Fish responses to multiple scales in coastal blackwater Atlantic Forest streams in Southeast Brazil

Abstract Environmental factors act at multiple spatial scales in a hierarchical manner to shape the organization of biota. However, the relative influence of different scale-related factors is poorly known, especially in Atlantic Forest Blackwater streams. Therefore, we herein aimed to evaluate local, landscape and spatial factors that shape fish assemblages in 14 blackwater restinga coastal Atlantic Forest streams under natural conditions and verify species occurrence patterns among four sub-basins during the low-precipitation season. When we combined local, landscape and spatial factors, variance partitioning explained a high proportion of variation in species matrix. Local variables pH and Total Dissolved Solids explained most of the variability, and these were the most important factors in determining fish community structure. Significant differences in fish assemblage structure among the four sub-basins were observed, and Mimagoniates microlepis, Phalloceros harpagos, and Hollandichthys multifasciatus were the species that most contributed to this dissimilarity. The important contribution of local predictors, the high number of endemic species herein recorded, the presence of an endangered species (Spintherobolus broccae), and near pristine conditions, may be used as baseline conditions for the assessment of similar environments.


INTRODUCTION
Metacommunity theory has been widely used to understand how the complex spatial structure of river networks mediates the relative influences of local and regional control on species composition.This framework considers that processes act at multiple spatial scales, and that community composition is subject to the interaction between local environmental conditions (both biotic and abiotic) and dispersal driven regional effects (Brown et al., 2011).Another interpretation is linked to the Niche Theory, which predicts that resources and conditions dictate the distribution of species over space and time (Hutchinson, 1957), which, together with local conditions and species' dispersion abilities influence community structure (Cottenie, 2005).Thus, multiscale approaches have been considered good tools to gain a better understanding of broader scales that affect the composition of aquatic communities (Frissell et al., 1986;Schlosser, 1991), and may help identify the most appropriate scales and environmental factors for the management of aquatic ecosystems (Roth et al., 1996).
Stream communities are largely determined by the structure and dynamics of the physical stream habitat, together with the pool of species available for colonization (Wevers, Warren, 1986).Conditions at smaller spatial scales are constrained by processes at larger spatial scales (O'Neill et al., 1989).As such, multiple spatial scales act hierarchically to shape the biota's organization (Compin, Ceréghino, 2007;Stephenson, Morin, 2009).At the local scale, for example, fish assemblage structure is regulated by both abiotic factors such as chemical characteristics of water and physical features of the stream channel, and biotic interactions, such as competition and predation (Gebrekiros, 2016).Since these factors determine conditions for the survival of these organisms, they may help predict the structure of fish communities (Gerhard et al., 2004;Terra et al., 2013).At the same time, however, both abiotic and biotic factors are subject to complex, dynamic and context-dependent mechanisms that affect fish communities.For instance, local variables in headwaters can largely control fish assemblages, but this depends on environmental heterogeneity and overall connectivity of the watersheds (Henriques-Silva et al., 2019).
Recent studies using variation partitioning have demonstrated a relatively small, but statistically significant, unique effect of spatial variables on the similarity of aquatic communities (De Bie et al., 2012).However, the relative importance of spatial distance as a predictor of local community composition remains to be elucidated.Environmental factors that affect species distributions are also spatially structured, and they have become evident at large-scale ecoregion comparisons (Van Sickle, Hughes, 2000) and between drainage basins (Brown, Lomolino, 1998;Matthews, 1998), either historically by the differences in species pool, or dispersal limitations (Cottenie, 2005;Mykrä et al., 2007).Spatial context also determines the relative importance of local environmental conditions and dispersal of the fish community (Geheber, Geheber, 2016), as well as the similarity of assemblages among sampling sites, which decreases with increasing spatial distance (Legendre, 1993;Dray et al., 2006).In this sense, knowledge of connections among drainage basins will lead to an understanding of the major distribution patterns of fishes (Hocutt, Wiley, 1986) since physiography may govern fish distribution and abundance (Pflieger et al., 1981).
The Atlantic Rain Forest, which originally extended continuously along the Brazilian east coast, is today a patchwork of fragments (11.4 to 16%) (Ribeiro et al., 2009), but still considered a global biodiversity hotspot (Myers et al., 2000).Blackwater rivers and clearwater rivers are the main river types of the Atlantic Rainforest (Por, 1992).They harbor a high richness of fish species, consisting of 89 genera and 269 species, 70% of which are endemic (Abilhoa et al., 2011).The most important reason for this diversity and endemicity is the isolating effect exerted by mountains (upstream) and sea (downstream) among coastal river basins (Bizerril, 1994;Menezes et al., 2007).Half of these species are small in size and complete their life cycle entirely in streams (Abilhoa et al., 2011).
Most studies on community structure, distribution and influence of local variables on fish assemblages have been performed in clearwater Atlantic Forest rivers (Gerhard et al., 2004;Mazzoni et al., 2006;Terra et al., 2016;Gonçalves, Perez-Mayorga, 2016;Teshima et al., 2016;Wolff, Hahn, 2017;Gonçalves et al., 2018).Despite several studies on fish assemblages of blackwater rivers (Ferreira, Petrere Jr., 2009;Gonçalves, Braga, 2012;Ferreira et al., 2014;Esteves et al., 2019), an expanding basin-wide perspective would provide a better foundation for understanding the factors that govern fish assemblage structure in these rivers.This view is highlighted by the unique conditions of blackwater rivers, including high humic content, decreased pH, low nutrients (Por, 1992), low water velocities, fine-particulate substrate (Ferreira et al., 2014) and their origin at low altitudes in the Coastal Plain Forest or Restinga Forest.Restingas are herbaceous/shrubby coastal sand-dune habitats that represent one of the faces of the Atlantic Rain Forest (Marques et al., 2015).They occur over Brazilian coastal plains and are formed by marine sediments originated from the Quaternary (CONAMA, 1999).

Fish responses in coastal blackwaters
Lowland streams in coastal regions are also strongly influenced by proximity to the sea with faunal transitions occurring over distances of only a few kilometers on the coastal alluvial plains (Winemiller, Leslie, 1992).Thus, as observed for other studies, simple ecological processes such as dispersal, spawning migration and reproduction can lead to strong distance-decay relationships, whereby the similarity of assemblages decreases with increasing spatial distance of the sampling sites (Legendre, 1993;Dray et al., 2006).
Multiscale influences on Atlantic Forest fish communities performed in clearwater streams have shown a predominance of local (Terra et al., 2016) and spatial variables, explaining most of the variation in fish assemblages (Teshima et al., 2016;Gonçalves et al., 2019).However, in blackwater streams multiscale influences are still poorly known, providing the rationale for this study.We aimed to understand the contributions of environmental and spatial factors in describing the variance of fish assemblages in blackwater streams located in a region subject to the isolating effect of mountains and sea among coastal river basins.We predicted that local factors would be significant in assemblage structuring because in least modified watersheds, landscape variables can be expected to contribute little to explaining the fish assemblage structure (Barbosa et al., 2019).Also, because of the particular conditions of blackwater streams, it is expected that local conditions provide for rare organisms that are specially adapted to life in low-pH environments.As we selected an area considered small to medium in extension, we hypothesized that the contribution of environmental factors is larger than that of spatial factors, because the contribution of spatial factors is predicted to decline with decreasing geographical extent.More specifically, we herein aimed to determine a) the relative contribution of site-scale (local variables), landscape and spatial configuration in explaining variability of fish assemblages; b) the local and/or landscape variables that best predict stream fish composition, and c) the patterns of fish assemblages in different coastal sub-basins.The outcomes of this study are important for understanding how fish assemblages are structured in Atlantic Forest blackwater streams, directing efforts for conservation of aquatic biodiversity.

MATERIAL AND METHODS
Study area.We sampled 14 blackwater streams located in the coastal plain of São Paulo State (Southeast Brazil) where meandering rivers occur in a low declivity region.The sampled streams are of 1 st and 2 nd order and originate at low altitudes in a region between marine and coastal mountain (Serra do Mar) environments.Sites belong to four sub-basins: Itapanhaú, Itaguaré, Guaratuba, and Una (Tab.1).The first three subbasins are located in the Baixada Santista Hydrographic Basin (Ribeiro, 2018), while the Una is located on the northern coast sub-basin.All sub-basins are included in the Ribeira do Iguape Ecoregion sensu Abell et al. (2008).The criteria used to select the sampling sites were accessibility, adequacy for electrofishing, salinity ≤ 0.05 ppm and good conservation status of the streams.Thus, we considered only sites located in areas with well-preserved stretches of High Dry Restinga Forest and sub-montane and/or lowland rain forest where intense human activities are absent.Maximum urban occupation attained 10.9% at one site of the Itapanhaú River sub-basin because of the proximity of a sewage treatment plant (Tab.1).Most sites were located in the municipalities of Bertioga and São Sebastião, covering the Conservation Unit "Parque Estadual das Restingas de Bertioga" (PERB), a private area known as "Fazenda Jimbuíbas", and an area close to the "Rio Silveira Indigenous Reserve" in the municipality of São Sebastião (Fig. 1).Created in 2010, the PERB consists of 9,312.32 hectares (Fundação Florestal, 2018) and is a Conservation Unit housing 97% of the remaining Restinga Forest of Baixada Santista (Fundação Florestal, 2022).This vegetation refers to the set of physiognomically distinct plant communities under marine and fluvial-marine influence, occurring on coastal sandy deposits (Cerqueira, 2000).Ecological interest in the PERB stems from its rich diversity of environments, including the Itaguaré and Guaratuba sub-basins and the eastern segment of the Itapanhaú River sub-basin.Vegetation consists of Dense Ombrophilous Forest and Pioneer Formations (Instituto Ekos Brasil, 2008).The climate in the Bertioga region is characterized as tropical with precipitation throughout the year (Af).Average annual temperature is around 24ºC, and rainfall is 3,200 mm (Köppen, 1948).The highest average rainfall values occur in January, February and March, and the lowest in June and July, revealing a water surplus throughout the year.
The Una River sub-basin is located in the northern coastal Hydrographic Basin of São Paulo State, covering 12,060 ha.Approximately 80% of this basin's area is found within state park boundaries with important natural remnants of the Atlantic Forest (Instituto Ekos Brasil, 2008).The average temperature recorded for the low-rainfall season in this region (April to September) was 18.33 °C in 2016, along with cumulative precipitation of 593.06 mm (CIIAGRO, 2021).Sampling design and environmental variables.Data on fish, water and habitat characteristics were obtained during the period of low rainfall between the months of July and September 2016.This season is associated with a continuously falling water level and decreasing interconnection among habitats.In each stream, a continuous stretch of 100m was selected in the middle-downstream section, and the following local parameters were recorded at a single point along that reach: temperature (ºC), pH, conductivity (μS.cm -1 ), turbidity (NTU), dissolved oxygen (mg.L -1 ) and total dissolved solids (mg.L -1 ), using a Horiba model U-5000G multiparameter probe.In addition, subsurface water collection was performed for the analysis of total phosphorus (mg.L -1 ), dissolved organic carbon (DOC) (mg.L -1 ), total organic carbon (TOC) (mg.L -1 ) and total iron (mg.L -1 ), according to APHA/AWWA/WEF (2012).Water velocity was obtained using the float method (Marques, Argento, 1988), which provides an estimate of the maximum surface water velocity.For this calculation, we launched a float three times and measured the time required to pass through a 5 m section.
Landscape variables were measured using the percentage cover of urban areas (houses, roads, sewage treatment plant), native forests, areas under regeneration (shrubs, bushes and small trees), deforested areas (clearings, sandbanks, dirt roads), rivers and beaches within a 500-m radius of the sampling point.Measurements were performed according to Tibúrcio et al. (2016), using satellite images for the year of 2016 and measurement tools from Google Earth, v. 7.3.Spatial data were calculated considering watercourse distance between sampling sites as recommended by Landeiro et al. (2011).We used drainage shapefiles and performed measurements on QGis 3.4 (QGIS Development Team, 2018).
Fish community sampling.Fish sampling was conducted from downstream to upstream with electrofishing equipment (Honda EUi10 Generator, 1000 W, CA), exploring all types of microhabitats along the 100 m reaches (Mazzoni et al., 2000).Blocking nets were used to prevent the escape of the most mobile species.For each site, three successive removals were used (Zippin, 1958) since three passes have proven to be adequate for obtaining quantitative data in Atlantic Forest streams (Mazzoni et al., 2000).The reach length of 100 m was set considering previous studies that evaluated the efficiency of electrofishing in small streams of the Neotropical region in reaches varying from 50 to 80 m in length (Mazzoni et al., 2000;Teixeira-de Mello et al., 2014).Forty channel widths were recommended by Terra et al. (2013) to estimate species richness in clearwater rain forest streams, but we could not sample greater extensions because of downstream marine influences and/or inaccessibility of upstream sections.
We separated fish into plastic boxes, anesthetized in eugenol solution, fixed in a 10% formalin solution and transferred to 70% ethanol after 10 days.For species taxonomy, we used the identification keys of Britski (1972) and Britski et al. (1984).Specialists from the Museu de Zoologia of the Universidade de São Paulo (MZUSP) and Universidade Estadual de Campinas (UNICAMP) confirmed fish identification.Voucher specimens were deposited in the Museu de Zoologia of the UNICAMP (ZUEC).Data analysis.To compare fish assemblage similarity based on abundance data among sub-basins, we used a nonmetric multidimensional scaling analysis (NMDS).We applied the Bray-Curtis distance measure that is suggested for abundance data and considered the NMDS valid for a stress value < 0.2 (Clarke, Warwick, 1994).To test for statistical difference among the communities of the different sub-basins, we ran a non-parametric similarity analysis (ANOSIM) on the raw abundance data using 9999 permutations with a significance level of  = 0.05.In this test, the R value varies between 0 and +1.0.Values close to "1.0", suggest dissimilarity among groups, while values close to "0" mean low separation among groups (Clarke, Warwick, 1994).A value of p < 0.05 indicates that the R value observed was not randomly obtained.We also used a SIMPER analysis to evaluate which taxa contributed most to the dissimilarity among streams from the four sub-basins.All analyses were performed on PRIMER 7 software (Clarke, Gorley, 2006).
Variance partitioning.Because species composition and environmental variables are spatially structured, we used variance partitioning (pRDA) to evaluate the relative Fish responses in coastal blackwaters contribution of 13 local and six landscape variables, including the spatial data matrix (W matrix) as a conditional value within the RDA.This method evaluates how much of the change in species abundance can be attributed to the changes in environmental variables (at local and landscape scale), and spatial variables as well as how much variability is shared among them (Borcard et al., 2011).According to Landeiro et al. (2011), our spatial matrix was represented as the network distance, i.e., the distance between the sampling points following the watercourse.We extracted spatial variables for inclusion into the model through the Analysis of Principal Coordinates of Neighborhood Matrix (PCNM).Following the step-by-step procedures of Borcard et al. (2011), we generated the PCNM variables and then calculated the Euclidean distance between points following the watercourse, using the 'dist' function in the R programming environment v. 4.2.2 (R Development Core Team, 2022).Based on the resulting Euclidean distance matrix, a truncation threshold was calculated using the maximum value of the minimum spanning tree using the 'spantree' function (Sedgewick, 1990) from the 'vegan' package (Oksanen et al., 2022) in R. The truncation limit was based on the minimum distance connecting all sample points resulting from the 'spantree' and it was used as a reference to calculate a PCoA (Principal Coordinate Analysis) of the truncated matrix.From this PCoA, only the positive eigenvalues were retained for use as spatial predictors represented as PCNM variables (Dray et al., 2006;Borcard et al., 2011).
Prior to the statistical modeling, the three predictive matrices (local, landscape and spatial) were evaluated for the levels of collinearity among them.To accomplish this, we calculated Pearson correlation coefficients for all pairs of variables using the 'color' function in R and excluded one of the variables with r > 0.70 (Dormann et al., 2013).Then, the predictive matrices containing only less correlated variables (r < 0.70) were individually subjected to a 'forward selection' using the 'ordistep' function from the 'vegan' package (Oksanen et al., 2022) in R to select the most important predictive variables that would explain the species matrix (Blanchet et al., 2008).Thus, we selected the significant variables (p < 0.05) of the three matrices as the most important to integrate the final datasets (Blanchet et al., 2008).Accordingly, the final explanatory matrices used in the variance partition models and pRDAs (Partial Redundancy Analysis) were composed of the following variables: local matrix: pH and TDS; landscape matrix: areas under regeneration; spatial matrix: PCNM1 (axis 1 of the PCNM).
To check the relative importance of local, regional and spatial variables for community composition, we performed three variance partitionings using the 'varpart' function of the 'vegan' package (Oksanen et al., 2022).This function partitions the variance of the dependent matrix Y (e.g., species composition) against the predictive matrices using redundancy analysis (RDA) ordering (Borcard et al., 1992;Legendre, Legendre, 1998).The significance of the contribution of the individual fractions (local, regional and spatial matrices) that quantify the variance in species composition was evaluated by applying the 'anova' function in R on partial RDA models (pRDAs) calculated using the 'rda' function of the 'vegan' package.Prior to the analysis of variance partitioning and the pRDAs, data from the species composition matrix were standardized using the Hellinger transformation, thereby avoiding statistical bias imposed by high abundances of few species (Legendre, Gallagher, 2001).
We also evaluated the possible influence of spatial autocorrelation on our results by applying multiscale ordination (MSO) on the RDA models using the 'mso' function Mariana Landucci Giongo, Maria Letizia Petesse and Katharina Eichbaum Esteves from the 'vegan' package (Wagner, 2004;Borcard et al., 2011).For this, we applied the 'MSO' over a global RDA model containing all predictive matrices (local, landscape, and spatial) and over the individual pRDAs for each matrix.The MSO combines multivariate variograms with ordinations (simple or canonical), allowing the user to assess whether the variables of the predictive matrices exert a spatial autocorrelation effect on the response matrix and to identify in which distance classes this effect is significant (Wagner, 2003(Wagner, , 2004)).

RESULTS
Environmental variables and fish communities.The studied streams presented unique characteristics, having dark and acidic waters, high values of DOC, TOC and conductivity, low flow and a predominance of native forest, followed by areas under regeneration (Tab.2).
Predictors of fish composition.Variance partitioning analysis indicated that the three predictive matrices (local, landscape and spatial) together explained 58.6% of the variance in the composition of fish communities and the model was statistically significant (F = 2.73, P = 0.001).All fractions (local, spatial, landscape) had a significant influence on the variance of fish community composition, but the local variable had the highest R² adj (37%), explaining the greatest amount of variance in the fish community (Tab.5).Within this amount, the pure local variable explains 17% of the variance, and the rest is shared among the other components: [local + landscape], [local + spatial], and [local + landscape + spatial], which explained 4.2, 1.6, and 15%, respectively (Fig. 3).We    found no significant spatial correlation of residuals in pRDA models according to MSO diagnoses, indicating that our models have residual independence and homogeneity.
In Fig. 4 we performed an RDA with the best predicted variables for fish composition elected from the local, landscape and spatial matrices (pH, TDS, areas under regeneration and PCNM1).The percentage of variance explained by the first two axes was 30.6 and 20.14%, respectively.Some species, such as Acentronichthys leptos Eigenmann & Eigenmann, 1889, Gymnotus pantherinus (Steindachner, 1908), and H. multifasciatus, were associated with areas under regeneration and occurred in sites with low pH, characteristic of the Guaratuba sub-basin.On the other hand, M. microlepis, P. obtusa, Scleromystax barbatus (Quoy & Gaimard, 1824) and Characidium lanei Travassos, 1967 occurred mainly in sites with low total dissolved solids.Phalloceros harpagos occurred in sites with higher pH and Dormitator maculatus (Bloch, 1792) in sites with higher TDS.

DISCUSSION
We found that environmental variables at the local scale best predicted fish abundance of stream fish assemblages in blackwater streams.The two most important selected local factors in our study, pH and TDS are apparently related to the main characteristics of blackwater rivers which are rich in humic substances and form complexes with iron, aluminum, calcium and magnesium (Krachler et al., 2019).The pH is known to have a strong effect on fish communities (Jackson et al., 2001), indicating that it is probably the main environmental filter for them, as also observed by Gonçalves, Braga (2012) in blackwater Atlantic Forest streams.Total Dissolved Solid (TDS) levels, a measurement of inorganic salts, organic matter and other dissolved materials in water (Weber-Scannell, Duffy, 2007), were high when compared to blackwater rivers from the Amazon (Duncan, Fernandes, 2010).These conditions may be important in maintaining fish diversity, as changes in TDS may exclude less tolerant species and cause shifts in biotic communities (Weber-Scannell, Duffy, 2007).The lower importance of the "pure" landscape scale, compared to that of local scale, has been commonly reported in several other neotropical streams (Lemke, Suárez, 2013; Mariana Landucci Giongo, Maria Letizia Petesse and Katharina Eichbaum Esteves Junqueira et al., 2016;Terra et al., 2016;Barbosa et al., 2019), while other studies found contrasting results concerning their relative importance (Weigel et al., 2006;Ferreira et al., 2007;Cunico et al., 2012).Although these studies were performed in different river types, it is known that local and landscape factors may be context-dependent by their relationship to the spatial scale investigated (Junqueira et al., 2016) and the degree of system preservation (Dias, Tejerina-Garro, 2010).Thus, the higher importance of local variables in our study may be explained by the relatively uniform landscape conditions since streams were, in general, located in areas with well-preserved vegetation.Also, as organic acidity of naturally acidic waterways is reliant on the input of large amounts of allochthonous organic matter such as Dissolved Organic Carbon and leaf litter (Holland et al., 2012), linkages between local factors as the riparian flora and the aquatic environment are of fundamental importance.
Shared variation between local and landscape factors was also observed in our study, suggesting that landscape variables had a complementary effect in explaining variance in fish assemblages.In addition, areas under regeneration represented the only landscape predictor that influenced our model, and seems to be the one giving the best evidence of human-induced changes, especially in the Guaratuba sub-basin, where fine substrate was dominant.Landscape variables are known to create and maintain local habitats, influencing sedimentation and water regime, oxygen, and habitat properties, consequently impacting biodiversity (Blevins et al., 2013).For instance, native forests play a vital role in physical channel heterogeneity by supplying large wood, branches, litter and roots to aquatic systems (Allan, 2004;Jackson et al., 2015), creating mesohabitats and microhabitats along stream channels, as well as increasing marginal and instream habitat complexity (Harvey et al., 2018).On the other hand, modification of natural conditions can contribute to the input of unconsolidated substrate to the stream channel (i.e., siltation) (Wantzen, Mol, 2013).
Spatial effects were significant, but the "pure" spatial fraction did not explain the variation in the composition of the assemblages.This could be explained by the different processes by which metacommunities are organized and connected, including such mechanisms as species sorting, mass effects and dispersal limitation (Heino et al., 2015).Also, spatial processes, which can be important at very large and very small spatial extent, depend on fish dispersal ability and the influence of local and landscape predictors (Heino et al., 2015).Our study covered a 65 km stretch along the coast, an area that might be considered small to medium, covering regions "across streams" and "across drainage basins".Considering that a possible connection among adjacent localities may occur during the rainy season, some species may have sufficient dispersion capacity to generate a 'mass effect', whereby high dispersal rates homogenize communities to some degree at neighboring localities, irrespective of their abiotic and biotic environmental conditions (Heino et al., 2015).This hypothesis may be confirmed by the presence of several Characidae species, which perform reproductive migrations, as described by Mazzoni et al. (2004) and Mazzoni, Iglesias-Rios (2007) for Atlantic Forest streams.More specifically, they have found evidence of a group of fishes termed "Long Movement group", which moved at least 6 km within 60 days.Nevertheless, another mechanism such as dispersal limitation, which increases with spatial distance between sites, may also structure fish communities and, consequently, explain highest dissimilarities between the most distant sites as Vila Agaó and Bora Bora 2.

Fish responses in coastal blackwaters
In this respect, our analyses should be viewed with caution as they represent an ecological snapshot, and seasonal or more sampling-intensive studies would help to clarify the variability of the spatial and temporal correlates.Also, since sampling sites in the present study were at sea level, spatial factors and environmental variables are not comparable to those of most studies performed in clearwater coastal Atlantic Forest streams, studied along longitudinal gradients (Ferreira et al., 2014;Terra et al., 2016;Gonçalves et al., 2019).However, Gonçalves et al. (2019) highlight the importance of dispersal processes in Atlantic Forest streams, particularly downstream, which seem to have a strong influence on community organization.
Differences in community composition were confirmed by the NMDS, which showed distinct fish composition among the four sub-basins, especially between the Guaratuba and both Itaguaré and Itapanhaú sub-basins, indicating dissimilarities within the same ecoregion (Ribeira do Iguape) sensu Abell et al. (2008).Although some authors state that a similarity among ichthyofauna is expected to occur between ecoregions (Hughes et al., 1987;Rohm et al., 1987), others imply that local factors within an ecoregion are more important (Van Sickle, Hughes, 2000;Herlihy et al., 2006).Fish species may show distinct preferences for different habitats, giving rise to distinct fish assemblages, which may be related to such variables as depth, pools and riffles, and substrate, as well water quality, such as oxygen, pH, conductivity and temperature (Gebrekiros, 2016).
Streams from the Guaratuba sub-basin were characterized by the presence of fine substrate and very acidic conditions, and dissimilarities between this sub-basin and both Itaguaré and Itapanhaú sub-basins were attributed mainly to the characid M. microlepis.This species was dominant in our study and is among the most abundant species in Atlantic Rain Forest streams (Esteves, Lobón-Cerviá, 2001;Barreto, Aranha, 2005), occurring both in clear and blackwater streams (Weitzman, Menezes, 1994).As this is an insectivorous surface feeder forming schools, usually associated with deep pools (Esteves, Lobón-Cerviá, 2001), its occurrence in regions with lower TDS indicates a preference for the most distant sites from the sea, essentially because proximity to the sea results in higher levels of water conductivity (Por, 2004) and, hence, high related TDS values.
Phalloceros harpagos, H. multifasciatus, and D. maculatus also contributed to the dissimilarities among sub-basins, indicating their association with different environmental conditions.Phalloceros harpagos has been defined as an indicator of deeper areas, like pools, favoring colonization owing to higher amounts of trunks, branches and leaf litter (Rezende et al., 2010), which were common conditions within the studied streams.On the other side, D. maculatus can be found in streams considered as transitional environments between the estuarine-riverine and the estuarine mixing zones (Abilhoa et al., 2011).The occurrence of a group of species associated with very acidic streams and fine sediment in the Guaratuba basin (H.multifasciatus, G. pantherinus, A. leptos) indicates a strong adaptation to these harsh conditions, which were also influenced by the proximity of areas under regeneration.
In conclusion, results showed differences in composition and structure among subbasins that revealed distinct ecological requirements of some species, suggesting, in turn, environmental heterogeneity among streams determine the wider structural patterns of ichthyofauna.They also suggest that the contribution of local predictors to these blackwater streams holds the greatest importance in maintaining their fish assemblages, especially given that 60% of the recorded species are endemic, including  (ICMBio, 2018).Also, the Atlantic Forest concentrates the largest number of threatened fishes, representing 35% of the total endangered species at the national level (Castro, Polaz, 2020).This situation draws attention to the importance of consistent conservation efforts, especially when coupled with the paucity of knowledge that exists about blackwater Atlantic streams located in restinga areas and the presence of rare organisms especially adapted to life in low-pH environments.In addition, restinga forests are considered highly fragmented, and little attention has been given to ensure the maintenance of their and processes (Marques et al., 2015).Nonetheless, many of the studied streams were located within a conservation area (PERB) harboring 98% of the remnants of the restinga forests found within the Baixada Santista Hydrographic Basin (Banzato et al., 2012).Thus, the nearly pristine conditions of our study seem to be essential for conserving fish species, and to achieve this, the heterogeneity found among sub-basins seems to be important in maintaining fish biodiversity.
Environmental modifications, such as land use, may directly affect the characteristics of water (e.g., pH level), influencing fish preferences for reproduction habitat (Dei Tos et al., 2002) and determining fish growth (Ferreira et al., 2001).Thus, alterations in the studied streams could result in the elimination of fish species which are adapted to low pH and, consequently, modify fish composition.From a conservation perspective, then, this study may provide subsidies for species management in other blackwater Atlantic Forest streams and may be used as a baseline condition for the assessment of similar environments.Maintaining spatial heterogeneity and connectivity between stream reaches is also recommended in order to preserve the metacommunity dynamics of fish in these restinga remnants, which are under constant human pressure.

FIGURE 1 |
FIGURE 1 | Location of the study area in the coastal region of the state of São Paulo, indicating the Baixada Santista and Northern Coast basins and the distribution of the sampling sites in the Itapanhaú (JP1, JP2, VA), Itaguaré (MP, PM, GU), Guaratuba (P1-P6), and Una (BB1, BB2) sub-basins (Adapted from Esteves et al., 2019).In detail, sites of the Guaratuba sub-basin and main landscape types.

Atlantic
Forest blackwater streams.a = fraction of exclusive explanation of local variables; b = fraction of exclusive explanation of landscape variables; c = fraction of exclusive explanation of spatial variables; d = fraction of shared explanation between local and landscape variables; e = fraction of shared explanation between landscape and spatial variables; f = fraction of shared explanation between local and spatial variables; g = fraction of shared explanation between local, landscape and spatial variables; h = residuals; df = degrees of freedom; R² = r-squared; R² adj = adjusted r-squared; p = p-value of F.

FIGURE 3 |
FIGURE 3 | Venn diagram showing the variance partition of local, regional and spatial variables in explaining the composition of fish communities.a = fraction of exclusive explanation of local variables; b = fraction of exclusive explanation of landscape variables; c = fraction of exclusive explanation of spatial variables; d = fraction of shared explanation between local and landscape variables; e = fraction of shared explanation between landscape and spatial variables; f = fraction of shared explanation between local and spatial variables; g = fraction of shared explanation among local, landscape and spatial variables.

FIGURE 4 |
FIGURE 4 | Redundancy analysis (RDA) explaining the relationship between selected local, landscape, and spatial variables and fish species.Vector lines indicate the relationship of significant environmental variables to the ordination axis.TDS -Total Dissolved Solids; AR -Areas under Regeneration; PCNM1 -Spatial Component represented by the Principal Coordinates of Neighbor Matrices.Species are indicated by black dots.Acronyms according to Tab. 3 Mariana Landucci Giongo, Maria Letizia Petesse and Katharina Eichbaum Esteves ni.bio.br| scielo.br/ni

TABLE 2 |
Maximum, minimum, mean and standard deviation (SD) of local and regional environmental variables obtained for 14 blackwater Atlantic Forest streams during the period of low rainfall (June-September/2016) along the coast of São Paulo State.The most important variables selected for the variance partitioning procedure (pRDA) are indicated.Mariana Landucci Giongo, Maria Letizia Petesse and Katharina Eichbaum Esteves ni.bio.br| scielo.br/ni

TABLE 4 |
Analysis of similarity percentage (SIMPER) for fish species with pairing among sub-basins, showing the species that determine group dissimilarity.Subscripts indicate sub-basins containing the highest abundance of each species.Only higher-contributing species are shown.Ita= Itapanhaú, Itg = Itaguaré, Gua = Guaratuba, and Una sub-basins.*Indicates groups with significant dissimilarities (p<0.05).