Interconnectedness during high water maintains similarity in fish assemblages of island floodplain lakes in the Amazonian Basin

We conducted a study to test the hypothesis that interconnectedness among island floodplain lakes and the adjacent Solimões River during the flood stage of the hydrologic cycle is enough to maintain similarity in fish species assemblages. Gill net samples were collected during high and low water periods for three consecutive years (July 2004 to July 2006) in four lakes on Paciência Island. Two lakes, Piranha and Ressaca, are connected to the river all year, and the other two, Preto and Cacau, which are in the center of the island, are isolated during low water periods. The abundance, species richness and evenness of the fish assemblages in these lakes did not differ according to their relative positions or the season of the hydrological cycle, which confirmed our hypothesis. However, fish abundance during the dry season was greater than in the flood season. Apparently, the short period of full connection between the lakes is enough to allow the colonization of all fish species, but not to cause similar abundances. Our study indicates that persistence of the species composition of island floodplain lakes is primarily due to the annual replenishment of fish to the lakes during the flood season.

Several studies have been conducted during the past few decades to evaluate the effects of abiotic and biotic factors on the structure and dynamics of fish communities (CHESSON 1986, DUNSON & TRAVIS 1991, TEJERINA-GARRO et al. 1998, SÚAREZ et al. 2001, 2004).In general, it has been accepted that spatial habitat diversity promotes species diversity in fish communities (HOEINGHAUS et al. 2003, POUILLY & RODRIGUEZ 2004), but temporal aspects, such as seasonal changes in the habitat, have been little studied thus far.
Several hypotheses have been advanced to explain the high fish diversity in the Amazon Basin.LOWE-MCCONNELL (1999) suggested three possibilities: 1) the age and size of the watershed, 2) the high spatial heterogeneity, which promotes a wide range of niches, and 3) the interchange of fauna through the influx of rivers from neighboring basins resulting from geological changes.The Amazonian floodplain is a very dynamic system in which the water level rises and falls seasonally with a range between maximum and minimum annual levels of approximately 15 m near Manaus, which is in the central area of the basin (FILIZOLA et al. 2002), creating markedly distinct terrestrial and aquatic phases within the same habitat (JUNK et al. 1989).This system has been called a "mobile ecotone" by JUNK & SOARES (2002) due to the movement of the terrestrial-aquatic interface during the course of a hydrological cycle and the resulting dramatic environmental changes in the system.
The hydrodynamic strength of the Solimões River causes large modifications in the river channel, destroying lateral banks by erosion.At the same time, the deposition of material is the main force acting in the formation of new islands and river banks.These floodplain islands contain several lakes, some of which are isolated from one another during the dry season.However, these lakes are interconnected during the flood season, and the duration of the interconnection depends on their distance from the river.Lakes near the shore may never completely lose their connection with the river as the water level falls, while those nearer the center of the island may be isolated until the peak of flood season, and the water level depends exclusively on rainfall and groundwater during the rest of the year.
This would be expected to result in the possibility of the formation of metacommunities (HOLYOAK et al. 2005) with relatively independent assemblages of fish species in which coexistence might depend mainly on biotic interactions, such as competition (AMARASEKARE & NISBET 2001), based on the definition of HOLYAK et al. (2005) that a metacommunity is a regional set of communities linked by dispersal.However, a critical question about the occurrence of such patterns is how much dis-

Interconnectedness during high water maintains similarity in fish assemblages of island floodplain lakes in the Amazonian Basin
persal is required for a system to be deemed a metacommunity (HOLYOAK et al. 2005).
In this study, we tested the hypothesis that the interconnectedness of lakes that occurs during the flood season is sufficient to assure similarity in their associated fish assemblages, regardless of their spatial position on an island and relative isolation during the low water period.

MATERIAL AND METHODS
We sampled fish in four lakes of Paciência Island (Figs 1 and 2): Ressaca Lake (03°18'54"S, 60°12'37"W), Piranha Lake (03°16'40"S, 60°13'19"W), Cacau Lake (03°96'36"S, 60°38'31"W) and Preto Lake (03°18'46"S, 60°13'8"W).This island is located in the lower stretch of the Solimões River, approximately 50 km from Manaus (Amazonas).Ressaca Lake and Piranha Lake are relatively more recent and remain connected to the Solimões River from the beginning of the flooding season to the end of drying season: therefore, these lakes will be referred to hereafter as type CON (connected).They were formed by the natural deposition of material over the last few decades.In contrast, the Preto and Cacau Lakes are located in the center of the island and are isolated from the Solimões River until the peak of the flood season, when the water is at its maximum depth, and will hereafter be referred as type ISO (isolated).
For three consecutive years, we sampled fish at each lake three times during each season (dry and flood).We assumed that the dry season corresponded to a depth of less than 2 m, as measured close to center of the lake, and the flood season corresponded to depths above 5 m.This criterion is consistent with that proposed by BITTENCOURT & AMADIO (2007) to define seasons of the hydrological cycle in the Central Amazon.Thus, we conducted three samplings during the dry season: October 2004, August 2005, andJanuary 2006; and three samplings for the flood season: July 2004, June 2005, andMay 2006.Six gill nets (20 m long x 2 m high, with 20, 30, 40, 50, 60 and 70 mm stretched mesh) were deployed at each lake for approximately 10 hours from 10:00 am to 8:00 pm to catch fishes exhibiting both diurnal and nocturnal behavior.The fish caught were identified to the level of species and deposited in the fish collection of the Federal University of Amazonas.
A mixed model of one-way ANOVAs with repeated measures for multiple samplings was used to evaluate differences in species assemblage parameters.The spatial position of each lake (CON = Ressaca and Piranha Lakes; ISO = Cacau and Preto Lakes) was used as the independent factor, and abundance, species richness, and evenness were used as response variables.Evenness was calculated as: E = H'/ln S.
After detection of violation of homogeneity of variance by Levene's Test, the abundance and species richness data were log-transformed to avoid heterocedasticy.We used Mauchly's test for sphericity, and when it was violated, we employed the epsilon of Greenhouse-Geiser to fit the degrees of freedom (GIRDEN 1992).We did not use lake area as a factor in our analyses due to its wide variation during the year as the water rose and fell and the full connection of these lakes when we sampled them during the flood season (Figs 1 and 2).
We performed a detrented correspondence analysis -DCA (HILL & GAUCH 1980) to identify potential patterns in fish community structure among the lakes.This analysis was chosen to avoid the arch effect, a common feature in the results of correspondence analyses (MANLY 2005).
To evaluate the stability of community dynamics in CON and ISO-lakes, we measured the rate of change in species composition, or species turnover, estimating the percent turnover as proposed by DAVIES et al. (2005): (C obs + E obs )/(S i + S i+1 ) x 100, where, C obs is the number of species present in the flood season but not in the dry season (i.e., the number of species observed to colonize with the flood pulse); E obs is the number of species absent in the flood season but present in the dry season (i.e., the number of species observed to migrate as the water level rises); S i is the number of species observed in the dry season; and S i+1 is the number of species observed in the flood season.

RESULTS
We captured 2,019 fish during the six sampling events, which were distributed among 92 species, 68 genera, 19 families and 6 orders (Tab.I).The order Characiformes was the dominant group, with 47 species belonging to seven families.Siluriformes was the second most frequent order, with 20 species belonging to 4 families.We did not detect any differences in species composition among the lakes in the DCA analysis (Fig. 3).
The greatest species richness and abundance were observed in Ressaca Lake (Tab.II), but the other three lakes exhibited no pattern related to their spatial position with respect to either of these two variables.In fact, the two ISO lakes, Preto and Cacau, had somewhat higher mean species richness than Piranha Lake (CON).Abundance, species richness and evenness were also relatively constant in Preto Lake.In this lake, the evenness was always higher than 0.60, indicating an assemblage that is quite stable from one season to another.The other three lakes showed great variability, including within the same season.
We detected no significant differences in fish abundance, species richness and evenness between the CON and ISO lakes (Tab.III).There was, however, a significant difference in the abundance between sampling periods, but we did not find effects of the interaction between the lake type and sampling periods.
We found a p-value of 0.07 for the time factor on species richness in the ANOVA.However, the sphericity condition was not accepted and we used the epsilon of Huynh-Feldt as a correction factor.This correction factor reduces the degrees of freedom and improves the probability of type I error.Thus, this result should be interpreted cautiously, and an effect of the hydrological season on the richness could have been present.

DISCUSSION
Our working hypothesis was not refuted because connectedness was a poor predictor of the structure fish commu- Figure 3. Detrented correspondence analysis with lake name and species (codes given in table I).nities that inhabit island floodplain lakes.The two lakes that were relatively isolated from the main river during low water (Cacau and Preto) did not systematically differ from those that remained connected to the river (Ressaca and Piranha) with respect to abundance, species richness, and evenness.
Assuming that a metacommunity is composed of several metapopulations, we extrapolate the definition of metacommunity proposed by SALE et al. (2006) as a system in which: However, it is necessary to identify the importance of each factor in the structuring of floodplain fish communities.SÚAREZ et al. (2001) studied fish communities in Pantanal lagoons and concluded that the most important factors influencing fish community organization were macrophyte cover, abundance of piscivorous species, and the depth of the lagoons.However, these authors conducted their sampling during the dry season, after the lagoons were isolated.In contrast, we conducted sampling during both the dry and flood seasons, and our results indicate a lack of patterns in the structure of the fish assemblages among lakes during these periods.Our findings, therefore, support the hypothesis that the annual period of connection between the lakes and with the river during the peak of the flood season is enough to assure a similar opportunity of occurrence for all species in the lakes, independent of their isolation during the dry season.Any interactive assortment of predators or other factors that may occur during lake isolation is, therefore, mitigated by connectedness during high water period.Additionally, SIQUEIRA-SOUZA & FREITAS (2004) studied floodplain lakes of the Solimões River and observed the occurrence of species typical of lotic environments swimming in schools during the flood season.
In systems of large rivers with adjacent forested floodplains, the flood pulse is quite predictable (JUNK et al. 1989) and floodplains represent spawning and nursery grounds for many fish species as a result of the occurrence there of increased heterogeneity of habitats, food supply, and shelter against predators (GOULDING 1980, LOWE-MCCONNELL 1999).Thus, dispersion during high water seems to be an adaptive strategy of some fish species to exploit the variety of habitats available in the system.Fish diversity in floodplain lakes is related to habitat availability (PETRY et al. 2003).For instance, FREITAS & GARCEZ (2004) studied the fish assemblages in the connection canals between lakes and rivers and found the predominance of species that are able to use canals to quickly colonize more favorable environments, such as lakes.
Studies conducted in the Amazon floodplain indicate that seasonal changes due to flood pulses modify the relative influence of lentic and lotic waters on fish community structure (JUNK 1984) and that fish communities in lentic systems seem to be temporally more stable than those in lotic systems (MERONA 1987).The constancy of our percent turnover estimates supports this idea, indicating that changes in species composition are relatively constant for all lakes that occupy the same island.Thus, we were able to observe the influence of seasonal changes on assemblage structure.In our opinion, the observed stability in fish assemblage composition of the studied island floodplain lakes is due to the annual replenishment of fish to the lakes during the flood season.We conclude that the absence of streams in the drainage basin of these lakes is a key factor in explaining this homogeneity in the fish assemblages.

Figures
Figures 1-2.Study area showing Paciência Island and the sampled lakes: Piranha (A), Ressaca (B), Preto (C) and Cacau (D), during the flood season (1) and dry season (2).Dark channels indicating flooded forest are evident in I as an indication of the interconnectedness among lakes and between lakes and the main channel of the river.These convert to terrestrial forest during the low water period (II).

Table I .
Sampled species in Ressaca, Piranha, Cacau and Preto Lakes.We also report common names (as available), codes employed in the DCA (Fig.2), and the number of individuals sampled in each lake.

Table I .
Continued.

Table II .
Mean and standard deviation of ecological parameters for both seasons of the hydrological cycle (Dry and Flood).Lake types: (CON) connected year-round to the river; (ISO) isolated from the river and other lakes during the low water period.