Marked population and phylogenetic structure are consistent with lineage diversification in geographic isolation in Arabian mangroves
Population structure analyses recovered remarkably high levels of divergence across Arabia, and identified groups of genetic clusters that mirrored the geographic distribution of mangroves across the various seas surrounding the Arabian Peninsula. A t-SNE analysis based on selectively neutral SNPs revealed a striking, widespread pattern of differentiation among almost all sampled populations, suggesting limited dispersal and long-term divergence between biogeographic regions, but also among sub-regional populations. Within the Red Sea, SNMF plots showed a pattern of genetic differentiation that separated the populations from the north, the center and the south of the basin into three distinct groups. SNMF scores also showed marked genetic differentiation between the northern and southern basins of the PAG, as well as considerable levels of shared ancestry between the populations of the southern PAG basin and the Sea of Oman, suggesting either incomplete lineage sorting or ongoing gene flow.
Phylogenetic inference revealed a generalized pattern of reciprocal monophyly among Arabian mangrove populations, and identified clearly differentiated phylogenetic lineages occurring in the Red Sea, the Arabian Sea and the PAG. Patterns of phylogenetic divergence among Arabian biogeographic regions have been previously reported for other marine taxa (34-36 ), congruent with the geographic distribution of the species and with physical barriers to gene flow (24, 36 ). However, few marine organisms in Arabia exhibit comparable levels of differentiation at such fine geographic scales as the gray mangrove, suggesting a process of genetic diversification occurring within the historically enclosed water bodies of the Red Sea and the PAG. Neutral population structure and phylogenetic relationships reported here support the hypothesis of mangroves finding refuge within the enclosed water bodies of the Arabian Peninsula during glacial cycles of the Pleistocene, where lineages could diverge in geographic isolation. Geological and paleoclimatic data suggest that areas of the Red Sea basin could indeed have acted as glacial refugia for mangroves (22 ), as it has been proposed for other marine endemic lineages of the region (21, 24 ). This scenario is particularly striking in the case of the PAG, where the presence of potentially suitable habitats was arguably very limited during glacial periods, where only the estuary of the historic Shatt-Ur river outlet to the Sea of Oman near the Strait of Hormuz had been thought to occur (23 ). Interestingly, populations from the southern basin of the PAG appeared as more closely related to the populations of the Sea of Oman, on the other side of the Strait of Hormuz, than with those of the northern PAG basin. This sequence of cladogenetic events also better fits a scenario of differentiation in glacial refugia where populations in the north would have become isolated earlier as the shore lines retreated, as opposed to a scenario of recent colonization throughout the PAG after the LGM, where northern populations would represent the most recent phylogenetic split (e.g. 35 ).
Demographic inference supports lineage divergence in (cryptic) glacial refugia and limited gene flow between Red Sea populations Two analyses comparing three and six different demographic models were run with fastSIMCOAL2 to identify the most likely scenario under which mangroves from the Red Sea and the PAG diverged, respectively. For the Red Sea, AIC revealed that the most likely scenario was consistent with lineage splits occurring during a period spanning the two last glacial cycles, suggesting that mangroves may have colonized the Red Sea basin roughly after the Strait of Bab al-Mandab opened to the Indian Ocean ca. 400,000 years ago (22 ) and persisted thereafter. Glacial cycles in the Red Sea are characterized by changes in the eustatic sea level and connectivity, but also by periods of desiccation and sharp changes in salinity and temperature. Therefore, the results reported here suggest that the levels of population and phylogenetic diversity found in the Red Sea are the product of historical events involving changes in aspects of population history such as effective size and geographic isolation, but also environmental conditions (21, 36 ). Moreover, the model with the highest likelihood for the Red Sea revealed relatively small effective population sizes, and no signs of recent gene flow among populations, a pattern that was further supported by the TreeMix analysis. This observation is unexpected as, unlike the PAG, the Red Sea lacks potential geographic barriers. The Red Sea presents a number of circulation features, particularly a system of meridional eddies in the Central Red Sea (37 ) that act to create an oceanographic discontinuity, which may limit propagule dispersal. However, large-scale high levels of surface circulation resulting from the Red Sea overflow and the Gulf of Aden intermediate water cells (16, 38 ) suggest that marine currents are not limiting migration among mangrove populations along the basin. Moreover, mathematical models simulating mangrove propagule dispersal across the global oceans have shown moderate to high dispersal densities for the Red Sea (39 ).
In the case of the PAG, the model that better fit our data revealed a scenario of simultaneous cladogenesis dating to 32,725 years, prior to the LGM and the later infilling of the enclosed sea. These results support the hypothesis of lineage diversification within the enclosed water body of the PAG during a period of low sea connectivity. Furthermore, the estimated splitting time reveals the presence of cryptic glacial refugia within the PAG, which was nearly completely drained during the LGM (23 ). Drastic changes in environmental conditions during glacial maxima, particularly in terms of salinity and temperature may have presented exceptional physiological challenges to resident marine life. Until now, survival through glacial periods has not been documented for any marine taxa in the PAG. However, conditions of intertidal zones in marine swamps, while extreme, may have remained locally suitable for gray mangroves. Lambeck (23 ) reconstructions of historic shorelines at times of the LGM show the potential development of lagoons and lakes in several localities, with the main ones corresponding to what are now northern and southern basins, as well as in the Strait of Hormuz. Indeed, today A. marina occurs in a number of inland, non-tidal saline or even freshwater lakes (e.g.40, 41, 42 ) under conditions that could resemble those of the PAG during glacial periods. Also, previous examples of marine cryptic glacial refugia have been documented for taxa such as seaweeds, rays and snails in places like southwest Ireland, the northern Brittany-Hurd Deep area of the English Channel, the Azores islands or the northwest Iberian Peninsula (21, 43-45 ), suggesting that the persistence of benign local conditions in marine environments during glacial periods may be relatively common. In addition to a split occurring prior to the last glacial maximum, the best fitting demographic model revealed potential gene flow between PAG basins and between the southern basin and the Sea of Oman. The TreeMix analysis also detected significant levels of gene flow between both terminal and ancestral branches, suggesting that some migration could occur as mangrove populations became isolated and also after the LGM, once the infilling of the PAG started. Potential barriers such as the narrow waterway of the Strait of Hormuz or the marine eddies between Qatar and Iran separating the northern and southern basin may, therefore, not be as limiting for mangrove propagule dispersal as has been proposed for other species (35, 46 ).
Our analyses reveal thereby a process of lineage diversification in the mangroves of Arabia, which differentiated in geographic isolation during glacial periods, followed by a phase of secondary contact within the main water bodies bordering the peninsula. Although taking place before the LGM, the Arabian mangrove system represents a strikingly fast process of diversification, occurring at temporal scales of tens to a few hundred thousand years. Generation time for the gray mangrove has been estimated at 20 years in tropical areas (31, 47 ), and to range from 25.6 to 53.8 years in high-latitude populations of the PAG (48 ), which implies that the diversification process may date back to as few as 12,000 generations or less. High levels of population divergence are expected at the edge of species’ ranges, where reduced population sizes and pronounced habitat fragmentation can lead to rapid differentiation due to drift in isolation and divergent selection (10 ). In addition, and despite long flotation periods during which propagules remain viable, a relatively low dispersal capacity has been reported for A. marina (49 ), which may also promote genetic differentiation among peripheral populations. Indeed, higher genetic structure in the margins of the species distribution compared to core populations of the gray mangrove of the West Pacific has been previously reported (50 ). Moreover, increased rates of speciation have been documented in other mangrove systems undergoing periodic cycles of isolation and gene flow due to fluctuations in sea level during glacial periods (31 ), similar to what may have occurred in the enclosed water bodies of the PAG and especially the Red Sea. Overall, the results reported here reveal mangrove peripheral populations as potentially important sites for speciation, and suggest that such populations may be of particular conservation value as significant drivers of intraspecific diversity and sources of evolutionary innovation (10, 11 ).
Redundancy analyses reveal multi-loci adaptive divergence driven by environmental extremes and distinct genotype-environment association patterns in Arabian mangroves In an RDA based on population allele frequencies, four constraining variables approximating key environmental parameters for mangrove ecology explained 18.3% of the total variance in population SNP frequencies. 668 SNP outliers were identified based on their contribution to the genotype-environment association patterns. We also identified 51 functionally annotated genes associated with one or more SNP outliers. Among the functions associated to detected candidate genes, at least eleven seem particularly relevant for adaptation to the extreme and variable habitats of the seas around the Arabian Peninsula, including regulation of responses to multiple abiotic stresses such as drought, salinity and low and high temperature extremes; stomatal regulation; modulation of germination, growth and flowering in response to abiotic stresses; and Cu intake regulation and Cd tolerance (specific relevant functional genes and further details on reported functions are provided in the Supplementary Information).
Our genotype-environment association analysis also revealed strikingly different patterns of covariation between genetic variance and environmental parameters. The first two axes of the RDA based on allele frequencies recovered highly divergent association patterns between the Red Sea, the Arabian Sea and the PAG, yet limited variation within these regions, as showed by the RDA scores. However, a complementary RDA based solely on the previously identified 668 SNP outliers and computed over individual genotypes revealed considerable adaptive divergence among Red Sea populations (Figure 6), which differentiated along a gradient of minimum temperatures from south to north. The Red Sea does present major latitudinal, climatic differences along its basin. Towards the south, strong solar insolation results in air and sea surface temperatures exceeding the 40º and 30º C respectively, yet minimum temperatures rarely drop under 20º C. In turn, salinity increases northwards due to limited precipitation and high evaporation, and minimum annual air temperatures regularly drop below 10º C, together with a reduction in nutrient supply and productivity (17 ). As a result, the northern, central and southern portions of the Red Sea are considered to be distinct biotopes (15, 16, 51 ). Mangroves in these areas are characterized by tall (> 10 m canopy height) in the southern Red Sea, dwarf trees (< 3 m canopy height) in the Central Red Sea, and somewhat taller (> 3 m canopy height) mangrove trees in the northern Red Sea (17 ).
In turn, our analyses revealed reduced levels of within-group divergence driven by environmental factors for the PAG plus the Sea of Oman lineage. The most distal population Qurm from the Sea of Oman showed the most divergent association signals along a gradient of salinity maxima and to a lesser extent, of temperature minima. Remaining populations within the clade showed limited differences, especially for those of the PAG. Reduced divergence in patterns of genotype-environment association may reflect less pronounced difference in habitat conditions across PAG basins. Northern and southern PAG basins present large-scale differences in parameters like salinity or temperature maxima and minima. However, projections of georeferenced environmental data suggests that conditions at the coastal, specific sites inhabited by mangroves are seemingly less differentiated between both subregions (35, 52 ).