Channel network structure determines genetic connectivity of landward–seaward Avicennia marina populations in a tropical bay

Abstract Mangrove ecosystems along the East African coast are often characterized by a disjunct zonation pattern of seaward and landward Avicennia marina trees. This disjunct zonation may be maintained through different positions in the tidal frame, yielding different dispersal settings. The spatial configuration of the landscape and coastal processes such as tides and waves is expected to largely influence the extent of propagule transport and subsequent regeneration. We hypothesized that landward sites would keep a stronger genetic structure over short distances in comparison with enhanced gene flow among regularly flooded seaward fringes. We tested this hypothesis from densely vegetated A. marina transects of a well‐documented mangrove system (Gazi Bay, Kenya) and estimated local gene flow and kinship‐based fine‐scale genetic structure. Ten polymorphic microsatellite markers in 457 A. marina trees revealed no overall significant difference in levels of allele or gene diversities between sites that differ in hydrological proximity. Genetic structure and connectivity of A. marina populations however indicated an overall effect of geographic distance and revealed a pronounced distinction between channels and topographic setting. Migration models allowed to infer gene flow directionality among channels, and indicated a bidirectional steppingstone between seaward and nearest located landward stands. Admixed gene pools without any fine‐scale structure were found within the wider and more exposed Kidogoweni channel, suggesting open systems. Elevated kinship values and structure over 5 to 20 m distance were only detected in two distant landward and seaward transects near the mouth of the Mkurumuji River, indicating local retention and establishment. Overall, our findings show that patterns of A. marina connectivity are explained by hydrological proximity, channel network structure, and hydrokinetic energy, rather than just their positioning as disjunct landward or seaward zones.


| INTRODUC TI ON
Mangroves represent structurally and functionally characteristic forests, predominantly along tropical and subtropical coastlines and mostly occupying sheltered (low-gradient) tidal flats in estuaries, deltas, and lagoons. Exposed to the dynamic conditions of these intertidal environments (e.g., tidal flooding, salinity fluctuations), mangrove trees and shrubs display a number of adaptive strategies such as salt-secreting glands, aerial roots, and the production of hydrochorous propagules (i.e., dispersal units) (Tomlinson, 2016).
Transport of these propagules allows for gene flow within and between mangrove populations, and determines the ability of species to track climate-driven changes in the spatial distribution of suitable habitat. Estimates of gene flow and knowledge on the factors that determine the distribution of genetic diversity is not only of theoretical interest, but can be useful to inform management and conservation of these coastal ecosystems (Balbar & Metaxas, 2019;Carr et al., 2017;Nakajima et al., 2017;Pujolar et al., 2013;Schwarzbach & Ricklefs, 2001).  Rabinowitz (1978) proposed that the interacting effects of water depth with species-specific propagule traits ("tidal sorting") might explain the differential distribution (zonation) of mangrove species along the tidal gradient. Similarly, the position of populations relative to open-water channels and the spatial arrangement of these channels and their tidal currents may determine rates of hydrological connectivity and influence population genetic structure (Hughes et al., 2009;Pilger et al., 2017;Sander et al., 2018;Thomaz et al., 2016). However, while "riverscape genetics" is an active field of research and the importance of tidal inundation, dispersal traits, and establishment in determining mangrove forest structure (intertidal zonation) has been studied extensively (e.g., Clarke et al., 2001;Jiménez & Sauter, 1991;Rabinowitz, 1978;Sousa et al., 2007;Wang et al., 2019), few studies have linked these factors to the local (<10 km) and fine-scale (i.e., within-population) spatial genetic structure of mangroves.
Previous studies found correlations between genetic differentiation and geographic distance (e.g., Binks et al., 2018;Cerón-Souza et al., 2015;Mori et al., 2015), a correlation known as "isolation by distance" (Rousset, 1997;Wright, 1943). However, aspects such as founding history, variations in dispersal traits, and interactions with spatially heterogeneous landscapes (i.e., spatial variation in transport resistance) have challenged the explanatory power of this model (e.g., Dodd et al., 2002;Millán-Aguillar et al., 2016;Wee et al., 2014). As a result, alternative hypotheses have been proposed that account for dispersal limitation (isolation-by-dispersal limitation; Orsini et al., 2013), the effect of ecological and geographical barriers (isolation-by-barrier; Ricketts, 2001), or incorporate resistance surfaces that reflect landscape properties ("roughness") (isolation by resistance; McRae, 2006). For example, fine-scale spatial genetic structure was observed in two Avicennia species along the Brazilian coast and explained by restricted pollen and propagule dispersal (Mori et al., 2015). Cisneros-de la Cruz However, fine-scale spatial genetic structure was not observed in all of the estuaries studied, and where absent, was explained by the recent recolonization of areas that are cleared for coastal development (Ngeve, Van der Stocken, Menemenlis, et al., 2017). While these studies help clarifying the role of propagule (dispersal) traits and interactions with the spatial complexity of the landscape, only a handful of studies considered the role of intertidal position and local hydrological system. Based on a preliminary genetic analysis, Dahdouh-Guebas et al. (2004) found few allele frequency differences between landward and seaward Avicennia marina (Forsk.) Vierh. (Acanthaceae) zones in a Kenyan mangrove forest, indicating that there might be less genetic interchange between these intertidal zones than within each zone. Recently, Chablé Iuit et al. (2020) studied the genetic diversity and structure of R. mangle in the southern part of Quintana Roo (Mexico) and reported that the fine-scale spatial genetic structure reflects contemporary processes such as restricted propagule dispersal and local hydrology.
The goal of this study is to characterize the genetic structure and diversity of the mangrove species A. marina in a coastal bay in Kenya, focusing on the effect of intertidal position and the structure of the area's channel network. More specifically, we aim to (a) analyze the genetic structure between seaward and landward mangrove patches, positioned along a same transect perpendicular to a channel; (b) estimate patterns of connectivity in the light of channel network structure; and (c) examine the fine-scale spatial genetic structure of mangrove patches located along the same channel. The local-and fine-scale spatial genetic structure of A. marina can be K E Y W O R D S Avicennia, gene flow models, genetic structure, mangrove, microsatellites hypothesized to maintain higher levels of connectivity among the more regularly flooded seaward sites and higher kinship values (relatedness) with a stronger structure over short distances in higher intertidal (landward) sites. We test this hypothesis of confinement in landward sites versus open connectivity between seaward sites using densely vegetated transects in a well-documented mangrove area (Gazi Bay, Kenya). The study site and species present an ideal case to undertake this study, given the disjunct (landward-seaward) pattern of A. marina in the area and the regional setting that is characterized by a series of open-water channels. Throughout this manuscript, we use the term "landward" to refer to "higher intertidal" and "seaward" for "lower intertidal." It is important to note that since channels have different orientations relative to the coastline, seaward does not necessarily mean oriented toward the sea. Instead, the terms "landward" and "seaward" reflect different geographical proximities (distant vs. close, respectively) to one of the three major water channels that cross the area's mangrove forest.

| Study area
For the purpose of this study, data were collected in a mangrove forest in Gazi Bay (4°26′S, 39°30′E), about 45 km south-southwest of Mombasa (Figure 1). Climate conditions in the region are influenced by monsoon winds, with long rains during the southeast monsoon (March-July) and short rains during the northeast monsoon (November-December) (Kitheka et al., 1996). Gazi Bay consists of a shallow tropical water system characterized by a mangrove forest that covers >600 ha (Hemminga et al., 1994).  (Gallin et al., 1989), of which the former four are most abundant (Neukermans et al., 2008). The area's hydrological network is characterized by three major channels: Kinondo, Kidogoweni, and Mkurumuji (Figure 1), crossing the mangrove forest in the eastern, central, and southwestern part of the bay, respectively. In contrast to the Kinondo tidal creek, which lacks direct riverine input, the Kidogoweni river estuary receives surface freshwater input from the Kidogoweni River in the northern part of the bay (Kitheka, 1997).
River discharge has seasonal variation, peaks in the wet season, and is higher for the Mkurumuji River than for the Kidogoweni River (Kitheka et al., 1996). Due to the riverine input, salinity in the Kidogoweni river estuary varies greatly, from 2 to 38 PSU, while salinity in the Kinondo tidal creek fluctuates between 22 and 38 PSU, and with salinity maxima (38 PSU) found in the upper parts of these channels during the dry season (Kitheka, 1997). Previous studies in the growth zone of A. marina revealed that salinity can also fluctuate strongly over the course of a tidal cycle (Tonné et al., 2017) and seasonally . Besides riverine influence, the water circulation in the area is controlled predominantly by the strong semi-diurnal tides that enter the bay via a ca. 3.5-km-wide entrance (to the Indian Ocean) in the south (Kruyt & van den Berg, 1993), with a spring tide range of 3.2 m and neap tide range of 1.4 m (Kitheka, 1997). These tides cause strong and reversing currents that are characterized by relatively stronger ebb than flood currents (tidal asymmetry), allowing for net export (Kitheka et al., 1996).

| Study species
Avicennia marina is the most widely distributed of all mangrove species, found across the Indo-Pacific, between latitudes 25°N and 38°S. It has been shown that A. marina is able to grow and reproduce across a relatively broad range of climatic, saline, and tidal conditions (Duke et al., 1998). Salt-excreting glands in its leaves allow the species to better tolerate high salinities compared to other mangrove species (Clough, 1984). In a sedimentation experiment in Gazi Bay, Okello et al. (2014) showed that A. marina trees respond, and may Phenological research in the study area revealed that propagule fall peaks during the wet season (April-May; Wang'ondu et al., 2010).
Shade intolerance and high predation rates on its propagules are believed to limit the distribution of A. marina across the intertidal zone (Smith III, 1987).
The propagules of A. marina consist of a single embryo surrounded by a thin pericarp (Tomlinson, 2016). Reported flotation and viability times for A. marina propagules are relatively short, spanning a couple of days to weeks (Clarke et al., 2001;Clarke & Myerscough, 1991;Steinke, 1986). However, it should be mentioned that the duration of the experimental trials on which these findings are based may be too short to obtain meaningful frequency distributions of these propagule traits, and should ideally Floating periods in other Avicennia species exceed several months (Alleman & Hester, 2011;Rabinowitz, 1978;Van der Stocken et al., 2018), and previous studies reported that the buoyancy of A. marina propagules varies greatly among estuaries (Steinke, 1986

| Sample collection
A total of 457 A. marina individual trees were sampled during July 2017 in eight locations (Table 1)

| Genetic analyses
Prior to population and individual-based data analysis, we tested for genotypic disequilibrium, potential null alleles, and overall resolution of the selected ten microsatellite markers in A. marina. A linkage test between all pairs of loci (1,000 permutations) gave no genotypic disequilibrium at the 0.05 level using FSTAT (v.2.9.3) (Goudet, 2001).
No scoring errors, large allele dropouts, or null alleles were indicated An additional hierarchical AMOVA was performed and F-statistics were calculated, considering three channels (Kinondo, Kidogoweni, and Mkurumuji) as regions, and using 999 random permutations.
Pairwise genetic differentiation (F ST ) was used to produce a PCoA at population level and together with a pairwise geographic Euclidean distance to perform a Mantel test using 1,000 permutations in GenAlEx (v.6.5). Pairwise genotypic differentiation was used to produce a PCoA at individual level. The overall F IJ kinship coefficient (Loiselle et al., 1995) for all sites of A. marina in Gazi Bay was estimated for five mean distance classes at 0.27, 0.44, 1.12, 1.78, and 4.2 km, as were automatically generated when requesting an equal number of pairwise comparisons within each class by SPAGeDi 1.5a (Hardy & Vekemans, 2002) and using the whole sample as a reference. These distance classes represent threshold values as indicated by a Mantel test. Two zonation groups (seaward and landward) and two age groups (young and mixed older) were tested for differences in their A R , H O , H E , F IS, and F ST using 1,000 permutations in FSTAT.
The F IJ kinship coefficient was estimated between reciprocal pairs of seaward and landward sites using SPAGeDi. An assignment of individuals to their "self" population or to another population was done with the "leave-one-out" option in GenAlEx.
A Bayesian clustering analysis at individual level was carried out in STRUCTURE version 2.3.4 (Pritchard et al., 2000) using an admixture model with correlated allele frequencies. The model ran 20 iterations for each K value from 1 to 8; the burn-in period was 100,000 with 500,000 Markov chain Monte Carlo (MCMC) repeats. The optimal K value was inferred with the ΔK statistic (Evanno et al., 2005), from LnP(K), and the Puechmaille (2016) method using Structure Harvester (Earl & von Holdt, 2012) and CLUMPAK (Kopelman et al., 2015), calculated with StructureSelector (Li & Liu, 2018). The software BARRIER 2.2 (Manni et al., 2004)

| Genetic diversity levels
In

| Differentiation between sites
Avicennia marina within Gazi Bay showed an overall AMOVA-  Mean F IS taken from AMOVA. Standard errors are provided between brackets  The connectivity among the mouth of three major water channels in Gazi Bay was best supported from a unidirectional steppingstone model (Table 5b)  1S -*** *** *** *** *** *** ***

| Landward-seaward sites
We found similar levels of genetic diversity for landward and seaward stands, which show no overall significant differentiation be- Tidal asymmetry in the area is characterized by weaker incoming flows than outgoing flows, promoting net export of matter from the mangrove system (Kitheka et al., 1997). This asymmetry in kinetic energy may result in an asymmetry of propagule deposition potential.
Propagule deposition potential may be higher for propagules transported by the weaker flood tide, from seaward to landward stands, Another explanation for these results could be related to different predation rates in landward and seaward sites. In our study area, for example, grapsid crabs (particularly Neosarmatium africanum) were shown to clear nearly 100% of the propagules in landward stands (Dahdouh-Guebas et al., 1997 with fast consumption of A. marina propagules compared to propagules from other mangrove species, and this particularly under A. marina canopy (Van Nedervelde et al., 2015). Hence, despite the large number of propagules produced in A. marina (Clarke, 1992), high predation rates in landward stands in our study area may strongly reduce the number of potential migrants from landward to seaward zones. Overall, these asymmetries in tidal currents (strength and directionality), hydroperiod, and predation rates, are consistent with our MIGRATE results, indicating higher seaward-to-landward than landward-to-seaward migration. A further explanation could be a possible effect on directionality of gene flow caused by pollinator movements . Pollen flow could not be tested from our data on adult trees, as such analysis would require a different design including mother trees and their propagules. However, considering present results, we assume the effect of pollen flow to be of minor importance because connectivity between nearby landward sites was not supported by any MIGRATE model. Pollination over rather short distances has also been suggested by . In their study in the Sydney region (Australia),

| Channel network structure
Instead of a consistent disjunct landward-seaward zonation, spatial genetic patterns within Gazi Bay reflect the local channel network architecture. Our genetic analyses (F ST , PCoA, STRUCTURE, and BARRIER) revealed significant genetic differentiation between sites that are situated along different water channels (Kinondo, Kidogoweni, and Mkurumuji). High longitudinal (i.e., along-channel) connectivity was found for the Kidogoweni channel but not for Mkurumuji (it was not tested along the Kinondo channel, where we considered only one location). Overall, this is consistent with findings that stream channels may act as corridors for dispersal (Johansson et al., 1996;Schmiedel & Tackenberg, 2013) and that connectivity in water-dispersed species is typically low between sites that are not well connected hydrologically (Hughes et al., 2009). The absence of high connectivity along the Mkurumuji most likely reflects the lower influence of tides in this channel compared to Kidogoweni, due to the channel's orientation perpendicular to the directionality of tidal currents. In contrast, strong semi-diurnal tides entering the bay from the south (Kruyt & van den Berg, 1993) may greatly influence dispersal dynamics along the Kidogoweni channel that is oriented predominantly north-south.
The role of river network structure in shaping the genetic variation within and between populations has been commonly investigated for a broad range of freshwater organisms, including fish (Shao et al., 2019;Thomaz et al., 2016), insects (Finn et al., 2006(Finn et al., , 2007, and plants (Sander et al., 2018). In these studies, the observed population genetic patterns are generally described using four connectivity models that predict how populations with different life history traits and dispersal capabilities interact within their structured riverine habitat (Finn et al., 2007;Hughes et al., 2009). Generally, the explanatory power of the river network is expected to be stronger for species with no or limited capacity for terrestrial (among-stream) movement and short floating abilities, and in riverine systems with permanent downstream and confined (unidirectional) water flow (Tonkin et al., 2017). As in riverine networks, dispersal traits, and the physical and hydrodynamic characteristics of the mangrove landscape are important factors regulating dispersal ( Van der Stocken et al., 2019). However, in contrast to riverine systems where the transport of passive propagules is predominantly downstream (Tonkin et al., 2017), coastal processes such as tides and waves add to the complexity of dispersal in these coastal settings (Di Nitto et al., 2008). For example, in the Normanby River estuary (Northeast Australia), Stieglitz and Ridd (2001) found that mangrove propagules were moving upstream from the mangrove fringe, trapped in an axial convergence generated by a density-driven circulation Indeed, in the lower reaches of estuaries, tidal discharge may greatly exceed river discharge (Pritchard, 1956), and drive gene flow upstream during flood tide and downstream during ebb tide. This tidal effect likely explains the overall isolation by distance within the bay, the absence of strong genetic differentiation between mangroves along the same channel, as well as the absence of greater genetic diversity downstream of confluences as predicted in systems dominated by riverine regimes (Thomaz et al., 2016).
The outcome of our MIGRATE analysis supports the hypothesis of channels being connected in a stepwise manner instead of an extensive mixing regime throughout the entire bay. Both unidirectional steppingstone models correspond to the dominant directionality of flood and ebb tide acting through the 3.5-km-wide entrance in the southern part of the bay, slightly more reflecting tidal flood than ebb.
This pattern may potentially be explained by the slower flood tide that may transport propagules gradually upstream within channels, while the stronger ebb tide may result in lower chances of deposition/retention and hence establishment. The younger and diverse stand near the mouth of the Mkurumuji (7S) was not a subset of the older mangrove along Kidogoweni (5S), supporting the abovementioned flood-mediated directionality from south to north. Significant genetic differentiation among channels may result from the tidal asymmetry in the region, characterized by stronger ebb than flood tide and resulting in a net export (Kitheka et al., 1996). This net transport out of the bay may reduce the chance of propagules from one channel being transported out of this channel and then upstream into another channel. Direct among-channel transport has been observed in Amazonian studies, where channel overflow during rainy seasons may facilitate long-distance propagule transport (De Campos et al., 2013). However, while fluctuations in water level over the course of the tidal cycle may allow for regular channel overflow in our study site, direct lateral connectivity between the Kinondo and Kidogoweni channels may occur only during times of elevated tidal height or extreme events. In the three different channels, the potential for among-channel connectivity decreases for sites that are located more upstream as the physical distance between the channels increases (Tonkin et al., 2017). In a way, this is illustrated also by the shape of the mangrove forest's inland fringe, which fans out along the channel network and diverges toward the channel heads ( Figure 1).

| Fine-scale genetic structure within Avicennia stands
The overall Sp value (ca. 0.009) found for A. marina in Gazi Bay is among the ranges reported for outcrossing trees in general (Vekemans & Hardy, 2004  ) may reduce or even nullify its fine-scale structure. On the contrary, elevated kinship values and fine-scale structure were detected within a distance class of 5 to 20 m, only in the two Mkurumuji sites (7S and 8L). Presence or absence of fine-scale spatial genetic structure is most likely due to differences in the relative orientation of both channels with regards to the direction of tidal currents. Kidogoweni is positioned near-parallel to the direction of tidal currents. Hence, the mangrove patches along this channel are much more exposed to the tidal currents than patches along the Mkurumuji channel, which is oriented more or less perpendicular to the tidal currents entering the bay in the south. Even though freshwater discharge can be high for Mkurumuji during the wet season, the riverine energy flux is confined predominantly within the channel with limited overflow, reducing the chance of propagules from 8L to be exported to open waters.

| CON CLUS ION
As a conclusion, the genetic diversity levels were comparable between seaward and landward A. marina mangrove patches and revealed no overall genetic differentiation between these spatially disjunct zones. Gene flow appears to be governed by incoming tides from seaward to nearby landward sites, perpendicular to a channel.
The genetic structure of A. marina within the bay corresponds to the channel network structure, and channel connectivity was most supported by unidirectional steppingstone models corresponding to the dominant directionality of flood and ebb tide. A fine-scale spatial genetic structure was absent for mangrove patches located along the north-south oriented and wide Kidogoweni channel, but was clearly present along a less tidally influenced channel. Overall, our findings show that patterns of A. marina connectivity are explained by hydrological proximity, channel network structure, and hydrokinetic energy, rather than just their positioning as disjunct landward or seaward zones.

ACK N OWLED G M ENTS
We are grateful to the Flemish Interuniversity Council for Cooperation and Development (VLIR-UOS) for the VLIR-ICP scholarship of Abbie Allela Akinyi. The authors acknowledge financial support of the Flemish Interuniversity Council-University Development Cooperation (VLIR-UOS) through the TEAM project "Transboundary coastal processes and human resource utilization patterns as a basis for a Kenya-Tanzania conservation area initiative (Trans-Coast)" (ZEIN2016PR425). We thank the Vrije Universiteit Brussel for research funding (BAS42). Many thanks to Donald Maringa whose help was invaluable in collecting the samples for the study. Finally, we sincerely thank the anonymous reviewers for their helpful comments and suggestions during the review process.

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

DATA AVA I L A B I L I T Y S TAT E M E N T
Sequence data can be found in NCBI GenBank (https://www.