Communities at the extreme: Aquatic food webs in desert landscapes

Abstract Studying food webs across contrasting abiotic conditions is an important tool in understanding how environmental variability impacts community structure and ecosystem dynamics. The study of extreme environments provides insight into community‐wide level responses to environmental pressures with relevance to the future management of aquatic ecosystems. In the western Lake Eyre Basin of arid Australia, there are two characteristic and contrasting aquatic habitats: springs and rivers. Permanent isolated Great Artesian Basin springs represent hydrologically persistent environments in an arid desert landscape. In contrast, hydrologically variable river waterholes are ephemeral in space and time. We comprehensively sampled aquatic assemblages in contrasting ecosystem types to assess patterns in community composition and to quantify food web attributes with stable isotopes. Springs and rivers were found to have markedly different invertebrate communities, with rivers dominated by more dispersive species and springs associated with species that show high local endemism. Qualitative assessment of basal resources shows autochthonous carbon appears to be a key basal resource in both types of habitat, although the particular sources differed between habitats. Food‐web variables such as trophic length, trophic breadth, and community isotopic niche size were relatively similar in the two habitat types. The basis for the similarity in food‐web structure despite differences in community composition appears to be broader isotopic niches for predatory invertebrates and fish in springs as compared with rivers. In contrast to published theory, our findings suggest that the food webs of the hydrologically variable river sites may show less dietary generalization and more compact food‐web modules than in springs.


| INTRODUC TI ON
Studying ecological communities across environmental gradients is a key step toward predicting how complex communities respond to their abiotic environment, providing insight into how they persist over time (Dunne, Saleska, Fischer, & Harte, 2004;Thompson, Dunne, & Woodward, 2012). Changes along environmental gradients can alter which taxa are present and their abundance (community composition) and the trophic interactions between taxa (food-web structure). Food-web analysis includes all species in the system and the movement of energy between those species (Cole et al., 2006;Thompson, Brose, et al., 2012). By including all species present and their interactions, we can understand how whole ecosystems respond to stressors such as invasive species (Ho, Bond, & Lake, 2011), drought (Ledger, Brown, Edwards, Milner, & Woodward, 2013), or climate change (Binzer, Guill, Brose, & Rall, 2012). Instead of focusing on effects on individual species, food-web approaches can account for emergent and interactive effects of stressors on whole communities (Layer et al., 2011).
Analysis of carbon and nitrogen isotopes quantifies the relative trophic positions of species, which allows the structure of the food web to be characterized and emergent food-web variables to be quantified (Post, 2002). For example, food-chain length (i.e., the number of trophic transfers from basal resources to top predators) is a commonly studied response metric that can assess the factors that support high trophic-level species, overall community complexity, and biodiversity (McHugh, McIntosh, & Jellyman, 2010). Similarly, the ecological niche occupied by functional groups and species can be quantified using the "isotopic niche" concept, where the isotope composition of an organism's tissue reflects what it consumes (Jackson, Inger, Parnell, & Bearhop, 2011;Newsome, Martinez del Rio, Bearhop, & Phillips, 2007). The size and shape of the niche occupied by taxa or functional groups can be a sensitive measure of community trophic characteristics under varying levels of abiotic stress (McGill, Enquist, Weiher, & Westoby, 2006).
In aquatic ecosystems, hydrology is a key ecological driver which can influence community composition both directly through disturbance effects (Bunn & Arthington, 2002;Lake, 2003) and indirectly through impacts on basal resource availability (Takimoto & Post, 2013;Townsend et al., 1998). Natural disturbance, such as the extreme high flow events and droughts experienced by desert rivers, has been hypothesized as a key driver of patterns of diversity generally, although empirical evidence remains weak for classical ecological hypotheses such as the "intermediate disturbance hypothesis" (Connell, 1978;Fox, 2013). More disturbed habitats may be more likely to contain species with broad diets, including omnivores (e.g., Menge & Sutherland, 1987). This hypothesis is supported by empirical studies of early successional communities, where trophic generalists are more common (e.g., Brown & Southwood, 1983;Kullberg & Scheibe, 1989). Applying stable isotope analysis to communities' subject to differences in flow permanence allows us to assess how disturbance may influence total niche width and isotopic niche size (Jackson et al., 2011).
The effects of disturbance on food-web characteristics have long been of interest to ecologists, particularly from a perspective of understanding food-web attributes that may promote resilience and resistance (e.g., Kondoh, 2003;Krause, Frank, Mason, Ulanowicz, & Taylor, 2003;Wootton, Parker, & Power, 1996). It has been hypothesized that more variable habitats will have shorter food chains, due to stochastic loss of numerically rare top predators (McHugh et al., 2010;Pimm & Lawton, 1977). Experimental support for this found negative effects of hydrologic variability on food-chain length in North American rivers (Sabo, Finlay, Kennedy, & Post, 2010).
However, Walters and Post (2008) found that a low-flow disturbance altered the body size characteristics of communities but not the length of the food chain. A recent meta-analysis also found little evidence of an effect of disturbance on food-chain length (Takimoto & Post, 2013).
Indirect effects of disturbance on diversity through impacts on the basal resource supply have received relatively less attention (Wootton, 1994a(Wootton, , 1994b; but see Townsend et al., 1998). Across aquatic ecosystems, and particularly in river systems, basal carbon from in-stream autochthonous (e.g., algal and macrophyte) and allochthonous (e.g., riparian and floodplain) sources are fundamental processes underpinning food webs (Bunn, Davies, & Winning, 2003;Cole et al., 2006).
Carbon isotope analysis allows for tracking basal resources such as algal or terrestrial carbon through food webs (Post, 2002). Isotope studies of primary productivity have provided unique insights into the relative influence of carbon sources, where flow variability can play an important role in facilitating inputs of terrestrial riparian and floodplain carbon (Bunn et al., 2003;Hadwen, Fellows, et al., 2010;Robertson, Bunn, Boon, & Walker, 1999).
Two aquatic ecosystems in arid Australia provide an opportunity to analyze whether differences in flow permanence and abiotic variability influence community composition and food-web structure.
In northern South Australia, the Great Artesian Basin (GAB) forms a series of highly stable and extremely isolated groundwater springs (Mudd, 2000;Murphy, Breed, Guzik, Cooper, & Austin, 2012), while the Lake Eyre Basin (LEB) is a largely ephemeral surface water basin that includes some of the most hydrologically variable rivers in the world (Kotwicki & Allan, 1998;McNeil, Schmarr, & Rosenberger, 2011). Western LEB waterholes range from semi-permanent to highly ephemeral, and the flow variation is coupled with variation in physiochemical properties; for example, salinity can range from near fresh during high flows to hypersaline in recession phases (Costelloe, Grayson, McMahon, & Argent, 2005;Costelloe & Russell, 2014). Biodiversity studies have shown that assemblages have been highly conserved in arid Australian waterholes over time, with more persistent waterholes being key ecological refugia (Davis, Pavlova, Thompson, & Sunnucks, 2013). Furthermore, studies have shown that some taxa are reliant on GAB waterbodies and show high springand catchment-specific endemism (Murphy, Adams, & Austin, 2009;Murphy et al., 2012), and permanent springs across arid Australia broadly tend to be characterized by endemic species in contrast to the riverine communities (Davis et al., 2018). This suggests that present-day assemblages are likely to be the product of long-term physical differences between waterbodies and their connectivity within the landscape, and the contrasting abiotic regimes provide an opportunity to study their influence on community characteristics.
We integrated community composition analysis with stable isotope approaches (Layman et al., 2012) and contemporary methods to quantify isotopic niche size (Jackson et al., 2011). This approach was used to explore how physical variability influences aquatic foodweb characteristics (Thompson, Brose, et al., 2012). Specifically, we sought to determine whether the differences in the abiotic characteristics of river and spring habitats influence: 1. Community composition, where we predicted that river environments would be characterized by broadly dispersed, highly mobile species, whereas spring habitats would contain taxa with poorer dispersal abilities and have higher local endemism.
2. Basal resource base, where we predicted that river environments may have a broader resource base due to lateral connectivity with riparian and floodplain energy sources.
3. Isotopic niche size of the community, which we predicted would be larger in more disturbed sites due to selection for generalist taxa. F I G U R E 1 Map of the 11 study sites in northern South Australia, within the western Lake Eyre Basin. Both spring (sites 1-5) and river (sites 6-11) sites span multiple river systems that feed into Lake Eyre, with the northern sites (4, 5, 7, 11) within the Neales River Catchment, and the remaining sites in smaller creek catchments that flow into the lake's southern lagoon. 2013-2014 rainfall records at towns/stations shown are available in Appendix S1 (Figures A1-A3) 4. Food-chain length, where we predicted that food chains would be shorter in the more physical-disturbance prone river food webs compared to the more stable spring food webs.

| Study location
Sites were selected to represent permanent groundwater springs (n = 5) and intermittent river waterholes spanning multiple catchments in the western Lake Eyre Basin (n = 6; Table 1, Figure 1).
Spring sites chosen were the largest accessible springs in the area, while waterholes chosen were the largest known accessible pools to maximize the chance that sites were not dry during sampling. Both aquatic habitat types span a range of sizes and to ensure consistent areas were sampled, entire pools/spring outlets were sampled where they were <50 m length and a maximum 50 m reach was sampled for larger sites (e.g., Algebuckina, The Bubbler). In April 2014, invertebrate and fish communities were sampled as well as basal resources, including plant material from riparian and aquatic vegetation, phytoplankton, and algae. At sampling, all river sites were nonflowing pools, with physiochemical properties showing high among-and within-site variation relative to springs during the study period and measured parameters (e.g., temperature, conductivity, salinity) ranging from below that of spring sites to significantly higher (Table 1) being almost entirely dry prior to that rainfall.

| Stable isotope sampling and processing
For basal resources, three replicate samples (predominantly leaves) were collected from all plant species observed at sites and frozen.
After transport to Monash University, samples were rinsed with reverse osmosis (RO) water and oven-dried for 1 week at 60°C.
Samples were then ground to a fine powder using a mortar and pes- Gastropods, which were common, had their shells removed before processing, as carbon-based precipitates reflect the isotopic ratios of the inorganic environment, whereas their soft tissue reflects their diet (Post, 2002). Where enough biomass was available, three independent (distinct individuals/groups of individuals) replicates were analyzed. If insufficient biomass was sampled, samples were pooled to give a mean value per species/site. Replicates were dried and ground before analysis. For fish and crayfish sampling, differences in the depth between sites did not allow a standardized approach to be used. In springs, unbaited box traps (45 cm × 25 cm × 25 cm, 5 cm aperture, 1 mm mesh) and hand dip nets (20 cm × 25 cm, 1 mm mesh) were used. At river sites, three fkye nets (single 5 m wing, 60 cm drop, 19 mm mesh, deployed overnight) and a seine net (7 m × 2 m, 6 mm mesh) were also used, water depth permitting.
Individuals were removed and euthanized by immersion in a solution of clove oil and water sourced from the site, before freezing. Where enough individuals were collected, three independent samples of muscle tissue were rinsed, dried, and ground before analysis. If there were insufficient individuals/biomass to provide three independent samples, three samples were analyzed where possible and a mean isotope value reported.

| Stable isotope analysis
Samples were analyzed for carbon (carbon-12-carbon-13; hereafter δ 13 C) and nitrogen (nitrogen-14-nitrogen-15; hereafter δ 15 N) isotope ratios at the Water Studies Centre (Monash University) on an ANCA GSL2 elemental analyzer interfaced to a Hydra 20-22 continuousflow isotope ratio mass-spectrometer (Sercon Ltd.). The precision of the stable isotope analysis was ±0.1‰ for δ 13 C and ±0.2‰ for δ 15 N. Stable isotope data are expressed in the delta notation (δ 13 C and δ 15 N), relative to the stable isotopic ratio of Vienna Pee Dee Belemnite standard (R VPDB = 0.0111797) for carbon and atmospheric N 2 (R Air = 0.0036765) for nitrogen.

| Statistical analysis
All statistical analyses were conducted using R Statistical Package 3.5.1 (R Core Team, 2018). To analyze community composition, we used invertebrate taxa presence/absence data by site in multidimensional scaling (MDS) analysis. Invertebrate data (cf. fish) were used as we were able to apply a more strictly standardized sampling methodology across sites and because the desert goby (Chlamydogobius eremius) is known to be the only species present in any spring site.
Only taxa caught using kick/sweep sampling were used in this analysis, while all taxa (including those captured using other methods, e.g., Cherax destructor) are used in isotope analysis. PERANOVA hypothesis tests (adonis function, vegan package; Oksanen et al., 2007) tested for differences in composition associated with habitat type.
We also tested for an effect of region as genetic isolation between northern (e.g., Algebuckina) and southern (e.g., Warriner Creek) catchments has been detected in previous studies (see Mossop et al., 2015). As compositional differences were found between springs and rivers, the contribution of each taxa to the effect of habitat type was estimated using Bray-Cutis dissimilarity (SIMPER function).
We considered taxa with a percentage contribution greater than its standard deviation to be substantially contributing to the habitat differences in composition, and their taxa-specific NMDS scores were calculated.
In addition to the descriptive analysis of stable isotope plots, the mean δ 13 C, and total number of taxa. One-way ANOVAs compared each measure between spring and river sites.

| RE SULTS
We found a significant effect of habitat type on invertebrate community composition (F 1,8 = 12.1370, p = .001, Figure 2) and no effect of region. Eleven invertebrate taxa were found to be substantially contributing to the habitat effect via SIMPER analysis, and their NMDS scores were calculated (Table 2). Rivers were primarily associated with Coleoptera taxa, whereas springs were associated with several groupings, including arachnids, gastropods, and isopods Standard ellipse plots (Figure 4a) and the high among-site variation in δ 15 N and δ 13 C values relative to river sites (Table 3) indicate variability in carbon sources and isotopic niches between springs.
Isotope plots of river sites (Figure 3f-k) and mean δ 13 C values (Table 3) show animal communities clustered around −20‰, and standard ellipse plots (Figure 4b)  For the animal communities at springs and river sites, there were no overall differences in SEAc, δ 15 N mean/range, δ 13 C mean/range, or total taxa richness based on habitat type (Table 3) The SEAc of functional groupings (Table 3) showed that the isotopic niche space occupied by predatory invertebrates and fish was larger in springs than river sites (F 1,7 = 18.09, p = .0038), despite spring sites tending to have fewer predatory invertebrate taxa and fish.
Although this should be interpreted cautiously as two spring sites did not have sufficient taxa to calculate SEAc for this group. There was no significant habitat difference for nonpredatory invertebrates.

| D ISCUSS I ON
Invertebrate assemblages across a wide geographical range in the western LEB were strongly associated with their local habitat type. Springs were linked with taxa that have low active dispersal capability and are less likely to be adept at coping with physically variable environments (i.e., isopods and gastropods), whereas rivers were linked to Coleoptera taxa with the capacity for active flight and dispersal (Boda & Csabai, 2013). Temporal patterns in hydrology and connectivity have been shown to be an important driver of invertebrate assemblages and diversity in the arid Cooper Creek (Marshall, Sheldon, Thoms, & Choy, 2006), so the possible flow event within 2 months of sampling may have been a factor driving the observed similarity between river waterholes. Directly following a flow disturbance and the re-wetting of dry or drying waterholes, species succession can favor dispersive invertebrates, and in this case, Coleopterans may represent early colonizers (Sousa, 1979;Ward, Tockner, Arscott, & Claret, 2002).
While dispersal strategies appear to allow certain taxa to persist across more physically variable LEB rivers, in temperate streams and isolated pools on the Iberian Peninsula, the opposite pattern was observed, with lentic waterbodies linked to dispersive Coleopteran species (Ribera & Vogler, 2000). Nonetheless, the mechanism underlying this is consistent with what we observed, as Ribera and Vogler's isolated pools were also more ephemeral and required frequent migration for aquatic species, so may be considered more like LEB waterholes. Consistent findings from the Pilbara region in northwest Australia show that the local physical conditions and connectivity drive invertebrate assemblages in arid waterholes (Moran-Ordonez et al., 2015). These results reinforce the ecological importance of  Murphy et al., 2012). This is also true of terrestrial wolf spiders (Araneae, Lycosidae), which are dominant invertebrate predators in GAB springs and were commonly detected in aquatic invertebrate samples (Gotch, Adams, Murphy, & Austin, 2008).
Patterns in fish distributions similarly support the dual ecological significance of permanent and ephemeral habitats, where mound springs are key refuges for the poor swimming desert goby, which shows genetic structuring separated according to the north and south catchments of Lake Eyre (Mossop et al., 2015). Conversely, more actively dispersive species such as spangled perch (Leiopotherapon unicolor) and the Lake Eyre hardyhead (Craterocephalus eyresii) are widespread in hydrologically variable riverine habitats across the LEB without reliance on the spring habitats (McNeil et al., 2011).
Food webs showed distinct similarity across river habitats, with the isotopic niche of animal communities overlapping and appearing clustered around −20‰ δ 13 C. This suggests autochthonous algal food sources (particularly the algal benthic mat) may be important primary producers in rivers during no flow periods, although our sites were also affected by a relatively recent flow event, which can enrich the δ 13 C of benthic algae (Hadwen, Spears, & Kennard, 2010).
Our results are consistent with those from the eastern Lake Eyre Basin where benthic algal carbon was highly productive and a key carbon input during no-flow periods in Cooper Creek (Bunn et al., 2003).

For Finniss Creek, Margaret Creek, Warriner Creek, and
Algebuckina, the δ 13 C range of fauna was relatively large, ranging to above −25‰. This suggests that terrestrial vegetation may also be providing significant inputs, and it is expected that both terrestrial and autochthonous sources are providing carbon inputs into rivers (Robertson et al., 1999). The apparent importance of autochthonous carbon to river sites is consistent with international models (see Thorp & Delong, 1994) and studies of Australian rivers under prolonged low-flow/ no-flow periods (Bunn et al., 2003;Gawne et al., 2007). The extreme hydrological variability of LEB rivers likely facilitates floodplain connection and terrestrial inputs, but for the majority of time, these rivers exist as nonflowing isolated pools and no-flow droughts can persist for multiple years (Kotwicki & Allan, 1998). Our findings suggest that consistent autochthonous algal productivity is an important for aquatic communities to persist despite the unpredictable flow regime.
Across springs, the basal carbon sources supporting food webs appear to vary based on their local environment. An unusual example is the filamentous algae in The Bubbler, which had an extremely low δ 15 N value. This has been observed elsewhere, where phosphorus limitation in lichens and mangroves encourages the uptake of isotopically depleted atmospheric ammonia (Fogel et al., 2008), suggesting nutrient limitation and atmospheric nitrogen fixing are potentially influencing basal productivity at this spring. This is also potentially influencing productivity by phytoplankton/ FPOM sources at Blanche Cup, which is likely a significant source of basal carbon to this food web. Gastropods appear particularly important to spring food webs, as they are hugely abundant (usually >1,000 per sample) and our own qualitative observations of desert goby gut contents (the only vertebrate species occupying spring sites) suggest that gastropods make up a significant portion of their diet. An important role of gastropods demonstrates that inputs from biofilm and algae are key basal sources underpinning these food webs, which further emphasizes the value of stable isotopes in exploring patterns of primary productivity, particularly in aquatic ecosystems, and importance of basal resources to aquatic community structure.
Contrary to our hypotheses, food webs from contrasting habitat types did not appear to have overall differences in their structural attributes. Neither the SEAc nor δ 15 N range was larger across the physically stable springs, suggesting that the buffer against extreme hydrological variability provided by groundwater does not necessarily influence these general food-web measures. While the disturbance hypothesis predicts that longer food chains are more sensitive to perturbations (McHugh et al., 2010), our data are more consistent with analysis showing that disturbance does not appear to have an overarching effect of food-chain length (Takimoto & Post, 2013). This is analogous with theoretical food-web modeling, which has suggested that larger complex food webs will be more sensitive to perturbations (May, 1972). More recent modeling instead suggests that larger complex food webs are not inherently unstable, by incorporating characteristics of empirical communities such as body size scaling (Brose, Williams, & Martinez, 2006) and adaptive behavior (Valdovinos, Ramos-Jiliberto, Garay-Narvaez, Urbani, & Dunne, 2010). Finally, the two largest δ 15 N ranges were detected at the largest spring and river sites, suggesting that the area/volume of the aquatic habitat is likely to play a role in the food-chain length. The size and spatial heterogeneity of stream ecosystems has been shown to influence food-web structures such as food-chain length (Spencer & Warren, 1996;Thompson & Townsend, 2005; consistent with hypotheses such as the Productive Space Hypothesis, Schoener, 1989), and among-site variation in physical structure may significantly influence the variability in spring and river food webs.

ACK N OWLED G M ENTS
We acknowledge the traditional owners of the Kati Thanda region and appreciate their knowledge offered. We thank all project contrib-

CO N FLI C T O F I NTE R E S T
We declare no conflict of interests in relation to this manuscript.