Relationships Between Environmental Conditions And Fish Assemblages In Tropical Savanna Headwater Streams

Riparian vegetation plays an important role in providing energy to small watercourses and maintaining ecological processes through organic matter input and together with hydrological and geomorphological watercourse characteristics influence on fish assemblages. The goal of this paper was partitioning and quantifying the influence of riparian zone (type of riverbank substrate, bank slope, type of riparian vegetation cover and percentage of riparian vegetation cover on the main channel), physical habitat (stream channel width and depth, type of substrate and aquatic habitat in channel, water velocity and organic matter), water quality (turbidity, temperature, conductivity, pH, dissolved oxygen and chlorophyll concentration) and spatial variables (linear distances between sampled points) on fish assemblages (richness and abundance per species) in headwater streams of the Upper Paraná River basin, Central Brazil. For this purpose, it was performed a variation partitioning analysis between riparian, physical habitat, water and spatial variables sets and a Redundancy Analysis to quantify the influence of variables on the fish assemblages. Only the physical habitat and water quality variables influenced the fish assemblages (richness and abundance per species).

Fish assemblages-environmental conditions relationships. According to the broken stick criteria, there were two significant axes for the PCAs performed separately on the water quality (77% of the variation) and physical habitat (81% of the variation), eight axes for the PCA performed on the riparian zone (87% of the variation) and three significant MENs (Moran's Index = 0.01 for each one) for the spatial variables.
The multiple linear regression showed no significant relation between fish abundance and the variables of the four groups considered (r 2 = 0.566; F (12,27) = 2.536; p = 0.128). In contrast, a significant relation was observed between fish richness (r² = 0.784; F (12,27) = 5.865; p = 0.001) and the PCA-1 of the physical habitat variables (p = 0.005; Table 2). All of the other compartments did not display significant relationship ( Table 2).
The relationship between assemblage and physical habitat variables is detailed by the RDA (total variance explained by the two axes = 53.4%; F (10, 17) = 3.543; p = 0.003). The first axis (35.17%) was positively correlated with conductivity, dissolved oxygen and chlorophyll concentration and negatively correlated with water temperature, whereas the second axis (18.23%) was positively correlated with organic matter, channel depth, pH and channel width and negatively correlated with turbidity (Fig. 2). The characins Piabina argentea and Astyanax altiparanae and the scrapetooths Parodon nasus were related to high values of water conductivity, dissolved oxygen and chlorophyll concentration, whereas the characins Astyanax fasciatus and Astyanax scabripinnis and the poeciliid Poecilia reticulata were associated with elevated water temperature values. The scrapetooths Apareiodon ibitiensis, the headstander Leporinus microphthalmus, and the toothless Steindachnerina insculpta were associated with elevated organic matter and pH values and a large and deep channel stream. The characin Bryconamericus stramineus, the callichthyid armored catfish Aspidoras fuscoguttatus and the South American darter Characidium zebra were correlated with high values of turbidity (Fig. 2).

Discussion
The riparian zone does not display any significant influences on fish abundance or richness in the headwater streams sampled. Similar results using a different methodology were obtained for fish diversity 43 in 1 st to 3 rd order headwaters streams in the Amazon region. This result suggests a low influence of riparian vegetation removal, assessed indirectly in this paper by the variables of the riparian zone group (type and percentage of the vegetation cover), on fish assemblages. However, studies focused on this subject have stressed the influence of the riparian zone on fish assemblages in the Amazon (channel fragmentation, deforestation 44 ; mechanized agriculture 43 ), São Francisco (deforestation 42,45 ) and Paraná River basin (deforestation 45 ), the last two of which contain the same vegetation cover of the area sampled in this paper (i.e., Cerrado).
The spatial component also showed no significant influence on fish assemblages. The abundance and richness of plants and animals, including stream organisms, are spatially structured 45,46 because of the influence of geology, the local climate 30 and the watercourse position along a longitudinal gradient 32 , especially for 1 st to 3 rd order streams 33 . However, if the 1 st and 2 nd order streams sampled in this study were in the same geologic (a combination of Precambrian metamorphic rocks, continental sedimentary rocks and tholeiitic basalts 47 ) and climatic (tropical climate with a dry season) domain, a similarity of fish abundance and richness could be expected. It suggest that the influence of environmental conditions and resources appear to be more influent than the spatial process, even that the sample sites are located in different basins.
In this study, fish richness was influenced by physical habitat (stream channel width and depth, and organic matter) and water quality (conductivity, water temperature, pH, chlorophyll, dissolved oxygen, and turbidity) variables. These variables are known to structure not only fish assemblages 4,48,49 but also their specific attributes, such as richness [50][51][52][53] . The results agree with those reported for Amazonian 43 and Cerrado fish assemblages of 1 st to 3 rd order headwater streams 42 , although some previous studies did not separate the influence of physical habitat and water quality variables from those of the riparian zone, as was done in this paper. Additionally, these physical habitat and water quality variables are better predictors of fish assemblage variability than riparian or catchment variables 43 or land use and the geophysical landscape 42 in Amazon and Cerrado headwater streams, respectively.
The influence of water conductivity on fish assemblages, as observed in this study, was also reported for tropical 54 and temperate watercourses 51 . Conductivity is a surrogate or correlate of water productivity, which influences freshwater fish body condition 45 , because it measures the electrical conductivity resulting from the concentration of dissociated ions 55 . Fish species can prefer aquatic habitats with specific requirements, such as elevated values of water conductivity, dissolved oxygen and chlorophyll concentration (as seen in the scrapetooths Parodon nasus and the characins Astyanax altiparanae and Piabina argentea in the watercourses sampled). In the case of P. nasus, the relationship observed is explained because this species is found in riffles 56 where there are elevated levels of dissolved oxygen. Furthermore, P. nasus, a periphyton scraper that prefers rocky substrates where algae and bryophytes are abundant, is associated with waters with high conductivity because of eutrophication 57  disturbances such as pollution 58 , which elevates water conductivity, and displays adaptations (i.e., a projection of the lower lip increase oxygen capture from water surface) to survive in low concentrations of dissolved oxygen 56 . Finally, the characin P. argentea is a midwater swimmer described as an opportunistic generalist species abundant in disturbed watercourses (modified from lotic to lentic conditions) 59 that is also positively correlated to dissolved oxygen concentrations in streams of the Upper Paraná River basin 60 . The poecilid P. reticulata, an exotic species in Brazilian watercourses, and the characin A. fasciatus are tolerant to habitat alterations 57,61 . Additionally, A. fasciatus and A. scabripinis (to a lesser extent 62 ) are sensitive to water temperature because of the influence on their reproduction cycles 63 , whereas P. reticulata displays female-choice sexual selection 64 , fry production 65 , schooling behavior 66 , and aquatic surface respiration (ASR) to meet oxygen demand in hypoxic water 67 regulated by the water temperature. These relationships explain the affinity of these species for the water temperatures found in the streams sampled. However, this affinity, especially for P. reticulata and A. scabripinis, can change during the low-and high-water seasons, when both species are associated with low water temperature 68 .
The accumulation of organic matter, such as trunks and bundles of leaves, may be responsible for species coexistence in different habitats. This coexistence can occur because of the increase in habitat heterogeneity resulting from organic matter input 69,70 from the surrounding riparian zone or the transport of leaves and other matter from upstream to downstream [71][72][73][74][75] , which are then deposited in stream areas with low water velocity 76 . This seems to be the case in this study for the scrapetooths Apareiodon ibitiensis, a detritivorous species that scrape the algal film adhered on the surfaces of rocks and logs 77 , the toothless characin Steindachnerina insculpta, a bottom feeding fish 55 , and the headstander Leporinus microphthalmus, which, like other anostomids, feeds on sponges, detritus, insects, seeds, leaves, and filamentous algae, in the substrate 78,79 .
Additionally, the preference of these species for relatively large and deep streams can be related to their body length (A. ibitiensis = 11.3 cm, S. insculpta = 16.1 cm, L. microphthalmus = 11,8 cm 54 ), as reported for A. ibitiensis 80 . However, the results found can be influenced by local or regional modifications. For example, the fragmentation of a channel or watercourse and local/regional deforestation influence the organic matter inputs (leaves, trunks and stems in this case), habitat complexity and riverbed stability. This, in turn, influences fish richness, as pointed out for Amazonian headwater streams 44 .
Among the species sampled, the callichthyid armored catfishes Aspidoras fuscoguttatus, the characin Bryconamericus stramineus and the South American darter Characidium zebra are associated with high water turbidity. The callichthyid A. fuscoguttatus is a bottom dwelling species that swims near the watercourse substrate gathering food ("grubber excavating while moving" 81 ). This behavior can explain its ability to exploit the watercourse substrates, which are covered by fine sediments 56 that are transported by water, and its capacity to survive in streams that have remarkable seasonal oscillation in turbidity, with lower values during the dry period and higher values in the rainy period 82 . On the other hand, the characin B. stramineus is a predominantly insectivorous 83 active swimmer 84 that is abundant in shallow streams of the Upper Paraná basin with elevated turbidity 83,85 and water velocity 85 . The relationship of C. zebra with water turbidity is unexpected considering that it is an indicator species of pristine environments, with a sit-and-wait behavior for capturing prey 86 and rheophilic preferences that can be affected by high levels of suspended sediments in the water column and the resulting siltation of the substrate 54 .
Among the four groups of environmental variables considered, only those related to the physical habitat and water quality significantly influenced the richness of the fish assemblages. This influence is explained by the interaction of the fish assemblages with nine variables (conductivity, water temperature, pH, chlorophyll, organic matter, dissolved www.nature.com/scientificreports www.nature.com/scientificreports/ oxygen, turbidity, channel width and channel depth). These results indicate that local instream characteristics of headwater streams have more influence on fish assemblages than factors associated with the riparian zone in Cerrado river basin draining areas. The comparison between these findings and those from the Amazon River basin suggests that this influence exists regardless of the river basin and its vegetation cover (Cerrado and Amazon in this case).

Materials and Methods
Study area. Twenty-seven sites (one sample site per stream) of the 1 st and 2 nd order tributaries of the Meia Ponte River (seven streams; 2.7 to 10.2 km apart from each other), Piracanjuba River (14; 4.8 to 17.8 km) and Santa Maria River (six; 4.8 to 6.0 km) were sampled, all of which are located in the Southeast Region, Goiás state, Upper Paraná River basin, Central Brazil (Fig. 3, Table 3). Sampling was conducted between April and September   www.nature.com/scientificreports www.nature.com/scientificreports/ 2009, which corresponded with the dry season of the regional climate (Aw per the Köppen-Geiger classification). The Paraná River basin drainage is located on sedimentary deposits corresponding with the Paleozoic and Cenozoic and covered by basalt from the Jurassic-Cretaceous age 47 . The sampling stations are located on a combination of three types of rocks: i) Precambrian metamorphic rocks; (ii) continental sedimentary rocks; and (iii) tholeiitic basalts, which are abundant in the Paraná basin 47 . The vegetation cover of the Meia Ponte and Piracanjuba River basin was deciduous forest, and that of the Santa Maria basin was a semideciduous forest, all of which belong to the Cerrado (the Brazilian savanna biome).
In each stream, one 100-m site was selected according to its accessibility, marked and georeferenced (Garmin GPSMAP64. Each site was divided into 11 transects, one every ten meters, where the data collection for both fish assemblages and variables was performed. All sites were away from urban areas and were found in a landscape matrix formed mainly by pasture. The exception was site P17, which was surrounded by a sugarcane crop. The sites sampled had riparian vegetation covering the stream channel and at least one opening, which was intended for watering livestock or replaced by grass for feeding cattle (site P5), in the riparian cover along the site. The channel depth of the stream sites ranged from a minimum of 0.10 (P2 and P20) to a maximum of 0.53 m (P12), whereas the channel width ranged from 0.60 (P7) to 7.78 m (P14; Table 3). The predominant substrate in the sites sampled was sand, except in P4, P13, P19 (gravel) and P11 (rocky outcrops; Table 3). The predominant aquatic habitat type was lotic except in stretch P9. Upstream site P17 was located in a reservoir.
Sampling protocols. Sixteen environmental variables were measured in each site. Six variables were associated with physical habitat, six with water quality and four with the riparian zone (Table 4).
Riverbank substrate, riverbank slope, aquatic habitat, type of riparian vegetation cover and percentage of riparian vegetation cover were visually characterized at each transect (along both riverbanks) along with luminosity (photometer; Polaris), stream channel width (measuring tape), stream channel depth (graduated rope) and water velocity (flowmeter; General Oceanic 2030). At the initial, middle and final transects of each site, organic matter samples of the stream channel bed and water were collected to determine algae biomass and to measure the physical and chemical variables. www.nature.com/scientificreports www.nature.com/scientificreports/ Organic matter was collected using a Surber sampler (30 × 30 cm). In the laboratory, the samples were dried at 100 °C for 24 hours and weighed (SC2020 -Ohaus; 0.001 g) 87 .
Alpha chlorophyll concentration was used as a reliable and common proxy for the total phytoplankton biomass 88 , which may vary according to the degree of shading caused by riparian forests in headwater streams 19 . In the field, 25 L of water was filtered directly from the stream using a plankton net (mesh 1 μm) and a water pump (P835; Stihl). The product of the filtering process was placed in a 600 ml opaque bottle containing 1 ml of saturated magnesium carbonate. In the laboratory, the samples were filtered (cellulose ester membrane; porosity 0.45 μm) and quantified by spectrophotometry (spectrophotometer; Varian-Cary-50 CONC) 89 . The a, b and c chlorophyll concentrations were calculated following the Jeffrey and Humphrey equation 90 .
Water turbidity (turbidimeter; LaMotte 2020), temperature and conductivity (thermometer/conductivity meter WTW 3015i) and dissolved oxygen (DO-Lutron 5510) were measured at ~20 cm depth. The water turbidity, temperature, conductivity, dissolved oxygen and water velocity were measured at ~20 cm depth, whereas luminosity and air temperature were measured at ~20 cm above the water surface.
Fish were collected by shore electrofishing (electrofisher DC, 100-600 V plugged into a 220 V electric generator) modified from 91 ; that is, the site's length was 100 m and traversed only one time instead of being 50-80 m in length and traversed three times. Both modifications were performed based on the results of 92 , taking into account the logistics of the electrofishing gear used and displacement difficulties that occur along Cerrado streams because of physical conditions (e.g., trunks and steep stream bank). Four people collected samples for one hour in each site. The collected fish were placed in plastic bags, euthanized with a saturated clove oil solution and fixed in formalin (10%). All the bags were identified with tags containing the stream and site code. Fish was collected in the dry season when captures are more efficient because of lower water levels 93   www.nature.com/scientificreports www.nature.com/scientificreports/ water velocity and organic matter. The third consisted of water quality variables (average values): turbidity, water temperature, conductivity, pH, dissolved oxygen and chlorophyll concentration. The fourth consisted of variables related to the riparian zone (frequency values by category): riverbank substrate, riverbank slope, type of riparian vegetation cover and percentage of riparian vegetation cover in the channel. The fifth data matrix grouped the main spatial eigenvectors (MENs) 94 , which constitute a representation of the spatial process resulting from the analyses performed on the spatial data matrix (geographic coordinates) considering a linear distance (Euclidean distance) between sampling points. The MENs represent spatial autocorrelations (Moran's index) and can be used as a surrogate for the dispersion ability of species 94,95 . Significant MENs were considered those with Moran's index values < 0.05. All the procedures to obtain the MENs were performed in SAM macroecology software 96 .
To determine the influence of the variable groups (physical habitat, water quality, riparian zone and spatial) (environmental variables) on the fish (biotic structure), a variation partitioning analysis was performed. After that, each data matrix was transformed to a similarity matrix using a specific index (Bray-Curtis for fish species abundance and Euclidean distance for all the other data matrices) and nonparametric multidimensional scaling (NMDS) was performed 97 . Using the resulting NMDS, a correlation (Procrustes analysis 98 ) was performed separately between the fish assemblages and the physical habitat, water quality, riparian zone and spatial groups (9999 permutations 99 ).
To determine the relationship between the fish assemblages and the variable groups (physical habitat, water quality and riparian zone), two multiple linear regressions were performed: the first one was for fish species richness, and the second one was for fish species abundance. A principal component analyses (PCA) was performed