Species dispersal mediates opposing influences of a branching network on 1 genetic variation in a metapopulation 2 3

14 In nature, ubiquitous fractal networks can have two but opposing influences, by increasing 15 distal and confluent habitats, respectively, under raising branching complexity on 16 metapopulations’ genetic structure, although this remains poorly understood, particularly 17 regarding the roles of species-specific traits. In this study, we evaluated the integrated 18 influences of network complexity and species dispersal mode/ability on genetic divergence 19 among populations at the catchment scale, using a theoretical framework with empirical 20 genetic data from four sympatric stream macroinvertebrate species. Empirical patterns of 21 spatial genetic structure were attributed to dispersal ability and the species’ habitat 22 specialisation levels. Our theoretical evidence showed that both greater landscape connectivity 23 (via shorter watercourse distance) and greater isolation of distal habitats (e.g. headwater 24 streams) occur in the more-branched networks. These two spatial features have negative and 25 positve influences on genetic divergence, respectively, with their relative importance varying in 26 different species. Watersheds harbouring a higher number of local populations have larger 27 genetic divergence of metapopulations. Downstreamand upstream-biased asymmetric 28 dispersals dictate increases and declines, respectively, in genetic divergence. In addition, distal 29 populations (e.g. in headwaters) have higher genetic independence between themselves under 30 higher levels of downstream-biased asymmetry. A strong association between species features 31 and evolutionary processes (gene flow and genetic drift) mediates the pervasive influences of 32

In the Natori and Nanakita Rivers in northeastern Japan (integrated catchment area c. 1200 km 2 ; 116 City with a population of one million. Approximately 60% of this area is forested and 120 mountainous. Two major reservoir dams (Kamafusa and Okura dams) are located there. The 121 regional lowlands are farmlands (13%, primarily with rice paddy fields) and a mixture of 122 residential and commercial areas (11%). 123 For both empirical and theoretical evidence, we used genetic data of neutral amplified 124 fragment length polymorphism (AFLP) markers from four macroinvertebrate species in this 125 catchment (Watanabe et al., 2014). Three species were caddisflies (Trichoptera), namely, 126 Hydropsyche orientalis, Stenopsyche marmorata and Hydropsyche albicephala, while the 127 fourth was a mayfly, Ephemera japonica (Ephemeroptera). In this integrated catchment, the 128 species distributions vary considerably, from the widespread H. orientalis to the narrowly 129 distributed E. japonica (Fig. 1). These species have similar ecological functions in river 130 ecosystems by feeding on fine organic matter (< 1 mm diameter). Approximately 18 to 20 131 individuals collected at each sampling site were genotyped (128 to 473 polymorphic AFLP loci 132 for each species). Based on the locus-specific genetic differentiation across this catchment, 133 non-neutral loci identified by DFDIST (Beaumont & Nichols, 1996) and/or BayeScan (Foll & 134 Gaggiotti, 2008) were removed, and 98 to 449 neutral AFLP loci for each species (Fig. S1) 135 were retained and used for this study. Detailed protocols on the identification of non-neutral 136 loci are described in our previous report (Watanabe et al., 2014). is obtained from the inverse logit transformation as follows: 163 where part a or b describes the autocovariance, with the variance Before the simulations, we created artificial river networks with varying branching 190 complexities (Terui et al., 2018). The river networks were made up of nodes with scale length 191 e, with each node representing a local population. These nodes were assigned to be either 192 branching (or an upstream terminal) or non-branching with a probability of P or 1 -P, 193 respectively. As a series of non-branching nodes terminated at a branching (or terminal) node. 194 The individual segments (watercourse stretches) were the geometric random variables with 195 branching probability P. Before merging the segments to create a river network, the drawing 196 process was repeated until the targeted number of notes (the number of local populations) and 197 an odd number of segments were reached. To create the river network, these segments were put 198 together as a pool merged hierarchically as follows (Fig. S2): Step 1): One segment was 199 randomly selected as the root and its upstream end was merged with the downstream end of 200 another two random segment selections. In this status, the semi-complete network had two 201 unmerged upstream ends each for the next possible merger.
Step 2) Two more segments were 202 randomly selected and their downstream ends were merged together to the random draw one of 203 two (or even more at subsequent steps) unmerged upstream ends of the semi-complete network. 204 Step 3) Step 2 was repeated until there were no available segments in the pool. and S. marmorata than in the other two species with narrower habitat distributions (Fig. S1). 244 The pairwise genetic difference between empirical local populations tended to increase with 245 their watercourse distances throughout the four macroinvertebrate species (Fig. S4). Despite 246 substantial variation in the scale parameter, amplifying the isolating effect of distance across 247 study species (Fig. S5), there was a consistent decline in the genetic correlation between 248 populations (the covariance Ω divided by the variance ; see Formula 4) with 249 the increasing distance between local populations (Fig. S6). In addition, there was a greater 250 decline in the genetic correlation with distance in the widely distributed H. orientalis than in 251 the other species. 252 253

River-branching influence 254
We describe changes in the two landscape spatial configurations (fraction of any two local 255 populations being streamflow-disconnected in all combinations and mean watercourse distance 256 between local populations) with the increasing branching probability (P) in river networks (Fig.  257 2). Situations in which any two local populations are streamflow-disconnected (e.g. in different 258 tributaries) across metapopulations occur at higher rates in heavily branched river networks. ) of the river network across four species (Fig.  265 3). Species-specific responses to the influence of river branching were identified. For example, 266 increased branching probability decreased the genetic divergence of the metapopulation for 267 three caddisflies (H. orientalis, S. marmorata and H. albicephala), but in the mayfly E. 268 japonica, the opposite response (higher genetic divergence) occurred. In addition, both a low 269 level and variation of genetic divergence are less likely to occur in the generalist H. orientalis 270 than in the other three speices. The findings showed that the metapopulation size was 271 positively correlated to genetic divergence in all species. According to the GB modelling 272 results, the relative importance of streamflow-disconnected habitats, compared to the landscape 273 connectivity via a shorter watercourse distance, was higher in the mayfly E. japonica than in 274 the other three caddisfly species (Table 1). 275 The genetic performances, varying across species, were illustrated by how these model 276 parameters related to upstream and downstream dispersals take effect on the genetic 277 divergences (Fig. 4 and 5). Branching complexity has various impacts on genetic divergence,  (Fig. 4). These, in turn, indicated higher and lower isolation 283 effects of watercourse distance between local populations, respectively. Lower genetic 284 divergence levels occurred in more-branched networks when there was higher variance related 285 to upstream movement (ߪ ଶ ) than downstream movement (ߪ ଶ ) (Fig. 5). In other words, the 286 populations in the distal branches (e.g. headwaters) have relatively strong genetic covariation 287 between themselves, particularly in complex river networks. In addition, river branching has 288 the opposite (positive) influence on the genetic divergences when ߪ ଶ is lower than or equal to 289 ߪ ଶ (Fig. 5). 290

291
In this study, we explored the integrated role of landscape architecture and species 292 ecological strategy in shaping genetic divergence at neutral loci. We compared the landscape 293 genetics of sympatric macroinvertebrate species in river networks, based on our Bayesian 294 model, explicitly accounting for the effects of evolutionary processes among components of 295 metapopulations on the spatial genetic structure. This model indicated that river-network 296 connectivity predicted spatial genetic structures in four macroinvertebrate species. In addition, 297 their empirical structuring patterns were determined by the species' intrinsic factors 298 parameterised in this model. In this case, these factors can be associated with dispersal ability 299 and mode, species distribution and effective population size (associated with the genetic 300 variance in our model) in characterising relationships between genetic divergence and 301 landscape connectivity, as shown in the discussion below (see the subsequent section 302 'Importance of species' intrinsic factors'). 303 In our simulations, these intrinsic factors could cause varying levels of overall genetic 304 differentiation in river networks and induced increased river branching to have different or 305 even opposite effects. Moreover, greater landscape connectivity (via shortened watercourse 306 distance) and higher distal habitat isolation (e.g. headwater streams) simultaneously occur in 307 more-branched river networks and have countervailing influences on genetic divergence; they 308 also have different levels of relative importance across these sympatric species. This can 309 provide extensive insights into other complex networks (e.g. highly fragmented landscapes or 310 those with corridors via ocean and atmospheric circulation). Our empirical and theoretical 311 results highlight the fundamental importance of considering species' biological traits, which 312 make different contributions to genetic connectivity, for the successful management of 313 ecological corridors. 314 315

River branching and metapopulation genetic divergence 316
In dendritic river networks, our simulation results showed a species-dependent change in 317 global genetic differentiation levels occurring with increases in network complexity and the 318 number of local populations in a metapopulation. We theoretically showed that the differential 319 downstream and upstream gene flows we considered in the model can act together to generate 320 such relationships. Our finding that increased populations in the river network enhanced 321 genetic differentiation is consistent with previous theoretical evidence ( Thomaz et al., 2016). 322 River branching's role has been documented, to some extent, in riverscape genetics, when 323 higher genetic diversity is observed in downstream populations than in upstream ones 324 to low downstream-biased asymmetry and the opposite (negative) influence occurring under 332 increased river branching. In addition, our comprehensive consideration of various branching 333 river network topologies in simulations helped us to demonstrate the existence of opposing 334 influences co-occurring under branching complexity. In one early theoretical study, not 335 considering dispersal asymmetry (analogous to equal downstream and upstream dispersals in 336 our study), the dendritic network structure was also documented to promote low genetic 337 distances under high riverscape connectivity (Labonne et al., 2008). 338 339

Importance of species' intrinsic factors 340
The river networks' architecture can be one important extrinsic factor for explaining the 341 observed and simulated genetic patterns, but there was strong variation among species with 342 different intrinsic factors in our study. This finding was also previously observed; for example, 343 two sympatric salmonid species were found to have remarkably different spawning locations, 344 mating systems and population sizes, and these biological traits mediated the influences of 345 riverscape features shaping their dispersal and genetic divergence in the Clark Fork River in 346 the USA (Whiteley et al., 2004). For each upstream and downstream dispersal tendency in our 347