Homogenization of Fish Composition Among Mesohabitats Driven by Stream Degradation from Urban Land Use

The heterogeneity along the course of streams beget habitat features that are highly different and strongly inuence the composition of sh assemblages. Stream stretches such as ries, runs, and pools are particularly distinct in physical structures and water ow, with expected differences in the identity and body shape of the species that occupy these habitat units. However, how land is used in the adjacent areas of these aquatic environments also changes the habitat characteristics and, therefore, the sh composition from each stretch. In this context, we collected data from both rural and urban streams to assess how these land-use types inuence the species composition and their body morphology among mesohabitats. Differences in body morphology were evaluated using Analysis of Variance (ANOVA) on the mean of Compression Index (CI) weighted by the species abundance for each sampled site. The differences in species composition were assessed using permutational multivariate analogous (PERMANOVA) and Indicator Value (IndVal). Urban streams showed a signicantly decreased sh diversity combined with no differences in body morphology of sh and homogenization of species composition among mesohabitats. Importantly, we could infer that mesohabitats inuence the body shape of sh and, consequently, species composition in less disturbed streams. However, the lower sh diversity in more imperiled streams led to the homogenization of sh composition among mesohabitats. These patterns constitute important contributions for evidence-based management and restoration of streams, as the presence of different mesohabitats is not enough to overcome the effects of urbanization on sh assemblages.


Introduction
The habitat-species relationship is a focus of ecological studies whereby habitat characteristics determine the presence and abundance of species (Smokorowski and Pratt 2007). In headwater streams, especially ow velocity, water depth, and substrate composition are factors that in uence the distribution of sh species along the stream channel ( (Pardo and Armitage 1997). Each of these areas can be de ned as apparently uniform and visually distinct habitat units, which can be classi ed into (i) ri es, where the slope is increased and the water ow is turbulent and fast, usually with a bedrock substrate, (ii) runs, in which the water ow is also fast, but less turbulent than ri es, and (iii) pools, with slow ow and deep water, usually with a ne substrate (Pardo and Armitage 1997). Studies have shown that these distinct characteristics of mesohabitats can select sh species according to their ecological requirements (e.g. diet, reproductive strategies, and substrate preference) ( However, little is known about the effect of different land uses on the sh composition among mesohabitats. Understanding how land use affects the diversity and composition of sh species in mesohabitats is crucial for the conservation of streams since these habitat units are responsible for maintaining the biodiversity of sh in these ecosystems (Hitchman et al. 2018) and, therefore, they are frequently used in management and restoration initiatives in highly altered basins (Wade et al. In this context, we collected data from ri es, runs, and pools of both rural and urban streams to assess how these land-use types in uence the species composition and their body morphology among these distinct mesohabitats. We predict that urbanization would have a su ciently negative effect on environmental conditions, mainly related to the physical conditions of the stream channel so that the diversity and the composition of sh assemblages would differ from rural ones. The physical changes in streams can lead to the loss of specialist species, especially those ri e-speci c (Berkman and Rabeni 1987), resulting in a homogenization of species composition among mesohabitats. Therefore, we predict that both the species composition and body morphology of sh would not differ among urban mesohabitats.

Study site
We carried out this study in ten headwater streams (1 st to 3 rd order, sensu Strahler 1957) belonging to the Pirapó River basin (Upper Paraná River system). The Pirapó River presents a drainage area of 5,000 km², with approximately 168 km of extension to its mouth, in the Paranapanema River (Pagotto et al. 2012). Streams were sampled within the rural and urban areas of the municipality of Maringá, Paraná State, Brazil (Fig. 1). A mixture of temporary croplands (mainly soy, corn, and sugar cane) and urban areas predominate in the landscape, which stands out among the three most populous cities in the Paraná State. This region integrates the transition between tropical and subtropical climate and is classi ed as a permanently hot humid rain climate zone, Cfa (h), according to the Köppen scale (Alvares et al. 2013). Mean annual temperature varies between 16° and 20°C, with January being the hottest and wettest month and July the month with coldest and driest records. In general, the annual rainfall rate in the region exceeds 1,000 mm (Alvares et al. 2013).

Field sampling
Sampling was carried out in April and May 2017. We selected 30 sampling sites, encompassing three mesohabitats (ri es, runs, and pools) from ve rural (Atlântico, Lombo, Queçaba, Romeira, and Roseira) and ve urban (Miosótis, Maringá, Mandacarú, Morangueira, and Guaiapó) streams (Fig. 1). Before the collection day, we visited several stretches along each stream to visually select the mesohabitats according to the description by Rincón (1999), with the following characterization: (i) ri es presenting fast and turbulent waters, with a substrate composed mainly of large rocks; (ii) runs with relatively fast waters, but deeper and less turbulent than ri es; (iii) pools presenting deep and slow water, with ne sediment the most common substrate. The prior visual selection of the mesohabitats ensured the standardization of their quantity, with each stream presenting a ri e, run, and pool. Also, each mesohabitat had to be at least ten meters long, since this length was used to standardize the size of the sample units, where the following variables were recorded: current velocity (with a JDC electronic owmeter, model Flowatch FL-K2), channel depth and width, proportions of ooded vegetation, canopy shading and substrate type (sand, rock, clay, and arti cial substrate), dissolved oxygen (O 2 ; DIGIMED, model DM-4P), pH (DIGIMED, model DM-22) and electrical conductivity (DIGIMED, model DM-32). As there was no strictly aquatic vegetation in any stream, we considered the roots, trunks, and branches of the riparian forest as ooded vegetation. Also, construction waste and trash were treated as arti cial substrates.
The stream width was measured in three transects, comprising upstream, downstream, and in the intermediate portion along each mesohabitat. The other variables were measured on the right and left margins and in the middle of each of the three transects, comprising nine collection points for each mesohabitat. We quanti ed the proportions of ooded vegetation, canopy shading by riparian vegetation, and substrate type using a 0.25 m 2 wooden square subdivided into 25 squares smaller than 0.01 m 2 and estimated their values from the sum of the lled squares. After quantifying these variables, we calculated their averages to characterize the mesohabitats according to their environmental conditions. We collected sh using electro shing with a 2,500W alternating current generator operated at 500V and 2A through successive passages deployed for 30 minutes in each mesohabitat. To optimize sampling effort and reduce sh escape, we blocked each mesohabitat downstream using a 5mm mesh size. We anesthetized the captured sh with benzocaine as an immersive solution for at least 10 minutes or until the opercular movement ceased. Subsequently, we placed the specimens in vials containing 4% diluted formaldehyde, to be transferred to 70% alcohol 72 hours later. We counted and identi ed the individuals according to Graça

Morphological measurements
Two morphological measures related to the body shape of the sh species were obtained, namely body height (BH) and body width (BW). BH and BW were taken from ten adult individuals of each species using a digital caliper (0.01 mm approximation), and for species that did not have such abundance, all individuals were measured. From the average of these two morphological measures for each species, the Compression Index (CI) was calculated by dividing the BH with BW. Thus, species with the highest CI values have a laterally compressed body, while the lowest values are for species that have a dorsoventrally depressed body (Winemiller 1991).

Data analysis
Fish diversity was assessed by species richness and the Shannon-Wiener index. The statistical differences in these assemblage attributes were assessed with a two-way ANOVA, including the interaction of the land use types and mesohabitats in the model. For these ANOVAs, we removed the pool site of the Lombo stream, as it was an outlier in these analyzes, presenting one and zero values for species richness and Shannon-Wiener index, respectively (Table S1 in Supplementary Information). Because each stream is considered three times in the analysis (three mesohabitats), stream identity was used as a blocking factor (additive factor) in ANOVAs and in all the models made below, to control its effect on the model variances To assess the differences in species composition among mesohabitats, an abundance sites × species matrix was subjected to Hellinger standardization (Legendre and Gallagher 2001), from which the Bray-Curtis distance was calculated. From this distance matrix, a Principal Coordinate Analysis (PCoA) was used to visually assess the mesohabitat ordination regarding the differences in relative abundances of species. Subsequently, the 12 environmental variables collected were standardized for zero mean and unit variance and correlated with the PCoA scores using the env t function of the vegan package. This function considers the environmental variables as the dependent variables that are explained by the ordination scores, and each dependent variable is analysed separately. Only the signi cant variables (p<0.05 based on 999 permutations) were added to the ordering graph.
The statistical differences in PCoA ordination according to land use types and mesohabitats were evaluated by Permutational multivariate ANOVA (PERMANOVA based on 999 permutations). Pairwise tests (based on 999 permutations) were then performed to assess the differences between all combinations of land use and mesohabitat factors. The assumption of homogeneity of variances was evaluated and met by permutation tests of multivariate dispersion (PERMDISP based on 999 permutations).
The indicator species of each mesohabitat were evaluated by the procedure adopted by Dufrêne and Legendre (1997) considering the probability <5% as indicator values (IndVal, based on 999 permutations) using the labdsv package (Roberts 2019). For this purpose, we regarded the combination of the levels between land use and mesohabitat factors on a matrix of species abundance. For both PERMANOVA and IndVal, rare species (with no more than 1-2 occurrences) were removed because they do not in uence IndVal p-values and disproportionately in uence the separation of points in PERMANOVA. Thus, seven species were removed from these analyzes, which were carried out with 19 species (Table 1).
To assess the difference in the body shape of the sh species among mesohabitats, a matrix was generated with the average values of the Compression Index (CI) weighted by the species abundance for each sampled site. For this, the CI values of each species were multiplied by the abundance matrix of the 19 species used in PERMANOVA and IndVal using the SYNCSA package (Debastiani and Pillar 2012). Then, an ANOVA was performed on the mean IC values using the interaction between the factors land use and mesohabitat as predictor variables. The assumptions of normality and homogeneity of variance were evaluated and met for all ANOVAs by the Shapiro-Wilk and Levene's tests, respectively.
To verify whether the composition of sh species was related to the proximity among sampling sites, we performed a Mantel correlogram with the hydrological distance among all sites using the mpmcorrelogram package (Matesanz et al. 2011). The species composition matrix was generated through the Bray-Curtis distance calculated on the Hellinger standardization of the species abundance. The hydrological distance was obtained from the calculation of the distance, in kilometers, between the collection points in the QGIS program (QGIS Development Team 2018), using the stream network of the Pirapó River basin, downloaded from the website of the Instituto Água e terra do Paraná (IAT 2021), and the geographical coordinates of the collection points. We performed all analyses in the R program (R Core Team 2020). The PCA and MANOVA were performed using the stats package, while the vegan package (Oksanen et al. 2020) was used to running the ANOVAs, PERMANOVA, PERMDISP, and env t function. All graphics were generated by the ggplot2 package (Wickham 2016).
The rst two axes of the PCoA explained 55.22% of the total sh abundance variation (Fig. 3). The rst axis separated P. reticulata (-0.32) and H. cf. nigromaculatus (-0.22) from the other species. These species mainly separated the sampling sites according to the land-use type, with urban points positioned left in the panel, while rural points were located right in the gure panel. The second axis mainly represented the species Gymnotus inaequilabiatus (-0.33), Phenacorhamdia tenebrosa (-0.31), Hypostomus strigaticeps (-0.25), Psalidodon aff. fasciatus (0.27), P. bockmanni (0.37), and P. aff. paranae (0.42). These species mainly separated the mesohabitats of the rural streams, with most of the pool points in the negative portion and ri e points in the positive portion of this axis (Fig. 3). Table 2, rural mesohabitats showed higher proportions of canopy shading, ooded vegetation, and dissolved O 2 .

As described in
On the other hand, urban mesohabitats showed higher values for electrical conductivity, channel width and sand in the substrate.
However, only seven variables showed a signi cant relationship with the PCoA scores (Fig. 3). Electrical conductivity was positively related to urban mesohabitats, while O 2 was related to rural ones. Also, the ow velocity and the rocky substrate was related to the ri es, while the depth, clay and sand with the pools.
Even excluding the rare species, some sh species showed exclusivity for a particular mesohabitat in rural streams, such as G. inaequilabiatus for ri es and P. bockmanni for pools (Table 1). However, due to the low abundance of these species, they were not signi cant in the IndVal analysis. According to IndVal, Impar nis mirini (Pval <0.01) and P. tenebrosa (Pval = 0.01) showed a preference for ri es and Astyanax lacustris (Pval = 0.02) for pools in rural streams. On the other hand, in urban streams, H. cf. nigromaculatus (Pval = 0.04) showing a preference for ri es and P. reticulata (Pval = 0.01) for runs.
The difference in the body shape of the sh species among mesohabitats depended on the land use type (Fig. 4). The ANOVA performed on the average of the Compression Index (CI) showed signi cance for the interaction between the factors land use and mesohabitat (F = 4.72, P = 0.024). Tukey's post hoc test revealed a difference among mesohabitats of rural streams, with ri es presenting the lowest mean values of CI, while pools the highest mean values. In other words, in rural streams, species with a dorsoventrally depressed body were more common in ri es, while in pools the most common was the laterally compressed body. However, this difference did not occur among mesohabitats in urban streams.
Lastly, the mantel correlogram revealed that the values of the correlation coe cient were not associated with the distance classes (P > 0.05). These results indicate that the composition of the sh assemblage is not related to the proximity among sampling sites, which should not deserve special attention in the analyses.

Discussion
Our results show that the land use type in uenced the environmental conditions and the sh assemblage in streams, with the urban environment having negative effects on the physical and chemical conditions and decreased species diversity. The lower species richness in urban streams resulted in the homogenization of sh composition among mesohabitats. These ndings are similar to a general pattern for urban streams, in which the loss of sensitive species and the dominance by tolerant and nonnative ones (Ruaro et al. 2018(Ruaro et al. , 2019Marques et al. 2020) lead to homogenization of the assemblage between these ecosystems (Walters et al. 2003; Cruz and Pompeu 2020), a pattern known as the "urban stream syndrome" (Walsh et al. 2005).
The environmental conditions of rural streams, such as larger portions of canopy shading, ooded vegetation (trunks and roots), and stable substrates (rock and clay) are characteristics of less disturbed environments Zeni et al. 2019). In these streams, the composition of sh species differed among mesohabitats (Fig. 4), mainly between ri es and pools. In the ri es, G. inaequilabiatus showed exclusivity, and I. mirini and P. tenebrosa were considered indicator species. These species presented a dorsoventrally depressed body, which is characteristics associated with fast water (Gaston et  The dorsoventrally depressed body of sh allows better exploitation of resources by species in the ri es because this body shape decreases the high energy cost associated with maintaining the position in the water column in fast waters, due to the hydraulic drag exercised over a large body surface area (Webb 1984(Webb , 1988. In contrast, the laterally compressed body is more e cient for pools because this body shape provides greater maneuverability for species (Werner 1977;Gerstner 1999), allowing e cient exploration of more structured lentic environments. Therefore, the difference in body morphology of sh led to the heterogeneity in the species composition between ri es and pools of streams.
The changes in environmental characteristics and the sh assemblage in urban streams led to environmental and biotic homogenization among mesohabitats. Since mesohabitats were previously selected, differences in water ow and depth occurred among urban mesohabitats in the same way as in rural ones. However, the presence of sand and arti cial substrate and the absence of ooded vegetation in most of the mesohabitats led to the environmental homogenization among these habitat units in urban streams (Cunico and  Urban streams showed a signi cant decrease in species diversity and were dominated by two species (i.e. P. reticulata and H. cf. nigromaculatus). Poecilia reticulata showed a clear dominance in the urban environment, representing 81.87% of the total individuals that were collected in these streams. This species was introduced in many Neotropical streams and thrived due to its high reproduction rate (Gomiero and Braga 2007) . Therefore, the high abundance of H. cf. nigromaculatus in urban mesohabitats may have occurred due to the possible increase in benthic algae, which is the main food resource of this species (Casatti 2002), caused by the entry of light in environments with few canopy shading (Taylor et al. 2004). Poecilia reticulata and H. cf. nigromaculatus were indicated by the IndVal analysis as indicators of runs and ri es urban mesohabitats, respectively. However, both were the most abundant species in all urban mesohabitats (Table 1). Therefore, the loss of mesohabitat-speci c species and the lack of preference for one of these habitat units for species that were present in urban streams led to the homogenization of the sh composition (Ruaro et al. 2019).
Indeed, P reticulata presented intermediate values for the compression index, which may explain its high relative abundance in all urban mesohabitats since this body shape is not characteristic of ri es or pool species. Also, even in the ri es, there are microhabitats in which the ow velocity is lower, as in the margins and between large rocks, which allows the presence of species with compressed laterally body without great losses of energy to remain in the water column (Casatti and Castro 2006).
On the other hand, H. nigromaculatus presented one of the lowest values for CI, a body shape that is not commonly found in pools. However, during the collections, we noticed that these species were captured close to the tiles and concrete structures in the urban pools, classi ed here as arti cial substrates. Therefore, we can deduce that these structures served as a hiding place for these species, in the same way as the rocks in the ri es.
A common response of the sh assemblage to the homogenization among habitats is the increased local richness, but the loss in the regional species pool (Rahel 2002;Nielsen et al. 2019). This same pattern could be observed here, with a signi cant decrease in species richness and diversity in urban streams, but between rural and urban mesohabitats this difference was less clear. The difference between the pools was due to the greater number of rare and exclusive species in this mesohabitat in rural streams (four of the seven rare species excluded from the IndVal and PERMANOVA analyzes; Table 1), so the loss of these species caused a signi cant decrease for the urban environment.
It is important to mention that rural streams are subtly closer to the mainstem river which, given the ecological theory of the hierarchical dendritic in river networks (Tonkin et al. 2018), may enhance the diversity of shes in these sections compared to headwater reaches in our urban streams (Carvalho et al. 2021). Although this caveat may be a limitation to our results -and we take the opportunity to hypothesize future insights dealing with this theoretical framework in our studied streams -several studies have pointed out abrupt environmental and hydrological differences in the same streams from the Pirapó River basin, mentioning severe consequences to their associated sh assemblages (e.g. Cunico  . Therefore, we are prone to a rm that the differences evidenced to the mesohabitats of the urban and rural streams evaluated here have a really strong association with the impacts of urbanization.
In short, we can conclude that in streams with less disturbed environmental conditions the distinct characteristics between ri es and pools in uence the body morphology of sh and, consequently, the species composition. However, in highly altered environments the loss of species leads to homogenization of sh composition among mesohabitats. These patterns constitute important contributions for evidence-based management and restoration of streams, as the presence of different mesohabitats, which in theory should promote greater diversity of species, is not enough to overcome the effects of urbanization on sh assemblages.

Page 14/19
Phenacorhamdia tenebrosa (Schubart, 1964) Pten 24 3 Rhamdia quelen ( Figure 1 Location of rural (1-5) and urban (6-10) sites where mesohabitats (ri es, runs, and pools) were sampled Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 3
Ordination, generated by the Principal Coordinate Analysis (PCoA), of the sampling sites according to the 19 species collected in mesohabitats from rural and urban streams. The species codes are described in Table 1. The length of the arrows indicates the strength of the relationship between environmental variables (E.conductivity = electric conductivity, Velocity = ow velocity) and PCoA scores.