Multi-scale Homogenization of Caddisfly Metacomminities in Human-modified Landscapes

The multiple scale of stream networks spatial organization reflects the hierarchical arrangement of streams habitats with increasingly levels of complexity from sub-catchments until entire hydrographic basins. Through these multiple spatial scales, local stream habitats form nested subsets of increasingly landscape scale and habitat size with varying contributions of both alpha and beta diversity for the regional diversity. Here, we aimed to test the relative importance of multiple nested hierarchical levels of spatial scales while determining alpha and beta diversity of caddisflies in regions with different levels of landscape degradation in a core Cerrado area in Brazil. We used quantitative environmental variables to test the hypothesis that landscape homogenization affects the contribution of alpha and beta diversity of caddisflies to regional diversity. We found that the contribution of alpha and beta diversity for gamma diversity varied according to landscape degradation. Sub-catchments with more intense agriculture had lower diversity at multiple levels, markedly alpha and beta diversities. We have also found that environmental predictors mainly associated with water quality, channel size, and habitat integrity (lower scores indicate stream degradation) were related to community dissimilarity at the catchment scale. For an effective management of the headwater biodiversity of caddisfly, towards the conservation of these catchments, heterogeneous streams with more pristine riparian vegetation found within the river basin need to be preserved in protected areas. Additionally, in the most degraded areas the restoration of riparian vegetation and size increase of protected areas will be needed to accomplish such effort.


Introduction
Human-modified landscapes became a widespread habitat worldwide (Harden et al. 2014), leading to biotic homogenization of marine, terrestrial, and freshwater ecosystems (Solar et al. 2015;Olden et al. 2006). Freshwaters are amongst the most threatened ecosystems by human activities, which influence both water quality and biodiversity (Vörösmarty et al. 2010). Headwater streams are unique environments within a hydrographic basin, easily influenced by the surrounding landscape and changes in environmental conditions that affect the riparian forest structure (Meyer et al. 2007;Clarke et al. 2008Clarke et al. , 2010. The carrying of sediments, nutrients, and other agrochemicals may affect the entire freshwater ecosystems, severely affect native species populations, causing local extinctions, and facilitate their invasion by exotic species (Siqueira et al. 2015;Olden et al. 2006;Rahel 2002). The fact that the biodiversity of these ecosystems became so threatened (reviewed by Rahel 2002) reflects the particular spatial organization of stream networks of increasing hierarchy within the hydrographic basin, which catch the percolation of water flow from the landscapes (Heino et al. 2015).
Headwater streams have highly diverse, exclusive, and essential habitats sheltering the communities responsible for processing organic matter and for the maintenance of ecological processes within the entire basin (Merritt and Cummins 1996;Meyer et al. 2007;Minshall et al. 1992). For instance, the vegetation that covers streams, compose natural corridors that facilitate the dispersal of both aquatic and terrestrial species (Brown and Swan 2010). Changes affecting riverine landscapes (e.g., the removal of native forests and agriculture intensification) cause several alterations in streams and riparian vegetation environmental conditions, especially within headwaters that have high dependence on their adjacent land areas (Griffith et al. 2002;Allan and Castillo 2007). Although headwaters are amongst the most threatened ecosystems affected by soil-use change at the landscape scale, habitat degradation, water quality loss, and climatic change (Heino et al. 2012), they are usually neglected as conservation priorities and the design of protected areas (Clarke et al. 2010), especially in the Neotropical region.
The ecological processes operating in streams are usually determined by environmental characteristics occurring at different spatial scales (Brown et al. 2011;Dray et al. 2012). Consequently, a deep understanding of the organization of diversity patterns within stream metacommunities is necessary to interpret the scale at which these environmental characteristics act (Whittaker et al. 2001;Cushman and Mcgarigal 2002). Additionally, considering the stream network structure may help on the understanding of the variation of the biological communities' spatial patterns and how their heterogeneity is reflected in several environmental factors in multiple scales (Roque and Trivinho-Strixino 2001;Ligeiro et al. 2010; see Heino et al. 2015). This understanding may also allow the development of viable procedures for managing the riparian vegetation along streams' courses, towards more effective management/conservation strategies for the protection of stream biodiversity. Such better understanding may help conservation-planners identifying the geographic scales and regions that better represent the river basin's biodiversity (Clarke et al. 2010;Jost et al. 2010;Lessard et al. 2012;Ligeiro et al. 2010;Tylianakis et al. 2006).
Most studies regarding aquatic ecosystems focus on diversity at the streams scale and their habitats [e.g., microhabitat/substrate, and segments of rapids, and backwaters (Ligeiro et al. 2010;Heino et al. 2013)], disregarding the influence of processes at larger geographic scales on local community patterns. This scenario reinforces the need for more studies quantifying the effect of stream network on scale-dependence diversity especially in headwaters (Clarke et al. 2010;Heino et al. 2015).
According to Whittaker (1972), the diversity of a given region (γ diversity) may be decomposed into α and β diversity, which, respectively, represent local diversity and changes in species composition among sites. Among the many ways of quantifying α and β diversity (Whittaker 1972;Wilson and Shmida 1984;Jurasinski et al. 2009), the additive partition has been successfully used in the ecological literature (Wagner et al. 2000;Chandy et al. 2006;Ávila et al. 2011;Flohre et al. 2011). This approach allows understanding the contribution of α and β diversity in terms of effective number of species (Jost et al. 2010) in each scale and allows to define the spatial scale in which regional diversity would be maximized and maintained (Hepp and Melo 2013). The β diversity may also be assessed using dissimilarity in species composition between all pairs of sites sampled within a region (e.g., using Jaccard or Bray-Curtis; Wilson and Shmida 1984;Anderson et al. 2011). Any dissimilarity measure could be used as response variable to be plotted or modeled against both environmental and spatial predictors through powerful multivariate methods (see Anderson et al. 2011;Legendre and Legendre 2012), each with its own advantages. However, despite the existence of different ways to account for α and β diversity (Whittaker 1972), the understanding the effects of scale driving community patterns and also the underling processes at multiple spatial scales in metacommunities have failed to link observed patterns to dynamic mechanisms (Chandy et al. 2006;Crist et al. 2003;Freestone and Inouye 2006). Elucidating the mechanisms shaping regional diversity became an essential step to advance stream metacommunity theory (Brown et al. 2011), and lead freshwater research objectives to new and most predictive grounds (Heino et al. 2015).
The aquatic insect community represents the great majority of headwater stream invertebrates and is closely related with the stream's environmental features (Stendera and Johnson 2005;Mykrä et al. 2007). The order Trichoptera, for example, comprises taxa with high microhabitat selectivity and high sensitivity to local and regional environmental variation and so it is widely used to evaluate the integrity of aquatic ecosystems (Rosenberg and Resh 1993;Galbraith et al. 2007;Chakona et al. 2008;Gombeer et al. 2011;Baccaro et al. 2012;Ruiz-García et al. 2012;Baptista et al. 2013). From a metacommunity perspective and considering the structure of streams, most of these organisms may be considered intermediate dispersers, constrained within catchments and sub-catchments, and strongly controlled by environmental features, which are the most plausible mechanism driving their β diversity in multiple spatial scales (see Heino et al. 2015). Therefore, since the spatial autocorrelation of the environmental features from sampling sites near to one another is high (Legendre 1993;Diniz-Filho et al. 2003;Shurin et al. 2009), it is expected that the relationship between environmental heterogeneity and caddisfly β diversity increase at larger spatial scales (e.g., catchments) (Clarke et al. 2008). However, mechanisms underlying spatial variation of β diversity in stream metacommunities may also suffer from the intensity of human activities in the regional hydrographic landscape at multiple scale (Heino et al. 2015), mainly because instream habitat homogenization (Solar et al. 2015;Olden et al. 2006).
In this study, we analyzed the effect of the nested structure of headwater habitats inserted into hierarchically superior levels (streams, sub-catchment, and catchment), to assess the relative importance of both α and β diversity of caddisfly metacommunity from region that differ greatly in land use intensity (see Carvalho et al. 2009). We tested the following set of hypotheses: i) α, β, and γ diversities are, in average, higher in the catchment scale, reflecting the variation in environmental conditions of streams at broader scale; ii) α, β, and γ diversities are lower in intensively human-modified landscapes (i.e., streams from Paraná basin), reflecting the effect of both biotic and local habitat homogenization; iii) local environmental predictors related to landscape degradation are the most important drivers of metacommunity homogenization, because caddisflies suffer mainly from small scaled constraints on local habitats, which scale up to catchments landscapes, what may lead to better predictions of conservation status of stream diversity patterns.

Study Area
We collected 48 streams within the Paraná and Tocantins river basins (24 sites each), both representing the two most important basins of the Cerrado of Goiás state, Brazil. We used Landsat TM satellite images to select the sampling sites and define the scopes of the drainage networks (e.g., sub-catchments, catchments; see details below) (Fig. 1a). These river basins are located within the Brazilian Cerrado savanna, in central Brazil. This biome is considered one of the biodiversity hotspots and has high socioeconomic and conservation conflicts due to farming expansion (Myers . The studied basins have different land-use and exploitation histories, especially due to topography and soil quality. The Tocantins river basin comprises areas of higher altitude and slope that hinder the development of technified agriculture systems (e.g., that make intense and massive use of several farming machines) in all of its extension and, consequently, it holds larger areas with pristine Cerrado remnants (Carvalho et al. 2009). On the other hand, in the Paraná river basin, the areas have low slope, what allows the full development of technified plantations of several crops (e.g., soy, sugarcane, and cotton).

Sampling Design and Biotic Data Collection
We used a hierarchical nested design to sample the caddisfly specimens (Insecta: Trichoptera) in these 48 streams ( Fig.  1b) from both river basins, considering three hierarchical levels, from the highest to the lowest one: catchments, subcatchments, and the streams, following the terminology of Allan and Castillo (2007). Our experimental design was defined to represent the hierarchical organization of the stream network, given that our broadest scale of analysis reflects differences prevailing at the basin scale which includes four catchments and eight sub-catchments with three streams in each. This sampling design is similar to the one used by Clarke et al. (2010), and allows a better comprehension on the processes structuring the diversity of headwater streams. Such processes act on different spatial scales within the dendritic network of the catchments studied.
We selected the streams considering their size (maximum width of 5 m), riverbed features (stones or gravel with leaves as substrate) and presence of riparian vegetation. We carried out the sampling in the dry period of 2010 (July and August) to acquire the largest representation of the aquatic insect community (Diniz-Filho et al. 1998;Bispo et al. 2001).
In each stream, we drew a 100 m transect, divided into five 20 m segments. These segments were used as pseudoreplication to determine diversity in the stream-scale, and as subunits to build taxa accumulation curves to determine sample sufficiency within each hierarchical level (see details below). We sampled the insects in leaf substrates using a Surber sampler (area of 0.092 m 2 and mesh opening of 250 μm) in each one of the five segments and in two portions of substrate (riffles and pools). In each stream, the approximate sampling area was 1 m 2 per stream (2 substrates × 0.092 Suber's area × 5 segments = 0.92 m 2 ). We separated the collected material in the field and fixed it in 5% formalin. In the laboratory the material was transferred to 80% alcohol. Taxonomic identifications were made to genus level using taxonomic keys (Wiggins 1977;Merritt and Cummins 1996;Domínguez and Fernández 2001;Pes et al. 2005), under an optical stereomicroscope.

Spatial and Environmental Predictors
We lumped the environmental predictors we used into three groups: (1) local water features (e.g., pH, turbidity, electrical conductivity, water temperature, dissolved oxygen, and discharge); (2) habitat integrity index (HII), representing the environmental conservation state at a range of 50 m around the stream (see Nessimian et al. 2008); (3) set of landscape variables representing a classification of the main land use and vegetation cover in an area with a buffer around each creek. We selected these environmental variables because they show strong relationship with landscape degradation around streams associated with human activities (Nessimian et al. 2008). With them, we applied the HII protocol proposed by Nessimian et al. (2008), where the lower is the recorded score, the higher the environmental degradation observed in the stream. Such degradation is related to changes of the stream channel and/or riparian vegetation that occurred within 10 m surrounding the stream, what affects the availability of vegetation cover and substrates within the stream. The HII also correlates with others environmental factors of high quality of pristine streams. Pristine streams in this region have pH ranging from 5.5 to 7 (slightly acidic to neutral), low turbidity (low amount of total dissolved solids), and low ionic conductivity, which are related to small input of inorganic chemical from leaching riparian zones around the streams. Dissolved oxygen is, generally, higher in pristine streams than streams with low vegetation cover. We included the landscape predictors to reflect the status of the vegetation cover around the sampled streams up to 400 m. For these metrics, lower or negative values indicate pastures or exposed soils, respectively, while higher scores indicates high vegetation cover and vertical heterogeneity.
We obtained our landscape metrics from Landsat TM images with a resolution of 30 × 30 m composed of seven spectral bands and performed image composition with three georeferenced images and recorded bands (TM5, TM4, TM3). Then, we obtained the vegetation cover data from these images using the normalized difference vegetation index (NDVI) and a soil use classification. We extracted the NDVI through image processing in ArcGIS Desktop ® 10.1 (Esri). This index evaluates the presence and activity of photosynthetic vegetation and is related with photosynthetic biomass and percentage of ground cover. We assembled image mosaics to classify the soil use (e.g., agriculture, pasture, deforested area, forest, and Cerrado savannah) of areas surrounding the streams at 100, 200, and 400 m. We determined two large groups to summarize the variables (Natural: forest and savanna; Degraded: agriculture, exposed soil, and pasture area) from the land use initially observed.

Data Analysis
We carried out an independent-samples t-test for each variable studied to test whether the two basins have different local and landscape environmental characteristics. Thus, it was possible to locate the variables with values differing among basins and to better understand environmental variation at the broadest spatial scale represented by the Paraná and Tocantins basins.
We built species accumulation curves to indirectly assess the influence of land use intensification on regional diversity at different spatial scales. We used the stream levels as replicates to draw stratified curves within each scale (subcatchment and catchments; see Fig. 1b), where the samples were added sequentially without replacement and the confidence intervals calculated after 1000 permutations of individuals among samples.
In addition, we carried out an analysis of additive partitioning of species diversity to evaluate the relationships between diversity components at different spatial scales. The data were organized according to the following hierarchy: streams (α), among streams (β 1 ), among subcatchments (β 2 ), among catchments (β 3 ), and among basins (β 4 ). Thus, we partitioned the total diversity into these five components, with the model: γ = α + β 1 + β 2 + β 3 for the analysis of each basin, and γ = α + β 1 + β 2 + β 3 + β 4 for the analysis of both basin jointly. The α diversity was calculated as the average species richness per sample, and was expressed as a percentage of the number of species at the above γ level along the hierarchy Veech et al. 2002;. The β diversity components were calculated as the difference in the average number of species between each specific scale and the summed number of species shown above (β 1 , β 2 , and β 3 ) for each basin and jointly for both basins (which includes β 4 ). We used the PARTITION software version 2.0 (Veech and Crist 2007) to conduct the additive partition. This procedure completely randomizes individuals and taxa simultaneously for all levels in the hierarchy , allowing the statistical evaluation of the different components of the biological diversity through a complete randomization procedure.
We performed a Permutational Multivariate Analyses of Variance [PERMANOVA; (Legendre and Anderson 1999;Anderson 2001a;McArdle and Anderson 2001)] to test the effect of the hierarchical spatial scales on the variation of the Trichoptera community structure and on environmental factors and to directly assess the constraint imposed by the environmental covariates on community structure. This approach allows us to evaluate the changes in composition among sites, considering the hierarchical nested structures from streams to basins (Legendre and Anderson 1999). We partitioned the F statistics from the sum of squares and the crossed products matrix (SSCP matrix) to compare the difference among groups under the nested design of the analysis of variance table. We tested for significance using randomization tests confined within each sub-nesting level (Anderson 2001a). That way, streams, catchment, and subcatchment levels of each river basins were included as random factors nested at higher scales (streams < subcatchments < catchments < river basin, respectively), while basins were fixed factors. We only used the stream level as random sampling unit nested within each sub-catchment as true sample replicates (n = 3), which corresponded to the 48 observations. This enables test robustness even with a limited number of sample units given our balanced design through the hierarchy (see details in Anderson 2001b; McArdle and Anderson 2001). The procedure described was the same as for the PERMANOVA carried out with the Trichoptera community data (using the semi-metric Bray-Curtis index), as well as, for the Euclidean Distance of each environment. The same procedure applies in the incorporation of the covariables to the PERMANOVA of the community dissimilarity procedure. We obtained the significance of the PERMANOVA tests with 9999 random permutations.
We used a principal component analysis (PCA) for each set of predictors (except for the HII) to reduce data dimensionality and collinearity , retaining the first three PCA axes, which had eigenvalues larger than the overall eigenvalue average, for interpretation (Kaiser-Guttman criterion; Peres-Neto et al. 2005). We measured the range of variation for each set of predictors along the stream network hierarchy using the PERMANOVA based on the Euclidean distance of the original predictors' variance.
We also used PERMANOVA for the log-transformed abundance data of the community, based on the Bray-Curtis dissimilarity index in two ways. First, we considered only the nested hierarchical levels of the river network and quantified its influence on the Trichoptera community. Second, we separately inserted each of the three sets of predictors as environmental covariates, and used 999 permutations to test for significance. The environmental covariates represent physical and chemical, and habitat integrity and landscape characteristics, respectively. In a first round, we randomized streams within its respective sub-catchments. Then, we randomized subcatchments within its respective catchments, and, catchments within the two basins studied. Finally, we tested the effect of the basin level randomizing all streams within each basin (Anderson 2001b; see details in Supplementary Material).

Environmental Characterization
Sub-catchments along the two basins had different environmental characteristics, especially regarding dissolved oxygen and vegetation cover at a distance of 150 and 1000 m from the stream. These two variables had higher values in the Tocantins basin, reflecting its better preservation status due lower levels of mechanized agriculture, steeper relief, and recent occupation in comparison to the Paraná basin (Table 1). On the other hand, turbidity and percentage of land use at 400 m from the streams had higher values in the Paraná basin, indicating greater intensity of land use, higher nutrient leaching to the drainage system and higher landscape degradation.

Community Structure
We collected 6,593 caddisfly, pertaining to 26 genera and 11 families in the 48 streams of the Tocantins and Paraná basins. The most abundant genera were Smicridea (Hidropsychidae), with 3,892 specimens occurring in all of the sampled streams (81.08 ± 93.71; mean ± standard deviation) and Chimarra (Philopotamidae), with 921 specimens occurring in 25 streams (19.19 ± 42.64). However, some genera, were rarely sampled, such as Metrichia (Glossosomatidae) and Austrotinoides (Ecnomidae), with three and seven individuals, respectively, occurring in only three streams.

Taxa Accumulation Curves
We observed that only in the Tocantins basin the caddisfly diversity was equivalent to the accumulation curve in all sampling sites ( Fig. 2a and b, Table 1). Such pattern is reflected on the accumulation curves of the lower hierarchical levels of the river network, once there was higher regional diversity in the catchments of the Tocantins basin (note that region concept used here is scale dependent to meet our goals). The catchments within the Tocantins basin had genera richness values ranging from 20 to 25 genera, while the catchments within the Paraná basin have from 15 to 19 genera. This result may be a consequence of the higher level of degradation of Paraná basin. The accumulation curves in the sub-catchment scale showed high species richness variation for both basin (Fig. 2c and d). However, while the sub-catchments within the Paraná basin had up to 15 genera (minimum of 10), in the streams from the Tocantins basin, the taxa richness ranged from 10 to 24 genera. These results suggest that a larger area would be needed to maintain a greater number of Trichoptera taxa (i.e., gamma diversity) within the Paraná basin, which has suffered greater environmental impact than Tocantins basin (Table 1).

Additive Partition
The additive partition, considering both basins, showed that α and β 1 (among streams) diversities represented 34% and 24% of the total diversity (Fig. 3). However, it was not different from the expected after 9999 randomizations. The β diversity was higher than the expected for the scales sub-catchment (β 2 ), catchment (β 3 ), and basin (β 4 ), despite being proportionately higher for sub-catchment and catchment (19 and 20.1%) than for the basin scale. Distinct patterns emerged when we additively partitioned diversity separately for the two basins. The diversity observed at the stream-scale (β 1 ) contributed in genera richness to the total diversity only for the Tocantins basin. Finer scales (sub-catchment and catchment) contributed to the total diversity of both basins, as observed by the comparison with the null model. The variation in taxonomic composition among the catchments (β 3 ) provided around 33% of the total diversity within the Paraná basin, and was followed by the diversity among sub-catchments (β 2; 17%). However, in the Tocantins basin, the main contribution to regional diversity was provided by the variation of faunal composition among streams (β 1 ), followed by sub-catchment (β 2 ), and catchment (β 3 ) (27%, 20%, and 11%, respectively).

The Influence of the Hierarchy on the Community and on Environmental Factors
The different sets of environmental predictors showed a spatial variation, especially in the sub-catchment scale, for water physical-chemical features, and for landscape variables (Table 2). We used these sets of predictors as covariates to establish the effect of environmental variables and of the hierarchical nestedness on community variation in different spatial scales. The results of the PERMANOVA also showed a greater beta diversity variation among catchments (Table 3), which is consistent with the diversity partition results, and indicates the influence of environmental predictors (especially of physical and chemical variables, and of the habitat integrity index) upon the caddisfly diversity. These results differ from our initial predictions that environmental predictors would influence the community on different spatial scales. What we actually observed was that this influence occurred only at the catchment level (Fig. 4). In our PER-MANOVA, we used additive partitioning and observed that the caddisfly composition was constrained by factors that act at multiple spatial scales. Fig. 3 Ratio of observed and expected diversity partitioned into alpha and beta diversity in different scales (α = alpha diversity at stream, β 1 = beta diversity at stream, β 2 = beta diversity at sub-catchment, β 3 = at catchment, and β 4 = basin scale). The asterisks indicate the ratio of randomized samples with more taxa than the expected for each hierarchical level

Discussion
Our results indicate that variations in both α and β diversities are scale-dependent and that the level of regional degradation affects diversity at multiple geographical scales. Within the Paraná basin, landscapes experiencing stronger degradation from agriculture and pastures had lower average altitude and slope than those in the Tocantins basin, and community varied mainly among the catchment scale. On the other hand, the Tocantins basin had a greater proportion of vegetation cover, which possibly had a heavier contribution on the caddisfly regional diversity and was associated to the variation in taxonomic composition among streams, but also at the sub-catchment scale. The agriculture intensification homogenizes aquatic habitats, and exerts scale-dependent effects on diversity within the regional basin (Rahel 2002). The scale-dependent pattern and the environmental predictors were consistent among the different approaches and confirmed our expectations of smaller diversity of Trichoptera community in impacted areas at multiple spatial scales. Human-modified landscapes leads to the reduction of local and regional taxa richness and the β diversity of aquatic communities, through strong homogenization effects in stream habitats (Colzani et al. 2013;Solar et al. 2015). We found through species accumulation curves and multivariate analyses a selective loss of sensitive taxa in landscapes with intense human activity (mainly agriculture), whereas more widely distributed and resistant groups, especially of the family Hydropsychidae, remained. These result are also supported other studies in temperate and tropical ecosystems (see Harding et al. 1998;Flynn et al. 2009;Laliberté and Legendre 2010;Barragán et al. 2011;Flohre et al. 2011;Colzani et al. 2013). Regional diversity is lower in areas with higher agriculture intensity (e.g., Paraná basin), when compared with regions with larger amounts of conserved areas (e.g., Tocantins basin). This is most apparent in sub-catchments, catchments, and basins when species accumulation curves were evaluated. However, there is little variation in stream scale α diversity. We showed that the main contribution to regional diversity comes from the high β diversity among the spatial scales studied, especially for larger scales, as expected. Nevertheless, the communities have regionally important variations in composition in areas where land use impact is less intense, including in the streams level (e.g., Tocantins basin).
The β diversity among catchments contributed to the biggest of fraction of the Trichoptera regional diversity, but the importance of each scale varies according to land use intensification between basins. The β diversity at all spatial scale were higher at the most preserved region (Tocantins basin), including among streams, sub-catchments, and the catchments. Conversely, the β diversity in the Paraná basin was greater than expected at the catchment and subcatchments scales, but did not differ from the expected for streams reach. In addition, α diversity did not differ from expected by chance among streams in any basin, while representing 30-40% of the regional diversity. Our findings suggest that the degree of variation in species composition (i.e., β diversity) was higher at catchment than at stream or sub-catchment scales, which is essential to maintain regional diversity. Still, in the Tocantins (more preserved basins), the sub-catchment scale was more relatively important than at the Paraná, where only the catchment holds substantial fraction of the regional diversity. Most of the regional diversity variation was related to the variation of beta diversity in scales higher than the streams (sub-catchment and catchment). This pattern is probably due to the higher environmental heterogeneity and the restriction imposed to organism dispersal when moving to larger spatial scales (Leibold et al. 2004;Shurin et al. 2009). Moreover, it is important to notice that the stream-scale β diversity was higher than expected only in the Tocantins basin. This higher beta diversity may be a result of the better conservation status of the Tocantins basin, which was certainly exhibited a larger local and among-stream heterogeneity even at this scale. As hypothesized and in agreement with other studies Stendera and Johnson 2005;Jost et al. 2010), we were able to show that the diversity in all spatial scales decreases with intensification of land use by agriculture by comparing among less and more disturbed basins. Furthermore, and more importantly, conservation at the basin scale could be feasible only when considering the addition of streams at a broader spatial scale, and that the conservation of stream biodiversity should be achieved by protecting whole catchments.
Our results also indicate the influence of environmental heterogeneity on the diversity of aquatic insects (Brown Fig. 4 The variation of the Trichoptera community composition (beta diversity) among hierarchical scales (sub-catchment, catchment, and basin), constrained by three sets of environmental covariates, and the unconstrained community. The asterisks (*) represent the scale in which each covariate predictor significantly explained community variation (significance level p < 0.05) 2003; Heino and Mykrä 2008;Ligeiro et al. 2010;Brown et al. 2011;Heino et al. 2015). Agriculture intensification causes severe changes in aquatic environments with habitat loss and changes in water characteristics (Griffith et al. 2002;Chakona et al. 2008;Zhou et al. 2012). Environmental change at landscape scale occurs in aquatic systems within more degraded regions, and this reflects in the homogenization the aquatic fauna diversity. Thus, to achieve a diversity gain in human-modified areas an expansion in the geographical extent is needed to represent fully regional diversity (Harding et al. 1998;Linke et al. 2007).
Historically, studies regarding the effects of human impacts on streams concern local diversity and disregard the nested hierarchy of headwaters streams. Nevertheless, streams are important components nested into higher hierarchical levels of hydrographic basins (Clarke et al. 2010;Finn et al. 2011), and make up the dendritic networks forming natural barriers for the dispersal of many organisms (Brown et al. 2011), but specially non-flying organisms. However, taxonomic turnover among streams, sub-catchments, catchments, and even among basins represent a considerable fraction of the regional diversity in these ecosystems, as demonstrated in this study. Besides, our findings suggest that α, β, and γ diversities tend to decrease in areas with greater agriculture intensification. Our results support the findings of other authors for communities of animals and terrestrial plants in agricultural areas (Flohre et al. 2011) and are pioneer in identifying the scales of diversity variation associated with the main factors that determine such patterns.
Agriculture intensification changes the environmental conditions of streams, including micro-habitat availability and water physical-chemical parameters, allowing process like eutrophication, which directly affect aquatic communities (Griffith et al. 2002;Allan 2004;Petersen and Masters 2004;Niyogi et al. 2007;Death and Collier 2010;Zhou et al. 2012;Hunt et al. 2017a). Cook et al. (2017) reported that the increase of input of nutrients that leads to eutrophication may cause a homogenization of aquatic macroinvertebrate communities, with temporal reduction of β diversity. The combined effects of the amount of vegetation and the local effects of water quality, habitat size, and habitat integrity (HII) varied along the nested hierarchy of the studied hydrographies. We found that only the catchment scale significantly explained the variation of the Trichoptera community dissimilarity. The environmental covariates best related to such variation were water quality, channel size and habitat integrity locally, which explained the variance among catchments.
We were able to identify the relative importance of different sets of predictors along the spatial hierarchy, and observed that the effect of heterogeneity of local environmental predictors on the catchment scale is responsible for the variation in Trichoptera species composition. We obtained similar results regarding the importance of the different scales studied in structuring communities by combining additive partitioning approaches to multivariate techniques based on distance matrices. For these organisms, changes in local water characteristics and in the riverside may determine community variations, especially due to its high habitat selectivity and sensitivity to natural or anthropogenic changes, as previously stated in several studies (Mackay and Wiggins 1979;Buss et al. 2004;Chakona et al. 2008;Ruiz-García et al. 2012).
The factors that determine the variation in species richness and in Trichoptera community composition in streams have been linked water chemistry, substrate availability, and flow variation (Costa and Melo 2007;Galbraith et al. 2007;Ligeiro et al. 2010;Landeiro et al. 2012). Besides water features, environmental changes at larger scales related with vegetation cover and land use (Moraes et al. 2014;Galbraith et al. 2007;Hunt et al. 2017b), or simply with spatial variation, may influence the composition of the community of organisms with low or intermediate dispersal abilities (Leibold et al. 2004;Van de Meutter et al. 2007;Löbel and Rydin 2009). Our study shows results different from Galbraith et al. (2007), who found an equal contribution of local and landscape characteristics in determining shifts in the Trichoptera community composition. Our results suggest that landscape modification may interact with local factors at broader spatial scales, but may have contrasting effects on the organisms in headwater metacommunities. Besides, the Trichoptera compose an important link for the trophic dynamics of aquatic ecosystems and have higher biomass in comparison to other aquatic insects (Flint et al. 1999). Trichoptera is a very diverse order among the aquatic insects, with nearly three thousand species currently known in the Neotropics (Morse 2016). However, this order is still poorly studied in Brazil, and innumerous species are still undescribed and, those that are lack biogeographic information (Linnaean and Wallacean shortfalls, respectively; Whittaker et al. 2005;Lomolino 2004). This imply that most studies (as ours) carry out analyses only with genera level in Brazil, but have the advantage to potentially be most representative from other biogeographic regions, given that we rely part of our analysis on proportional diversity partitioning schemes (in addition to dissimilarities based on paired species composition).
Our results confirm the expectations of high beta diversity in headwater streams (Finn et al. 2011), and point out the need to develop effective conservation strategies in these lotic ecosystems (Erős et al. 2011;Juen and De Marco 2011;Kanno et al. 2012). In this direction, several streams in the same basin would be needed to be incorporated into the protected catchments to enable effective conservation and management of headwaters streams. Additionally, we showed that β diversity in the most impacted region would be maximized only at the larger spatial scales (catchmentscale), especially if it comes from more intensively humanmodified landscapes. Thus, in this region, only larger areas would be able to preserve a greater proportion of the regional diversity of caddisfly. Additionally, for an effective management of the headwater biodiversity of caddisfly, towards the conservation of these catchments, heterogeneous streams, with more pristine and conserved riparian vegetation within the river basin are needed to be considered in future conservation plans and future design of protected areas in order to increase conservation targets. Finally, we would also suggested that the riparian vegetation of degraded streams, especially those in the most affected catchment, must be recovered in order to revert the loss of aquatic species (caddisfly included) and improve overall conservation efforts.