In many developing countries, a large proportion of untreated raw sewage is released into aquatic environments (Macedo and Sipaúba-Tavares 2010), increasing the load of organic matter and pollutants in rivers, which are considered as the main drivers of artificial eutrophication in these environments (Tundisi and Tundisi 2008). Nutrients from human activities, when released into the water, contribute to the rapid growth of algal blooms and aquatic plants, altering the physico-chemical and ecological conditions of aquatic systems (Pereira and Mercante 2005, Hicks et al. 2016, Baeta et al. 2017). Among the consequences of an increase in primary productivity is the rapid reduction in water oxygen levels, which has drastic effects on fish and invertebrate communities (Macedo and Sipaúba-Tavares 2010, Baeta et al. 2017). In addition, changes in primary productivity in response to pollutants affect directly the diets of these aquatic consumers (Cabana and Rasmussen 1996, Esteves and Aranha 1999), and may also be responsible for promoting the local extinction of specialist and less tolerant species.

By favoring primary productivity, environmental pollution of aquatic systems may homogenize the type of resources available to organisms in higher trophic levels. This process of homogenization in aquatic systems has been also described in several taxonomic groups such as diatoms, zooplankton and macroinvertebrates (Lougheed et al. 2008). Such changes in the balance of available resources may consequently affect the food web since changes in nutritional composition or abundances of basal food sources can induce shifts in primary consumers or even their exclusion (Hall 2004). However, the effects of this homogenization of producer communities on upper trophic levels remains unclear.

According to ecological theory, generalist species are less sensitive to environmental change than specialists as they have the capacity of varing their diet according to the availability of resources present in their respective habitats (Tundisi and Tundisi 2008). Therefore, trophic plasticity is an important strategy to allow species to tolerate changes in environment condition (Gerking 1994, Wootton 1999). Species of the American fish Astyanax Baird & Girard, 1854 are well known for their broad geographic distribution and their ability to inhabit environments with differing levels of preservation, including highly polluted environments (eg., Souza and Lima-Júnior 2013, Carvalho et al. 2015). Astyanax is composed of approximately 100 species that are distributed from the southern United States to northern Argentina (Eigenmann 1921, Géry 1997, Weitzman and Fink 1983). Most species have omnivorous feeding habits, with diets composed of animal and vegetable items, of both autochthonous and allochthonous origins (e.g., Esteves 1996, Vilela et al. 2002, Cassemiro et al. 2002, Bennemann et al. 2005). In addition, some species of this genus present generalist feeding habits and high trophic plasticity in response to environmental changes and resource availability (Lobón-Cerviá and Bennemann 2000, Carvalho et al. 2015), which increases their chance of survival in disturbed habitats (Menezes et al. 2007). However, congeneric species (species of the same genus) may respond differently to changes in the aquatic environment and the availability of resources.

One way to identify how distinct species respond to changes in the environment is by comparing their feeding habits in regions under differing levels of human disturbance (e.g, Carvalho et al. 2015). Accordingly, analyses of stomach contents and stable isotopes (carbon and nitrogen), can be used simultaneously for robust and reliable assessment of feeding habits (e.g., Carassou et al. 2017, Connan 2017). Stomach contents analyses provide useful taxonomic information on consumed prey items (Beaudoin et al. 1999). However, there are often uncertainties in the identification of such items, due to the different stages of digestion of food items, and that not all ingested items are in fact assimilated into biomass (Manetta and Benedito-Cecílio 2003). Stable isotopes analyses, on the other hand, provide information on the energy flow in food chains (Peterson and Fry 1987, Kling et al. 1992). The nitrogen (δ¹⁵N) isotope is consistently fractionated throughout the trophic web, allowing researchers to make inferences about the trophic relationships of consumers with their diet (Vander Zanden et al. 1997, Post 2002, Vanderklift and Ponsard 2003). The carbon (δ¹³C) isotope, in turn, allows to delineate the energy flow in environments that present several foods with different carbon values (Manetta and Benedito-Cecílio 2003).

Based on this information, we aimed to evaluate how trophic characteristics of two congeneric species, Astyanax lacustris (Lütken, 1875) and Astyanax taeniatus (Jenyns, 1842), change across an environmental pollution gradient. The diet and the trophic niches occupied by these two species were evaluated in different regions of a highly disturbed Neotropical river basin, the Rio das Velhas, south east Brazil. The main source of disturbance in this river basin is the discharge of untreated domestic and industrial sewage from a large nearby urban conurbation. We tested the following hypotheses: 1) Under natural (undisturbed) conditions, the congeneric species occupy different trophic niches, and consequently present little food overlap; 2) However, along a gradient of pollution, due to the simplification (homogenization) of the available resources, and due to their high trophic plasticity, both species will increase their food overlap and will present more similar isotopic signatures.

The study was conducted in the Rio das Velhas Basin, southeast Brazil, with sampling sites located in the main channel of the Rio das Velhas. The Rio das Velhas is the largest tributary of the São Francisco river Basin (Alves and Pompeu 2001), and is located entirely in the territory of Minas Gerais state (CETEC 1983), covering 51 municipalities. The basin, with a drainage area of 29,173 km², has an average annual flow rate at its mouth of 300 m³/s and average width of 38.3m (CETEC 1983). The Rio das Velhas is of significant economic and social importance. Its upper course is located at the Metropolitan Region of Belo Horizonte (MRBH), the third largest urban conurbation in Brazil, with almost six million inhabitants, and is the main water supply.

Eight sites were sampled along the Rio das Velhas channel (RV-01 to RV-08), which were divided into three regions (upper, middle and lower course). The upper course of the Rio das Velhas (Upper RV) corresponds to the region with the best water quality (RV-01 and RV-02). The middle course (Middle RV) is in the region with the greatest influence of the MRBH, characterized by the discharge of large amounts of domestic and industrial sewage (RV-03, RV-04, and RV-05). The lower course (Low RV), in turn, is the most distant region from the MRBH and is close to the river mouth (RV-06, RV-07, and RV-08). In this region the river partly recovers its quality, due to the presence of numerous well preserved tributaries (Alves and Pompeu 2001) (Table 1, Fig. 1).

Two sewage treatment plants (STP), Arrudas and Onça, were also sampled to obtain complementary samples of the suspended material to obtain the isotopic composition of the raw sewage. All the sites were sampled between the years 2015 and 2016, in the dry (May to August) and wet (October to January) seasons (Table 1, Fig. 1).

The information about degradation level of sampling sites was obtained through data from literature (Feio et al. 2015). The sites RV08, RV11, RV10, RV12, RV13, RV14, RV15 and RV16 (Feio et al. 2015) were considered as the correspondents of RV-01, RV-02, RV-03, RV-04, RV-05, RV-06, RV-07 and RV-08, respectively (Table 1). Degradation levels range from I (preserved) to IV (degraded).

Data about water quality, hypereutrophic condition, toxic contamination and pressure factors in the study sites were accessed through IGAM's website (http://portalinfohidro.igam.mg.gov.br), which monitors water quality quarterly at several points across the Rio das Velhas Basin. The IGAM monitoring sites: BV001, BV139, BV105; BV137; BV141, BV150, BV151 and BV149, were considered as sampling points: RV-01, RV-02, RV-03, RV-04, RV-05, RV-06, RV-07 and RV-08, respectively (Table 1). Mean values of conductivity, dissolved oxygen, total ammoniacal nitrogen and total phosphorous were obtained from IGAM measurements carried out in the years 2015 and 2016. The hypereutrophic condition, toxic contamination and pressure factors acting in the study sites were extracted from the quarterly report of the year 2017.

Captures of specimens of A. lacustris and A. taeniatus were carried out with gillnets with mesh sizes of 2.4, 3.0 and 4.0 cm between opposing nodes and with cast nets, seines and sieves. A total of 137 individuals of A. lacustris and 103 individuals of A. taeniatus was sampled in the three regions. The captures with gillnets represented 63% of sampling. For the stable isotope analyses, we collected at least five samples of each species at each sampling site (whenever possible). In the field, dorsal muscle was removed for large specimens and for small the whole fish was used removing the digestive tract. All samples were kept frozen until laboratory processing to avoid decomposition and deterioration of the material. In the laboratory, the fish samples were lyophilized for 24 hours, ground to fine and homogeneous powder using mortar and pestle and stored in eppendorf tubes.

The individuals that were not selected to stable isotopes analyses were fixed in formalin (10%) in the field, washed in water after fixation and transferred to alcohol (70%) in laboratory. Individuals predated or in high stage of decomposition were discarded. The remain individuals were used to stomach contents analyses in laboratory, where they had their stomach contents carefully removed. The same individuals were not used for both isotopic and stomach contents analyses because the stomach contents were analyzed following the results of stable isotope analyses, when we detect the need for complementary information.

Whenever possible, we collected five samples of all basal food resources available at each sampling site: periphyton, filamentous algae, suspended matter, fine particulate organic matter (FPOM) from sediments, vegetation (grasses and riparian vegetation), coarse particulate organic matter (CPOM), and aquatic macrophytes. Complementary samples of the suspended material were made at the sewage treatment plants to obtain the isotope signature of the raw sewage.

Samples of algae, aquatic macrophytes, vegetation and CPOM were collected at all sites where they were present, stored in plastic bottles and kept frozen until laboratory processing. Filamentous algae and aquatic macrophytes were collected manually in each site where they were present. Leaves from pasture (grasses) and from the natural riparian vegetation were manually collected along river banks in each site, with the most common species being prioritized at the site. The CPOM was randomly collected from leaf litter deposits in the streams.

Liquid samples, like periphyton, suspended matter (including sewage samples) and sediment, were collected at each site and kept frozen until processing in laboratory, where they were filtered using a filtration device attached to a vacuum pump using calcined quartz fiber filters (Whatman^® QMA quartz filters). The periphyton was collected by scraping rocks with a brush and placing the material in a plastic bottle with distilled water. FPOM samples were collected from sediment deposits revolving in each sampling site and stored in plastic bottles. The suspended matter presented in the sampling sites and at STPs were collected with a phytoplankton net (0.45 mm mesh) deployed for a period of three minutes at each sampling site.

In the laboratory, all basal resource samples were dried in an oven at 60 °C for 48 hours and then ground with a mortar and pestle and stored in Eppendorf tubes.

The contents of 44 stomachs of A. lacustris, and 31 stomachs of A. taeniatus were analyzed in total. Food items were weighed (0.001 g accuracy/ wet weight) and identified under stereomicroscope to the lowest taxonomic category possible. The frequency of occurrence (Fi = number of times item i occurred, divided by the total number of stomachs) and the relative weight (Pi = sum of the weight of item i divided by the sum of the weight of all items) of each item were obtained. The food index (IA), proposed by Kawakami and Vazzoler (1980), was then calculated for each species and region, according to the formula: IAi = (Fi.Pi)/ΣFi.Pi, where, IAi = food index of item i; Fi = frequency of occurrence of item i, and Pi = weight of item i.

The degree of overlap in food items between species was calculated using the simplified Morisita index (Morisita-Horn index) (Krebs 2014), according to the formula below: C_H = 2ΣPij.Pik/ΣP²ij + ΣP²ik, where, C_H = Simplified Morisita Index of overlap (Horn 1966) between species j and species k; Pij = Proportion resource i is of the total resources used by species j; Pik = Proportion resource i is of the total resources used by species k, and n = Total number of resource states (I = 1, 2, 3, ... n).

For the food items characterization, “detritus" was considered dead particulate organic material, “sediment" included inorganic particles, and “plant remnants" were related to fragment of terrestrial vegetation.

A total of 42 samples of A. lacustris, 47 samples of A. taeniatus and 703 basal resources samples were sent to the Center for Nuclear Energy in Agriculture (CENA) at University of São Paulo (USP) for isotopic analysis. About 2–5 mg of dry animal tissue material and approximately 5–10 mg of basal resources samples were selected for analysis.

To determine the isotopic ratio, a mass spectrometer system in the Continuous-flow (CF-IRMS) mode was used with a Carlo Erba elemental analyzer (CHN 1110) coupled to a Delta Plus mass spectrometer (Thermo Scientific). Results were expressed as relative difference of international reference standards, in the delta notation (δ ‰), and calculated using the following formula: δX = [(R_sample /R_standard)^-1] x 10³, where X is ¹³C or ¹⁵N and R represents the isotopic ratio ¹³C/¹²C or ¹⁵N/¹⁴N (Barrie and Prosser 1996).

Differences in isotopic ratios of δ¹³C and δ¹⁵ N of consumers and resources between the three regions were tested using one-way analysis of variance (ANOVAs) when the normality and homoscedasticity assumptions were met. The nonparametric Kruskal-Wallis test was used for data with non-normal distribution. When significant differences (p < 0.05) were observed, means were compared using the post-hoc Tukey's test. We also tested if isotopic signatures of A. lacustris and A. taeniatus presented variation between the dry and wet season, using t-tests (normal distribution) and Mann-Whitney tests (non-parametric). These analyses were performed in the software Statistica 6.0 (Statsoft 2004).

To evaluate the trophic structure of A. lacustris and A. taeniatus populations, individuals of the two species were plotted in the bi-plot space according to the isotopic values of carbon (x-axis) and nitrogen (y-axis) in each region (Fig. S1). Source contributions to the A. lacustris and A. taeniatus diet were estimated for the three regions based on stable isotope data analyzed through Bayesian stable isotope mixed models (Moore and Semmens 2008, Parnell et al. 2010), specifically using the MixSIAR package in R (Stock and Semmens 2016a). For both analyses, only the autochtonous sources, algae and periphyton, and the allochtounous sources, leaves of riparian vegetation and grasses were taken into account. The samples of sewage were also considered to sites “Middle RV" and “Low RV" since they are located in the area under influence of pollution. We used Markov chain Monte Carlo sampling based on the following parameters: number of chains = 3; chain length = 100,000; burn in = 50,000; thin = 50 and model 4 (Resid*Process) error structure (Stock and Semmens 2016b). Diagnostic tests (Gelmin-Rubin, Heidelberger-Welch and Geweke) and trace plots were examined for model convergence. The fractionation values used for consumers were 0.4 ± 1.3 ‰ for Carbon and 2.54 ± 1.27‰ for Nitrogen (Vanderklift and Ponsard 2003, Post 2002). Both the graphical representation and the partition analysis were done using the MixSIAR package in the R programming environment (Stock and Semmens 2016a).

The isotopic niches of A. lacustris and A. taeniatus in both regions (Upper RV, Middle RV and Low RV) were quantified based on standard ellipse areas (SEA – expressed in ‰²) through use of the Stable Isotope Bayesian Ellipses package in R (SIBER, Jackson et al. 2011). The standard ellipse area (SEA) represents the core isotopic niche space and it is a proxy of the richness and evenness of resources consumed by the population (Bearhop et al. 2004). All measures were “bootstrapped" (n = 10,000, indicated by the letter “b") to compare groups with different sample sizes. A small sample size correction (indicated by the subscript letter “c") was applied to SEA to increase the accuracy of the comparisons, enabling the comparison of niches of populations with different sample sizes (Jackson et al. 2011). The SEAc allows to calculate the degree of niche overlap (in percentage, where 100% indicates total overlap) and can be used as a quantitative measure of diet similarity among different species (Hill et al. 2015).

Food overlap between the two congeneric species was low in the least-disturbed region (upper Rio das Velhas), confirming our first hypothesis, that closely-related sympatric species diverge in their trophic niche to allow coexistence. In this study, the species A. lacustris and A. taeniatus presented high trophic plasticity in response to pollutants, increasing their food overlap and presenting similar isotopic signatures in the heavily polluted areas. Such aspect confirm our second hypothesis, that human disturbance promotes homogenization of fish species' diets. Despite plant and insect remnants were the predominant items in the stomachs of A. lacustris and A. taeniatus, algae and periphyton were also important food items (especially in lower sites). The importance of autochthonous resources as food items was highlighted in the partition analysis, which indicated that algae (in polluted regions) and periphyton (in least-disturbed region) were the most assimilated resources for both species.

The variation in δ¹³C and δ¹⁵N compositions and in stomach contents of A. lacustris and A. taeniatus along the Rio das Velhas highlight their generalistic habits and high trophic plasticity (Manna et al. 2012, Carvalho et al. 2015) probably as a result of resource availability found in the aquatic environment (Lobón-Cerviá and Bennemann 2000). The A. taeniatus also changes its δ¹⁵N compositions between seasons (with enriched values in the dry season) which can be due changes on trophic levels (Vander Zanden et al. 1997), but also can occur in response to a higher enrichment in δ¹⁵N values of resources in dry seasons.

The predominance of insects and plant remains in their stomach contents as well as the consumption of algae/periphyton are in agreement with the literature (e.g. Andrian et al. 2001, Casatti et al. 2003, Bennemann et al. 2005, Souza and Lima-Júnior 2013). It is likely that periphyton (in least-disturbed sites) and algae (in more degraded sites) are also being consumed indirectly through the consumption of insects. Castro et al. (2016) also observed a trend of changes in macroinvertebrates assimilation of algae and periphyton between degraded and preserved environments, which reinforces our statement. These changes in the type and proportion of autochthonous resources that sustain the two species are probably due to changes in environmental conditions as a result of pollution. Periphyton (or biofilm) is defined as an integral and independent micro-ecosystem in aquatic ecosystems, harboring biotic components (like algae, fungi, bacteria, protozoans, metazoans) and abiotic components (like substrata, extracellular polymeric substance and detritus) (Wu 2016). These organisms occur on the surface of rocks and submerged vegetation (Tundisi and Tundisi 2008), in environments with good water quality and greater presence of rocks and wood that favor the proliferation of periphyton. On the other hand, an increase of nutrients in the aquatic environment triggers a marked increase in algae (Tundisi and Tundisi 2008). Therefore, the expected greater abundance of algae in areas under the influence of pollution and of periphyton in sites with better environmental conditions, explain the shifts on basal resources assimilated by fish species.

In this study, stomach contents and stable isotopes analyses showed that there is a tendency to niche overlapping in A. lacustris and A. taeniatus in the presence of pollutants. The percentage of niche overlap observed by stomach contents and stable isotopes analyses were not the same, which is expected since not all items found in fish stomachs are assimilated (Manetta and Benedito-Cecílio 2003). In addition, the items consumed only occasionally or accidentally by individuals are observed on stomach contents, but will not be reflected on isotopic composition of fish. The greater overlap observed in the middle and lower course of Rio das Velhas could be due to the lower heterogeneity and resource availability in impacted sites (Gutiérrez-Cánovas et al. 2015). Fish tend to exhibit greater selectivity and specialization in the resources consumed in heterogeneous aquatic ecosystems, while in environments with few resources (or predominance of a single resource), fish tend to share the same food items (Knoppel 1970, Hurlbert 1978). Although we did not measure algae abundance, it is expected that in polluted sites populations reach high densities (e.g. Lata Dora et al. 2010, Macedo and Sipaúba-Tavares 2010), becoming an important food source consumed either directly or indirectly by generalist species, which can explain the higher niche overlap between A. lacustris and A. taeniatus in the middle and lower regions.

Trophic niche amplitude differed between regions. In the undisturbed region (Upper RV), both species had a broader trophic niche on the horizontal axis (niche with great carbon range). This trend is expected in food webs in which there are multiple basal resources with varying δ¹³C values, enabling niche diversification at the base of a food web (Layman et al. 2007), which indicates that A. lacustris and A. taeniatus feed on a greater range of resources under natural conditions. On the other hand, in the most polluted region (Middle RV), the two species presented a narrow carbon range and a large nitrogen range (more vertical trophic niche). The narrow carbon range may be occurring in response to the restriction and homogenization of available food resources (the opposite of what has been observed in preserved sites). A larger range in δ¹⁵N sometimes suggests more trophic levels and thus a greater degree of trophic diversity (Layman et al. 2007), however, probably this is not the explanation to the niche verticalization observed in this study, but the greater enrichment observed in the basal resources. This verticalization of the trophic niche has been found in fish (De Carvalho et al. 2017) and macroinvertebrates (Castro et al. 2016) in environments impacted by other anthropogenic activities (sugarcane).

The enriched nitrogen values of fish and resources especially in the middle section are probably related to the influence of sewage effluents, since δ¹⁵N values of domestic wastes ranges between 7‰ to 38‰ (Dailer et al. 2010). Domestic wastes are nitrogen enriched especially because of isotopic fractionation during nitrification and volatilization in the case of ammonium, or denitrification in the case of nitrate (Nikolenko et al. 2018). Therefore, the uptake of enriched δ¹⁵N by primary producers are reflected in the entire food web (McClelland et al. 1997). Changes in δ¹⁵N values along the pollution gradient, with a particularly large increase in regions affected by sewage effluent, were similar to those reported in studies of macroinvertebrates (e.g. Morrissey et al. 2013, Pastor et al. 2014, Baumgartner and Robinson 2016), and primary producers (e.g. MCclelland and Valiela 1998, Cole et al. 2004, Wang et al. 2016). Together, results of these studies, including ours, support the finding that high δ¹⁵N values are good indicators of anthropogenic stress in aquatic systems.

Therefore, stomach contents and stable isotope analyses were very useful to evaluate the effects of the presence of pollutants in the trophic ecology of two congeneric species. It was possible to observe that even where species originally present different feeding habits (verified through the analysis of the stomach contents), food webs were mainly based on autochthonous items, such as algae and periphyton (verified through the isotopic analysis), assimilated directly and indirectly through aquatic insects. The presence of pollution, besides triggering increased food overlap between A. lacustris and A. taeniatus, also promoted an enrichment in δ¹⁵N values of fish and resources. The δ¹⁵N values of fish seems to be an effective means to detect anthropogenic impacts in aquatic ecosystems. In addition to providing important information on species biology, our work contributes to elucidate one of the 100 key ecological issues (Sutherland et al. 2013): How do resource pulses affect resource use and interactions between organisms?

We thank the Agência Peixe Vivo and the Comitê de Bacia Hidrográfica do Rio das Velhas (CBH, Rio das Velhas) for the project financial support and the Projeto Manuelzão for the logistical support. Thanks to Aline J. Grossi (Universidade Federal de Lavras, UFLA) and Luiza Hoehne (Universidade Federal de Minas Gerais, UFMG) for the support on processing of samples. Thanks also to the Benthos Ecology Laboratory (UFMG) and the Laboratory of Fish Ecology (UFLA) that allowed processing of samples and infrastructure, and to the Centro de Energia Nuclear na Agricultura (CENA) for their support and partnership in the isotopic analysis. PSP received a research grant and a research fellowship from the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq 303548/2017-7) and from the Fundação de Amparo à Pesquisa do Estado de Minas Gerais (FAPEMIG PPM-00608/15).This study was financed in part by the Coordenação de Aperfeiçoamento Pessoal de Nível Superior (CAPES) – Finance code 32004010017P3. The manuscript underwent grammar revision by Alistair Campbell.