Recent studies have shown that the spatial turnover of bacterial communities, that is, beta-diversity, is determined by a combination of different assembly mechanisms, such as species sorting, that is, environmental filtering, and dispersal-related mechanisms. However, it is currently unclear to what extent the importance of the different mechanisms depends on community traits. Here, we implemented a study using a rock pool metacommunity to test whether habitat specialization of bacterial taxa and groups or their phylogenetic identity influenced by which mechanisms communities were assembled. In general, our results show that species sorting was the most important assembly mechanism. However, we found that a larger fraction of the variation in bacterial community composition between pools could be explained by environmental factors in case of habitat generalists, that is, taxa that were widespread and abundant in the metacommunity, compared with habitat specialists, that is, taxa that had a more restricted distribution range and tended to be rare. Differences in assembly mechanisms were observed between different major phyla and classes. However, also here, a larger fraction of the variation in community composition among pools could be explained for taxonomic groups that contained on average more habitat generalists. In summary, our results show that species sorting is stronger for the most common taxa, indicating that beta-diversity along environmental gradients can be adequately described without considering rare taxa.
Recently, much progress has been made to integrate our understanding of the drivers of spatial turnover of communities, including those of bacteria (Hanson et al., 2012; Lindström & Langenheder, 2012). There is conclusive support that species sorting, that is, ‘filtering by local environmental conditions’, is important in structuring bacterial metacommunities (Langenheder & Ragnarsson, 2007; Van der Gucht et al., 2007; Logue & Lindström, 2010; Langenheder et al., 2012). Spatial factors can also explain variation in community composition indicating that, for example, dispersal-related mechanisms such as mass effects and dispersal limitation can be important as well (Langenheder et al., 2012; Lindström & Langenheder, 2012). Moreover, it has been shown that bacterial communities can be neutrally assembled, because abundances of taxa at the local scale are often well correlated with the occurrence and abundance of taxa in the metacommunity (Sloan et al., 2006; Östman et al., 2010), in particular in those which lack strong environmental gradients (Östman et al., 2012). It has also been shown that different mechanisms act simultaneously during the assembly of bacterial communities (Ofiteru et al., 2010; Langenheder & Szekely, 2011; Stegen et al., 2012). Hence, it is possible that different parts of bacterial communities are potentially assembled by different mechanisms depending on intrinsic properties or traits (Green et al., 2008; Burke et al., 2011; Barberan et al., 2012). The importance of spatial factors, for example, has been related to life-history traits related to dispersal, such as the ability to form resting stages (Bissett et al., 2010). The endeavor to understand how life-history traits influence the relative importance of different assembly mechanisms is not limited to bacteria. Accordingly, it has been found that body size and dispersal mode are important traits that influence differences in the importance of dispersal-related assembly mechanisms in aquatic organisms, and more specifically, that small species and good dispersers are less prone to be affected by dispersal limitation (Vanschoenwinkel et al., 2007; De Bie et al., 2012). It has also been shown for zooplankton communities that habitat generalists, that is, taxa that occur evenly distributed in a wide range of different habitats, are to a greater extent assembled by dispersal-related mechanisms, while habitat specialists that are more restricted in their habitat range are to a greater extent assembled by species sorting (Pandit et al., 2009).
Even though it has been shown that most functional traits are phylogenetically dispersed in bacteria (Martiny et al., 2013), there also seem to be traits that reflect life history and general attributes that are phylogenetically conserved at deeper levels, and consequently, it has been suggested that there is some ecological coherence at high taxonomic ranks (Philippot et al., 2010). Examples include growth rates, copiotrophy vs. oligotrophy (Fierer et al., 2007) as well as niche breadth along environmental gradients (Lennon et al., 2012). Thus, if some traits differ among phyla and classes, for example, one may hypothesize that there are also going to be differences in assembly mechanisms among them. Even though this has been observed (Barberan & Casamayor, 2010), our understanding about if and why this is the case is currently very limited.
Here, we carried out a rock pool metacommunity study with several specific aims. First, we wanted to test whether differences in habitat specialization of bacterial taxa result in differences in assembly mechanisms among them. Second, we investigated whether habitat specialization differed depending on the phylogenetic affiliation of taxa, resulting in differences in assembly mechanisms among phyla or classes. Specifically, we aimed to investigate whether there are differences in the relative importance of species sorting compared with dispersal-related assembly mechanisms depending on habitat specialization. In congruence with findings from zooplankton communities (Pandit et al., 2009), we would expect habitat specialists to be strongly assembled by species sorting because they have specific environmental requirements that can only be found at a few sites. On the contrary, species sorting should be weaker for habitat generalists because they can cope with a wide range of environmental conditions and therefore occur in a large number of locations.
Materials and methods
On June 30, 2009, 50 rock pools located on the island of Gräsö close to the Swedish Baltic Sea Coast (60°29.910′N, 18°25.768′E) were sampled. All pools were located within an area of approximately 3000 m2, and the distance to the coast ranged between 0.2 and 16.2 m (Fig. 1). At the time of the sampling, all pools represented separate water bodies. Temperature, maximum depth, length, and width were recorded for each pool, and conductivity was measured using a WTW conductivity meter (Cond 3210) with a TetraCon 325/C measuring cell (WTW Weilheim, Germany). We took plankton-net-filtered (> 250 μm) water samples that were transported back to the laboratory under cooled conditions and immediately processed for analyses of water chemistry, bacterial and flagellate abundance measurements, and analysis of bacterial community composition. Chlorophyll a (Chl-a), pH, absorbance, dissolved organic carbon (DOC), total phosphorus (TP), and total nitrogen (TN) concentrations were analyzed, and pool volume was calculated as described earlier (Langenheder & Ragnarsson, 2007). Flagellate abundance was determined using an epifluorescence microscope after staining with DAPI as described earlier (Langenheder & Jürgens, 2001). Cells with a clear signal under blue light excitation were classified as autotrophs and not counted, so that we only included flagellates with either no or very weak autofluorescence signal that are either heterotrophs or mixotrophs (from now onwards called heterotrophic nanoflagellates, HNF). Zooplankton (> 250 μm) were collected in the field by filtering volumes ranging between 0.5 and 5 L depending on zooplankton density through a plankton net and were preserved with 70% ethanol. Smaller zooplankton that passed through this net were collected later back in the laboratory using a 100-μm filter. Zooplankton were identified and counted using a dissective microscope and classified into the following categories: copepods (> 100 μm; mainly Cyclops spp. and Diaptomus spp.), Daphnia spp. (> 250 μm; D. magna, D. longispina, and D. pulex), and small ZP (all remaining zooplankton with a size between 100 and 250 μm).
Bacterial community composition analysis
Bacterial community composition was determined by 454 sequencing of the bacterial 16S rRNA gene. Following removal of zooplankton with a 100-μm plankton net, water samples were filtered onto 0.2-μm, 47-mm Supor-200 filters (Pall Corporation, Port Washington, NY) and stored at −80 °C until processing. DNA was extracted using the Power Soil DNA Isolation Kit (MO BIO, Carlsbad, CA), and the 16S rRNA gene was amplified by PCR using the adaptor-linked bacterial forward primer 341f and the adaptor- and sample-specific barcode-linked reverse universal primer 805R and the PCR conditions described in Langenheder & Szekely (2011).
The amplicons for 454 pyrosequencing were also prepared and sequenced as described before (Langenheder & Szekely, 2011). Low-quality, ambiguous, and chimeric sequences were evaluated and removed using AmpliconNoise and Perseus (Quince et al., 2011). Remaining clustering was performed into 3% dissimilarity OTUs using UCLUST (Edgar, 2010) and OTUs were classified using the naive Bayesian classifier (Wang et al., 2007) implemented in MOTHUR (Schloss et al., 2009). Only sequences affiliated with at least 95% similarity to bacterial sequences recognized in RDP at least at phylum level were included in the analyses. Sequences used in this study had an average size of 226 ± 6 bases and have been deposited to the NCBI sequence read archive under accession number SRR583945. Before the statistical analyses described below, the samples were standardized to the sample with lowest total sequence number (322 sequences) by excluding OTUs with a relative abundance of < 0.31% (for determination of this threshold, see details in Langenheder & Szekely (2011).
Definition of habitat specialists and generalists
Habitat specialization was calculated according to Pandit et al. (2009) using Levins' niche width (B) index (Levins, 1968):
where pij is the proportion of OTU j in rock pool i, and N is the total number of pools. B describes the extent of habitat specialization based on the distribution of species abundances in the metacommunity without taking the spatial location and environmental conditions in a local community into account, which is a prerequisite for the independence of the statistical analyses described below. Hence, OTUs with a high B occur at evenly distributed abundances along a wide range of habitats and can therefore be classified as habitat generalists. Meanwhile, OTUs with a low B have uneven distributed abundances among pool and can therefore be considered as habitat specialists. Accordingly, the lowest possible B value is 1 for OTUs present in only one pool, that is, singletons. To compare B with other general measures of species' distributions and rarity (Rabinowitz, 1981; Rabinowitz et al., 1986; Gaston, 1994), we calculated for each OTU (1) its occurrence in the metacommunity by dividing the number of pools in which it was detected by the total number of pools (N = 50), (2) its average regional abundance, that is, the average relative abundance in all 50 rock pools, and (3) its average local abundance, that is, its average abundance in the pools where it was detected. Because these data were not normally distributed, we used Spearman rank order correlations (ρ) to test whether there were significant correlations between B of OTUs and their occupancy and mean regional and local abundances, respectively. B varied between 1 and 16.8. To test whether community assembly mechanisms differed depending on habitat specialization of taxa, we divided the community into subsets that would differ in the degree of habitat specialization of their OTUs in a similar fashion than done for zooplankton (Siqueira et al., 2012) and bacterioplankton (Logares et al., 2013) previously. We split up the data set along arbitrary cutoff values (B < 3, B 3–6, B 6–10, B > 10) and used partial redundancy analyses (RDA) as described below (see ‘Statistical Analyses’ below) to test whether there were differences in the underlying assembly mechanisms between the subsets. To account for differences in information content among the different matrices that were used for the RDAs, the abundance of each species was weighed by dividing it with ∑pi (1 − pi), where pi is the proportion of sites occupied by the ith species (Siqueira et al., 2012). Further, because singletons are known to introduce noise into multivariate analyses, we excluded them from the B < 3 data set (we also performed the analyses with the singletons included, which yielded very similar results).
We also calculated the average value of B for all OTUs of the following bacterial phyla: Bacteroidetes, Actinobacteria, Proteobacteria, and Cyanobacteria, which are known to be dominant in freshwater and brackish water (Newton et al., 2011) and which were also the dominant phyla in the pools sampled here with all of them having more than 20 OTUs. We further split Proteobacteria into its most abundant classes (Alpha-, Beta-, and Gammaproteobacteria) as these all contained more than 20 OTUs each and had significantly different B values.
There were considerable differences in conductivity among the 50 pools (range, 112 and 20 200 μS cm−1, Fig. 1), and there was a major gap in this salinity range because there were no pools with intermediate conductivity between 512 and 2980 μS cm−1. To take into account that this phenomenon might have had a strong overall structuring effect on the bacterial assemblages, we also split up the data set to separately consider freshwater pools (conductivity range, 112–512 μS cm−1, 23 pools) and brackish pools (conductivity range, 2980–20 200 μS cm−1, 27 pools). We used the same niche width categories as above, with the difference that we pooled the B 6–10 and B > 10 categories due to the low number of OTUs in the latter. The difference between the communities of freshwater and brackish pools was analyzed by ANOSIM (analysis of similarity, Clarke, 1993), which was performed with 999 permutations using vegan package in R (Oksanen et al., 2012).
For the bacterial community composition data, we initially performed detrended correspondence analyses (DCA), which showed a length of the first gradient > 4 for all data sets, thus an unimodal gradient in the OTU data, which opts for the use of canonical correspondence analyses (CCA; ter Braak & Smilauer, 2002). Accordingly, CCA was used to visualize differences in community composition in dependence of environmental conditions. However, because there is currently no way to perform partial CCA analyses including an adjustment of R2 values according to Peres-Neto et al. (2006), we used a chord transformation to linearize the data and analyzed it further using partial redundancy analysis (RDA; Ramette & Tiedje, 2007; Legendre & Birks, 2012). This approach had the aim to determine the fraction of variation in community composition that can be explained by local environmental and spatial variables. If only environmental variables, but not spatial variables, are significant, this would indicate species sorting, whereas the opposite case, where only spatial but not environmental variables are significant, would indicate that dispersal-related assembly mechanisms are important (Cottenie, 2005; Langenheder et al., 2012). To select environmental variables to be included, we first implemented a PCA with all measured environmental and morphometric variables to be able to sort out those which covaried in explaining differences in environmental conditions among pools, resulting in a list of nine environmental variables (E) that were included as initial environmental parameters in all RDAs that were implemented (conductivity; Chl-a, TP, and DOC concentration; pool volume; HNF, copepods, Daphnia spp., and small ZP abundances). Environmental parameters were normalized and standardized by log and Z-score transformations (Leps & Smilauer, 2003). Spatial variables (S) were obtained using the principal coordinates of neighbor matrices (PCNM) procedure (Griffith & Peres-Neto, 2006), resulting in 11 spatial variables for all pools and 5 and 6 when freshwater and brackish pools were analyzed separately, respectively. We also included distance from the coast as an additional spatial variable. When the analyses were carried out separately for different bacterial phyla and classes, we excluded ‘empty pools’, that is, pools that were not occupied; for example, if no taxon belonging to the Gammaproteobacteria could be detected in a pool, this pool was excluded. RDA analysis was implemented as follows: First, we tested separately whether E and S were significant. If that was the case, we went on with a forward selection procedure as described by Blanchet et al. (2008) to select a subset of environmental variables that contribute significantly to explain variation in community composition among sites ([E], [S]). We further used variation partitioning (Borcard et al., 1992) to determine how much of the variation could be attributed to [E] and [S]. Initially, we tested whether [E] and [S] contributed significantly to the explanation of variation in community composition among rock pools. Thus, we calculated the fractions [E|S] (pure environmental variation), [S|E] (pure spatial variation) and [S-E] (shared variation). In all models, significance testing was carried out with 999 Monte Carlo permutations under the reduced model, and all R2 values were calculated and adjusted as described by Peres-Neto et al. (2006). All analyses were carried out using CANOCO 4.5 (Biometrics, Wageningen, the Netherlands).
Pearson product-moment correlations (r) were used to test whether average values of B calculated for a phylum/class were correlated with the amount of variation in community composition that could be explained in total ([E+S]) and by [E|S] and [S|E], respectively.
Bacterial community composition
Of 76 802 good-quality sequences, after denoising and chimera removal, 61 590 could be assigned with at least 95% similarity to known bacterial phyla. The 50 rock pool samples analyzed contained on average 1260 identified sequences. After removal of chloroplast sequences and standardization, sequences were grouped into 543 OTUs with an average of 31 OTUs per sample (range: 13–65).
Based on phylogenetic classification, OTUs could be affiliated to 15 different phyla. Proteobacteria, Bacteroidetes, Actinobacteria, and Cyanobacteria occurred in more than half of the pools each and had an average regional abundance of 41% (abundance range in individual pools: 17–75%), 37% (4–80%), 15% (0–34%), and 3% (0–15%), respectively. Among Proteobacteria, Alpha-, Beta-, and Gammaproteobacteria were present in more than half of the pools with 17% (0–70%), 23% (0–71%), and 3% (0–20%) average regional abundance, respectively. Regarding habitat specialization, more than half of the OTUs (318) were singletons (B = 1), while 136 had a B values large than 1, but smaller than 3. Sixty-three OTUs were found in the B 3–6; 19 in the B 6–10; and 9 in the B > 10 category.
Assembly mechanisms in dependence of taxonomic affiliation and habitat specialization
Conductivity and HNF abundance were the factors that were significant in explaining variation in community composition when all OTUs in the 50 rock pools were considered. Similarly, when the different habitat specialization categories were analyzed separately, conductivity was always the most important factor, and in several cases, also the abundance of HNF had a significant effect (Table 1). Moreover, when different phyla/classes were analyzed, conductivity was the most important environmental factor for all of them. In addition, flagellate abundance could explain a significant proportion of the variation in composition of Bacteroidetes and Alphaproteobacteria, whereas Daphnia abundance and total phosphorus were significant in case of the Actinobacteria and Cyanobacteria, respectively (Table 1). The strong overall structuring effect of conductivity became also obvious in the CCA, which clearly separated pools according to their conductivity along the first axis (Supporting Information, Fig. S1) and by the significant ANOSIM result (R = 0.6462, P < 0.001). B was significantly correlated with occurrence (ρ = 0.887, P < 0.001) and mean regional abundance (ρ = 0.773, P < 0.001) and, although less strongly, with mean local abundance (ρ = 0.443, P < 0.001; Fig. 2).
Table 1. Summary of the partial RDA analyses for all OTUs, different habitat specialization categories (B > 10, B 6–10, B 3–6, and B < 3) and phylogenetic groups in all rock pools
*P < 0.05; **P < 0.001.
#OTUs: number of OTUs detected, B: average Levins' niche width, occurence: number of rock pools (of a total of 50) occupied by taxa belonging to a certain habitat specialization category or phylum/class (equals the number of pools included in the partial RDA analyses), [E], [S]: fraction of total variation in community composition explained by environmental factors E or spatial factors S. [E|S] and [S|E]: these fractions were calculated by partial RDA in cases when both E and S were significant and identify pure environmental ([E|S]) and pure spatial ([S|E]) effects, respectively. Shared: variation that is shared between environmental and spatial variables. Sign E: significant environmental factors identified by forward selection; Cond: conductivity; Daph: abundance of Daphnia; HNF: abundances of flagellates; Sig. S: significant spatial factors identified by forward selection; Dist: distance from sea; EV: positive eigenvectors derived from a PCNM analysis.
B > 10
B < 3
There were no differences in assembly mechanisms if all OTUs or the different habitat specialization categories were considered: in all cases, there were significant pure environmental ([E|S]) as well as pure spatial ([S|E]) effects. However, strikingly, the total amount of explained variation was much lower for OTUs with a low B compared with those with a larger B for environmental as well as spatial factors (Table 1, Fig. 3).
When freshwater and brackish pools were analyzed separately a slightly different picture arouse. For freshwater pools, neither environmental ([E]) nor spatial factors ([S]) were significant, the only exception being a significant spatial effect in case of the B < 3 category (Supporting Information, Table S1). For brackish pools, the results were similar compared with what was found when all pools were analyzed together; that is, environmental factors ([E]) were significant for all OTUs as well as for all the different B categories. One difference was, however, that spatial factors ([S]) were significant only for two B categories, and pure factors were not significant at all (Table S1).
There were differences in average habitat specialization among bacterial phyla/classes with highest values for B observed for the Actinobacteria and lowest values for the Gammaproteobacteria (Table 1). Significant pure environmental effects ([E|S]) were observed for all except the Gammaproteobacteria, whereas significant pure spatial effects ([S|E]) were observed for Betaproteobacteria and Actinobacteria (Table 1), but not for any of the other groups. There was a significant positive correlation between habitat specialization of phyla/classes and the amount of variation that could be explained by pure environmental factors (r = 0.811, P < 0.001) as well as the total amount of explained variation (r = 0.951, P = 0.0036), whereas pure spatial factors [S|E] were not significant (r = 0.656, P = 0.158; Fig. 4).
The major finding of this study is that much of the variation in bacterial community composition could be explained by environmental factors for taxa or phyla that were habitat generalists, indicating stronger species sorting processes during the assembly of widespread taxa with evenly distributed abundances. Our results contradict our initial hypothesis, namely that habitat generalists are more likely to be assembled by dispersal-related mechanisms because they can tolerate a wide range of environmental conditions. One previous study on zooplankton has indeed shown that habitat generalists are to a greater extent assembled by spatial factors compared with environmental factors (Pandit et al., 2009). Similar to our study, they used Levins' niche width index to classify taxa as either habitat generalists or specialists and showed that both groups contained both species that were rare or common at the local scale, that is, had low or high average abundances in local communities. In our case, habitat specialization was strongly correlated with different estimates of rarity (Rabinowitz, 1981; Rabinowitz et al., 1986; Gaston, 1994), such as regional abundance and occurrence and, albeit weaker, local abundance (Fig. 2). This indicates that most taxa that we defined as habitat generalists were common and abundant, whereas those which we defined as habitat specialists tended to be rare, at both the local and regional scale, that is, metacommunity scale. Hence, our results are conform with observations found in zooplankton and phytoplankton communities that showed no differences in assembly mechanisms between taxa, which were common or rare at the regional scale (Heino & Soininen, 2010; Siqueira et al., 2012). However, the overall amount of variation that could be explained in general tended to be higher for habitat generalists compared with habitat specialists (Fig. 3), suggesting that spatial turnover or beta-diversity can be sufficiently described, and more importantly explained, by the most common taxa at a given point in time. Furthermore, it indicates that biogeographical patterns in rare taxa are more difficult to detect and predict.
It has been suggested for microbial communities that taxa that have a local relative abundance > 0.1% are called to be common (Reid & Buckley, 2011). Because our study only considered OTUs that occurred at a relative local abundance > 0.31%, it has to be stressed that the OTUs that we defined as habitat specialists were not locally rare per se, but only in relation to more common OTUs. Moreover, regionally rare species are often considered as a source of noise in multivariate analyses (Clarke & Green, 1988) and can lead to low degrees of explained variation. We tried to take this statistical problem into account by removing all species that occurred only at one location of the metacommunity, as well as by adjusting the information content of the matrices prepared for the different habitat specialization categories. Thus, it might also be possible that less of the variation could be explained for habitat specialists because they are stochastically assembled. However, it could also be possible that habitat specialists were to a greater extent affected by environmental variables that we did not measure, such as the density of a specific predator, symbiotic species, or concentrations of micronutrients, whereas habitat generalists respond to the major and strongest environmental variables (Lennon et al., 2011), such as conductivity in our case. Thus, our results do not imply that there are no differences among taxa in dependence of habitat specialization with regard to their environmental requirements, but simply show that there is no evidence that the overall mechanisms underlying their spatial distribution patterns are different and that the spatial turnover among bacterial communities is mainly driven by common species in a similar fashion than for other planktonic communities (Heino & Soininen, 2010). Thus, there is accumulating evidence that large-scale patterns in both alpha- and beta-diversity are in general mainly driven by common taxa that respond to environmental variation (e.g. Jetz & Rahbek, 2002; Lennon et al., 2004, 2011; Heino & Soininen, 2010).
In general, our study highlights that little additional information is going to be gained by in-depth sequencing of local communities aiming at including also the rarest taxa in snapshot studies. Instead, future studies on microbial biogeography need to focus more explicitly on the temporal metacommunity dynamics as well as on different dispersal pathways between local communities to obtain a complete picture of the assembly of bacterial metacommunities. Such an integration of temporal and spatial dynamics is also needed to fully understand the ecological role of rare taxa. It has been suggested that rare taxa serve as ‘seed banks’ from which taxa can be recruited if environmental conditions change and that then become dominant (Fenchel et al., 1997; Finlay, 2002; Pedrós-Alió, 2006). For example, it has been shown that rare taxa in upstream habitats have the potential to become dominant members in downstream environments, presumably due to species sorting (Crump et al., 2012).
Differences in assembly mechanisms were observed among phyla/classes, namely the Bacteroidetes, Alphaproteobacteria, Betaproteobacteria, Gammaproteobacteria, Actino-bacteria, and Cyanobacteria, which were the dominating groups in the rock pools (Table 1). For all phyla/classes, except the Gammaproteobacteria, environmental factors were important in structuring bacterial communities among pools, indicating that species sorting was important. Pure spatial factors were only significant in cases of the Betaproteobacteria and Actinobacteria, indicating that dispersal-related assembly mechanisms, such as mass effects or dispersal limitation (Cottenie, 2005; Ng et al., 2009), could have played a role in the assembly of those groups. The interpretation of spatial factors is, however, ambivalent, because they are based on indirect measurements related to the spatial distance among pools, which probably does not appropriately reflect the connectivity and thus dispersal potential among pools. For this, we would have to take direct indicators of connectivity into consideration, such as rivulets connecting pools in case of rain. Moreover, spatial effects might mask unmeasured environmental factors that are spatially autocorrelated or historical effects, such as past dispersal events or past environmental conditions, and care needs therefore to be taken in their interpretation (Lindström and Langenheder 2012). It is, however, interesting to note that spatial factors were significant for the two phyla/classes with the widest niche width (Table 1), which is in congruence with findings that show that dispersal-related assembly processes should be stronger for habitat generalists compared with specialist (Pandit et al., 2009). Interestingly, our results are remarkably similar to those observed in the only previous study that has looked at assembly mechanisms for different phyla in lakes (Barberan & Casamayor, 2010), even though that study was carried out at a much larger scale and with a much more restricted selection of environmental variables. The only difference was that pure spatial effects were significant in case of the Gammaproteobacteria in the study of Barberan & Casamayor (2010), whereas we observed neither significant environmental nor spatial effects. This difference may arise from the distinct number of environmental factors considered. Nevertheless, the deviating role of the Gammaproteobacteria in both studies is interesting and might be related to the notion that this class is primarily composed of opportunistic taxa with a feast-and-famine life style that become abundant sporadically in nutrient-rich environments or microniches, but are otherwise relatively rare (Pernthaler & Amann, 2005).
We further tried to link differences in assembly mechanisms and/or the strength of different mechanisms to the average niche width of the phyla/class (Fig. 4). In congruence with the results from the analyses based on all OTUs, we found also here that environmental factors could explain much of the variation in community composition for phyla/classes that contained on average more habitat generalists compared with those which contained more habitat specialists. On the contrary, a similar but weaker and insignificant relationship was observed between habitat specialization and spatial factors. Hence, species sorting processes were stronger for groups, such as the Actinobacteria and Betaproteobacteria that contained a high proportion of habitat generalists, compared with groups such as the Alphaproteobacteria or Bacteroidetes that consisted of a higher proportion of habitat specialists. More studies are, however, needed to investigate whether this pattern can also be observed in other environments that are dominated by different phyla as the ones we studied here. A recently developed framework suggests that it is possible to use the complexity of a trait to predict at which phylogenetic depth it is conserved (Martiny et al., 2013). Accordingly, complex traits encoded by many genes are conserved at a relatively deep phylogenetic level, whereas phylogenetic dispersion is typically found for simple traits (Martiny et al., 2013). Thus, our finding here may suggest that habitat specialization is a complex trait that is conserved at the phyla or class levels, similar to studies that have found that adaptation to soil moisture is highly conserved at the phyla level (Lennon et al., 2012) and that different phyla can be classified as either copiotrophs or oligotrophs (Fierer et al., 2007).
Conductivity was the most important factor structuring metacommunity composition, which confirms the important role of osmolytic conditions (salinity, pH) in structuring bacterial communities (e.g. Fierer & Jackson, 2006; Lozupone & Knight, 2007; Tamames et al., 2010). There were, however, some striking differences when freshwater and brackish pools were analyzed separately, because variation in community composition in brackish pools continued to be determined primarily by environmental factors (mainly conductivity), whereas neither environmental nor pure spatial factors could explain differences in BCC among freshwater pools, no matter if all OTUs or different niche width categories were analyzed (Table S1). The latter finding is quite remarkable given that there were other strong environmental gradients, for example related to differences in the presence and abundance of grazers (Daphnia, HNF), phosphorus, chlorophyll a as well as DOC concentrations among the freshwater pools. Hence, our results might indicate that the rock pool metacommunity studied here is structured primarily by conductivity, which enforces a strong ‘environmental barrier’ onto taxa in the sense that there seem to be only very few ‘conductivity generalists’, that is, taxa that can tolerate a wide range of conductivities, but that other factors are of relatively minor importance. This might indicate that rock pools, which are systems that are spatially as well as temporally very heterogeneous, harbor taxa that are generalists with regard to other environmental factors than conductivity.
In summary, our results show that species sorting is the most important assembly mechanisms independent of whether taxa or phyla are habitat generalists that are common in the metacommunity or habitat specialists that are rare. However, we found that species sorting is strongest for habitat generalists, indicating that it is sufficient for studies that aim to describe and identify the mechanisms underlying beta-diversity of bacterial communities to focus on common taxa.
We like to thank Mercè Berga and Hannes Peter for their help during field sampling. This study was funded by the Swedish Research Council Formas, a Marie Curie Reintegration Grant, the Carl Tryggers foundation, and the Center of Metagenomic Sequence Analysis (CMS).