the species.
The European BR was only characterized by the latitudinal spatial factor for its more northerly location. Baía dos Tigres and Namibia were marked by the spatial trend in the southern part and also by low sea surface temperature. Latitude is highly correlated with sea surface temperature, but also involves a historical factor. It is understood that this spatial factor with its historical component is crucial in temperate BR, where the environmentally similar areas in the opposite hemisphere may be far away and beyond the species' potential for dispersal. This is less important in the tropical BR.
In a global scale, the processes that mainly determine the distribution and availability of the primary environmental resources, and therefore the species' distribution, are latitudinal and seasonal variation (Mackey & Lindenmayer, 2001 showing a disjunct distribution. The latter patterns are recognized as chorotypes in the analysis, but subordinate to the broader West African chorotype C6 (see Figure 3d). Not mentioned by Cosel and Gofas (2019) are 'bridge species, whose ranges cross the tropical/ extratropical boundary' as defined by Jablonski et al. (2013). Some of these either belong to the Euro-West African chorotype C2 or to the essentially European chorotype C5, or are nor recognized as part of a chorotype (see Appendix S1). Belanger et al. (2012) found that bivalve biogeographic structure can be predicted accurately by very few readily available oceanographic variables (temperature, salinity, productivity), with temperature alone predicting 53%-99% of the present-day structure along coastlines. This suggests that bivalve distributions are more likely to reflect large-scale oceanographic structure than smaller differences in local conditions. As Belanger et al. (2012) (Taviani, 2015), but surprisingly, data regarding bivalves are scarce; Cuerda (1987) reported seven tropical west African bivalves from the Pleistocene of Mallorca, of which six are today West African endemics and one (Eastonia rugosa) that is in progress inside the Mediterranean. This number is raised to 15 (5.7% of current West African endemic bivalves) by Bellomo (1998, unpublished doctoral thesis). Nevertheless, West Africa seems to contribute little to current faunal shifts reflecting the ongoing climate change (Vermeij, 2012). The 'Senegalese hosts' reported in the Mediterranean Pleistocene are too few to trigger profound differences in the position of faunal boundaries, but show how fast faunal elements can disperse and occupy new areas in reaction to changing environmental conditions.
Cape Verdes, which hold 108 species in our matrix, and Canaries, which hold 153 species, were nevertheless correctly assessed. Chaudhary et al. (2016) suggested that species richness is highest in mid-latitudes and dips near the Equator. Global bimodality is inter-

| Conservation issues
Our results report that 261 species present in the West African BR (60.8%) are endemic and that 33 of those are known from only a single West African OGU (7 in the Cape Verdes, 5 in southern Angola and 2 in the Gulf of Guinea islands). At a higher taxonomic level that reflects phylogenetic uniqueness, there are 19 genera (7.2%) endemic to the tropical West African BR. This calls for an evaluation of conservation perspectives for those taxa.
Species to be considered for conservation issues, in addition to being endemic, must be suitable both for the dissemination of their values to the non-scientific public, and for the follow-up and monitoring of a marine protected area (Ducarme et al., 2012). These may be 'flagship species' that serve as symbols to stimulate conservation awareness, and 'vulnerable' species that are known to be less resistant to environmental changes than others in the community.
In the first category, we would consider the ark-shell Senilia seni-  OGUs (Cape Verdes and southern Angola) that harbour local endemics, and would be even more critical if endemic gastropods (e.g. Conidae, see Peters et al. (2013), and the above-mentioned Trochita) were taken into account. Representativeness of MPAs can be summarized using the chorotypes; in this case, chorotype C6 (the tropical West African BR globally) is represented in existing MPAs but chorotypes C9 and C13 (the islands of Gulf of Guinea and southern Angola respectively) are not.

| CON CLUS IONS
This study brings overwhelming support for the recognition of a tropical West African coastal province (not including the Islands of St. Helena and Ascension), with a moderate species richness but with a proportion of endemic species ranking high among the major biotic regions of the World. This natural heritage is badly underrepresented in current conservation schemes, taking into account the scarcity of formal marine protected areas in West Africa.
The present analyses provide an objective regionalization, in which each cluster is individually assessed and could potentially be considered to represent a biogeographic unit entity (Olivero et al., 2013). In contrast to other analyses, the resulting number of biotic regions and the degree of fuzziness for partitioning do not need to be assigned a priori (Kreft & Jetz, 2010). This methodological approach, combining fuzzy logic and statistics, avoids arbitrary decisions in the definition of biogeographic regions. In addition, the robustness of classification has been tested in the analysis against changes in species composition (e.g. it has produced the same output when performed without 65 data-deficient species) showing a high stability.

ACK N OWLED G EM ENTS
JAC thanks Nicolas Puillandre for welcoming him to MNHN as an ERASMUS student in 2020. Funding for the APC of this paper was provided by RNM141 research group of Junta de Andalucía.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The supplementary material containing the detailed distribution data used in this work is available from <https://doi.org/10.5061/ dryad.00000 004w>. Distribution data used in the analysis: Also available from <https://doi.org/10.5281/zenodo.5653589>. Sources as described in the Section 2.