Introduction

The niche concept proposes that interactions between resources and consumers depend, in part, on how resources affect fitness components of consumers (^{Leibold, 1995}). Indeed, most direct effects of intraspecific food competition in fishes result in changes in population density, survival, and growth (Alanärä et al., 2001). Given such circumstances, understanding how mechanisms contribute to intraspecific partitioning has been a focus of ecological research. Thus, the estuarine fishes are excellent models because they can reduce the impact of competition by partitioning the resources along three main dimensions: habitat, trophic, and temporal segregation (^{Ross, 1986}). Study of the ontogenetic variations in fish diets has also enhanced our ability to understand coexistence strategies in an environment characterized by high amounts of microhabitats, such as estuaries (^{França et al., 2012}).

Estuarine fishes exploit a wide range of prey types, which are locally abundant in the environment, and this has frequently been used to infer trophic plasticity (^{Hostim-Silva et al., 1995}). Several studies of feeding habits in tropical estuarine fishes have reported shifts in prey consumption during the ontogeny (^{Pessanha, Araújo, 2001}), which has been associated with the optimal foraging theory proposed by ^{MacArthur, Pianka (1966}). However, differences in the physiological abilities, with reference to swimming ability and mouth and jaw size, also result in an expansion in the niche width by differences in predation ability and food requirements to be exploited as fish grow (^{Ward et al., 2006}). Analogous changes are documented in neotropical silversides (Atherinopsidae), including features of gut structure and function (^{Logothetis et al., 2001}; ^{Horn et al., 2006}), trade-offs between energy allocation in maintenance and somatic and reproductive growth (^{Billerbeck et al., 2001}; ^{Chizinski et al., 2007}), trophic ecomorphology (^{Cassemiro et al., 2003}; ^{Ross et al., 2006}) and the relation between fish size and feeding selectivity (^{Unger, Lewis, 1983}, ¹⁹⁹¹).

The Brazilian Silverside, Atherinella brasiliensis (Quoy & Gaimard, 1854), is a common estuarine resident and generalist species and is abundant in estuaries (^{Neves et al., 2006}; ^{Alves et al., 2016}) and others nearshore shallow water ecosystems, as such beaches (^{Chagas, Junior, 2013}; ^{Lacerda et al., 2014}; ^{Favero, Dias, 2015}), coastal lagoons (^{Andreata, 2012}) and bays (^{Paiva Filho, Gianinni, 1990}; ^{Pessanha, Araújo, 2001}; ^{Carvalho, Spach, 2015}). Furthermore, A. brasiliensis, show its important in a key structuring role in estuarine food webs (^{Campos et al., 2015}; ^{Figueiredo, Pessanha, 2016}), because generalist species are able to feeding and exploit quantitatively on prey from multiple basal nutrient sources (^{Claudino et al., 2015}). Generally, zooplankton, insects, diatoms form the principal diet of Brazilian silversides in estuaries (^{Contente et al., 2011}; ^{Alves et al., 2016}).

Here we aimed to (1) provide a detailed description of the diet and (2) identify potential competition for food resources between size classes of this species and (3) to relate the morphological traits to the use of resources, identifying potential prey/ fish morphology relationships. We discuss how differences in habitats preferences and foraging strategies among body sizes are major means by which A. brasiliensis avoid direct overlap in resource use in a tropical estuary, and morphological traits which are decisive in this strategy.

Material and Methods

Study area. The Mamanguape River estuary is located on the north coast of the Brazil and is part of the Environmental Protection Area of Barra de Mamanguape (Fig. 1). The regional climate is classified by Köppen as As-type (hot and humid). The mean rainfall recorded in the area is between 1,750 and 2,000 mm annually, and the mean temperature is approximately 24-26ºC. There is a well-preserved forest mangrove in the area, which grows around the primary channel and tidal creek and extends to 600 ha, in addition to Atlantic Forest remnants (^{Rocha et al., 2008}). Endangered species, such as the seahorse, Hippocampus reidi, and the West Indian manatee, Trichechus manatus, are also found in this estuary (^{Mourão, Nordi, 2003}).

Fig. 1 Study area. Map of the study area with locations of the sampling sites in the Mamanguape River Estuary, Brazil: Mud flat; Tidal Creek 1 and Tidal Creek 2. 

Three sampling sites were established along an estuarine salinity gradient and distinct physiographic features (Fig. 1): (1) Mudflat: unvegetated tidal bottoms and found in protected estuaries characterized by lower influence of waves, fine sediment (somewhat muddy), and that during low tide extend 1.22 km, (2) Tidal Creek 1: is a mangrove channel wide and shallow (0.72 m), which is bordered by a mangrove forest (5.3 km), substrate is a sandy type that forms sandbars exposed during the low tide, (3) Tidal Creek 2: is a mangrove canal located in the most central part of the estuary, bordered along its entire extension (1.4 km) of mangrove, deeper (0.82 m) and muddy sediment. This research was conducted under SISBIO Collection of Species Permit number 24557-1 issued by ICMBio, Brazilian Environmental Agency.

Sampling and data analysis. In order to have a more representative survey of the ichthyofauna, samplings were carried out during both the wet season (February to July 2011) and the dry season (August/2011 to January/2012). The fish were sampled using a beach seine (10.0 x 1.5 m; 8-mm mesh size). The seine hauls were 30-m long, parallel to and close to the shore, and were taken out to a depth of approximately 1.5 m. Each haul covered an area of approximately 30 m². This procedure was replicated three times in each sampling occasion. The fish collected were later identified, measured (total length in mm), and weighed (g). Euthanasia was performed by immersion of fishes in a Eugenol solution (300 mg L-1). After death, we fixed the fishes in 10% formalin for 15 days and then we moved them to 70% alcohol. Temperature, salinity and transparency were measured, using a thermometer, an optical refractometer and Secchi disc, respectively.

To investigate the distribution of size classes of A. brasiliensis in estuary, the aptured specimens were defined according to ^{Jensen (1997}). The individuals were divided into the following size classes: early juveniles (TL₁ = < 50 mm), juveniles (TL₂= 51-80 mm ) and adults (TL₃= > 80 mm); the fishes were defined as juveniles if they were less than 1/3rd asymptotic length, and adults were determined to be approximately 2/3rd of the asymptotic length (^{Jaxiom-Harm et al., 2012}). The maximum length was obtained from FishBase (http://www.fishbase.org/search.php). This method was applied with the aim of standardize the size of classes up to have a better comparison between them. Voucher specimens (LEP 098) were deposited into the ichthyological collection of the Laboratório de Ecologia de Peixes of the Universidade Estadual da Paraíba (LEP-UEPB).

Permutational analysis of variance (PERMANOVA) was used to determine any significant differences in fish density and biomass and other environmental variables for the different habitats and sampling periods (^{Anderson et al., 2008}). Two fixed factors were selected: habitat (three levels: Mudflat, Creek 1 and Creek 2) and season (two levels: dry and wet). Prior to the analyses, the environmental data were log transformed using log₁₀(x + 1). The data collected for density and biomass were square-root transformed. Euclidean distance matrices were calculated for the univariate variables.

To analyse each diet, the frequency of occurrence (%F), the percentage number (%N) and the volume (%V) of different food items were calculated (^{Hyslop, 1980}), and the Index of Relative Importance (IRI) was subsequently applied for the quantitative analysis of the stomach contents (^{Pinkas et al., 1970}). Each dietary item was identified to the lowest possible taxon. Although the volumes of unidentifiable materials were also calculated, these were not considered valid dietary categories and were not, therefore, included in a subsequent dietary analysis. For food items that could not be counted, a value of 0.1 was given for their number (%N) when they were present in the diet to offset distortions in the index (^{Abdurahiman et al., 2010}).

The PCO (Principal Coordinates) ordination was applied to obtain a graphical representation of the size classes, using the volume values. A matrix of food items was constructed to reduce the number of samples to facilitate the detection of feeding patterns, as described by ^{Schafer et al. (2002}) and ^{Platell, Potter (2001}). This means that the dietary data for a single individual contained large numbers of zero values, giving rise to instabilities in the calculation of similarities at an individual level, which greatly reduced the effectiveness of multivariate analyses of dietary data. This problem was efficiently minimised by averaging the dietary data (%V) for pooling stomachs contents of 3 to 5 individuals to produce a new series of replicates for a given factor (i.e. size classes or season).This ordinations were performed on Bray-Curtis similarity matrices to determine the extent to which individual factors influenced the dietary composition. Similarity percentage analysis (SIMPER) was used to determine which items were most responsible for the Bray-Curtis similarity within groups. The multivariate analyses were performed with the PRIMER software package, version 6.0 (^{Clarke, 1993}).

Feeding strategies was analyzed following the graphical method of ^{Costello (1990}), modified by ^{Amundsen et al. (1996}). In this method, the prey-specific (%V), defined as the average volumetric contribution of a food item (only for individuals that had this food item in their stomach), was plotted against the frequency of occurrence (%FO), and then interpreted with respect to the position within the graph. The position of the food items in the diagram represents the feeding strategy of the predator in terms of specialization or generalization. The Shannon-Wiener diversity index (^{Krebs, 1989}), which corresponds to dietary breadth, was used for each species using a volumetric data of feeding.

Functional traits. Sixteen morphological variables were measured based on ^{Keast, Webb (1966}), ^{Gatz Jr (1979}), ^{Watson, Balon (1984}), and ^{Beaumord, Petrere Jr. (1994}). A numeric vernier calliper (0.1 mm precision) was used to measure: total length (TL), standard length (SL), body height (BH), mean body height (MHB), body width (BW), head length (HL), head height (HH), relative eye height (ERH), pectoral fin length (PFL), pectoral fin width (PFW), caudal fin height (CFH), caudal peduncle length (CPL), caudal peduncle height (CPH), caudal peduncle width (CPW), mouth width (WM) and mouth height (HM). Morphological measurements were transformed following the method of ^{Schaefer (1992}) and ^{Pessanha et al. (2015}). Such transformations allow the removal of the body size effect, which standardises the measures, taking into account the effect of standard length on the analyzed morphological attributes. Morphological variations were examined using principal component analysis (PCA). PCA was performed on the mean values of 9 functional traits calculated for the fish species studied to evaluate the morphological affinities of each age classes.

Results

Environmental variables. Environmental data showed differences between sites and season (PERMANOVA, P< 0.001) (Tab. 1). There were no differences in temperature among sites and seasons. Salinity and transparency were different between sites and seasons (PERMANOVA, P< 0.001). Only transparency showed interactions between sites and seasons (PERMANOVA, P< 0.001).The highest salinity and water transparency were recorded in Mudflat and Tidal Creek 1, and the highest values were recorded during the dry season.

Distribution and abundance. During the study, the highest abundance and biomass were found in sites situated at the entrance of the estuary (Mudflat and Tidal Creek 1), and PERMANOVA (P< 0.001) indicated significantly spatial differences (Fig. 2). However, temporally, these descriptors were not significantly different between hydrologic seasons.

Fig. 2 Box-plot of spatial and temporal variation of number of individuals (CPUE average ±SE) and Biomass (±SE) of A. brasiliensis collected Mamanguape river estuary in Brazil. Bold lines indicate medians, hinges indicate the 25^th and 75^th percentiles, whiskers indicate the largest and smallest observation within a distance of 1.5 the box size. White bar= Wet season and Dark Gray bar= Dry season. 

All individuals caught had a total length (TL) of between 16-130 mm (Fig. 3). The age structure of A. brasiliensis differed among sites. The smallest individuals (small juveniles) showed the highest frequency (46%) in Mudflat, whereas the largest individuals (adults) were mostly in Tidal Creek 2.

Fig. 3 Size distributions of Atherinella brasiliensis at sampling areas (Mudflat, Tidal Creek 1 and Tidal Creek 2) in de Mamanguape estuary river. Number of individuals (n) is listed. Age Class: White bar= small juveniles; Light Gray bar= juveniles, and Dark Gray bar= adults. 

Spatial, temporal and ontogenetic shift on diet. We analyzed a total of 1099 stomachs, of which 259 were empty (23.6%). The overall diet composition was dominated by Infauna, zooplankton and insect preys (IRI= 74.7%, 20% and 3.8%, respectively). The diet of A. brasiliensis consisted mainly of Gastropoda (IRI= 60.3%), Ceratopogonidae larvae (9.5%), Decapoda larvae (7.6%), Calanoida (7.4%), Cyclopoida (3.5%), Hymenoptera (2.8%), and Brachyura (2.7%).

Significant variations were found in the diet of A. brasiliensis between sites (PERMANOVA: F_{2, 220}= 16.57; P<0.001) and size classes (PERMANOVA: F_{2, 220}= 8.92; P<0.001), but not hydrological periods (PERMANOVA: F_{2, 220}= 0.99; P=0.61). We detected considerable spatial variation in the main prey items among sites (Fig. 4). For instance, in Mudflat caught fish had more Gastropoda and Ceratopogonidae larvae in their stomachs, whereas Calanoida, Cyclopoida, and Decapoda larvae were abundant at Tidal Creek 1, and Hymenoptera and Polychaeta dominated at Tidal Creek 2 (Fig. 4). Our data suggest ontogenetic changes in dietary compositions at each site, with reduced contributions made by zooplankton items (Calanoida, Cyclopoida and Decapoda Larvae) for the early juveniles(TL₁), whereas the importance of benthonic organisms (Polychaeta, Brachyura, Ceratopogonidae larvae and Gastropoda) tended to increase in larger individuals (TL₃) (Fig. 4).

Fig. 4 Volumetric Percentage (%V) per habitat (a) of the most important food items in the diet each size classes (TL₁= small juveniles; TL₂= juveniles and TL₃= adults) of the Mamanguape river estuary, northeastern Brazil. (b: Mudflat, c: Tidal Creek 1 and d: Tidal Creek 2). 

By principal coordinates analysis (PCoA) we identified spatial and ontogenetic patterns within the dietary samples of A. brasiliensis (Fig. 5). The Mudflat samples lay to the lower part of the diagram due to the similarity of the diets, whereas samples from Tidal Creek 1 and 2 lay to the upper part of the diagram. By SIMPER analysis, we detected high dissimilarity and identified those items that most contributed to the A. brasiliensis diet for each site and age class. Gastropoda, Ceratopogonidae and Calanoida (Mudflat), Decapoda larvae (Creek 2) and Hymenoptera (Creek 3) made the greatest contributions to the dissimilarity between sites.

Fig. 5 Principal coordinate analysis ordination (PCO) coded by habitat (a) and size classes (b) for Atherinella brasiliensis in the Mamanguape River estuary, Brazil. Symbols Habitat: Mud flat (Full Black Triangle); Tidal Creek 1 (Gray Square); Tidal Creek 2 (Black Circle); Symbols size classes: Small juveniles (Open Black Square); Juveniles (Light Gray Circle) and Adults (Dark Gray Circle). 

In relation to size classes, the smallest and largest individuals of A. brasiliensis formed groups in the lower part of the diagram, whereas the intermediate sized fish formed a very broad group in the upper diagram (Fig. 5). By SIMPER analysis, we detected greater volumes of zooplankton items (Calanoida and Cyclopoida) in the diet of the smallest individuals (TL₁), whereas Gastropoda and Ceratopogonidae were more important in the diets of largest individuals (TL₃), and Hymenoptera and Calanoida were important items to intermediate sized fish (TL₂).

Functional traits. When combined, the Principal Components 1 and 2 (PC1 and PC2) explained 38.7% of the variation in variation in functional traits; PC1 explained 21.4% and PC2 explained 17.3% (Tab. 2) The high scores of PC1 were associated with RMW, RHL, and CPCI, whereas low scores were associated with IVF. High scores of PC2 were associated with IVF and RH. The PCA pattern shows that positive scores are associated with the small size class (plotted in the right part of the diagram), whereas adults were correlated with negative scores (plotted in the left part of the diagram) (Fig. 6).

Fig. 6 Ordination diagram from the first two principal components for the functional traits of Atherinella brasiliensis in Mamanguape river estuary. Symbols size classes: Small juveniles (Open Black Square); Juveniles (Gray Circle) and Adults (Black Circle). 

Niche overlap and feeding strategy. Feeding overlap occurred between the first two age classes at Mudflat and Tidal Creek 1 (0.70 and 0.94, respectively), whereas there was consistent overlap between age classes at Tidal Creek 2. In the Amundsen diagram, the results show generalized feeding habits and broad feeding niche; the preys were located in the lower left and bottom parts of the diagram (Fig. 7). Shifts in the main resource amongst the smaller and intermediate age classes (small juveniles and juveniles) were observed, and comprised largely of Calanoida and Cyclopoida. Decapoda larvae and Hymenoptera contributed most to the diets of adults.

Fig. 7 Feeding strategy for Atherinella brasiliensis in Mamanguape river estuary: A= Mudflat; B= Tidal Creek 1 and C= Tidal Creek 2. Food items: Cyc, Cyclopoida; Cal, Calanoida; Dec, Decapoda; Decl, Decapoda larvae; Hym, Hymnoptera; Egf, Fish eggs; Cer, Ceratopogonidae larvae; Ost, Ostracoda; Gas, Gastropoda; Pol, Polychaeta; Esc, Scale. (TL₁= small juveniles; TL₂= juveniles and TL₃= adult). 

Discussion

Brazilian silversides in the tropical estuary exhibited marked spatial abundance and ontogenetic diet shifts. Overall, A. brasiliensis was abundant and widely distributed throughout the estuary, despite their higher abundance in Mudflat. These results are consistent with the findings of ^{Alves et al. (2016}) in this estuary. Three characteristics of A. brasiliensis were identified as factors important to its success in estuarine areas: salinity tolerance (^{Souza-Bastos, Freire, 2011}), slow rates of growing with positive alometric coeficients and short-lived life cycle (^{Contente et al., 2011}; ^{Carvalho et al., 2017}). Furthermore, in accordance with the findings of other authors, the Brazilian silverside showed higher abundance in habitats less affected by the hydrodynamic process, such as extensive sand (^{Pessanha, Araujo, 2001}) and mud flats (^{Campos et al., 2015}), and also in sites which provide more availability of habitats, as such tidal creeks and mangroves (^{Neves et al., 2006}; ^{Fávaro et al., 2007}). As a consequence, these sheltered habitats can be used as refuges from predation and/or contribute significantly to high food resource. Most of the Brazilian silversides used these areas in the Mamanguape estuary as regular nursery habitats. The physiological capacities (^{Hostim-Silva et al., 1995}; ^{Neves et al., 2006}) and reproductive strategy (^{Fávaro et al., 2003}) of Brazilian silversides might also influence and drive their success in estuarine areas.

In the present study, higher proportions of adults at creeks and small juveniles at mudflats were identified as a main spatial difference between age classes of A. brasiliensis. These occupations of different habitats support the importance of shifts in niche dimension during the life cycle of fishes. Several studies have shown that A. brasiliensis changes its habitat use throughout its life stages (^{Santos et al., 1999}; ^{Fávaro et al., 2003}; ^{Chagas, Junior, 2013}), suggesting that deeper waters are preferred for feeding and initiating gonad maturation, and shallow habitats for depositing eggs when spawning (^{Cortinhas et al., 2016}). These patterns appear to be associated with ecophysiological factors, as described by ^{Sheaves et al. (2014}), interpreted as a result of abilities to access and use different nursery areas or use nursery grounds in different ways.

The diet of A. brasiliensis was based on zooplankton (Calanoida, Cyclopoida and Decapoda larvae), allochthonous items (Hymenoptera and others insects), and benthic organisms (Gastropoda, Polychaeta and Ceratopogonidae larvae). These results are in agreement with the findings of other authors (^{Ringuelet, 1942}; ^{Paiva, Schefer, 1982}; ^{Bemvenuti, 1990}; ^{Rodrigues, Bemvenuti, 2001}), and most of them show a great contribution of these items as nutritional resources for the fitness of individuals (somatic and reproductive growth) and thus contribute to ecological success of populations. Additionally, zooplankton have a higher energy content, explained by the ability of planktonic invertebrates to store large amounts of lipids (^{Whitfiled, 1985}); benthic invertebrates and insects (by adults) make feeding more energetically attractive to fishes and, thus, results are consistent with optimal foraging theory. However, our results share some dissimilarities with ^{Alves et al. (2016}). One reason for this could be that we analyzed the diet of fish caught in the tidal creeks, while those authors only analyzed fish from the main channel of the river Mamanguape.

Our results indicate that A. brasiliensis consumes a large variety of prey, reflecting their ability in foraging in diverse habitats. Nevertheless, the same prey categories were consumed at multiple sites. In creeks, zooplankton represented the main diet item but it was consumed less in Tidal Creek 2, followed by insect and Polychaeta, whereas in Mudflat the principal prey item was Gastropoda. According to ^{Figueiredo, Pessanha (2016}), the nature of available prey which inhabits these habitats in this estuary depends on the environmental characteristics, such as substrate and salinity gradient. Therefore, our findings reinforce that spatial heterogeneity is an important factor that influence fish feeding habits. With regards to the macrofauna and zooplankton in Mamanguape estuary, some studies have shown a distribution pattern at different spatial scales (^{Nobrega-Silva et al., 2016}; ^{Moura et al., 2016}; ^{Linden et al., 2017}), which has an important role in this food web (^{Campos et al., 2015}; ^{Claudino et al., 2015}; ^{Figueiredo, Pessanha 2016}).

Size-related dietary changes also were observed during the study, with indications of ontogenetic niche shifts: the diet of the smallest class was based on zooplankton, whereas larger individuals consumed mainly benthic macroinvertebrates. The PCO analysis reflected these progressive ontogenetic differences, demonstrating that the diet composition changes between age classes. Our results also support previous research regarding the transition from zooplanktivorous to zoobenthivorous, as reported for other atherinopsids (^{Bemvenuti, 1990}; ^{Chizinski et al., 2007}; ^{Chagas, Junior, 2013}). Second, differences in foraging abilities reflect an increase in the efficiency of handling procedure principally to feed on benthic preys (Polychaeta, Brachyura, and Gastropoda), followed by changes swimming ability to capture of preys (^{Hourston et al., 2004}; ^{Ross, 2006}). ^{Contente et al. (2011}) provides overviews that body form in Brazilian silversides can influence feeding performance, due to a forked caudal fin which is ideal for maneuverability and mobility in preys capture. ^{Billerbeck et al. (2001}), in a study of swimming performance in Menidia menidia (Atlantic silverside), found evidence of trade-offs among locomotor performance and substantial differences in intrinsic growth rates and food consumption. Thus, ontogenetic changes might allow Brazilian silverside to maximize their energy storage in order to grow quickly, making them less vulnerable to predation, which is consistent with our findings.

Our PCA analysis corroborated these shifts, which are principally related to changes in the mouth and head size, thus providing evidence of the importance of intraspecific resource partitioning. Intraspecific variation in consumer body size increases the range of acceptable prey sizes and, thus, can alleviate intraspecific competition (^{Bolnick et al., 2011}). Intraspecific variation in consumer body size increases the range of acceptable prey sizes and, thus, can alleviate intraspecific competition. We found that the smallest juveniles of A. brasiliensis were associated with Mouth aspect ratio (MAR) and Relative Height (RH), which have characteristics such as narrow mouths and large aperture, followed by the higher capacity for making vertical turns in the water column, facilitating the capture of small-size prey such as Calanoida and Cyclopoida. Zooplanktivores are filter-feeders that engulf a large amount of water and then collect the small suspended particles (^{Faye et al., 2012}). For adults, wide Relative Head Length (RHL) and Relative Mouth Width (RMW) are also considered indicative of similarity in food acquisition of items that inhabit the substrate, such as Gastropoda and Polychaeta. Segregation by size and habitats among estuarine fishes minimizes intraspecific niche overlap by increased resource partitioning (^{Schlosser, 1987}).

In conclusion, our data support the hypothesis that differences in the diet between size classes and habitat selection by A. brasiliensis are strategies for avoiding intraspecific competition, clearly related to abundance and accessibility of resources in the estuary. Furthermore, the generalist and opportunistic feeding strategy of A. brasiliensis allow sharing resources through consumption of different amounts of the same prey groups. This often produces asymmetries in consumer-resource dynamics that are driven primarily by population density and, to some extent, by size and age classes, which contribute to the partitioning of resources.