Fish complementarity is associated to forests in Amazonian streams

The functional structure of communities is commonly measured by the variability in functional traits, which may demonstrate complementarity or redundancy patterns. In this study, we tested the influence of environmental variables on the functional structure of fish assemblages in Amazonian streams within a deforestation gradient. We calculated six ecomorphological traits related to habitat use from each fish species, and used them to calculate the net relatedness index (NRI) and the nearest taxon index (NTI). The set of species that used the habitat differently (complementary or overdispersed assemblages) occurred in sites with a greater proportion of forests. The set of species that used the habitat in a similar way (redundant or clustered assemblages) occurred in sites with a greater proportion of grasses in the stream banks. Therefore, the deforestation of entire watersheds, which has occurred in many Amazonian regions, may be a central factor for the functional homogenization of fish fauna.


Introduction
The functional diversity of a community can be greatly influenced by the loss or addition of species with different traits from most species (i.e., functionally unique) (Cianciaruso et al., 2013).These changes may occur due to different processes, and deforestation has been associated with decreases in functional diversity in different communities (Tilman et al., 1997;Dolédec et al., 2006;Flynn et al., 2009;Barragán et al., 2011).The consequences of these changes can be dramatic, especially in areas of high biodiversity, such as the Amazon (Barletta et al., 2010), one of the most important biomes of the planet due to the extent of its rainforests and drainage network (Krusche et al., 2005).Approximately 735,000 km 2 of the 5 million km 2 that comprised the original Amazon Forest biome have been deforested in Brazil until 2013 (Instituto Nacional de Pesquisas Espaciais (INPE), 2014).This phenomenon is particularly alarming in the state of Rondônia, which has the second highest deforestation rate in Brazil (772 km² in 2013), and in 2006 approximately 65.9% of the state area had been cleared (INPE, 2010).
Deforestation at the watershed or at the riparian buffer scale, affect stream characteristics at the local scale (Cruz et al., 2013), such as flow, depth, substrate composition, litter amount, stability of stream banks, and structural complexity (Gorman & Karr, 1978;Lorion & Kennedy, 2009;Casatti et al., 2009).Considering that the influence of these variables on species occurrence depends on their functional traits (Goldstein & Meador, 2005;Teresa & Casatti, 2012), it is presumable that the effects of deforestation on the functional structure of communities are mediated by changes at finer spatial scales.

2
The functional structure of communities is commonly measured through the variability in functional traits (i.e., functional diversity; Mouchet et al., 2010), which may demonstrate complementarity or redundancy patterns (Falk et al., 2006).High functional complementarity occurs in communities with higher functional diversity than expected by chance (Blüthgen & Klein, 2011).Conversely, functional redundancy is the occurrence of functionally similar species which have less functional diversity than expected by chance (Loreau, 2004).The occurrence of complementary or redundant communities may reflect the differential influence of environmental filters (Poff et al., 1997).For example, in highly degraded streams, where the harsh environmental conditions filters species through their traits, so that species with a given set of traits can only survive, it is expected that coexisting species would be functionally more similar (functionally redundant communities).Conversely, higher resource availability and habitat complexity in pristine streams may provide favourable conditions to functionally distinct species to coexist, forming communities with higher functional complementarity.
We tested the influence of environmental variables on the functional structure of Amazonian stream fish communities in watersheds with different degrees of deforestation.We expected to find communities functionally more different in stream reaches embedded in watersheds with higher amounts of forests.

Material and Methods
Study area.This study was conducted in the rio Machado basin (Fig. 1), which drains the most populated area of Rondônia, Northern Brazil, with a total catchment area of 75,400 km 2 .The rio Machado is approximately 1,200 km long (Fernandes & Guimarães, 2002) and is formed by the confluence of the Comemoração and Pimenta Bueno rivers.Along its course, it also receives the Rolim de Moura, Urupá, Jaru, Machadinho, and Preto rivers and flows into the right bank of the rio Madeira (Ballester et al., 2003).This region has many terra firme streams, which are intermittent during most of the dry season (Fernandes & Guimarães, 2002).This region has been altered since 1970, with settlements along the highway BR-364.The watersheds that form the rio Machado basin are covered by forests (mature and secondary, ranging from 0 to 100% of coverage) or grasses which are used as pasture for cattle ranching (Fernandes & Guimarães, 2002).Due to this mixed degree of forest cover conditions, the rio Machado basin represents a suitable model for studying the biological consequences of human activities, such as habitat loss and simplification, on diverse aspects of fish ecology, notably on the functional diversity.Samplings were conducted in streams with different degrees of forest cover, from highly degraded to entirely forested, like those inside the protected areas, such as Jaru Biological Reserve and Rio Preto-Jacundá, Castanheira, and Aquariquara Extractive Reserves.
Watersheds selection.We generated the drainage network and the watersheds using the hydrological model S.W.A.T. (Soil and Water Assessment Tools) and satellite images of MDET SRTM (90 x 90 m resolution) from NASA (available at www.usgs.gov) to select the watersheds to be sampled.In order to standardize the stream order (2 nd to 4 th orders sensu Strahler, 1957), we selected watersheds with areas between 1,500 ha and 5,000 ha that represented the forest coverage variation in the watersheds (from 0 to 100% of forests).Overall, we sampled 75 streams reaches (one per watershed), 80-m long, that were definitively selected in situ after following these criteria: accessibility and authorization by the owners, maximum depth of 1.5 m, and the presence of perennial watercourses.We conducted the fieldwork in August and October of 2011 and in June and July of 2012.These months are characterized by low rainfall and in both years the hydrological regime was similar (Agência Nacional das Águas (ANA), 2009).

Environmental variables.
As environmental variables we considered landscape and local attributes.The landscape variable was represented by the proportion of forests in the watershed, which was obtained for each site (see Table 1 for procedures).The amount of forests in the watershed influences not only habitat characteristics (Krusche et al., 2005;Gonçalves Jr. & Callisto, 2013), but also diversity patterns (Poole & Downing, 2004), and it is a good surrogate for the watershed's conservation status.
The local variables were obtained during the fieldwork.In each reach, we measured five local variables associated to fish habitat (see Table 1 for the details of how each variable was obtained): percentage of grasses in the riparian banks; percentage of submerged roots in the riparian banks; percentage of consolidate substrate; percentage of large wood debris on the stream bottom; and average depth (Table 1).(Jensen, 2000) in the software ERDAS 9.2.
Local scale: calculated from (at least 20 m) measurements obtained in each stream reach Grasses in the stream banks (%) GRA 35.02 ± 38.00 Percentage of the reach bank extension that was covered by marginal grasses derived from surrounding pasture entering the water.For this calculation, both stream sides were computed.
Submerged roots in the stream banks (%) ROO 3.43 ± 5.87 Percentage of the reach bank extension that presented roots derived from riparian trees entering the water.For this calculation, both stream sides were computed.
Consolidate substrate (%) CSU 2.11 ± 3.68 Percentage of gravel and cobbles (particles with 2-256 mm in size) on the bottom of each stream reach (following the classification of Krumbein & Sloss, 1963).
Large wood debris on the stream bed (%) LWD 11.35 ± 10.77 Percentage of fallen branches and trees, representing large wood debris, on the stream bed of each reach.
Fish data and ecomorphological traits.To collect fish, firstly we used two blocking nets (2 mm mesh) to isolate the stream reach.Two people collected fish using the most appropriate technique according to the reach characteristics.
A hand seine (2 mm mesh) was used for portions without marginal vegetation with a sandy or clay bottom; a dip net (2 mm mesh) was used for portions with trunks, branches, and gravel.The sampling effort was standardized in one hour for each reach.Fish were fixed in 10% formalin and transferred to 70% ethanol.Voucher specimens were deposited at the fish collection of the Departamento de Zoologia e Botânica (DZSJRP), Universidade Estadual Paulista, São José do Rio Preto, Sao Paulo, Brazil (for voucher numbers, see Appendix).
We considered ecomorphological traits related to habitat use as functional traits.From the set of 139 species (Appendix) sampled in the 75 streams, we measured 137 species, except for Potamotrygon orbignyi and Synbranchus marmoratus that were excluded from this analysis due to the absence of pectoral fins.We took 11 measurements from each specimen, which were used to calculate six ecomorphological traits (Table 2) related to adaptations to water flow, swimming ability, and position in the water column, following Gatz (1979), Mahon (1984), and Watson & Balon (1984).We obtained linear measurements, area, and width with a stereomicroscope (Zeiss Discovery V12 SteREO), coupled with an imaging software (AxioVision Zeiss) and digital caliper to the nearest 0.01 mm.For larger species, we obtained areas of fins and body by drawing their profiles on graph paper (Beaumord & Petrere Jr., 1994).
Functional structure.We calculated the net relatedness index (NRI) and the nearest taxon index (NTI) for each fish assemblage by using the functional dendrogram.To obtain the functional dendrogram we assembled a standardized matrix of ecomorphological traits (with zero mean and unit variance) by species and used the function "dist.ktab" in the software R (R Development Core Team, 2011), based on the distance matrix obtained by the generalization of Gower's distance.We used the unweighted pair-group method using arithmetic averages (UPGMA) clustering method (Pavoine et al., 2009).NRI and NTI were originally described by Webb (2000) for phylogenetic diversity and are considered relevant to represent the functional structure (Hidasi-Neto et al., 2012).We decided to use these indexes because they are based on presence/absence and, therefore, more sensitive to rare species that are more vulnerable in the degradation context.Positive values of NRI and NTI indicate functional redundancy and negative values indicate functional complementarity.The NRI and NTI correspond, respectively, to the standardized effect size of functional diversity indexes MPD (mean pairwise distance) and MNTD (mean nearest taxon distance) (Webb, 2000), multiplied by -1 and calculated in relation to 1,000 randomly generated communities using an independent swap algorithm, maintaining the observed species richness and occurrence frequency in the null communities (Gotelli & Entsminger, 2001).For this analysis, we used the functions 'ses.mpd' and 'ses.mntd' in the R (R Development Core Team, 2011) package 'picante' (Kembel et al., 2010).
Data analysis.We used a partial regression analysis to relate the landscape and local variables (explanatory variables) with the NRI and NTI (response variables).Prior to the analysis, we standardized the explanatory variables (with zero mean and unit variance).In order to guarantee spatial independence of data (Legendre & Fortin, 1989;Legendre & Legendre, 1998), we evaluated the spatial autocorrelation in the residuals generated in the partial regressions described previously.New partial regressions were carried out using the regression residuals as response variable and the spatial filters as predictor, taking the effect of environmental variables into account.The spatial filters were generated by eigenvector-based spatial filtering approach (Griffith, 2003) based on a matrix of fluvial distance among all pairs of sampled reaches.The spatial filters with significant spatial structure as measured by Moran's I coefficients, at the first distance class, higher than 0.5) were retained.We performed these analyses in the software SAM (Rangel et al., 2010).Lower values indicate fishes inhabiting fast waters.It is directly related to the ability to perform vertical spins (Gatz, 1979).
Index of ventral flattening IVF Middle line height divided by maximum body height.
Low values indicate fishes inhabiting environments with high hydrodynamism, able to maintain their position even when stationary (Hora, 1930).
Relative area of pectoral fin APF Pectoral fin area divided by body area.
High values indicate slow swimmers, which use pectoral fins to perform maneuvers and breakings, or fish inhabiting fast waters, which use them as airfoils to deflect the water current upwards and thereby, maintain themselves firmly attached to the substrate (Mahon, 1984;Watson & Balon, 1984).
Pectoral fin aspect ratio PFA Maximum length of the pectoral fin divided by its maximum width.

Relative eye position EP
Distance from the middle of the eye to the base of the head, divided by head height.

Fineness ratio FC
Standard length divided by the square root of the maximum height of body, multiplied by the maximum body width.
The influence of body shape on the ability to swim; values from 2 to 6 indicate low drag, the optimum ratio for swimming efficiency is 4.5 (Blake, 1983).
In order to identify the set of environmental variables that discriminate streams, we used the distance based Redundancy Analysis (dbRDA, as described by Legendre & Anderson, 1999).In dbRDA, a Principal Coordinate Analysis (PCoA) is used to extract the principal coordinates of a calculated matrix of distances.These principal coordinates are Euclidean representations of the distances and are suitable for analysis by linear models.Due to this, and because significance testing is by permutation, there was no need for an assumption of normality (Anderson, 2006).We conducted dbRDA in the Primer 6 software (Clarke & Gorley, 2006).In the resulting biplot, we identified a posteriori the stream reaches according to NTI values, and informed the most important variables.

Results
The partial regression with the NRI and NTI showed that explanatory variables only explained the NTI.The variables that significantly explained the NTI were the percentage of forest cover in the watershed, the percentage of grasses in the stream banks, and depth (Table 3), indicating that most of variation in functional diversity can be explained by the combined effects of landscape and local environmental predictors.The residuals from these regressions did not presented spatial structure, since the correlation between spatial filters and regression residuals were non-significant (P > 0.51).This indicates that there was no spatial autocorrelation in our database, which would inflate the type I error.
The first two axes of dbRDA accounted for 51.9% of the explained variation.The coefficients for linear combinations of environmental variables in the formation of dbRDA coordinates indicated that the percentage of forest cover in the watershed (axis 1 = 1.623, axis 2 = -0.680), the percent of submerged roots in the stream banks (axis 1 = 0.034, axis 2 = -0.008), the percentage of grasses in the stream banks (axis 1 = -0.016,axis 2 = -0.002),and depth (axis 1 = -0.003,axis 2 = 0.056) were the variables that contributed the most for stream variation.
By pooling the partial regression with the dbRDA results (Fig. 2), it is shown a gradient in which the more complementary communities were located in watersheds with higher proportions of forests.The more redundant communities were located in stream reaches with large amounts of grasses in the stream banks.

Discussion
As predicted, stream reaches in the most forested watersheds encompassed the more functionally complementary assemblages regarding fish habitat use.On the contrary, streams with a greater proportion of marginal grasses in stream banks were represented by more redundant assemblages.Therefore, local and landscape features influenced habitat use by stream fish.This relationship was mediated by functional traits, as revealed by the relationship between functional traits and environmental variables, and highlighted the importance of the habitat structure of streams in determining the patterns of functional diversity and composition.
The forest cover, a landscape predictor, was related to the proportion of submerged roots in the stream banks, a local variable.This relationship revealed the hierarchical influence of landscape features on streams habitat structure.In this same vein, the grasses gradient was the opposite of that for forests.Two implications can be inferred from this fact.First, the deforestation in the rio Machado basin has also probably affected the riparian zone.Otherwise, the riparian forests would control the amount of grasses growing in the stream banks (Bunn & Kellaway, 1997), and this variable would be of less importance for stream structure.Second, the deforestation dynamics in the region and the development of pasture for livestock, despite starting in the 1970's, has been severe enough to promote the functional redundancy of fish communities, as demonstrated here.
The greater complementarity in forested stream reaches can be attributed to the occurrence of species with functionally unique traits, a characteristic of complementary assemblages (Petchey & Gaston, 2002).The occurrence of these species is probably due to the availability of shelter, food resources associated to the riparian vegetation, and litter packs (Carvalho et al., 2013).Accordingly, functionally unique species tend to be lost with the removal of vegetation in the watershed (Devictor et al., 2008).If we assume that functionally unique species perform functions not carried out by other species (Mouillot et al., 2011(Mouillot et al., , 2013)), these results suggest that vegetation removal, one of the major threats to biodiversity in the region, could potentially impair ecosystem structure and functioning in streams (Turner, 1996;Laurance et al., 1998).
In our study, the NRI was not explained by the environmental variables, contrary to NTI.To explain such results we must understand the properties of these indexes.NRI is an index more sensitive to species present in deep branches of the dendrogram, i.e., functionally distinct species, whereas the NTI is more sensitive to variations towards the tips of the functional dendrogram (Webb, 2000;Hidasi-Neto et al., 2012).Our results show that communities along the environmental gradient were equally represented by species from different branches of the functional dendrogram (and then NRI did not vary).However, the number of species within each branch varied along the environmental gradient and, thus, they were detected by NTI.
Our results reinforced the need to preserve native forests, not only in the vicinity of streams, but also in the whole watershed because their forest elements can be transported downstream (Ferraz et al., 2005;Galas, 2013).Forest cover in the watershed influences habitat use by fish in streams and, consequently, the overall functional diversity of fish assemblages.The removal of forest can be a severe environmental filter (in the sense of Kraft et al., 2015) because it favors generalist species at the expense of functionally unique species, and therefore increases functional redundancy, at least on a reach scale.
Appendix.Species registered in the sampled streams, their voucher number and abundances (N).Potamotrygon orbignyi and Synbranchus marmoratus were not included in the present analysis.Classification follows Reis et al. (2003); except for Serrasalmidae that follows Calcagnotto et al. (2005) and Parauchenipterus porosus that follows Buckup et al. (2007)

Fig. 1 .
Fig. 1.Sampled sites along the rio Machado basin and the three main types of soil coverage (left).Hydrography of the rio Machado basin and flow direction of the rio Machado (right).

Fig. 2 .
Fig. 2. Biplot resulting from the distance based Redundancy Analysis with seven variables (landscape and local).The proportion of forest cover in the watershed, the proportion of grasses in the stream banks, and depth significantly explained the NTI (nearest taxon index) in the studied communities and therefore are represented here.Each community is identified by circles with different sizes according to the NTI values.

Table 1 .
Scales, variables, codes, mean ± standard deviation, and explanation of how each variable was obtained.

Table 3 .
Results from the partial regression analysis, including NRI and NTI as dependent variables.For variables codes, see Table1.Bold numbers of P indicate variables that significantly explain the functional indices.