Oceanographic connectivity explains the intra-specific diversity of mangrove forests at global scales

Significance

Among the multiple drivers shaping the intra-specific diversity (i.e., genetic diversity) of mangrove forests, the role of connectivity driven by ocean currents has not been well defined at global scales, despite its potential role in regulating population gene flow. Using biophysical modeling and empirical genetic data, we show that oceanographic connectivity can explain population genetic differentiation in mangroves, regardless of the species, region, and genetic marker, overcoming classical approaches based on geographic distance. Ocean currents can either create barriers that isolate populations and increase differentiation levels or promote long-distance dispersal responsible for first settlements and broadscale biogeographical patterns. These findings are relevant for mangroves’ biogeography and evolution, as well as management strategies considering climate change and genetic biodiversity conservation.

Abstract

The distribution of mangrove intra-specific biodiversity can be structured by historical demographic processes that enhance or limit effective population sizes. Oceanographic connectivity (OC) may further structure intra-specific biodiversity by preserving or diluting the genetic signatures of historical changes. Despite its relevance for biogeography and evolution, the role of oceanographic connectivity in structuring the distribution of mangrove’s genetic diversity has not been addressed at global scale. Here we ask whether connectivity mediated by ocean currents explains the intra-specific diversity of mangroves. A comprehensive dataset of population genetic differentiation was compiled from the literature. Multigenerational connectivity and population centrality indices were estimated with biophysical modeling coupled with network analyses. The variability explained in genetic differentiation was tested with competitive regression models built upon classical isolation-by-distance (IBD) models considering geographic distance. We show that oceanographic connectivity can explain the genetic differentiation of mangrove populations regardless of the species, region, and genetic marker (significant regression models in 95% of cases, with an average R-square of 0.44 ± 0.23 and Person’s correlation of 0.65 ± 0.17), systematically improving IBD models. Centrality indices, providing information on important stepping-stone sites between biogeographic regions, were also important in explaining differentiation (R-square improvement of 0.06 ± 0.07, up to 0.42). We further show that ocean currents produce skewed dispersal kernels for mangroves, highlighting the role of rare long-distance dispersal events responsible for historical settlements. Overall, we demonstrate the role of oceanographic connectivity in structuring mangrove intra-specific diversity. Our findings are critical for mangroves’ biogeography and evolution, but also for management strategies considering climate change and genetic biodiversity conservation.

The distribution of intra-specific genetic diversity can be structured by demographic processes that enhance or limit effective population sizes (1, 2). For instance, historical climate change fluctuations and conservation strategies can create refugial areas of long-term persistence where populations were allowed to accumulate higher and rarer genetic diversity (3), compared to where unsuitable climatic conditions or human-induced disturbances led to bottlenecks and drift (4, 5). Population connectivity can further structure the distribution of intra-specific diversity. For coastal species like mangrove forests, the regional patterns of oceanographic currents can restrict dispersal and gene flow, leading to the formation of highly structured populations that hold the signatures left by historical changes (6).

Mangroves are halophytic plants that form ecosystems with exceptionally high ecological and socioeconomic value (7) across tropical and subtropical coastlines, with distribution ranges typically restricted by low temperatures and reduced precipitation (8). The dispersal ecology of the approximately 80 species of mangroves (18 plant families) includes buoyant seeds, seedlings, and fruits (hereafter referred to as propagules; (9) that exhibit contrasting floating periods, from several days to months (10), which can strongly determine the scales of population connectivity and gene flow. However, the role of oceanographic connectivity in the distribution of intra-specific diversity of mangroves has not been well defined. Most studies to date have used isolation by distance models (11, 12) under the major assumption of higher genetic differentiation with increasing geographic distance (13), or illustrative maps of ocean currents to describe population connectivity (14). These approaches can’t capture the complexity of tidal, topographic, and wind-driven oceanographic processes (15), which produce asymmetric (directional) gene flow between populations (14) with high temporal and spatial variability. Additionally, the relationship between ocean currents and population dispersal is not linear, but rather positively skewed (16, 17), which is in line with mangrove’s strong genetic structure found across species’ distribution ranges (18).

More realistic estimates of oceanographic connectivity can be produced with biophysical modeling and Lagrangian simulations (17). These frameworks, mimicking the dispersal behavior of propagules advected by ocean currents, can provide estimates of connectivity (10) that generally match the expectations of independent genetic data (17, 19, 20). Studies on mangroves are still underrepresented, with oceanographic connectivity only estimated at regional scales and considering few species, namely Rhizophora mucronata (14), R. mangle (21), Laguncularia racemosa and R. racemosa (21, 22), and Avicennia species (12, 23, 24), precluding broad conclusions on the subject. One additional study using biophysical modeling provided key insights on the global dispersal and connectivity patterns of mangroves, yet it did not consider independent data to support conclusions (e.g., population genetic differentiation data and species-specific floating periods) (10).

We aim to fill this knowledge gap by asking whether oceanographic connectivity shapes the intra-specific diversity of mangrove forests at global scales. Simulations of oceanographic connectivity were performed with a biophysical model running across the global distribution of mangroves and tested against empirical genetic data compiled from the available literature. Overall, we hypothesize that oceanographic connectivity can better explain population genetic differentiation compared to more simplistic approaches based on isolation-by-distance (IBD) models, as previously found for other ecological groups (17, 25). We integrated graph theory (i.e., network analyses) to consider asymmetric multigenerational stepping-stone connectivity (17), and for the first time, centrality indices accounting for the effect of particular populations promoting gene flow across large water masses. Our results, supporting the role of oceanographic connectivity in mangrove’s diversity, have high relevance for biogeography, evolution, and climate change ecology (18), as well as for the new conservation targets of the post-2020 Global Biodiversity framework, which endorses the conservation of genetic diversity of wild populations, as well as the establishment of well-connected networks of protected areas (26, 27).

Results

The systematic literature search listed 540 potential studies, from which 43 reported genetic differentiation data for mangroves. These studies resulted in 59 datasets covering 20 species of the most common genera (Avicennia, Bruguiera, Sonneratia, and Rhizophora; SI Appendix, Table S1) and life strategies, i.e., viviparous and cryptoviviparous (9), 7 molecular markers, and 691 populations (SI Appendix, Table S1) spanning across the known distribution of mangrove forests (i.e., Indo-West Pacific and Atlantic-East Pacific; Fig. 1). On average, 11.71 ± 7.16 populations were sampled per study (range between 4 and 38 populations), with an average genetic differentiation of 0.31 ± 0.17 (range between 0.01 and 0.68, although estimates cannot be compared between studies, species or genetic marker). The average and maximum distance recorded between sampled populations per study were 1,361.51 ± 1,273.02 km and 12,149.20 km, respectively (SI Appendix, Table S1).

Fig. 1.

Global distribution of mangrove forests (dark contour; updated by UNEP-WCMC; version 1.4, March 2021; 28) where the biophysical modelling simulated oceanographic connectivity, and populations sampled for genetic differentiation data (yellow circles).

The biophysical model released a total of 21,891,914 particles from 11,989 source and sink sites over the 5-y period. These particles resulted in 6,106,904 connectivity events (average of 8.37 ± 29.68 per site) linking 561,247 pairs of sites. Longer propagule duration (PD) periods, i.e., up to 180 d, resulted in larger traveled distances. This pattern spanned an average distance of 55.12 ± 63.70 km with 10 d of propagule duration to 715.44 ± 765.59 km with 180 d (maximum of 10,582.96 km; Figs. 2 and 3). The number of connectivity events also increased with propagule duration, from 4,539,045 to 6,106,904 (Fig. 3). The probability of connectivity and dispersal distances showed a positively skewed relationship, regardless of the propagule duration (Fig. 2). Mapping betweenness centrality allowed identifying important hubs for stepping-stone connectivity. For reduced PD, connectivity hubs were mostly distributed along the coastlines of the western Atlantic and the Indo-West Pacific regions, while for increased PD, hubs were distributed in key islands and offshore regions that bridge distant populations (e.g., between continents) and along the Indo-West Pacific region (Fig. 3).

Fig. 2.

Probability of connectivity as a function of potential dispersal distance (km) inferred with biophysical modeling for (A) 10 d, (B) 70 d and (C) 180 d of propagule duration (10 d represent the minimum, 70 d the average and 180 d the 90th percentile of mangrove propagule duration; inferred by (29) to the closest 10th day). Black circles in A–C represent all connectivity events between pairs of sources and sink sites.

Fig. 3.

Connectivity of mangrove forests at global scale inferred with biophysical modeling for (A) 10 d, (B) 70 d and (C) 180 d (10 d represent the minimum, 70 d the average and 180 d the 90th percentile of mangrove propagule duration; inferred by (29) to the closest 10th day with information available on SI Appendix, Table S2). Betweenness centrality is mapped to depict important hubs for stepping-stone connectivity. Information on the distance traveled, probability of connectivity and number of connectivity events are also shown.

Isolation by distance models (IBD) were significant for 72.88% of the linear regression models (significant P values in 43 out of 59 models), with an average R-square of 0.33 ± 0.21 and a Person’s correlation of 0.55 ± 0.18 (Fig. 3 and SI Appendix, Table S1 and Supplement 2, (30)). Multigenerational stepping-stone oceanographic connectivity (OC) estimates considering centrality indices were significant for 94.91% of the linear regression models (significant P-values in 56 out of 59 models), with an average R-square of 0.44 ± 0.23 (up to 0.97) and a Person’s correlation of 0.65 ± 0.17 (Fig. 3 and SI Appendix, Table S1 and Supplement 2, (30)). The remaining non-significant models had marginal P values ranging between 0.06 and 0.08 (SI Appendix, Table S1). All individual comparisons between IBD and OC models at the study, species, and genetic marker levels are presented in Supplement 2 (30).

Comparing Akaike information criterion (AIC) values between approaches showed that 100% of the OC models (considering centrality) explained more variability in genetic differentiation data than the IBD. This resulted in an average gain in R-square of 0.18 ± 0.16 (up to 0.76) and in Person’s correlation of 0.21 ± 0.19 (up to 0.78; Fig. 4). The same general patterns were held when PD was varied to capture the uncertainty (SI Appendix, Table S3). The use of centrality indices enhanced the goodness of fit of 100% of the models, by an average R-square of 0.05 ± 0.07 (up to 0.42; SI Appendix, Table S1; Supplement 4, (30)).

Fig. 4.

Improvement in modeling performance by including oceanographic connectivity, considering the average propagule duration (PD) of each species (to the closest 10th day, with information available on SI Appendix, Table S2), over isolation by distance. Modeled species, region, genetic marker, propagule duration, number of studies, sampled populations (at the 8.5 km spatial resolution of the biophysical model), average of linear regression (R²) and Pearson´s correlation (r) of oceanographic connectivity, and the average (black circles) and minimum and maximum (gray circles) difference of R-square and Person’s correlation coefficient between oceanographic connectivity (considering centrality indices) and isolation by distance models. For results without centrality indices, please refer to SI Appendix, Table S1. Regions of mangrove occurrence in each study are abbreviated as IWP (Indo West Pacific) and AEP (Atlantic East Pacific). Genetic markers are shown as Mic (Microsatellite), Rgs (rubisco gene and spacers), SCNG (single-copy nuclear gene), SNPs (single nucleotide polymorphisms), RAPD (random amplified polymorphic DNA), AFLP (amplified fragment length polymorphism), and Ae (Allozyme).

Discussion

The global distribution of intra-specific diversity of mangrove forests can be explained by multigenerational stepping-stone oceanographic connectivity, regardless of the species, region, or genetic marker. The modeling approach, accounting for the temporal and spatial complexity of oceanographic processes, as well as key ecological dispersal traits, improved the ability to explain genetic differentiation when compared to more simplistic, yet widely used isolation by distance models. The integration of centrality indices derived from graph theory showed the potential role of particular sites as hubs for connectivity (Fig. 3), which may bridge populations across large water masses, facilitating gene flow and therefore reducing genetic differentiation levels.

The biophysical modeling based on high-resolution ocean circulation data coupled with graph theory showed how taxon-specific traits may strongly impact the potential for population dispersal, as well as the overall levels of genetic differentiation (SI Appendix, Fig. S1). Mangrove species have contrasting dispersal periods (29) that can result in oceanographic connectivity events at scales differing by more than one order of magnitude (from an average of 55.12 km to 715.44 km between 10 and 180 d of propagule duration). The number of population connectivity events also increased with dispersal potential, a result that is somewhat expected in biophysical modeling simulation connectivity of advected passive particles (31). However, the relationship between increasing dispersal periods and potential connectivity did not show a linear relationship, but rather a positively skewed one, independently of the propagule duration considered, which can result in connectivity events of up to 10,582 km. This is in line with theoretical dispersal kernels (29), as well as empirical evidence suggesting that mangrove propagules do not tend to disperse far away from source locations (32, 10, 11), yet rare long-distance dispersal events can occur, as proposed across the Atlantic (33, 34), Indian (35), and Pacific Oceans (33, 35). These general traits of mangroves’ dispersal ecology may be key in driving intra-specific biodiversity levels. While limited dispersal potential explains the strong genetic structure found across species distribution ranges (36), the rare events at larger scales must be crucial during the first colonization of distant habitats (10).

Comparing the inferred patterns of oceanographic connectivity against empirical genetic data per species, study, and genetic marker corroborated our main hypothesis. Connectivity estimates explained genetic differentiation in nearly all tests performed (95%), and, compared to isolation by distance models, the approach improved the explanatory power by an average R-square of 0.18 ± 0.16 (up to 0.76) and Person’s correlation of 0.21 ± 0.19 (up to 0.77). This result builds upon existing studies performed with a limited number of species and regions (21, 37). Such an improvement in modeling genetic differentiation likely resulted from considering the combined effect of realistic oceanographic features (17, 38), the directionality of ocean currents (16), and the potential for stepping-stone connectivity (17). On the one hand, fronts, eddies, and bifurcations generate oceanographic barriers that can isolate neighboring populations (39) and increase their genetic differentiation (16, 17), as verified for Rhizophora species considered in our models in regions of Southeast Asia (14), Eastern Atlantic (22), Western Indian Ocean (39). On the other hand, the directionality of ocean currents can further play a role in the relative degree of gene flow, by promoting either unidirectional or bidirectional connectivity between pairs of populations (16, 40). Multigenerational stepping-stone connectivity is also key while performing comparisons against genetic data, because differentiation levels reflect both colonization and gene flow events acting across large temporal scales, and direct oceanographic transport between sampled populations is very unlikely. Finally, our analyses also considered the add-on effect of centrality indices, which improved all models explaining genetic data (average R-square of 0.06 ± 0.07, and maximum of 0.42). Regions of high centrality provided information about particular sites functioning as stepping-stones between populations and with a higher degree of reachability (41, 42), which may promote gene flow and reduce genetic differentiation levels [i.e., hubs for genetic connectivity (43)]. Conversely, regions of lower centrality are linked to dispersal barriers and, therefore, genetic discontinuities (16). Mapping centrality allowed verifying that the Indo-West Pacific concentrates most hubs for stepping-stone connectivity, matching mangroves’ hotspot of interspecific diversity (3), and that islands and offshore regions are key for long-distance dispersal events, bridging populations of species with longer dispersal periods.

Despite the general agreement found between oceanographic connectivity and population genetic differentiation, our models were not significant in 3 out of the 59 tests (yet P-values were marginally significant), and even in those that were significant, not all variability was explained (~43% variability explained, on average), most likely due to the effect of additional drivers not related to present-day oceanographic processes. For instance, the direction and intensity of ocean currents could have changed in the past (44), shaping distinct gene flow patterns and barriers to dispersal (e.g., the Central American Isthmus) that could have left a strong imprint in the distribution of intra-specific diversity (45). Alongside population connectivity, genetic diversity can also be structured by past demographic processes. The sea-level fluctuations during glacial-interglacial cycles of the Quaternary might have repeatedly shifted mangroves’ distribution, also influencing genetic diversity (2). More recently, ongoing climate changes (24, 46) and human-induced disturbance (47, 48) may have also impacted genetic levels owing to population size reductions in the face of unfavorable niche conditions or by founder effects at the leading edge of new colonizations (49). Alternative means of dispersal, such as wind and insect-driven pollination (21, 50), and human-induced reforestation (51), can also shape the distribution of genetic diversity in ways (i.e., distance and direction) that may contrast with those from the prevalent oceanographic currents. Finally, the available genetic data compiled by systematic search of the literature did not cover all mangrove species (20 out of ~80 species), and the genetic markers used could have also biased the estimates of population differentiation. For instance, some markers like mitochondrial DNA have reduced resolution and can be prone to selecting specific genotypes (52), potentially confounding the actual degree of population connectivity.

Overall, our biophysical modeling shows that oceanographic connectivity explains the population differentiation of mangrove forests at global scales. This is of high relevance not only for biogeography and evolution (15, 29, 31) but also for climate change ecology since ocean currents can modify the link between climate change and biogeographical shifts (53, 54). The future persistence and expansion of mangrove populations, as well as their gene pools, may become a process primarily dependent on the regional patterns of ocean currents, which may facilitate or hinder distributional shifts depending on their agreement with temperature, precipitation (53, 55, 56) and salinity changes (57). Global conservation strategies can also benefit from our findings. The highlighted importance of stepping-stone processes and central populations acting as hubs for gene flow can pave the way for future research pinpointing priority conservation areas for mangroves under the post-2020 global biodiversity framework [e.g., high-betweenness hot spots; (58, 59), a new framework that targets the conservation of genetic diversity of wild populations, as well as the establishment of well-connected networks of protected areas (27)].

Methods

Biophysical Modeling and Graph Theory.

Simulations of oceanographic connectivity between populations of mangrove forests were produced at global scales with a biophysical modeling framework (Supplement 5; (41, 60)) previously validated against genetic and demographic data for numerous coastal species of different ecological groups (e.g., fish, invertebrates, and macroalgae; (16, 17, 54, 61). In this framework, passive particles representing mangrove propagules were advected in a virtual ocean by the patterns of daily ocean currents data derived from the Hybrid Coordinate Ocean Model (HYCOM), a high-resolution hindcast of the northward and eastward components of ocean currents (spatial resolution: 0.08°, approx. 6 to 9 km), which can resolve complex oceanographic processes such as ocean eddies filaments, fronts, and meandering currents (38).

The simulations run across the global distribution of mangrove forests [as inferred from (28) and updated by The United Nations Environment Programme World Conservation Monitoring Centre (UNEP-WCMC); version 1.4, March 2021] for the 5-y period between 2015 and 2019. Individual particles were released on a daily basis from source sites located 8,5 km apart (matching the resolution of the HYCOM model) and were allowed to drift on the ocean surface (mangrove propagules are buoyant and can remain on the ocean surface (62) for up to 180 d (90th percentile of mangrove propagule duration; 29) until they eventually ended up on sink sites, or got lost in the open ocean. The position of individual particles was determined hourly by bilinear interpolation over the velocity fields to smooth trajectories. All trajectories were aggregated to produce matrices of the pairwise probability of connectivity between the source and sink sites by dividing the number of particles released from site i that reached site j, by the total number of particles released from site i. Graph theory was used to capture asymmetrical multigenerational stepping-stone connectivity between sites (31, 41). The graph nodes were structured by the source and sink sites and the edges by the average probability of connectivity per year, i.e., to account for the inter-annual variability of oceanographic patterns (43). Floyd–Marshall’s algorithm found the shortest paths between sites by minimizing the sum of log-transformed probabilities, and their stepping-stone probabilities were determined with a product function (17, 63).

The directionality of oceanographic currents was considered using the index proposed by (16), under the assumption of bidirectionally connected populations having higher gene flow and, therefore, lower genetic differentiation compared to unidirectionally connected populations. Additionally, centrality graph-based indices were determined to consider their potential role in population connectivity (41), namely betweenness and closeness centrality. Betweenness centrality measured the influence of each node on the overall population connectivity, identifying stepping-stones linking nodes and aggregating the overall network (i.e., clusters), while closeness centrality measured the reachability of each node by the length of the shortest path between that node and all other nodes (42). Sites with higher betweenness and/or closeness centrality may promote higher levels of gene flow across populations and therefore reduce genetic differentiation (42).

Testing the Role of Oceanographic Connectivity.

Genetic differentiation data for mangrove forests were compiled through a systematic search of the literature. The Web of Science platform was accessed in December 2022 with the search criteria “genetic diversity and mangrove*” or “genetic differentiation and mangrove*” or “phylogeography and mangrove*”. The studies resulting from the search were reviewed (main article and SI Appendix), and only those reporting estimates of pairwise population genetic differentiation were selected. Information about the sampled species, the location (coordinates) of populations, the number of individuals sampled per population, the molecular markers used in genetic analyses and the pairwise differentiation matrices were retained. These data were then revised based on up-to-date taxonomic information. Specifically, Pelliciera rhizophorae data were separated into P. rhizophorae and Pelliciera benthamii (64) according to the known distribution of each species, while the data of Rhizophora stylosa refer to R. stylosa/mucronata, as these two were grouped into a single species (65).

Because genetic differentiation data cannot be compared, statistical tests exploring the role of oceanographic connectivity were performed individually per species, study, and genetic marker. Three competing linear regression models were performed per dataset by fitting genetic differentiation against 1) a null model of isolation by shortest marine distance between populations (i.e., excluding landmasses), 2) a multigenerational stepping-stone oceanographic connectivity model based on probability estimates, and 3) a multigenerational stepping-stone oceanographic connectivity model considering the role of betweenness and closeness centrality. Pairwise genetic differentiation levels were averaged whenever two or more sampled populations fell within a site (i.e., from where the biophysical modeling runs; sites located 8.5 km apart). Oceanographic connectivity per species was obtained by restricting the outputs of the biophysical model to the known PD of each species (SI Appendix, Table S2). This information was averaged whenever the same species had multiple PD periods reported in the literature, and for species with no PD reported, the average PD of the genus was considered (SI Appendix, Tables S1 and S2; 66). Datasets were discarded whenever the number of observations was below eight (i.e., pairs of connected populations), following recommendations for linear regression (Supplement 2 and 3; (30, 67)).

Betweenness and closeness centrality were considered in the regression models as the average of each pair of sampled populations.

The improvement in explaining genetic data by adding oceanographic connectivity was determined with the AIC, the goodness of fit (adjusted R-square), and Pearson’s correlation coefficient (Fig. 3).

A sensitivity analysis (SI Appendix, Table S3) showed how the uncertainty of PD estimates did not influence the capacity of the biophysical modelling to explain population genetic differentiation. This analysis was performed per study, species, and genetic marker by testing the improvement of adding oceanographic connectivity when using additional PD values around the average PD of species. In particular, ± 30 PD days (10-d steps) were considered, as this is the average difference found in the literature for multiple PD periods reported for the same species (SI Appendix, Table S2).

All analyses were performed in R (R Development Core Team, 2018) using the R packages “bigmemory,” “data.table,” “dismo,” “doparallel,” “geosphere,” “gstat,” “igraph,” “raster,” and “vegan.” The source code for biophysical modeling is available in Supplementary Information 5 (60).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Supplementary Information 2, 3, and 4 described in the Methods and Results section are openly available in Figshare (30). The source code for biophysical modeling (Supplement 5) may be accessed on GitHub (60).