Physical habitat as predictor of fish trophic structure in Brazilian Atlantic rainforest streams

Trophic structure of fish assemblages is one of the most sensitive indicators of changes in streams environments. Since it is crucial to understand the response of trophic groups to habitat alteration, our study aimed to address this research gap by assessing the influence of substrate composition, meso-habitat variability, and bank stability, on the richness, biomass, and number of individuals of carnivores, invertivores, omnivores, and herbivorous-detritivores. Using an electrofishing device, we sampled 13 Atlantic rainforest streams reaches in a degradation gradient, located in the upper Paranapanema river basin. Sample points were ranked using a physical habitat index. More pristine streams had high availability of twigs, trunks, rocks and boulders in the substrate, great meso-habitat variability, and the presence of roots, trunks, and rocks in the margins. Canonical correlations between habitat characteristics and trophic groups explained more than 90% of data variability. Richness and number of individuals of invertivores increased in more preserved stream reaches, while richness of carnivores and number of individuals of omnivores decreased. These results demonstrate that trophic structure varies according to level of degradation, and that invertivore richness represents the best indicator of fish trophic structure responses to physical habitat alterations in streams.


INTRODUCTION
The study of trophic structure is an important tool for evaluating fish assemblage structure (Angermeier, Karr, 1983;Nimet et al., 2015). Since environmental filters act by selecting assemblage species according to their traits, the effects of physical habitat characteristics on trophic structure may be a more accurate predictor than taxonomic composition . Additionally, investigating different aspects of trophic structure can help detect patterns and predict assemblage responses to different environmental conditions, which are fundamental aspects to biodiversity conservation (Cruz et al., 2013).
Habitat characteristics promote alterations in trophic structure (Cunico et al., 2011) and in-stream habitat homogenization, often due to anthropogenic causes, is expected to promote less diverse trophic structures (Zeni, Casatti, 2014). Several habitat elements have been shown to influence fish assemblages in streams, including substrate composition and flow heterogeneity (Hugueny et al., 2010;Carvalho, Tejerina-Garro, 2015). Channel morphology variables such as substrate, depth, and velocity alter trophic structure as large carnivores tend to predominate in large pools or streams (Schlosser, 1982). According to Lorion, Kennedy (2009), the impacts caused by the conversion of natural forest to pastures could alter trophic structure by increasing the density of herbivorous-detritivores fish. Additionally, Cross et al. (2013) showed that streams subjected to siltation, a form of environmental simplification in which there is homogenization of the substrate composition, can consequently decrease the diversity of trophic groups. Omnivorous fish tend to thrive after habitat alteration (Karr, 1981), especially in cases where larger substrate particles are replaced by smaller particles (Berkman, Rabeni, 1987). Furthermore, substrate composition can alter food availability for fish, since elements such as branches, leaves, and rocks create microhabitats not only for algae and fungi (Gordon, 1993) but for aquatic insects as well (Reid et al., 2010;Luek, Rasmussen, 2017). Angermeier, Karr (1983) also suggest that large scale alterations in stream structure may have serious impacts on biodiversity due to the effects of shading, siltation, and changes in available food types. However, identifying the effects of environmental change on the ichthyofauna may depend on the assemblage structure descriptor selected (Casatti et al., 2009a;Krause et al., 2013;Nimet et al., 2015;Feld et al., 2016;Vilmi et al., 2016).
In order to use trophic structure as a viable alternative to taxonomic approaches, the relationships between trophic structure and habitat characteristics must be well understood (Vitule et al., 2017). Therefore, we aimed to answer the question: Does substrate composition, meso-habitat variability, and bank stability correlate with richness, biomass, and number of individuals of each trophic group? We expected to detect variation in trophic group responses to habitat characteristics, especially for invertivores and herbivorous-detritivores in response to substrate composition due to the dependency of most species in these groups on substrates as foraging sites. In addition, when considering possible replacements of species by others from different trophic groups in response to environmental changes, we expected to detect a stronger response in trophic group richness than in number of individuals and biomass.

MATERIAL AND METHODS
Sampling sites. Sampling was carried out once at each of the 13 stream reaches (P1 -P13) between August and December in 2010 and 2011. The sampling sites were located in the southwest region of São Paulo state, Brazil, in streams located on the Upper Paranapanema River (Fig. 1). All sampled streams were of 3 rd or 4 th order (Strahler, 1957), and were selected due to accessibility and dimensions that allowed the use of electrofishing methods. The streams were located closest to the cities of Pilar do Sul and São Miguel Arcanjo in areas surrounded by natural vegetation, Eucalyptus spp. plantations, and/or pastures. The natural vegetation covers 15% of the total area of the upper Paranapanema river basin. The vegetation was primarily composed of fragments of Atlantic rainforest, especially from phytophysiognomys tropical semi-deciduous forest and floodplain forests. However, there were also some Brazilian savannah fragments, corresponding to 0.3% of the area (Secretaria do Meio Ambiente, 2005). The region presents a humid tropical climate (Type Aw according to Köppen classification) with annual precipitation of 1,200 mm. Geologically, two structural units predominate: the Atlantic Orogenic Belt and the Paraná Sedimentary basin. The altitude varies from 600 to 800 m and the region has mainly low hills with slopes between 20 to 30% (CBH-ALPA, 2016).
Sampling. Based on previous analyses of sampled sites located within the same basin  Peressin, Cetra, 2014 andPeressin et al., 2018 for more details), we considered the following habitat characteristics: substrate composition (SC), meso-habitat variability (MH), and bank stability (BS). We established three equidistant transverse transects across each stream and recorded depth, velocity, and substrate composition in one-meter intervals along each transect. In these same transects, we recorded the composition of both sides of the river. Classification of the SC, MH, and BS parameters was based on substrate elements (silt, twigs, rocks, etc.), velocity and depth combinations, and bank composition following the protocols adapted from Barbour et al. (1999) (Tab. 1). High SC scores indicate high availability of larger substrate particles such as twigs, trunks, and large substrate particles such as rocks and boulders. High MH scores indicate a great meso-habitats diversity such as alternations between shallow and deep reaches and also between high and low velocities. Elevated BS scores denote an increase of elements such as roots, trunks, and rocks in the river margins. Overall, high scores FIGURE 1 | Sampling sites located in the Paranapanema river basin, São Paulo state, Brazil. The Paranapanema river is highlighted in blue.

5/18
scielo.br/ni | sbi.bio.br/ni for these habitat parameters indicated more pristine streams while low scores indicated streams that were more affected by anthropogenic impacts. Fish sampling took place between 8 am and 5 pm, using an electrofishing apparatus. We sampled fish along a 70 m reach of each stream with a single electrofishing pass from downstream to upstream, without the use of block nets. This distance is considered sufficient to emcompass the diversity of meso and micro-habitats of tropical streams (Mazzoni et al., 2000). In a laboratory, fishes were identified to the lowest possible taxonomic level. Representative ichthyofauna specimens were deposited at the fish collection of Ichthyology Laboratory of the Department of Zoology and Botany of Universidade Estadual Paulista in São José do Rio Preto, Brazil (DZSJRP) under the catalog numbers: DZSJRP 13642-43, 13645-46, 13648, 13650, 13652-53, 13655-57, 13663-67, 13673-74, 13676-77, 13681-83, 13685-88, 13690-95, 13699, and 13701. Trophic groups. The trophic group of each fish species was assigned based on a review of available literature and diet analysis. We considered the following trophic groups: carnivores -species that consume mainly invertebrates and fish; invertivores -species that consume insects and other invertebrates such as molluscs, crustaceans, and annelids; omnivores -species that consume items of plant and animal origin, and may also include detritus; and herbivorous-detritivores -species that consume mainly detritus (particulate organic matter at different stages of decomposition) and algae, including periphyton.
Only 3 of the 4 pool types present (note: if fastshallow is absent, the score is less than with the absence of other pool types).
Only 2 of the 4 pool types present (note: the score is less if fast-shallow or slowshallow are absent).

Dominated by one pool type (generally slow-deep).
Score 20  Moderately stable; small areas of erosion exist but are infrequent and are usually stable; 5-30% of the margin affected by erosion.
Moderately unstable; 30-60% of the margin affected by erosion; high erosion potential at flow peaks.
Unstable; many eroded areas; frequent exposed areas (i.e., without cover); 60-100% of the margin affected by erosion.  We carried out stomach content analysis for 12 individuals of Neoplecostomus selenae, an endangered species (São Paulo, 2018) for which there was no diet data in the literature. We identified stomach contents to the lowest taxonomic level possible and then classified each item into the following alimentary categories: vegetal material, algae, detritus, insects, and other invertebrates. For each item per stomach sample, we assigned values according to the method of Degree of Food Preference (DFP) (Braga, 1999). Only stomachs with full contents were considered. To calculate the DFP, we assigned values of importance for each item recorded in a sample. If only one item type was present in the stomach, it was assigned a value of 4. For samples with more than one item type, the most abundant item was allocated the value 3, while the values 1 and 2 were assigned to the least abundant and the intermediate items, respectively. For samples with more than 3 items, multiple items were assigned the same value, following the same procedure described above.

Statistical analyses.
Considering the values assigned in the stomach content analysis, a DFP value was calculated for each alimentary category using the equation: DFP = S(i)/N, where S(i) is the sum of the values assigned to each item (i) in each stomach, and N is the total number of analyzed stomachs.
From SC, MH, and BS scores, we calculated a physical habitat index (PHI), following an adaptation from the method used in Casatti et al. (2006). Calculations of these metric scoring thresholds were based on the distribution of total values, with the lowest threshold established at the 25 th percentile and the uppest threshold at the 75 th percentile where the former would score 1 and the latter case would score 5. Intermediate conditions, i.e., between the 75 th and 25 th percentiles of the reference sites, would score 3. From the sum of metric scoring for each habitat parameter, we define a PHI for each stream. Again, sample points at the 25 th percentile were considered less pristine, points at 75 th percentile were considered more pristine and the remaining were considered in an intermediate condition.
To evaluate the relationship between habitat parameters and trophic groups, we conducted a canonical correlation analysis (CCorA). In this analysis, we used the three habitat characteristics as predictors for proportions within the four trophic groups at each sample point (Sherry, Henson, 2005;Legendre, Legendre, 2012). We ran three CCorA analyses: one with trophic group richness (S), another with biomass (B), and another for the number of individuals (N) in each trophic group. Because Wilks' λ represents the variance unexplained by the model, 1-λ yields the full model effect size in an r 2 metric. Wilks' Lambda values were calculated from the eigenvalues and converted to F statistics. With the F value we tested H0; the canonical correlations in the current row and all that follow are zero. For emphasis, structure (rs) and communality coefficients (h2) above 0.45 are in bold, following a convention in many factor analyses (Sherry, Henson, 2005). All analyses were performed using statistical software R version 3.0.0. We used the CCorA, Candisc, and Vegan R packages (R Core Team, 2018

RESULTS
According to PHI, sampled points P2, P6, P7 and P11 were considered more pristine, while P5, P8 and P10 were less pristine; the remaining were considered at intermediate preservation level (Tab. 2). The average width and depth of the sampling points were 4.64 and 0.34 m, respectively.
In total, 1,356 fishes were sampled, representing 31 species. Degree of food preference values for stomachs of Neoplecostomus selenae were 0.08 for vegetal material, 1.33 for algae, 2.33 for detritus and 0.58 for insects. Three species were considered carnivores, twelve invertivores, eleven omnivores, and five herbivorousdetritivores (including N. selenae) (Tab. 3). Hoplias malabaricus had the highest number of individuals and also presented the highest biomass for the carnivore trophic group. Similarly, Pimelodella avanhandavae had both the highest number of individuals and biomass for the invertivore group. For omnivores, Phalloceros reisi had the highest number of individuals, while Geophagus brasiliensis presented the highest biomass. Hypostomus ancistroides was the herbivorous-detritivore with the highest number of individuals and biomass (Tab. 3). In general, omnivores and invertivores were dominant in terms of richness and number of individuals in our streams. However, herbivorous-detritivores had the highest biomass and highest number of individuals in the P9 and P12 streams, respectively, while carnivores presented the highest biomass in P1, P5 and P13 (Fig. 2).   Stream order determined according to Strahler (1957).   The measured habitat characteristics were significantly related with trophic group richness. The CCorA analysis for richness yielded three functions (F1, F2, and F3) (r 2 c : F1 = 0.885; F2 = 0.678; F3 = 0.048) and the full model across all functions was statistically significant according to Wilks's (λ = 0.035) or F (12, 16.16) (= 3.4204, p = 0.012) criteria. Thus, for the set of three canonical functions, the r 2 type effect size is 0.965, which indicates that the full model explained a substantial proportion (96.5%) of the variance shared between the variables. The first two functions (F1 and F2) were retained because they explained most of the variability in the data. In F1, the MH variable was the primary contributor to the predictor synthetic variable. Considering the F1 coefficients, the relevant criterion variables were richness of omnivores (SOmn) and invertivores (SInv), which reverse signs. This indicates that an increase of mesohabitat diversity was followed by an increase in the omnivores richness and a decrease in the invertivores richness. As for F2, all habitat parameters were positively related to SInv and negatively related to carnivore richness (SCar). This suggests that streams with heterogeneous substrate (with boulders, rocks, trunks, and leaves), high mesohabitat diversity, and stable banks (with rocks, vegetation, and roots) present more invertivores species and less carnivores species (Tab. 4, Fig. 3). Since higher scores for habitat parameters indicate more pristine streams, these results suggest that physical habitat degradation due to human activities (land use change) can lead to a decrease in Habitat parameters were not related with trophic group biomass (Wilks's 1-λ = 0.85; p = 0.069) but were significantly related with trophic group number of individuals. When analysing the number of individuals, the full model across all functions was significant (Wilks's 1-λ = 0.94; p = 0.034). The SC and BS variables were the main contributors to the F1 function. The number of individuals for both omnivores (NOmn) and invertivores (NInv) varied significantly with the variables. However, these two trophic groups presented a contrasting relationship, indicating that streams with high SC scores (i.e., substrate that includes boulders, rocks, wood and leaves) and high BS scores (river banks that consist of rocks, wood and roots etc.) contain less omnivores and more invertivores, in number of individuals (Tab. 4, Fig. 4).

DISCUSSION
Our study results demonstrated that SC, MH, and BS influence trophic structure of fish assemblages, considering both richness and number of individuals. Invertivorous and omnivorous fish exhibit contrasting responses to habitat structure in terms of richness and number of individuals. Richness and number of individuals of invertivores increased in streams with larger substrate particles, meso-habitat variability, and bank stability, while the opposite was true of omnivorous fish. Similarly, opposite response patterns were revealed between invertivorous and carnivorous fish, but only for species richness.
Although the results of the biomass model were not significant, the same pattern was observed for the three trophic groups. Herbivorous-detritivores, on the other hand, presented no significant response to any of the habitat characteristics.
In stream habitats, changes in the banks can result in changes in flow (Vietz et al., 2018) which, in turn, can alter substrate composition (Shields et al., 1994), the latter being a key aspect for changes in fish assemblages (Hortle, Lake, 1983). Higher scores for SC, MH, and BS were generally associated with more intact environments since lower scores indicated a predominance of smaller particles in the substrate, such as sand and silt, more homogeneous flow, and exposed margins -all characteristics typical of more degraded environments (Kemp et al., 2011;Leal et al., 2016). Consequently, our results indicate a higher number of species and number of individuals of invertivorous fish in more pristine environments.
We found that the number of invertivore individuals increased with higher SC and BS scores, and that invertivore species richness was positively correlated to all measured habitat characteristics. Despite some nektonic species such as Piabina argentea and Piabarchus stramineus (Ferreira, 2007), invertivores in our study comprised mainly species associated with fast water and substrates composed of larger diameter particles. These include the nectobenthic Characidium spp. (Sabino, Castro, 1990) and heptapterid catfishes (Siluriformes: Heptapteridae) that practice substrate speculation (Casatti et al., 2009b). Benthic invertivores have often been associated with preserved streams (Teresa, Casatti, 2012;Krause et al., 2013). Furthermore, their feeding habits are frequently correlated with morphological characteristics, leading to dependence on substrate type (Ribeiro et al., 2016). Most invertivore species in our study had large pectoral fins, flattened dorsal-ventral bodies and wide mouths (Ribeiro et al., 2016). Such species tend to be benthic, rheophilic, and are sensitive to changes in substrate compositionespecially siltation which, in turn, leads to changes in flow; these species are generally associated with structurally complex and pristine environments (Teresa, Casatti, 2012;Cruz et al., 2013).
In more simplified environments, generalist species tend to prevail (Bozzeti, Schulz, 2004;Casatti et al., 2015). In our study, omnivorous fish decreased in richness, responding to an increase in SC, MH, and BS. However, since we used the proportion of trophic groups to represent trophic structure, an increase in the values of richness and number of individuals for invertivores can result in an apparent reduction in the participation of omnivores in the trophic structure, even though the latter does not present an actual reduction in the number of individuals. The omnivore group was mainly represented by species that present nektonic foraging habits and are able to forage in different environments (Sabino, Castro, 1990;Gomiero, Braga, 2008). Moreover, omnivore 13/18 scielo.br/ni | sbi.bio.br/ni species richness responded positively to MH. This group also included species with diverse habits, such as Astyanax spp. and Geophagus brasliensis, with deep, laterallycompressed bodies and terminal mouths; the genus Phalloceros, whose representative species are commonly observed in pools and slow waters and feed on the surface of shallow waters (Casatti, 2002); and Pimelodus maculatus, species associated with deeper parts of the water column (Lobon Cerviá, Bennemann, 2000).
The richness of carnivorous species decreased in streams with higher substrate diversity, mesohabitat diversity, and bank stability, i.e., more pristine streams. In our study, carnivores were represented predominantly by H. malabaricus and Rhamdia quellen, two of the largest species collected. Both are found mainly in waters with greater depth and slower water velocity, if compared with other stream species (Daga et al., 2012;Teresa, Casatti, 2012;Peressin, Cetra, 2014). As this meso-habitat generally has smaller substrate particles than high velocity meso-habitats, these species tends to occur in restricted conditions of SC and MH. For herbivorous-detritivores, on the other hand, a positive relationship with SC was expected, since many species from this group have specialized buccal morphology for grazing which allows the consumption of periphyton on hard substrates (Lujan et al., 2012). However, we did not detect a significant relationship between any of the trophic structure predictors of herbivorous-detritivores and the habitat characteristics measured. This may be explained by the fact that fish species in this group also consume detritus, which commonly occurs in almost all types of habitat, and it is therefore a food resource that is not restrictive (Peressin et al., 2018). Furthermore, the species with the highest number of individuals in this group was H. ancistroides, a species tolerant to reductions in environmental quality due to pollution (Smith et al., 1997), and it has been also considered a reliable indicator of anthropogenically altered environments (Oliveira, Bennemann, 2005).
The CCorA for richness explained greater amounts of variance in trophic groups than CCorA for biomass and number of individuals. Diversity indices generally ignore fish species identity and their roles in the community (Green, 1979). Consequently, species replacement may keep richness unchanged, preventing the detection of alterations and impacts on the fish assemblage (Scott, Helfmann, 2001). Trophic structure, on the other hand, incorporates species feeding habits, providing insights into ecological function and reflecting the role of species within an assemblage as well as their relationships with the environment. According to , in similar environmental conditions, different species of the same trophic group may replace each other. As a result, trophic structure would remain unchanged. On the other hand, as observed in this study, under physical environmental changes, species can be replaced by others with different feeding habits.
Invertivores species richness varied in response to all habitat parameters analysed in this study. In addition, invertivores richness and number of individuals were greater in more pristine streams, which had substrate composed by larger substrate particles, more variations between depth and shallow waters, more variations between high and low velocities, and more rocks and trunks along river banks. Furthermore, invertivore richness presented changes related to all measured habitat characteristics, and the richness canonical correlation explained a greater amount of variance in this data. Thus, we conclude that trophic structure can be a good indicator of environmental alteration, 14/18 scielo.br/ni | sbi.bio.br/ni and, more specifically, species richness of invertivores offers the best indicator of trophic structure responses to physical environmental changes in streams, and should be considered in further studies.

ACKNOWLEDGMENTS
We thank the Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) for the regular assistance to the research project "Study of physical and biotic integrity of headwater streams that cross the Atlantic Forest of São Paulo" (2009/53056 -8) granted to Mauricio Cetra and the master's degree scholarship awarded to the first author (2010/13758-0). We are grateful to Dr. Francisco Langeani Neto (UNESP -São of Rio Preto), for the aid in the identification of the species and Bruna Botti Cruz, Bruno Mello, Fernanda Ayumi Teshima and Rodrigo da Silva Almeida for help in the field. We are also greatful to two anonymous reviewers by suggestions to improve the manuscript. Natalie Swan translated the manuscript to English and Jessica Schulte reviewed for English clarity.