Biological interactions mediate context and species-specific sensitivities to salinity

Abstract

Toxicants have both sub-lethal and lethal effects on aquatic biota, influencing organism fitness and community composition. However, toxicant effects within ecosystems may be altered by interactions with abiotic and biotic ecosystem components, including biological interactions. Collectively, this generates the potential for toxicant sensitivity to be highly context dependent, with significantly different outcomes in ecosystems than laboratory toxicity tests predict. We experimentally manipulated stream macroinvertebrate communities in 32 mesocosms to examine how communities from a low-salinity site were influenced by interactions with those from a high-salinity site along a gradient of salinity. Relative to those from the low-salinity site, organisms from the high-salinity site were expected to have greater tolerance and fitness at higher salinities. This created the potential for both salinity and tolerant-sensitive organism interactions to influence communities. We found that community composition was influenced by both direct toxicity and tolerant-sensitive organism interactions. Taxon and context-dependent responses included: (i) direct toxicity effects, irrespective of biotic interactions; (ii) effects that were owing to the addition of tolerant taxa, irrespective of salinity; (iii) toxicity dependent on sensitive-tolerant taxa interactions; and (iv) toxic effects that were increased by interactions. Our results reinforce that ecological processes require consideration when examining toxicant effects within ecosystems.

This article is part of the theme issue ‘Salt in freshwaters: causes, ecological consequences and future prospects’.

1. Introduction

At both local and broad spatial and temporal scales, dispersal, stochastic demographic processes, speciation and deterministic niche processes influence patterns of community assembly [1–4]. At local scales, environmental filtering, or the restriction of fundamental niches into realized niche space depend on abiotic gradients and biotic interactions including competition [5,6]. Observational studies further suggest that anthropogenic chemical stressors may influence individuals, populations and communities at concentrations lower than laboratory-based toxicity experiments predict [7,8]. Toxicants often interact with other stressors [9,10], or with chemical [11], physical [12] and biotic ecosystem components, sometimes lessening [11] or strengthening their effects [13]. Among biological interactions, competition [14], predation [15,16] and parasitism [13] modify stressor effects on organisms. In this manner, local environmental gradients and stressors, including toxicants, can cause ‘context sensitivity’ in communities [17]. Therefore, differences between field and laboratory findings may arise from the complexities of a realized niche, including interactions between abiotic gradients, other stressors, biotic niche effects and ecological processes operating at multiple spatial and temporal scales [18]. Where ecological processes modify toxicant impacts [12,19], effect prediction is hindered, especially where species-specific and context-dependent responses occur [20].

Salinization is a globally important stressor in freshwater ecosystems [21]. Salinity is a component in all natural waters and is defined as the total concentration of dissolved inorganic ions in water or soil [22]. The dominant ions in anthropogenic and naturally saline waters include Na⁺, Cl⁻, Mg²⁺, Ca²⁺, HCO³⁻, and [23]. Natural salinization occurs from catchment weathering, sea-spray and salts transported by seawater evaporation [21]. Anthropogenic salinization can be caused by water harvesting, road de-icing, mining activities and changes to vegetation leading to water-table movement [21]. In freshwaters, salinity is a toxicant and physiological stressor causing direct lethal [24] and sub-lethal organism effects [25], which depend on ion proportions, their concentrations and organism sensitivities. These effects on organisms can reduce biodiversity [26], alter trait and community structure [27] and cause changes to ecosystem processes [28].

While it is well known that biotic interactions shape community composition, there has been little consideration of how interspecific biotic interactions may interact with organism tolerances/sensitivities to toxicants in populations and communities [29]. Increased tolerance to a stressor is expected to influence the outcome of biotic interactions and fitness in relation to that stressor [30]. Despite their expected importance, studies examining the interplay between deterministic abiotic filtering and biotic interactions on communities are rare, but are required to predict the effects of stressors within real ecosystems [29]. Here, we tested whether salinity effects were modified by interspecific biotic interactions between salt-tolerant organisms, collected from a high salinity site, and a community expected to be more salt-sensitive, collected from a low salinity site. Salinity effects were examined between two regimes of biological interaction using mesocosms: (i) interactions among salt-sensitive organisms only; and (ii) interactions between salt-sensitive and salt-tolerant organisms. Interactions within salt-sensitive communities were implicit, but could not be considered with the current study design. The design did allow several contrasting predictions to be made at the community and population levels. At the community level, both salinity and biological interactions between sensitive and tolerant organisms may be expected to influence community composition, reduce diversity and potentially homogenize communities. However, population responses are expected to be diverse (figure 1). Salinity effects may: (i) depend on interspecific interactions between salt tolerant and sensitive organisms (e.g. [31]); (ii) dominate through direct toxicity or physiological impairment, irrespective of biotic effects [32]; (iii) have effects that are partly related to direct toxicity or physiological impairment but are strengthened by biotic interactions [19,31]; by contrast (iv) biotic interactions may be dominant irrespective of salinity [33]; (v) salinity could cause positive responses irrespective of biotic interactions [34]; or (vi) toxicity and interspecific interactions may cause complex unexpected responses [35,36].

Figure 1.

Hypothetical effects and interactions associated without (blue line), and with (green line) interspecific biotic interactions between salt-sensitive and salt-tolerant communities/taxa. Effects have been linearized for simplicity, but could be nonlinear or exhibit threshold responses. Organism and community responses may: (a) depend on biological interactions between salt-sensitive and tolerant taxa; (b) be dominated by the toxic effects of the chemical stressor; (c) have effects that are partly chemical and partly biotic; (d) biotic interactions may have effects that dominate irrespective of the chemical gradient; (e) toxicants could cause effects that are beneficial irrespective of biotic interactions; or (f) exhibit complex unexpected responses. (Online version in colour.)

2. Methods

An outdoor mesocosm experiment was conducted in the austral winter (June–September) of 2017 at the University of Canberra, Canberra, Australia (ca 650 m elevation; 35°14′06.5″ S, 149°05′12.0″ E). Thirty-two mesocosms consisting of 1000 l Reln round troughs (1.5 m diameter; 0.8 m deep; figure 2b) were filled with de-chlorinated Canberra town supply water (ca 900 l per mesocosm). The town water comes from the Cotter River, the source location of the salt-sensitive taxa used in the experiments. Flow was maintained in experiments at ≈0.22 m S⁻¹ using a submersible 1400 l h⁻¹ pump. The experiment was an orthogonal design, with two levels of expected intensity of interspecific interactions: (i) between salt-sensitive taxa only and (ii) between salt-tolerant and salt-sensitive taxa. The first of these treatments included individuals only from the Cotter River, a low-salinity site (mean 30 µS cm⁻¹; 35°23′10.5″ S, 148°51′53.6″ E), which had taxa that ranged from salt-sensitive to tolerant (i.e. certain Odonate spp.). The second treatment included individuals both from the low-salinity Cotter River site, and individuals from Cunningham Creek, which had much higher salinity (mean 1600 µS cm⁻¹; Cunningham creek; 34°34′26.4″ S, 148°17′04.4″ E), which were assumed to be more salt tolerant compared to those from the Cotter River. Biotic treatments were exposed to a gradient of salinity treatments generated by synthetic marine salt (Ocean Natures). This resulted in electrical conductivity treatments: control (no added salt; mean ca 210 µS cm⁻¹; table 1), ca 500 µS cm⁻¹, ca 1000 µS cm⁻¹, ca 2500 µS cm⁻¹ and ca 5000 µS cm⁻¹. This salinity gradient was based on benign-harsh conditions from LC₅₀ values [32].

Figure 2.

(a) Salinity treatments across a measured conductivity gradient, crossed with two levels of expected biotic interaction. (b,c) Communities were subjected to treatment regimes in 32 recirculating 1000 l mesocosms. (c1) 1400 l h⁻¹ pumps, re-circulated water at approximately 0.2 m s⁻¹, along (c2) colonization tray sampling units.

Table 1.

Key physico-chemical parameters recorded at regular intervals during the mesocosm experiment (mean ± standard deviation). (For brevity not all comparisons are shown, measured salinity increased predictably within salinity treatments.)

	temperature	pH	oxidative/reductive potential	conductivity	turbidity	DO	DO%	alkalinity
	°C		mV	mS cm−1	NTU	mg l−1	% saturation	mg l−1
salinity treatment
control	11.75 ± 2.26	7.39 ± 0.62	226.14 ± 63.5	0.21 ± 0.16	2.55 ± 1.01	9.92 ± 1.99	94.74 ± 19.94	5.58 ± 1.97
500	11.52 ± 2.15	7.66 ± 0.45	202.65 ± 65.11	0.6 ± 0.13	2.92 ± 1.55	9.54 ± 2.23	91.88 ± 18.66	5.41 ± 1.3
1000	11.67 ± 2.68	7.47 ± 0.48	216.56 ± 67.71	1.12 ± 0.2	2.78 ± 2.36	9.84 ± 1.97	93.63 ± 19.58	5.25 ± 1.38
2500	11.46 ± 2.59	7.69 ± 0.51	208.07 ± 64.77	2.4 ± 0.64	2.72 ± 2.33	9.32 ± 2	87.44 ± 19.89	5.84 ± 1.89
5000	11.53 ± 2.17	7.69 ± 0.31	213.83 ± 49.34	4.79 ± 0.4	2.64 ± 1.38	9.67 ± 2.05	92.69 ± 20.62	5.96 ± 1.49
biotic treatments
salt-sensitive communities	11.79 ± 2.12	7.57 ± 0.45	214.77 ± 60.82	1.67 ± 1.64	2.6 ± 1.32	9.52 ± 2.1	90.89 ± 20.56	6.06 ± 1.83
salt-tolerant and sensitive communities	11.43 ± 2.41	7.56 ± 0.52	213.88 ± 66.03	1.73 ± 1.67	2.82 ± 1.93	9.81 ± 2.02	93.43 ± 19.57	5.2 ± 1.42

Forty-six days before starting the experiment, colonization trays (garden seedling trays; 360 mm long by 300 mm wide and 55 mm deep) were placed in the Cotter River near its confluence with Burkes Creek (−35°23′10.5″ S 148°51′53.6″ E). Trays had natural stream substrata of gravels, pebbles and cobbles translocated directly from the river bed, with existing biofilms and fauna, and were left in place for 44 days for further colonization. This colonization period was designed to minimize transplant and disturbance effects. Salt was added to mesocosms and then allowed to dissolve on the 22 June 2017, 12 h before the addition of colonization trays. Mesocosm colonization was supplemented with kick-net samples from riffle habitat from both the Cunningham and Cotter rivers which occurred from the 23–25 June 2017. Salt-sensitive treatments received two kick-net samples from the Cotter River, while salt-tolerant and sensitive treatments received one kick-net sample from the Cotter River and one from Cunningham Creek. Invertebrate densities differed between the Cunningham and Cotter sites, ascertained three weeks earlier with three replicate Surber samples per site. Mean densities at that time in the Cotter River were 4290 (±2600 s.d.), while Cunningham Creek had 14 900 (±10 200 s.d.). To standardize density differences, Cotter kick-net samples were collected from 3.3 m² of the river benthos and from 1 m² in Cunningham Creek. All Surber and kick-net samples were collected from undisturbed riffle habitat, moving in an upstream direction to avoid disturbing unsampled habitat. Kick-net samples and trays were randomly allocated to mesocosms.

Colonization leaf packs were placed with colonization trays in the Cotter River and transferred to mesocosms with trays. Leaf packs ensured the addition of shredders and were 15 g of River Red Gum (Eucalyptus camaldulensis) leaves collected from a location near the Yass River (−34°924413″ S, 149°179810″ E). HOBO Pendant temperature/light data loggers (UA-002-08) were deployed at the start of the experiment in a subset of mesocosms to record temperature variability among mesocosms. Weekly measurements were conducted of physico-chemical variables with a Horiba U-52 Water Quality Meter (IC-U52-2 m).

Mesocosms were left for 75 days before sampling on the 6–8 September 2017. One replicate was sampled from each of the 32 mesocosms on the same day, with sampling being consistent with respect to pump location to allow examination of velocity/turbulence effects. Velocities were recorded in a subset of mesocosms in relation to tray position. All invertebrates were sampled by carefully removing trays from the mesocosm using a kick-net working against the flow to collect dislodged invertebrates. All invertebrates were removed from the net, tray substrate and trays and stored in 100% ethanol. Taxa were identified to the lowest taxonomic unit (species where possible).

(a). Statistical analysis

All statistical analyses were conducted using R (R Core Development Team, [37]), in R Studio (RStudio Team, [38]). To remove the potential for confounding effects of tolerant taxa additions, all taxa that were known to occur from the Cunningham Creek were excluded prior to all analysis, with the exception of five taxa with less than 0.5% estimated community abundance in Cunningham Creek, typically comprising a single individual. This resulted in response data that was expected to be specific to the Cotter River, and thus were expected to be comparably salt-sensitive. Taxa common to both Cunningham Creek and Cotter River were assumed to be salt-tolerant, thereby allowing abundances of tolerant taxa, tolerant omnivores, tolerant predators and tolerant predator+omnivore densities to be calculated and used as covariates. The removal of Cunningham taxa data was necessary to allow for an unbiased estimate of biological effects on sensitive communities and populations.

(b). Community analysis

Hellinger transformation was used on community relative abundance datasets and was examined using non-metric multidimensional scaling (NMDS), partial redundancy analysis (pRDA) and variance partitioning. All abiotic and biotic predictors used in community analysis were standardized (0 mean, unit variance). Permutational multivariate analysis of variance adonis2 from the R package vegan were used to test for community differences among treatments. Mesocosm number was used as a block/random effect or was partialled out in permanova, pRDA, and variance partitioning results, with 200 permutation backwards stepwise used in model reduction. Two hundred permutations were used for model fractions in variance partitioning. NMDS was run for 200 iterations, Bray-Curtis was used as the measure of dissimilarity, and ‘ordisurf’ was used to fit general additive models (GAMs) [39]. Collinearity was set conservatively using Pearson's coefficients at (0.6). For all community analysis, data were transformed to per cent composition where design probably influenced organism densities.

(c). Single taxa and metric analysis

We further examined how the addition of salt-tolerant taxa influenced densities of the 20 most abundant salt-sensitive taxa, Ephemeroptera, and Plecoptera and Trichoptera (EPT) and total invertebrate densities. These data were analysed using a mixed-effect model with a lognormal hurdle distribution of salt-sensitive taxa densities. Modelling densities directly allowed comparisons of sensitivities (slopes) independent of differences in density. We fitted separate parameters for each taxon in each type of treatment (salt-sensitive communities and salt-tolerant and sensitive communities):

where: s is species and c for the type of community. A large number of species had 0 abundance for certain concentrations and experimental conditions, which led to our use of the hurdle model. A hurdle model assumes zero and non-zero data come from separate data-generating processes, such that positive densities are first conditional on an initial (Bernoulli) probability. Estimated intercepts and slopes for each taxon in each treatment were conditional on the probability of each taxon being observed. We tested whether salt effects differed among taxa when influenced by the presence of salt-tolerant taxa. The following assumes initial community composition of salt-sensitive taxa to be the same in each mesocosm (ignoring differences in density), therefore attributing reductions in taxonomic abundances to biotic interactions. Taxon intercept differences between treatments can be attributed to either or both of two processes: they had greater initial densities in the control community, or they have been reduced from competition with salt-tolerant taxa. However, the slopes only describe the effect of salinity after controlling for differences in initial densities and can be used to examine how biotic interactions influence salinity sensitivity. If tolerant taxa increase the effects of salinity, greater slopes are expected in the treatments including tolerant taxa, relative to the sensitive only community. An overall variance term ɛ was included to represent observation error. The model was fitted using the R package brms [40]. Sensitive community treatments were encoded 0, with 1 for communities subjected to tolerant taxa interactions. We further estimated biotic interaction coefficients, assessing how they differed between conductivity levels (electronic supplementary material 1). All parameters were given the default brms normal priors with mean 0, uniform Lewandowski, Kurowicka and Joe priors on all possible correlation matrices and Student t-distribution with 3 d.f. for the scale parameter [40]. We ran four Markov chains each of 4000 iterations, discarding half as burn-in. Convergence was assessed using chain traceplots and through calculating Rubin-Gelman statistics for each parameter, which all were less than 1.1 [41].

3. Results

(a). Physico-chemical differences

Physico-chemical variables changed consistently with salinity treatments and remained stable throughout the experiment. Conductivity (ANOVA, F_4,90 = 1423, p < 0.0001) and salinity (ANOVA, F_4,90 = 125.7, p < 0.0001) calculated by using electrical conductivity conversion changed consistently with salt addition, with minimal variation within salinity treatments and among the biotic treatments (table 1). Average water temperatures logged at 15 min intervals within 21 of the mesocosms, identified a mean temperature of 12.01°C, a mean minimum of 5.72°C and mean maximum of 23.09°C and were consistent with weekly meter readings (table 1).

(b). Community effects

Community analysis using non-metric multidimensional scaling identified patterns that were driven by both conductivity (GAM, p < 0.0001, 27% deviance explained) and the densities of tolerant taxa (GAM, p < 0.0001, 47.7% deviance explained; figure 3a). Partial redundancy analysis backwards model reduction of all candidate predictors identified patterns of community assembly were associated with velocity (F_1,89 = 2.35, p = 0.03), conductivity (F_1,89 = 3.91, p = 0.001), biotic treatments (F_1,89 = 2.5, p = 0.002), and the total abundance of salt-tolerant taxa (F_1,89 = 4.37, p = 0.001; figure 3b). These predictors collectively explained 12.6% of community variation. Predatory, and predatory + omnivore invertebrate abundances of tolerant taxa (i.e. abundances derived from tolerant taxa common to both the Cunningham and Cotter waterways) explained little variation and were removed in stepwise model reduction procedures. Permutational multivariate analysis of variance (adonis2) identified significant differences in assemblage composition with conductivity (F_1,94 = 5.75, p < 0.001) and biotic treatments (F_1,94 = 5.87, p < 0.001), and had a close to significant interaction (conductivity: biotic, F_1,93 = 1.53, p = 0.97), using mesocosm as a block (strata) effect. Individual multivariate homogeneity of group dispersions (betadisper) tests identified communities that became more variable when subjected to interactions with tolerant taxa (F_1,93 = 10.37, p = 0.0018), with no pattern identified by salinity (F_4,90 = 1.00, p = 0.41). Linear mixed effects modelling identified richness declined with increasing conductivity (t₂₈ = −2.44, p = 0.021), and was lower in tolerant-sensitive treatments compared to sensitive treatments (t₂₈ = −4.90, p < 0.0001), with no interaction between terms (t₂₈ = 0.102, p = 0.92).

Figure 3.

(a) NMDS with surfaces fit using general additive models of conductivity and the abundance of salt-tolerant taxa, (b) partial redundancy analysis (RDA) conditioned by blocks (mesocosms), constrained by conductivity, velocity, and the effects of biotic treatments and salt tolerant taxa abundance. (c) Variance partitioning further explored the combined and individual effects of significant environmental (conductivity, velocity and alkalinity), biotic predictors (salinity treatments and salt tolerant taxa abundance), and within treatment (mesocosm) effects on composition patterns. (Online version in colour.)

Variance partitioning of Hellinger-transformed relative community data identified physical variables explained 5.7% (alkalinity, conductivity and velocity) of community variance (adjusted R²), compared to 4.7% for biotic variables (biotic treatments and total salt-tolerant taxa abundances). Permutation tests (200) of individual fractions were significant for both physical environmental (F_3,88 = 2.36, p < 0.001), and biotic partitions (F_2,88 = 2.71, p < 0.001), while mesocosm was not significant (F_1,88 = 0.94, p = 0.50; figure 3c).

(c). Single taxa and metric responses

Taxa and metric specific responses were apparent based on hurdle model results, identifying differences between biotic treatments in relation to salinity as observed by slopes (figure 4) and standardized treatment slope coefficients (electronic supplementary material 1). Taxa on average were absent from the sensitive community 41% of the time, and 57% of the time in the community subject to interactions between tolerant and sensitive taxa. Numerous taxa appeared to show effects dominated by direct toxicity (figure 1b), such as Oecetis spp. (Trichoptera: Leptoceridae) and taxa within the Polycentropodidae family (Trichoptera). Responses of several taxa appeared to depend on interactions between biotic treatments and salinity (e.g. Conoescucidae spp., Trichoptera; figures 1a and 4). Some taxa appeared to respond to both direct toxicity and biotic treatment effects (e.g. Notalina fulva Kimmins (Trichoptera: Leptoceridae); figure 1c). Several taxa appeared to exhibit responses suggesting that biotic interactions were important irrespective of salinity (e.g. Agapetus AV 1 (Trichoptera: Glossomatidae), Gomphidae spp (Odonata); Newmanoperla thoreyi (Banks, 1920; Plecoptera); figures 1d and 4). Corynoneura spp. (Diptera: Orthocladiinae) appeared to increase in abundance in treatments mediated by interactions with tolerant taxa along the salinity gradient (figures 1f and 4). When differences in slope coefficients are examined, variable sensitivities are apparent and tolerant taxa appear to be less influenced by biotic treatments (electronic supplementary material 1).

Figure 4.

Predicted linear fits and 95% credible intervals for the 20 most common salt-sensitive taxa associated with conductivity and biotic treatments. (Online version in colour.)

Ephemeroptera abundance declined associated with salinity, with both direct salinity effects and in response to interactions with tolerant taxa (figure 5a). Salt-sensitivity appeared to increase in Plecoptera taxa in the presence of tolerant taxa, with little salinity effects in salt-sensitive only treatments (figure 5b). For Trichoptera and total invertebrate densities, direct toxicity effects predominated (figure 5c,d). This pattern was repeated within EPT, probably as a result of the large proportions of Trichoptera in EPT densities.

Figure 5.

Posterior coefficient estimates and 95% credible intervals for Ephemeroptera (a), Plecoptera (b), Trichoptera (c) and the total densities of all taxa (d). (Online version in colour.)

4. Discussion

Salinity is a growing and globally important threat to freshwater ecosystems [42]. Salinity effects on freshwater communities vary depending on a number of interacting variables, including co-occurring niche gradients, biological interactions and anthropogenic stressors [43,44]. We observed that salinity altered macroinvertebrate community composition, in both salt-sensitive treatments, and where salt-sensitive and salt-tolerant taxa were able to interact. Indeed, similar levels of community variation were explained by physical and biological factors (figure 3a–c). At the population level, however, salinity and tolerant-sensitive taxa interactions caused a range of species-specific and context-dependent responses (figures 4 and 5a–d). Clements & Kotalik [20] similarly identified organism responses to salinity can be species-specific and context-dependent. For example in our study, Austrophlebiodes pusillus (Ephemeroptera) and other Ephemeroptera taxa are sensitive to salinity [45–48], and declined with increasing salinity in the current study, which appeared to be moderately decreased with increasing biotic interaction strength (figure 5a); although competitive exclusion (i.e. a reduced intercept in tolerant-sensitive treatments) may have also caused this result. By contrast, Trichoptera, EPT, and total taxonomic densities declined at similar rates in both treatments, suggesting direct effects were most important across the gradient examined (figure 5a,c,d). At higher salinities, direct effects dominated community responses, resulting in reduced abundance, and altered community composition with almost complete loss of Ephemeroptera and much reduced Trichoptera abundance. We found direct effects of salinity on Plecoptera depended on biotic interactions with salt-tolerant taxa (figure 5b). Similar to the response of Plecoptera taxa that we found, Beklioglu et al. [31] identified low and environmentally relevant concentrations of the organic toxicant 4-nonylphenol had no observable effects on Daphnia magna Straus in single toxicity tests with abundant food, but had strong effects when coupled with effects that might be expected in a real ecosystem (i.e. food limitation and predator cues).

Standard toxicity test methods are structured to minimize control mortality and estimates of toxicity, therefore biological interactions are deliberately avoided [49]. Physical disturbance [50], food limitation [31], and chronic exposure [51], are similarly not typically considered in toxicity tests, although all influence toxicant effects [49]. There are many examples of biotic interactions influencing stressor effects, antagonistically [36], synergistically [35], through delaying recovery [19], with effects in both positive and negative directions [14,19,33,36,49,52,53]. Competitive interactions include exploitative competition for resources, interference competition mediated by aggressive interactions, and apparent competition where effects are indirectly modified through predators [54]. Liess [49] identified both interference and exploitative intraspecific competition caused an increase in pesticide effects in the cased caddis Limnephilus lunatus Curtis (Trichoptera). With limited environmental variability, increased intraspecific competitive effects may be expected given shared niche requirements. Indeed, intraspecific competition can be dominant among biotic processes [55,56]. Within landscapes, the dispersal of locally adapted genotypes is similarly likely to influence the outcomes of intraspecific competition [30]. However, communities subjected to greater variability in abiotic niche gradients (e.g. salinity here), and given organism fitness varies with these gradients, variation in the intensity of interspecific interactions is expected. This may be especially true where continuous dispersal of taxa known to be tolerant to this gradient is likely [4], which was imposed in our study through direct manipulation. In our study Lingora spp. and Agapetus spp. (both Trichoptera) and Newmanoperla thoreyi Banks (Plecoptera), Archichauliodes spp. (Megaloptera) had low densities in tolerant-sensitive treatments regardless of salinity, with effects suggestive of biological exclusion owing to tolerant taxa interactions (figure 1d). This is similar to the findings of Arco et al. [33], who identified that intraspecific competitive effects reduced treatment densities to carrying capacity during a 4 day pre-treatment phase and showed D. magna outcompeted the rotifer Brachionus calyciflorus through exploitative and interference competition. By contrast, Corynoneura spp. declined with increasing salinity in the salt-sensitive communities only, suggesting exposure to tolerant taxa may have caused complex indirect effects such as predatory or competitive release.

Order and family level effects were similar to taxa level responses in Ephemeroptera, but were more variable in Trichoptera and Plecoptera. For example, A. pusillus had similar responses to its order Ephemeroptera. However, there were differences in the two Plecoptera taxa examined (N. thoreyi and Dinotoperla fontana (Kimmins, 1951); figures 4 and 5). Trichoptera exhibited even greater variability in responses, including: densities influenced by salinity irrespective of interactions with tolerant taxa (e.g. total Trichoptera densities, Oecetis spp.); densities affected to a greater extent when exposed to tolerant taxa (e.g. Conoesucidae spp.); densities that may have been dominated by biotic interactions (e.g. Agapetus AV1); or densities that were largely unaffected by salinity or biotic treatments (e.g. Cheumatopsyche AV1).

Stressor effects can propagate throughout ecosystems, altering behavioural and trophic interactions with often unexpected outcomes [9,35,36,57]. Predator abundance was not manipulated in the current study, and top-down pressure was not different among treatments, with no detectable effects on any response examined. However, given variation in sensitivities within the assemblages examined here, and given that many predatory species were salt-tolerant, which is common [57], greater predatory effects may be expected with increasing salinity. Cañedo-Argüelles et al. manipulated the presence of a leech predator (Dina lineata) and salinity in mesocosms. The presence of this predator reduced herbivorous invertebrate abundance, leading to increased primary production, while salinity reduced taxon richness and caused significant changes to community composition within the benthos [58]. Salinity can also provide refugia from negative biological interactions. For example, Rogowski & Stockwell [59] identified both parasites and salinity were observed to have negative effects on pupfish (Cyprinodon tularosa), but high salinity caused a net benefit by reducing parasitism. By contrast, Piscart et al. [36] identified an acanthocephalan parasite (Polymorphus minutus) increased the acute salinity tolerance (LC₅₀) of the Gammarid amphipod Gammarus roeseli.

Within landscapes, population demographic stochasticity, speciation, and dispersal between patches influence local community dynamics [1–3]. At finer scales, coexistence [4], niche [5] and community theories [2] further suggest that patterns of assembly are shaped by dispersal, biological processes and physical gradients. Indeed, Carver et al. [60] used field and mesocosm approaches to show that insects select habitats for oviposition and colonization based on their salinity tolerances and habitat salinity, identifying behaviour influences organism distributions and abundances across salt-affected landscapes. Rico et al. [61] also identified toxicant effects are a single niche component among many, where hydromorphological and habitat parameters were also important in determining community composition. Our results support that biotic processes and abiotic environmental filtering differ among taxonomic groups and are collectively important determinants of community assembly. Furthermore, these local processes, biotic interactions and abiotic niche filtering, coupled with dispersal and stochastic effects can result in multiple stable equilibria [62]. Given this knowledge, we agree with Beketov & Liess [18] that inclusion of ecological theory such as meta-community, coexistence theory, macroecology and multiple stressor research are all necessary to advance understanding in ecotoxicology.

Low temperatures reduce the effects of salinity [63], biotic interactions [64] and probably their combined effects, therefore our results probably underestimate these effects because the current study was conducted during winter. To further explore patterns revealed in this experiment, the research presented here will be coupled with examination of trait-phylogeny-environment relationships to understand how traits, relatedness, and salinity may influence stream invertebrate communities. We expect trait-phylogeny-environment relationships to support that trait related fitness trade-offs occur associated with salinity tolerance [30].

5. Conclusion

Interspecific interactions can modify stressor effects and resulting patterns of community assembly. Studies that do not consider ecological processes such as biotic interactions may underestimate and fail to understand the true effect of a stressor in natural settings. Our results reinforced interspecific biological interactions both mediated salinity effects and were important on their own, irrespective of salinity toxicity, influencing taxa and community responses. Across the conductivity gradient examined, direct toxicity had a dominant effect on invertebrate densities. Salinity reduced the abundance and altered community composition, with almost complete losses of Ephemeroptera and other salt-sensitive species. Interspecific interactions between salt-tolerant and salt-sensitive taxa appeared to become more important as sensitivity to the toxicant increased. Several responses reported in other studies were identified here, supporting that species-specific and context-dependent effects may be widespread. In landscapes, ecological processes acting at differing scales are likely to contribute strongly to the effects of toxicants within ecosystems.

Data accessibility

Data are available in the Dryad Digital Respository: http://dx.doi.org/10.5061/dryad.n541d0t [65].

Authors' contributions

Overall design concept: B.J.K.; detailed design of the experiment: B.J.K., J.P.B., R.M.N., S.N., J.R. and R.T.; provision of funding: B.J.K., S.N. and R.M.N.; collection of data: J.P.B., J.R. and B.J.K.; data analysis: J.P.B., G.K.K.K., A.O.-N., production of first draft: J.P.B.; editing of manuscript and approval of final version: all authors.

Competing interests

We declare we have no competing interests.

Funding

The work reported here was funded by Australian Research Council Linkage Projects LP130100100 and LP160100093 awarded to B.J.K., and B.J.K., S.N. and R.M.N. respectively.