Spatio-temporal patterns of multi-trophic biodiversity and food-web characteristics across a river catchment

Accurate characterisation of ecological communities with respect to their biodiversity and food-web structure is essential for conservation. However, combined empirical study of biodiversity and multi-trophic food webs at a large spatial and temporal resolution has been prohibited by the lack of appropriate access to such data from natural systems. Here, we assessed biodiversity and food-web characteristics across a 700 km2 riverine network through time using environmental DNA. We find contrasting biodiversity patterns, with richness (α-diversity) of fish increasing towards downstream positions within the catchment, while freshwater bacteria and invertebrates having an invariant and minimal decrease in richness, respectively, with downstream position. Food-web characteristics, such as link density and nestedness, however, were relatively conserved across space, but varied over season. Patterns of biodiversity across major taxonomic groups are thus not directly scalable to food-web structures at the same spatial and temporal scales, indicating that effective conservation measures must consider them jointly.


Introduction 40 41
The study of biodiversity patterns 1-3 and the characterisation of food-web structures 4,5 are 42 essential, yet often disconnected goals in ecology. Understanding these patterns is not only of 43 fundamental interest, but also needed to predict stability, functioning and resilience of natural 44 ecosystems and to bend the curve of biodiversity loss in the context of anthropogenic pressures 45 including contemporary global change 6 . 46 47 Studies on biodiversity predominantly focus on analyses of a-, band g -diversity and possible 48 underlying fundamental drivers of their spatial or temporal patterns 7 . Freshwater rivers are 49 highly spatially structured systems 8-10 in which theoretical and empirical studies have identified 50 characteristic patterns of biodiversity for specific groups. For example, fish a-diversity has been 51 found to increase with distance downstream 11 , whereas headwaters often show high endemic 52 bacterial species richness 12 . Aquatic invertebrate biodiversity exhibits more complicated overall 53 patterns with disproportionately high biodiversity being found in headwaters 13 and a significant 54 increase in biodiversity linked to catchment size 14 . However, these group-specific biodiversity 55 patterns have been mostly studied in isolation from one another, although species are present 56 within the same system and trophically interact with each other. Indeed, recent theoretical work 57 shows that contrasting patterns driven by species' resource competition are possible 15 . Therefore, 58 to ensure optimal strategies for conservation and understanding of biodiversity patterns across 59 different organismal groups, an ensemble approach integrating major taxonomic and trophic 60 groups is crucial to reveal how species are linked through trophic interactions 16 . 61 62 Blackman et al.

5
Trophic interactions and food webs by definition encompass multiple groups of organisms. 63 Individual freshwater food webs are well-resolved 17 , and often exhibiting distinct features, such 64 as highly nested structures 18 and prevalent omnivory 17,19 . Nevertheless, food-web studies often 65 have a localised perspective due to methodological limitations of sampling food-web interactions 66 and organismal occurrence in a standardized and comparable manner across different places and 67 organismal groups [20][21][22] . Due to the same reason, these studies also tend to focus on simple spatial 68 and environmental gradients 23 or temporal change 24 when spatio-temporal influences should be 69 considered in conjunction 25,26 . This is particularly problematic in freshwater riverine ecosystems, To effectively conserve riverine biodiversity, we must encompass spatio-temporal variation of 77 multiple trophic levels to understand the underlying dynamics of both biodiversity patterns and 78 food-web characteristics 4,25,26 . In particular, molecular monitoring techniques may now provide a 79 suitable solution to break through the above-mentioned methodological constraints caused by 80 sampling based primarily on species sight or capture. Environmental DNA, or eDNA, is the 81 collection of DNA extracted from an environmental sample such as water, air or sediment 29 . By 82 collecting eDNA we can screen samples for multiple taxonomic groups via metabarcoding 30 , 83 thereby creating a biodiversity assessment suitable for food-web reconstruction. 84 85 space. With this approach, variation in food-web structures and function feeding groups emerge 131 from the spatio-temporal differences in genus composition (See Methods for further details). 132

133
To assess the relationship between α-diversity of each group (fish, invertebrates and bacteria) 134 and food-web structures with site location within the catchment, we ran linear mixed model 135 analysis. Drainage area (km 2 ) was log transformed to fit model assumptions. For each dependent 136 variable, drainage area and season were the fixed effects, while site was the random effect. We 137 determined the overall effect of both factors using analysis of variance and contrast testing of 138 estimated marginal means to determine the influence of seasonal changes on all α-diversity and 139 food-web elements (see Methods for details). To examine the effect of river distance on β-140 diversity we constructed a matrix of pairwise distances for sites that were connected along the 141 fluvial network and to examine the effect of river distance on β-diversity we performed a Mantel 142 test (see Methods and Supplementary information). 143 144 Spatial and temporal biodiversity patterns 145 In total we detected 374 genera across all organismal groups associated with freshwater, 146 including 12 fish genera, 80 invertebrate genera and 282 bacteria genera. When combining all 147 seasons, α-diversity (genus richness) ranged between 8-96 genera with all taxonomic groups 148 combined (Fig. 1). Over the different seasons, mean local α-diversity was 70 (range 10-92) in 149 spring, 48 (range 8-85) in summer, and 63 (range 19-96) in autumn (See Supporting Information 150 Fig. S1 and Table S3). We used mixed models to assess the influence of drainage area and 151 season on the local α-diversity of each group (Fig. 3, for model output see Supporting 152 Information Table S4-S6). Of the three taxonomic groups, only fish α-diversity significantly 153 increased with the size of the drainage area (p < 0.001, Fig. 3a and Supporting Information for 154 model output Table S4-S6), while for bacteria and invertebrates there was no significant  155 relationship (p = 0.670 and p = 0.239, respectively, Fig. 3c and 3e). There was, however, a 156 significant seasonal effect in α-diversity on invertebrate and bacteria genera (p < 0.001 for both 157 groups, Fig. 3d and 3f, see Supporting Information Table S4-S6). The influence of seasons can 158 be seen further in the contrast testing (Fig. 3b, d and f and Supporting Information Table S7 for 159 full contrast testing results), which shows invertebrate α-diversity in spring was significantly 160 higher compared to summer and autumn (p < 0.001, Fig. 3d), and bacteria α-diversity in summer 161 was significantly lower compared to spring and autumn (p < 0.001, Fig. 3f). Fish α-diversity 162 however did not vary significantly between seasons, with the mean number of 2 (range 0-6) fish 163 genera remaining constant (Fig. 3b).  Table S8). Further partitioned analyses on taxon replacement and loss revealed 172 contrasting patterns. Taxon replacement between sites increased over river distance for all 173 groups in most seasons (significant or marginally significant, see Table S8), apart from fish in 174 spring (Mantel statistics 0.07, p = 0.891) invertebrates in spring (Mantel statistics 0.066, p = 175 0.051) and bacteria in spring and summer (Mantel statistics 0.016, p = 0.34 and Mantel statistics 0.07, p = 0.068, respectively). In contrast, taxon loss was only found to significantly increase 177 over river distance for fish in spring and autumn (Mantel statistics 0.219, p = 0.001 and 0.059, p 178 = 0.05, respectively) and bacteria in summer (Mantel statistics 0.089, p < 0.05). 179 180 Spatial and temporal changes in food-web structure and functional characteristics 181 We examined commonly used food-web structural characteristics (link density, connectance, 182 nestedness, omnivory, coherence, number of links, modularity, and robustness), as well as 183 functional characteristics (functional diversity and redundancy). Functional characteristics were 184 examined by using the designated functional feeding groups (e.g., shredders and omnivorous 185 fish), based on specialised feeding behaviours (  Table S4 -S6 for full results), indicating that seasonal 194 variation is of more importance in food-web dynamics than site location within the riverine 195 network. To examine seasonal variation further, we carried out contrast testing, which showed a 196 range of seasonal change in food-web dynamics. In particular, spring had significantly higher 197 link density than summer and autumn (p < 0.001 and p < 0.001, respectively, Fig. 4a); autumn 198 had significantly lower connectance than spring and summer (p < 0.001 and p < 0.001, 199 respectively, Fig. 4b), and the same is true for nestedness (p < 0.01 and p < 0.01, respectively, 200 Studies of biodiversity and food-web assemblages in riverine networks are often constrained to 216 local scales or aggregated to a single time point, which in essence fails to capture the spatial 217 processes and the temporal fluctuations that together play a key role in community dynamics 218 present within a river network 8 . Our study is a first assessment utilising data derived from eDNA 219 metabarcoding to detect patterns of biodiversity and food-web characteristics across three major 220 taxonomic groups, namely fish, invertebrates and bacteria, in a whole river network at a spatial 221 and temporally large scale. In our study we find contrasting patterns of biodiversity across these 222 groups, which indicates different mechanisms may shape these organismal communities. By 223 using a metaweb approach, we showed a strong signature of seasonality in food-web structures 224 across the river network. We also showed that functional characteristics are influenced by both 225 spatial and temporal changes. Overall, our study supports the need to include both spatial and 226 temporal scales in order to understand changes in ecosystems, particularly as we see increased 227 effects of contemporary global change. 228 229 Aquatic biodiversity is subject to fluvial influences within a dendritic network, whereby spatial 230 patterns of α-diversity and β-diversity are known for key groups, such as fish, invertebrates and 231 microbes 11-14,34 . Our data is congruent with previous studies on fish diversity 11,33 , in that both α-232 diversity significantly increased downstream, and β-diversity (community dissimilarity) 233 significantly increased with river distance in all three seasons. However, patterns for 234 invertebrates and bacteria differed from the predictions, and exhibited no significant influence of 235 drainage size on α-diversity but were influenced by seasonal change. For invertebrates, seasonal 236 variation can be linked directly to the emergence of the non-aquatic adult life stage of several 237 macroinvertebrate genera we detected (Diptera, Ephemeroptera, Plecoptera and Trichoptera), 238 which takes place in the late spring and summer months 27 , and literally removes these organisms 239 from the aquatic food-web as they become air-bound and often end up in terrestrial food-webs. 240 For the spatial patterns, past studies often looked at spatial scales much larger or much smaller, 241 and thus disparity may be due to a mismatch in scale looked at here. For example, Finn and 242 colleagues 13 suggested headwaters harbour disproportionately high invertebrates compared to 243 lower points in the catchment, but their studied focused on small (1-2) and mid (3-4) stream 244 orders only, whereas the sites in our study ranged from small to much larger rivers (stream 245 orders 1-7, 35 ). Contrastingly, when looking at scales about 50-fold larger, Altermatt and 246 colleagues 14 showed the number of key aquatic invertebrate taxa (Ephemeroptera, Plecoptera and 247 Trichoptera) increased with catchment size, however they also found a combination of local 248 factors (catchment areas, drainage area, elevation and network centrality) had the greatest 249 influence on local invertebrate α-diversity. Possibly, the scale at which we study invertebrate α-250 diversity and contributing local factors falls between such small-scale vs large-scale perspective, 251 but instead we detected regional patterns where α-diversity remains relatively constant 252 throughout the catchment. Similarly, drainage area did not have a significant effect on bacterial 253 α-diversity, and β-diversity increased with river distance in summer only. We can therefore 254 postulate that aquatic bacteria are able to persist as they disperse through the catchment; 255 however, the number of bacteria genera (the α-diversity) that do persist are subject to seasonal 256 influences. Our findings expand on past studies, highlighting the influences of scale and 257 temporal variation on different groups, which is of utmost importance when trying to conserve 258 biodiversity and understand the differing drivers behind these patterns in a river network. 259 260 By resolving the fundamental trophic relationships among broad feeding groups, we established 261 a trophic interaction metaweb for the three taxonomic groups examined in this study. The meta-262 web with genus co-occurrence data thus allowed our investigation of both spatial and temporal 263 influences on the characteristics of local food webs. We found that the most significant factor 264 driving freshwater riverine food-web characteristics was season. All food-web structural 265 characteristics (apart from the number of links and robustness) were lowest in autumn, despite 266 overall α-diversity being lowest in summer. This indicates that the food-web structural variation 267 we detected is not merely reflecting the genera richness dynamics over seasons, but the genera 268 composition dynamics instead. In other words, the structures of the food webs are more 269 influenced by those genera absent in autumn. Indeed, previous studies on freshwater food webs 270 that have examined temporal changes have often found reduced productivity as the main driver 271 behind a declined food-web structure in winter vs summer 20,36 . Thus, the decrease in structure we 272 captured in this study could be the start of the productivity restriction seen in food webs over the 273 winter months. The less-connected and less-nested food-web structure also implies weaker 274 resource competition among consumers, which may be necessary for their coexistence in the 275 food web 37 , especially when competition becomes more costly with lower resource productivity 276 in autumn. 277

278
Moving from broad food-web patterns to examining functional feeding groups enables us to 279 study fundamental changes within the community and possible effects at the ecosystem level 17 . 280 Interestingly, and contrary to the biodiversity patterns and food-web dynamics, we see 281 significant effects of both drainage area and season on the functional diversity, which we found 282 to be higher in smaller drainage areas and in spring. The spatial part of such a pattern along the 283 hierarchical river network is demonstrated in the River Continuum Concept, which shows 284 increased diversity of invertebrate functional feeding groups at low stream orders 38,39 . Whereas 285 both drainage area and season influences functional diversity, season is the only significant 286 influence on functional redundancy, with fewer genera per functional feeding group observed in 287 summer rather than in autumn. These results indicate that in summer the genera absent are across 288 functional groups, whereas in autumn the genera absent include whole functional feeding groups 289 leading to the constriction of local food webs. The inconsistent temporal patterns of food-web 290 structure versus functional characteristics further imply that multiple feeding groups share 291 similar trophic roles in the food web (Fig. S2), though they each adopt specialised feeding 292 behaviour within the catchment and thus perform distinctive ecological functions (e.g., 293 shredders, collectors, filterers). In other words, the structure and function of a food web does not 294 necessarily match and synchronise (sensu stricto 22,40 ). These results are particularly encouraging, 295 because the patterns of both food-web structure and functional diversity are known to be 296 important for ecosystem health assessment and identifying potential vulnerability to 297 perturbation 17,26,41 . Addressing the consistency of their patterns across spatial and temporal scales 298 will likely lead to novel and comprehensive understanding of biodiversity and ecosystem 299 function loss due to environmental change. 300 301 In our study, the functional feeding groups were defined at the level of genus, while we expect 302 investigations at finer resolutions will be promising for future work that can reveal not only more 303 accurate patterns, but also the influences of sampling taxonomic resolution. Similarly, our 304 selection of focal taxa may have influenced the food-web patterns we detected. For example, 305 algae become an increasingly important resource when moving from allochthonous inputs in 306 headwaters to larger streams with increased light levels further down the catchment 38,39 . With the 307 selected three key taxonomic groups in riverine ecosystems, we present a broad and relevant 308 view capturing trophic roles from basal resources (cyanobacteria) to top consumers (piscivorous 309 fish), captured by three relatively broad metabarcoding markers. However, we also by default 310 excluded some further groups, such as algae or terrestrial taxa, which could have been relevant 311 as primary producers and as terrestrial-aquatic linkages, respectively. The choice of taxonomic 312 groups looked at was both methodologically defined, as well as driven by the goal to have an 313 overseeable and clearly aquatic-focused view on food webs. Thereby, it captures by default a 314 subset from the real-world food web in which more species are involved. 315 316 By using the eDNA technique, we gain three notable advantages: its scalability for monitoring 317 complex and large systems, its reusable nature, and its being a non-invasive method of collecting 318 biodiversity information 42,43 . Our understanding of how the information we ascertain from eDNA 319 sample collection has greatly improved in recent years due to studies on the hydrological 320 influences 44 and a general understanding of the rate of eDNA persistence in lotic ecosystems 45 . 321 However, the successful detection of taxa with eDNA is also linked to the ecology of individual 322 species 45 , and some seasonal variation in the detection of several taxonomic groups is known 46-323 49 . Therefore, it is possible that some taxa that were not detected in the colder season (autumn) in 324 this study were false-negative records. However, these non-detections are likely linked to low 325 abundance or low metabolic rates, and thus these species are, while not physically absent, at least 326 "relatively absent" in ecological terms. 327 328 In summary, our work showed we can construct comprehensible food webs at a large scale and 329 over time by using eDNA sampling and combining multiple markers. Based on the biodiversity 330 patterns we observed that spatial and temporal influences are different across groups. 331 Furthermore, temporal influences were the significant driver of change for commonly used food 332 web descriptors. Our approach is a first demonstration of the application of eDNA to a complex 333 river network for the reconstruction of food web patterns and can be easily replicated in other 334 systems worldwide. As biodiversity in freshwater systems face huge threats from anthropogenic 335 pressures, including global climate change, establishing vital information on the changes in 336 biodiversity and food web composition over spatio-temporal scales is essential for the detection 337 of these stressors in order to protect and conserve river systems.  Table  370 S8 for primer sequences). Positive controls (n = 6 per library prep, see Supporting Information 371 Table S9)  We calculated α-diversity (genus richness) at each site and compared β-diversity between sites 417 by using Jaccard dissimilarity using the betapart R-package 58 . We constructed a matrix of 418 pairwise distances between sites along the fluvial network with the igraph R-package 59 and 419 compared the similarity between β-diversity and river distance using the Mantel test with the 420 vegan R-package 60 . To further partition β-diversity into species loss and replacement, sites which 421 did not record genus from a target group were removed from the analysis of that group only. 422 423

Feeding group assignment 424
Feeding groups were determined based on literature, species inventories and targeted expert 425 knowledge 61,62 for fish and invertebrates. Genera of bacteria were included if the phyla they 426 belong to has a strong association with freshwater habitat 63 , bacteria were then broadly divided 427 into heterotrophic and cyanobacteria, the latter constituting a basal resource. A constant basal resource of detritus was also included in all food webs (See Supporting Information Table S2 for 429 assignment of genera to feeding groups). 430 431 Food-web structure 432 We constructed a metaweb based on known trophic relationships among feeding groups 433 (Supplementary Table S2), then defined local food webs using this metaweb at each field site, 434 based on co-occurrence of genera (nodes) and their interactions (links). The metaweb approach 435 has been applied to identify food-web characteristic change across spatial gradients in terrestrial 436 ecosystems 32 , and temporal changes in aquatic ecosystems 24 , but has yet to be used at a large 437 spatial scale in an aquatic ecosystem over time. To explore the change in functional characteristics within the food webs, we used the broad 466 feeding groups as described when constructing the interaction network (See Fig. 1 and Fig. S2). 467 Functional diversity was calculated as the number of feeding groups that had at least one genus 468 present in the sample. Functional redundancy was calculated as α-diversity divided by functional 469 diversity, reflecting the average number of genera within a feeding group. 470 471

Data analysis 472
We ran linear mixed model analysis to assess the relationship between group genus α-diversity 473 (bacteria, invertebrates and fish), food-web structural characteristics (coherence, connectance, link density, number of links, omnivory, modularity, nestedness and robustness) and functional 475 characteristics (functional diversity and redundancy) with site location within the catchment. 476 Drainage area (km 2 ) was log transformed to fit model assumptions. For each dependent variable, 477 drainage area and season were the fixed effects. We included site as a random effect to account 478 for repeated sampling of sites (e.g., genus α-diversity ~ drainage area + season + (1 | Site)). 479 Significance was calculated using the lmerTest R-package 67 , which applies Satterthwaite's 480 method to estimate degrees of freedom and generate p-values for mixed models (Table S4 and 481 S5). We then used anova() to see the overall effect of both factors: Drainage area and Season 482 (Table S6) and follow up contrast testing were carried out using the emmeans R-package 68 to 483 test for significant differences between seasons (Table S7).