INTRODUCTION

Headwater streams are small and open ecosystems highly interconnected to the terrestrial environment (^{Allan, 2004}). The land cover at catchment and riparian corridor scales can drive physical habitat structure, water quality (^{Heartsill-Scalley, Aide, 2003}; ^{Fernandes et al., 2013}; ^{Tanaka et al., 2016}), and food items availability to the stream biota (^{Sweeney, 1992}; ^{Carvalho et al., 2017}; ^{Manoel, Uieda, 2018}). Native forests provide allochthonous structures, like roots, leaves, and trunks that create different stream habitats and increase the supply of terrestrial food items, mainly insects and seeds, for stream fishes, for stream fishes (^{Lo et al., 2020}). However, anthropic-driven shifts in the land cover (e.g., deforestation followed by pasture/agriculture development) have homogenized physical habitat conditions by the loss of meso and microhabitats associated with forests (^{Walrath et al., 2016}; ^{Zeni et al., 2019}) and the invasion of exotic grasses (^{Zeni et al., 2019}). Moreover, agriculture expansion over Tropical Rainforest biomes (e.g., Amazon) has changed the energetic balance in streams by altering resource availability, since it shifts predominant resources from allochthonous (e.g., terrestrial insects and plants) to autochthonous (e.g., aquatic insects and algae) (^{Zeni, Casatti, 2014}).

These changes caused by human activities have the potential to influence several aspects related to the occurrence and persistence of fish assemblages, such as habitat use and trophic ecology (^{Chua et al., 2020}). In fact, ^{Chua et al. (2020)} found that morphological traits can mediate fish occurrence in streams under deforestation, since fish with superior mouths and small body mass were associated with marginal grasses and reduced wood debris. According to these authors, fishes with such morphological traits are more successful in the use of trophic resources (e.g., drifted insects) and habitats in oil palm streams. Thus, ecomorphology, that studies the relationship between body shape and patterns of habitats use and foraging behavior by fishes, represents an important tool to understand patterns and process in fish ecology and conservation (^{Bourke et al., 1997}; ^{Peres-Neto, 1999}; ^{Oliveira et al., 2010}; ^{Feilich, López-Fernández, 2019}). For example, depressed and elongated body with large pectoral fins, as in Loricariidae, facilitates the occupancy of bottom habitats with high water velocity (e.g., riffles) because these characteristics can deflect most of the dragging forces present in the high current (^{Casatti, Castro, 2006}; ^{Oliveira et al., 2010}). In a scenario of deforestation followed by the loss of riffles due to siltation, species with these morphological features will probably be locally extinct.

Besides ecomorphology, land use (^{Ferreira et al., 2012}) and instream habitat changes (^{Zeni, Casatti, 2014}), trophic aspects can trophic aspects can also be driven by be driven by species interactions (^{Barros et al., 2017}). For instance, ^{Barros et al. (2017)} showed that individuals of Hyphessobrycon aff. melazonatus Durbin, 1908 have trophic plasticity and change their diet when in the presence of other small characids. According to these authors, this is a vital mechanism to weaken deleterious species interaction that, ultimately, promotes species coexistence. Although tropical stream fishes occupy almost the entire spectrum of trophic niches that can occur in aquatic communities (^{Winemiller et al., 2008}) probably mediated by their general morphology (^{Abelha et al., 2001}), species coexistence, occurrence, and abundance can be determined by resource partitioning that minimizes interspecific competition (^{Silva et al., 2016}).

For closely related species, niche segregation (^{Bouton et al., 199}7) and adjustments to different environmental conditions (^{Manna et al., 2012}) can facilitate species coexistence. In the Neotropical region, this issue has been studied mostly for loricariids (e.g., ^{Manna et al., 2012}; ^{Leitão et al., 2015}; ^{Abilhoa et al., 2016}), a group that generally shows conservative morphology and similar ecological requirements. For characids, by contrast, morphology and ecology correlate to a lesser extent (for examples of morpho-ecological convergence see ^{Casatti, Castro, 2006}; ^{Mazzoni et al., 2010}) and coexistence mechanisms remain little known. Several authors consider resource abundance, feeding plasticity, and differential habitat use as primary explanations for characid species coexistence (^{Ceneviva-Bastos et al., 2010}; ^{Manna et al., 2012}; ^{Moraes et al., 2013}). Despite being the most abundant family in Neotropical streams (together with Loricariidae) (^{Lowe-McConnell, 1999}), characids members still pose unanswered questions regarding morphology and dietary relationships.

A morphologically similar pair of characids is represented by the so-called “tetras” Inpaichthys kerri (^{Géry, Junk, 1977}) and Hyphessobrycon vilmae (^{Géry, 1966}) (Fig. 1), from meridional Amazon headwater streams, where their relative abundances differ according to the stream conditions (^{Casatti et al., 2020}). In the present study, we developed a comparative analysis of ecomorphology and diet of I. kerri and H. vilmae and tested for environmental variables that can explain the abundance of these species in the set of streams where they occur. Considering that morphology resemblance would predict similar ecological requirements between closely related species (see ^{Feilich, López-Fernández, 2019}), we predict that I. kerri and H. vilmae would show similar habitat use and trophic patterns. However, because there is a limit to the similarity that allows species coexistence (^{MacArthur, Levins, 1967}), we may expect that each species will have their abundance maximized in different environmental conditions if high trophic and ecomorphological niche resemblance are observed.

FIGURE 1 |  Lateral view of A. Inpaichthys kerri, 34 mm SL; and B. Hyphessobrycon vilmae, 32 mm SL. Photos: A. R. Manzotti.  

MATERIAL AND METHODS

Study area and target species. We developed the study in the Aripuanã River basin, north of Mato Grosso State, Brazil. The Aripuanã River belongs to the Amazonian basin in the “Madeira Brazilian Shield” ecoregion, which has high richness and endemism of fish species (^{WWF, TNC, 2015}). The region is in the deforestation arc, where the native forests are being replaced by pasture since the 1980’s, with higher intensity occurring after the year 2000 (^{Casatti et al., 2020}). In this basin, we sampled 30 streams reaches (80 meters length) in independent catchments (Fig. 2) during the dry period in the region, in July and August 2017. Samples were done in the dry season because of the access to the streams and the low water volume, which guarantees a standardized sampling of the target species.

Inpaichthys kerri is a small fish characterized by the presence of an incomplete lateral line and a caudal fin without scales. It is a monotypic genus, characterized by having a superior mouth, more posterior dorsal fin, and relatively deep caudal peduncle (Fig. 1A). Inpaichthys kerri is endemic from Aripuanã River basin and inhabits relatively fast waters (^{Géry, Junk, 1977}). Hyphessobryon vilmae is also a small fish with a relatively elongated body, dorsal fin in the middle portion of the body, and small pectoral fins not reaching the insertion of the pelvic fins (^{Géry, 1966}) (Fig. 1B). There is no information on the habitat use or other ecological aspects of this species, but it belongs to the Hyphessobrycon heterorhabdus group (Fernando R. Carvalho, 2019, pers. comm.), in which the species uses marginal habitats while foraging items drifted by the current (^{Brejão et al., 2013}).

FIGURE 2 |  Location of the sampled streams in the Aripuanã River basin, Mato Grosso State, Brazil. White circles represent streams with Inpaichthys kerri, black circles represent streams with Hyphessobrycon vilmae, and gray circles represent streams with both species. 

Field sampling. We used upstream and downstream block nets and sampled fish species with a hand seine (2 mm mesh) and a dip net (2 mm mesh) for one hour by two collectors. We anesthetized collected fishes with eugenol (^{Lucena et al., 2013}), fixed them with formaldehyde (10%), and after 72 h, transferred them to alcohol (70%). All specimens are deposited in the fish collection of the Departamento de Zoologia e Botânica - Universidade Estadual Paulista, São José do Rio Preto, SP, Brazil - DZSJRP (voucher numbers presented in the Tab. 1).

We used a standardized protocol to obtain physical and chemical variables. We measured chemical variables related to water quality (temperature, conductivity, dissolved oxygen, and pH) with a multiparameter probe (YSI 556 MPS). We took one measure in the downstream section for each stream reach before the fish sampling. We visually estimated substrate composition (% of consolidated and unconsolidated particles), trunks (%), marginal grasses (%), and shrubs and trees (%) in a 10 m stripe of the riparian buffer both in both stream margins. Moreover, we obtained average depth, width, and water velocity (flowmeter) by measuring these variables at several points along different mesohabitats in the stream reach. We also calculated the physical habitat index (PHI, based on ^{Casatti et al., 2006}) that measures the physical habitat integrity based on stream physical attributes (hydrological, structural, and riparian characteristics). High PHI values indicate streams with high habitat diversity and stability, while low values indicate streams with some degree of habitat homogenization and instability (^{Casatti et al., 2006}). We standardized all our environmental variables (average = 1 and standard deviation = 0) and conducted a principal component analysis (PCA) to select the most important ones based on the highest eigenvalues (^{Legendre, Legendre, 1998}). Based on this selection, we used PHI, dissolved oxygen, water velocity, % of marginal grasses, and % of trees in the riparian buffer for further analysis (Tab. 1).

Ecomorphological patterns. We selected 30 adult individuals from I. kerri and H. vilmae with similar standard length and well-preserved fins from all the streams they occurred. We followed ^{Oliveira et al. (2010)} to obtain 31 morphometric measures and calculated 22 ecomorphological attributes as shown in the Tab. S1. We obtained lengths, widths, and heights with digital calipers. For specimens smaller than five centimeters, we obtained measurements in a stereomicroscope 4 Zeiss, SteREO Discovery V12, with the help of AxioVision/Zeiss image software. To evaluate the multicollinearity of ecomorphological attributes, we used the Draftsman Plot function in the Primer 6 & Permanova+ (^{Clarke, Gorley, 2012}). Most of the correlations were low, only three of them were superior to |0.80| as shown in the Tab. S2. Despite ^{Legendre, Legendre (1998)} considered |0.80| as a high correlation, we decided to keep these ecomorphological attributes in the analysis, since we were trying to identify the morphological differences between the species and these attributes perform different functions. After that, we ran a principal component analysis (PCA) with a Pearson correlation (from -1 to +1) to evaluate differences in the use of the ecomorphological space between I. kerri and H. vilmae. To test for differences in the ecomorphological patterns of I. kerri and H. vilmae, we ran an analysis of similarity (ANOSIM) with normalized (square root) data and Euclidean index, followed by the SIMPER analysis to obtain the percentage of dissimilarity, in the Primer 6 & Permanova+ (^{Clarke, Gorley, 2012}).

Trophic patterns. For the trophic analysis, we selected specimens from streams where more than five adult individuals were sampled. Thus, stomach contents of 80 adult specimens of I. kerri and 80 of H. vilmae were used to describe the diet of these two species. We identified the food items based on the lowest possible level of taxonomic classification. Due to the high degree of fragmentation and digestion, we were unable to identify the food items in levels lower than order. To evaluate if the number of analyzed stomachs were able to sum up the most of the food items variation, we run an ICE (Incidence-based Coverage Estimator) and Jackniffe 1 (based on the number of items that are present only in one stomach) in the EstimateS 9.1 software (^{Colwell, 2013}). The sample completeness was calculated considering the number of items observed (NO) by the number of items estimated (NE) multiplied by 100 (NO/NE*100) (^{Marques, 2019}). We obtained the frequency of occurrence (FO) for each food item, based on the number of times that each item appeared in the stomachs divided by the number of analyzed stomachs (^{Hyslop, 1980}). We also obtained the volume (V), based on the visually estimated volume occupied by the item considering the volume total of the stomach (^{Zeni, Casatti, 2014}). With the values of FO and V, we calculated an adapted feeding index (IAi) from ^{Kawakami, Vazzoler (1980)} for each species. After that, we calculated the trophic niche breadth following the Levins index (^{Levins, 1968}) standardized as proposed by ^{Hurlbert (1978)}. To calculate trophic niche overlap between the species, we used the Bray-Curtis similarity index in Primer 6 & Permanova+ software (^{Clarke, Gorley, 2012}).

To test for differences in the diet of I. kerri and H. vilmae, we ran an analysis of similarity (ANOSIM) with normalized (square root) data and Bray-Curtis index in the Primer 6 & Permanova+ (^{Clarke, Gorley, 2012}). The SIMPER analysis was applied to verify the items that contribute most to the observed differences between the species. Finally, in the same software, we used non-parametric multivariate form of Mantel test called RELATE routine to test if environmental variables influenced the species diet. For these analyses, items were grouped into the following categories: algae (filamentous and diatoms algae), aquatic insects (immature obtained and into the Coleoptera, Diptera, Ephemeroptera, Hemiptera, Lepidoptera, Megaloptera, Odonata, Plecoptera, and Trichoptera), aquatic invertebrates (Acari, Bivalve, Cladocera, Copepoda, Decapoda, Nematoda, Oligochaeta, and Ostracoda), periphyton, superior plants (roots, seeds, and leaves of plants), terrestrial insects (Coleoptera, Isoptera, Diptera, and Hymenoptera), terrestrial invertebrates (Acari and Araneae), animal fragment (animal fragment that was impossible to determine the origin), and detritus (particulate organic matter).

Occurrence patterns. To identify the environmental variables that influence I. kerri and H. vilmae occurrence and abundance, we developed generalized additive mixed models (GAMMs). Despite species samples did not include clustering effects, we used these models because they allow incorporating spatial autocorrelation and detecting nonlinear relationships between explanatory and response variables. Thus, I. kerri and H. vilmae abundances (response variables) were modeled against the environmental factors (explanatory variables: PHI, dissolved oxygen, water velocity, % of marginal grasses in the banks, and % of the riparian buffer area covered by trees) considering five structures of spatial correlation: exponential, gaussian, linear, rational, and spherical (^{Zuur et al., 2009}), and poisson distribution. We selected the best-fitted model based on the significant p-values (p< 0.05) for all variables that composed the model, posteriorly on the lowest AIC values (Akaike Information Criterion) (^{Akaike, 1974}), and also on the variance of the residual. To run the GAMMs, we used “mgcv” version 1.8-16 (^{Wood, 2017}) and “nlme” packages in the R 3.2.2 environment (^{R Development Core Team, 2011}).

RESULTS

The first two PCA axes from the ecomorphological data explained 45.6% of the variation in the species morphological patterns. The first axis highlighted the interspecific variation by separating I. kerri and H. vilmae, while the second one showed the intraspecific variation (Fig. 3). Inpaichthys kerri seems to have higher intraspecific morphological variation. Moreover, I. kerri showed higher and wider caudal peduncles, more dorsally located eyes, and larger dorsal, pectoral, anal, and caudal fin areas compared to H. vilmae (Tab. 2, Fig. 3). Both species had distinct ecomorphological patterns (ANOSIM R = 0.695; p-value = 0.0001) with an average dissimilarity of 56.6%.

FIGURE 3 |  Biplot resulting from the principal component analysis (PCA) from ecomorphological data, showing the position of 30 individuals of Inpaichthys kerri and 30 of Hyphessobrycon vilmae in the ecomorphological space, represented by the two first axis (PC1 and PC2) and analyzed from 22 ecomorphological attributes. Arrow indicates the variation of the most important attributes (see Tab. 2). 

We recorded 30 food items for I. kerri and 24 for H. vilmae with estimation and completeness values that indicate acceptable sampling effort in reporting fish diet (I. kerri ICE = 33 items, completeness = 91% / Jackknife 1 = 35, completeness = 86%; H. vilmae ICE = 25, completeness = 96% / Jackknife 1 = 27, completeness = 89%). Both species consumed mainly aquatic insects (IAi > 0.80) (Tab. 3), but H. vilmae showed higher dominance of this item. Despite the similar consumption of aquatic insects, I. kerri had broader niche breadth (β = 0.40) than H. vilmae (β = 0.23), with 55% of niche overlap. We did not find any relationship between species diet and environmental variables (Tab. 3). Despite that, both species had distinct trophic patterns (ANOSIM R = 0.166; p-value = 0.0001) with an average dissimilarity of 79.7%. The consumption of aquatic insect fragments, Chironomidae, filamentous algae, terrestrial Hymenoptera and terrestrial insect fragments contributed the most to this dissimilarity (Tab. 4).

Inpaichthys kerri and H. vilmae abundances responded to different environmental conditions (Fig. 4), both without a significant spatial autocorrelation (as shown in the Fig. S3 and Fig. S4, respectively; and as shown in the Tab. S5 and Tab. S6, respectively). The optimal environment for I. kerri (i.e., maximum abundance) is established in conditions with high water velocity, dissolved oxygen ranging from 7 to 9 mg/l, and low proportion of marginal grasses (GAMM; Adj. R² = 0.91; p < 0.05) as shown in the Tab. S5. For H. vilmae optimal conditions were on streams with high PHI, a high proportion of marginal grasses, and dissolved oxygen concentrations higher than 9 mg/l (GAMM; Adj. R² = 0.80; p < 0.05) as shown in the Tab. S6.

FIGURE 4 |  Relationships obtained from generalized additive mixed models (GAMM) among the environmental variables and Inpaichthys kerri (A) and Hyhessobrycon vilmae (B) abundance. 

DISCUSSION

Despite the overall morphological resemblance between I. kerri and H. vilmae, and contrary to our former predictions, ecomorphological and trophic patterns were different. Inpaichthys kerri has a broader and deeper caudal peduncle, more dorsal eyes, and larger dorsal, caudal, pectoral, and anal fin areas when compared to H. vilmae. Despite the similar consumption of aquatic insects and 55% of niche overlap, species showed different trophic patterns, with I. kerri feeding upon a lower proportion of aquatic insects, filamentous algae, and detritus and a higher proportion of terrestrial insects when compared to H. vilmae. Following the ecomorphological and trophic patterns, I. kerri and H. vilmae abundances were maximized in different environmental conditions.

Large dorsal, caudal, pectoral, and anal fins in I. kerri allow a better capacity of stabilization in deflections (^{Gosline, 1971}), acceleration (^{Balon et al., 1986}) and maneuverability potential (^{Watson, Balon, 1984}; ^{Wilga, Lauder, 1999}). Moreover, more dorsal eyes lead to a better ability to detect items drifted by the current in streams (^{Pouilly et al., 2003}). These morphological features improve the natatorial ability, which probably increase the success of I. kerri feeding strategy. This species exploits stream margins and invests in drift-feeding (i.e., to consume items carried by the current) in the center of the channel, where water velocity and the probability of finding drifted items are higher. Accordingly, the optimal environmental conditions for I. kerri encompass fast waters and banks covered by low amounts of marginal grasses. Such conditions refer to more preserved and forested streams. Forests, notably those in the streamside buffers, help to trap inorganic sediments (see ^{Sweeney, Newbold, 2014}, and authors cited therein), and consequently, avoid riffles to be buried by fine sediments (^{Rabeni, Smale, 1995}; ^{Barbour et al., 1999}). Indeed, when compared to H. vilmae, I. kerri occurrence was associated with streams located within more forested conditions (see Tab. 1).

Despite the ecomorphological differences, H. vilmae also occupies stream margins. However, with a lower natatorial ability compared to I. kerri, it is possible that H. vilmae invests less in the drift-feeding strategy and forages by picking food items close to the marginal vegetation. The abundance of H. vilmae was high in streams characterized by having copious amounts of marginal grasses, which is commonly associated with the conversion of native forests to pasture in many regions (^{Zeni et al., 2019}). The removal of the native forest, especially in the riparian area, reduces canopy shading and increases solar radiation, which together contribute to the invasion of exotic grasses from adjacent pasture (^{Bunn et al., 1997}; ^{Evangelista et al., 2017}).

The fact that differences in abundance between species were explained by specific stream conditions suggests that, at some level, environmental filtering is modulating the coexistence of these species. When co-occurring, which happened in four streams, one species was more abundant than the other one (see Tab.1). Thus, environmental conditions contribute to determining which of these species may exhibit higher performance. Another way in which environmental conditions can modulate species performance is through resource availability. Aquatic insects are equally crucial for both species; however, filamentous algae were more important to H. vilmae and terrestrial insects to I. kerri, which indicates their occurrences are related to more altered (pasture in the catchment and grasses in the stream margins) and more preserved (forests in the catchment and in the streamside buffer) streams, respectively (Fig. 5).

FIGURE 5 |  General view of streams where Inpaichthys kerri (A) and Hyphessobrycon vilmae (B) individuals were collected in the Aripuanã River basin, Mato Grosso State, Brazil. 

Both species can be associated with a more generalist trophic pattern (^{Baldasso et al., 2019}). A detailed analysis of the consumed items, however, revealed that I. kerri preyed on a higher volume of immature Trichoptera than H. vilmae. Trichopterans are relatively sensitive to environmental degradation (^{Pereira et al., 2012}), being more frequent and abundant in streams associated with native forests (^{Nessimian et al., 2008}). By contrast, pasture dominance can be associated with a more significant development of filamentous algae, which is favored by high light intensities and high nutrient concentrations (^{Vázquez et al., 2011}). ^{Zeni, Casatti (2014)} reported that the consumption of insects (especially Diptera) and algae can be associated with the presence of marginal grasses in altered streams, such as observed for H. vilmae. This finer observation of the trophic spectrum can reinforce the role of environmental conditions by mediating species coexistence by influencing the availability of fish feeding resources.

Therefore, morphology, trophic features and environment interact to determine I. kerri and H. vilmae abundances in streams of the Aripuanã River basin. Inpaichthys kerri occurred in streams with high water velocity (e.g., riffles) and preserved margins (i.e., without grasses) and fed mainly on aquatic insects, especially Trichoptera and terrestrial insects obtained through a drift-feeding strategy, maximized by its larger fin areas. On the other hand, H. vilmae occurred mainly in streams with high proportion of marginal grasses and preyed almost entirely on aquatic insects, probably from these grasses. The environmental features that maximize I. kerri abundances are probably associated with more forested catchments and riparian areas that can prevent grasses proliferation (^{Evangelista et al., 2017}; ^{Zeni et al., 2019}), while H. vilmae abundance is maximized in streams with high degree of forest conversion to pasture. In this context, we can argue that the deforestation in the Aripuanã basin could be a sharp threat to the persistence of I. kerri through time, since the optimal environmental conditions that sustain this species tend to be gradually replaced by the optimal conditions of H. vilmae. This prediction is especially alarming because I. kerri is an endemic species that can be locally extinct if deforestation keeps increasing, and the occurrence of this species may be restricted to conservation units and indigenous lands of the region. Thus, the natural history information obtained in this study can help understanding species function in the ecosystems they occur, and it can also represent essential contributions for more effective conservation measures.