Introduction

Thousands of small forest ponds and lakes shape the boreal landscape, contributing significantly to regional and global nutrient cycling, redistribution of carbon (Battin et al., 2009; Tranvik et al., 2009) and greenhouse gas emissions (Bastviken et al., 2011). While recent work has advanced our understanding of the composition and ecology of epilimnetic microbial communities in lakes (reviewed in Newton et al., 2011), the microbial communities of suboxic hypolimnia have been much less studied, particularly in forest lakes.

Forest lakes differ from the well-studied clearwater counterparts in several key characteristics. High concentrations of humic substances give these lakes their characteristic colour, which limits light penetration, and hence also restricts phytoplankton photosynthesis to the uppermost water column. At the same time, this chromophoric organic matter represents an energy subsidy for heterotrophic bacteria (for example, Cole et al., 1994; del Giorgio and Peters, 1994; Kankaala et al., 1996; Jones et al., 2001; Ask et al., 2009). Most boreal lakes are stratified with an oxic epilimnion and suboxic or even anoxic hypolimnion, separated by a transition layer (metalimnion). The stratification pattern may be disturbed by spring and autumn overturns, but many small boreal lakes are meromictic, with overturn only partial or totally lacking (Salonen et al., 1984; Gudasz et al., 2010). Hypolimnetic microbial communities in these lakes may therefore evolve in the absence of oxygen over periods as long as several years. Other characteristic features of boreal lakes are low pH and low inorganic nutrient concentration in the epilimnion while nutrient availability is much higher in the hypolimnion. Furthermore, methane produced in the oxygen limited hypolimnion and sediment may supply energy to micro-organisms, not only in the oxygenated part of the lake, but in the whole water column (Taipale et al., 2011).

From the major difference in the environmental conditions between epilimnion and hypolimnion, including oxygen and nutrient concentrations, it follows that their microbial communities are also different (Shade et al., 2008; Barberan and Casamayor, 2011). Even though there are several studies addressing bacterial communities in humic lakes (for example, Lindström, 1998; Burkert et al., 2003; Kent et al., 2004), the microbial diversity and community composition of the suboxic hypolimnetic layers remain poorly explored and the few studies available are typically restricted to very specific questions, sediment communities, or are based on methods that provide little if any information on taxonomy (for example, Kleinsteuber et al., 2008; Lehours et al., 2009; Berdjeb et al., 2011; Taipale et al., 2011) or diversity (Bent et al., 2007). One exception is the recent study by Barberan and Casamayor (2011), which used observations from multiple lakes to demonstrate that microbial communities in the hypolimnia are distinct from those in the epilimnia. However, the lakes included in that study had either high salinity or low content of dissolved organic carbon (DOC) and it is not certain that those results can be extrapolated to the numerous lakes of the boreal zone. So far, only one study has used clone libraries and sequencing of taxonomic marker genes to describe microbial communities in the hypolimnion of boreal lakes (Taipale et al., 2009). This study revealed the quantitative significance of members of the autotrophic green sulphur bacterial phylum Chlorobi, but was limited to a few samples and only probed the most abundant taxa as a result of limited sequencing depth.

En masse sequencing of 16S rRNA gene amplicons by 454 pyrosequencing now offers a tool for comprehensive studies of the members of bacterioplankton communities with the possibility of tracking their distribution patterns in time and space. When sequencing is coupled with other methods like analysis of stable isotopic composition of dissolved CH4 (δ13CH4) and statistical approaches such as local similarity analysis (LSA; Ruan et al., 2006), valuable information concerning the ecology of individual populations present in the community may be gained. This also enables the formulation of hypotheses on metabolic capabilities and the use of specific process-related methods that require prior information on the bacterial community before they can be applied.

Here, we present an extensive analysis of bacterioplankton dynamics in humic Lake Alinen Mustajärvi. Temporal dynamics of the microbial communities were visualized in association networks and subnetworks illustrating the linkages among bacterial taxa and between specific groups and environmental driver variables, while stable isotope data were revealing microbial processes, such as methane oxidation in the suboxic hypolimnion. Epilimnetic and hypolimnetic microbial communities of the lake were compared over 5 consecutive years. For comparison, data from four additional lakes that reflect the full range of DOC concentrations typical for the boreal zone were also included in the study. A special focus was put on temporal and spatial changes and on poorly known but abundant microbial groups such as candidate division OD1.

Materials and methods

Study site and sampling

Time-series samples were collected from Alinen Mustajärvi in southern Finland (61°12′N, 25°06′E), a 0.007 km2 head-water lake with maximum depth 6.5 m and an estimated volume of 31 × 103 m3. The catchment area is <0.5 km2 and consists of >90% coniferous forest and <10% peatland. The lake has icecover from late November until late April and it stratifies very steeply during summer. During the sampling period (from May 2006 until August 2010), the lake was only experiencing partial autumn overturns and no spring overturns. During stratifications, only the upper 1–2 m of the water column (the epilimnion) remained oxic. The natural DOC concentration in the epilimnion of the lake is around 10 mg C l−1, but in 2008 and 2009 it was experimentally elevated to 12 mg C l−1 by monthly additions of cane sugar. Effect of these additions is not in focus here and it is assumed to be equal for all layers. Water samples were taken with a 30-cm-long acrylic tube sampler (Limnos 2 L) and except for two samplings from under ice, samples were from open water season; thus, our results should not be extrapolated to winter season. Samplings were conducted twice in 2006, every other week in 2007–2008, once a month in 2009 and four times in 2010 making a total 44 time points. Samples for epilimnion, metalimnion and hypolimnion, respectively, were taken according to the oxygen concentration (Supplementary Figure 1). Epilimnion samples in 2006–2008 were pooled from 0, 1 and 2 m depths and in 2009–2010 from 0 and 1 m depths. Metalimnion samples in 2006–2008 were pooled from 3 and 4 m depths, and in 2009–2010 from 2 and 3 m depths. Hypolimnion samples in 2006 were from 5 m depth, in 2007–2008 pooled from 5 and 6 m depths and in 2009–2010 pooled from 4, 5 and 6 m depths. Water was passed through a 50-μm sieve to remove larger zooplankton. Pooled water samples were transported to the laboratory and kept at +4 °C until aliquots were processed by tangential flow filtration (Durapore cassette, pore size 0.22 μm; Millipore, Billerica, MA, USA) within 6 h of sampling. Concentrated particles retained in this ultrafiltration were frozen and freeze dried with an Alpha 1-4 LD plus (Christ, Osterode, Germany).

Analyses of pH, total and inorganic P and N, alkalinity, as well as chlorophyll-a (chla) were done using standard methods (http://www.sfs.fi/). Particulate organic carbon, DOC and methane concentration were measured as described in Kankaala et al. (2010), bacterial production as in Tulonen (1993) and δ13C-CH4 as in Kankaala et al. (2007). Samples for methane concentration and δ13CH4 measurement were taken every 1 m in 2008–2010. Methane concentration in the water column may diminish due to methane oxidation by methanotrophic bacteria, or because of diffusion to the atmosphere. Methane oxidation, like other microbial processes, prefers light isotope (δ12C), which leads to enrichment of the residual methane with heavier isotope (δ13C) (Whiticar, 1999). By determining the fraction of δ13C in methane, we could evaluate which of these processes contributed to the decrease in methane concentration in Alinen Mustajärvi and in which depths methane oxidation was occurring. Zooplankton samples were preserved with formaldehyde and the species composition and abundance of zooplankton were determined using an inverted microscope at × 100 magnification. The other four humic lakes were sampled twice in 2006 (May and October) from epilimnion, metalimnion and hypolimnion according to the oxygen concentration. The sampling and environmental characteristics of lakes Nimetön, Mekkojärvi, Valkea Kotinen, and Valkea Mustajärvi have been described in Kankaala et al. (2010). Summary of environmental data is given in Supplementary Table 1.

DNA extraction, PCR amplification, pyrosequencing and sequence quality control

The DNA extraction procedure was modified from protocol described by Griffiths et al. (2000). Briefly, 0.4 mg of freeze-dried material was homogenized with glass beads in a mixture of phenol-chloroform-isoamylalcohol (25:24:1) and hexadecyltrimethyl-ammonium bromide. After 5 min incubation on ice to allow humic acids dissolve into phenol-chloroform-isoamylalcohol, tubes were centrifuged. The upper aqueous phase was then re-extracted with chloroform-isoamylalcohol (24:1), precipitated with polyethylene glycol and dissolved in 50 μl of TE buffer (10 mM Tris (pH 8.0), 1 mM EDTA). DNA concentrations were measured with PicoGreen in a Qubit fluorometer (Invitrogen Corporation, Carlsbad, CA, USA) showing no difference between layers (χ2=0.0667, P=0.9672). Amplification of bacterial 16S rRNA genes (E. coli positions 341–805) was conducted using general bacteria primers 341F (5′-CCTACGGGNGGCWGCAG-3′) and 805R (5′-GACTACHVGGGTATCTAATCC-3′) (Herlemann et al., 2011). Primer 341F carried a 454FLX adaptor B at the 5′ end and primer 805R carried a 5-bp molecular barcode specific for each sample followed by a 454FLX adaptor A at the 5′ end. PCR and amplicon processing before sequencing was performed as described in Eiler et al. (2012), except for purification of PCR products with Agencourt AMPure XP purification system (Beckman Coulter, Danvers, MA, USA) and amplicon quantification with PicoGreen in a Qubit fluorometer. After PCR, amplicon concentrations were similar for all layers (χ2=4.8343, P=0.3047). Equal concentrations of amplicons from each sample were sequenced with Titanium chemistry from adaptor A, using a 454 GS-FLX system (454 Life Sciences, Branford, CT, USA) at the Institute of Biotechnology hosted by the University of Helsinki, Finland. The resulting reads carried the sample-specific molecular barcode and covered the entire V4 region of the 16S rRNA gene as well as flanking regions. The sequence runs also included samples that were not part of this study. A total of 380 512 reads were assigned to samples from Alinen Mustajärvi (called hereafter as time-series data) and 77 146 were assigned to samples from the other four lakes (Nimetön, Mekkojärvi, Valkea Kotinen and Valkea Mustajärvi, called hereafter as four lakes data). Ambiguous sequences were removed from the data set including reads with low quality as inferred from their flowcharts and those that did not carry the exact primer sequence. After implementation of these quality control criteria, remaining reads were denoised using AmpliconNoise Version 1.22 (Quince et al., 2011). AmpliconNoise implements algorithms that remove PCR and 454 pyrosequencing noise as well as the novel chimera removal tool Perseus. A total of 6201 reads was removed by Perseus using default settings (α=−7.5, β=0.5) for false positive detection. The 454 sequences have been deposited in the NCBI Short Read Archive under accession number SRP007933.

Operational taxonomic unit assignments

To assign reads into operational taxonomic units (OTUs), average linkage clustering was applied with a 97% sequence similarity cutoff. A representative sequence from each of the resulting 11 496 OTUs was classified using the naive Bayesian classifier (Wang et al., 2007) implemented in MOTHUR (Schloss et al., 2009) in combination with the Silva 102 database and the rdp6 taxonomic framework. In addition, the reads were annotated against a local freshwater bacterial sequence database that included almost 12 000 sequence entries (Newton et al., 2011). This classification system was designed to maintain the phylogenetic context by which epilimnetic freshwater bacterial gene sequences historically have been identified, clustered and named. Sequences that could not be annotated to at least the phylum level with the naive Bayesian classifier were aligned in MOTHUR using kmer for finding the template sequence and Needleman for aligning sequences against the SILVA102 database (Schloss et al., 2009). Aligned sequences were imported into ARB (Ludwig et al., 2004) and the quick parsimony option was used to add the aligned sequences to the small subunit reference tree included in the SILVA106 small subunit rRNA gene database. The resulting position in tree was used to annotate sequences to the phylum level.

Statistical analyses

To compare and perform statistics across samples, we used the perl script daisychopper.pl (available at http://www.genomics.ceh.ac.uk/GeneSwytch/Tools.html; Gilbert et al., 2009) to randomly resample all samples to the same size based on the sample with smallest sampling size. To include all samples and increase the number of reads after resampling, the number of reads in two samples with <1000 reads (E080826 and M070814 in the time-series data set) were multiplied by two after which the minimum read number was 1139. Simulation of the impact of this doubling was conducted (data not shown) in which all samples were resampled to 570 reads and then doubled; this had no significant affect on the Bray–Curtis indexes compared with the original data resampled to 1139 reads.

All statistical analyses were conducted using R (R Development Core Team, 2011; http://www.R-project.org/) and the vegan package. Non-metric multidimensional scaling plots of a Bray–Curtis distance matrix were used to visualize dynamics in community structure (β diversity) using an OTU abundance matrix based on 757 OTUs, each represented by at least 20 reads in the total data set. Kruskal–Wallis test, Spearman's Rank correlation, paired Wilcoxon's test, and permutational MANOVA were used to determine relationships between and within community components and physical and spatial variables. For Spearman's Rank correlation tests between DOC and taxonomic groups, only samples from 2006 from Alinen Mustajärvi and the four other lakes were included. The inverse Simpson index was used for diversity evaluation.

Using the resampled data matrix from the time-series data set and cutoff value P<0.05, time-dependent paired occurrences of each OTU with >20 reads in epilimnion and hypolimnion were analysed with Wilcoxon's signed-rank tests (Wilcoxon, 1945). False discovery rate was used to measure the proportion of false positives in multiple testing (Storey et al., 2004). This non-parametric test allowed identification of OTUs characteristic for the epilimnion or hypolimnion. To compare the dynamic properties of the communities among layers, permutational analysis of multivariate dispersions (also called MJ Anderson's permutated analysis of beta-dispersion) was applied on Bray–Curtis based dissimilarity matrices (Anderson et al., 2004). This analysis tests for homogeneity of multivariate dispersion among groups of samples, hence reveals if communities from different groups differ in their taxa variability. To summarize seasonal patterns among OD1-related OTUs, all OTUs were divided into 10 clusters using k-means clustering with the Hartigan–Wong algorithm (Hartigan and Wong, 1979).

Network analysis was conducted with 138 OTUs (minimum 50 reads), 33 zooplankton taxa and 15 environmental state variables which were normalized using normal score transformation before analysis. Significant correlations were identified based on a P-value of <0.01 and a false discovery rate of <0.05, coefficient greater than 0.3 or smaller than −0.3 among OTUs and between OTUs and environmental variables (biological and physicochemical data) using LSA as implemented in R. Other parameters for the LSA were set to 1000 permutations and a maximum delay of 0 as sampling points were not evenly spaced. The resulting LSA matrix was translated into an association network using Cytoscape 2.6.3 (Shannon et al., 2003). Cytoscape depicts data sets as nodes (environmental variables and OTUs) connected by edges (significant correlations) that denote the character (positive or negative correlation) of the relationship. Network analyses have been used in previous studies on both marine and freshwater bacterioplankton communities (see, for example, Fuhrman and Steele, 2008; Fuhrman, 2009; Shade et al., 2010; Steele et al., 2011; Eiler et al., 2012).

Results

Community composition in different layers and among lakes

Sequencing of 122 time-series samples from Alinen Mustajärvi yielded a total of 287 123 high quality reads (range 1139–4945 sequences per sample). For the 24 samples from the four lakes data set, 49 225 high quality reads were obtained (range 1680–2356 sequences per sample).

After resampling both data sets by daisychopper, 1139 sequences remained for each sample. Even though this diminished the read number per sample to only approximately one order of magnitude higher than in typical clone libraries, in total there were still 166 000 reads kept in the analysis, which is two to three orders of magnitude higher than collected by most clone library surveys. In the resampled time-series data set, these sequences clustered into 5055 OTUs (range 74–392 OTUs per sample). The number of OTUs detected in the hypolimnion was higher than in the epilimnion (paired Wilcoxon's test; P<0.001, inverse Simpson's index was on average 18.1, 26.6 and 38.0 for epilimnion, metalimnion and hypolimnion, respectively). In the four lakes data set, resampling left a total of 3564 OTUs (range 66–484 per sample). The highest number of OTUs was found in Lake Nimetön and the lowest in Lake Valkea Mustajärvi. These lakes also had the highest and lowest DOC concentration, respectively. Correlation analysis revealed a positive relationship between DOC concentration and the number of OTUs found for each sample (Spearman's rho=0.39; P<0.05). More specific OTU distributions can be found in Supplementary Table 2.

As expected, the microbial community composition changed gradually from epilimnion to metalimnion and hypolimnion as visualized in the non-metric multidimensional scaling plot (Figure 1). Disparity between layers was especially clear in the time-series data set (Lake Alinen Mustajärvi), but was also obvious in the four other lakes. Permutational MANOVA confirmed the presence of distinct communities in each layer (pseudo-F=13.4, P<0.001) and that communities differed between lakes (pseudo-F=5.1, P<0.001) as well as along season (pseudo-F=10.7, P<0.001).

Figure 1
figure 1

Non-metric multidimensional scaling ordination visualizing the Bray–Curtis distances of all 122 samples from the time-series data set from Lake Alinen Mustajärvi and 24 samples from the four lakes data set. The five lakes are represented by different symbols (▪=Nimetön, ▵=Mekkojärvi, +=Valkea Kotinen, •=Alinen Mustajärvi, =Valkea Mustajärvi) and the three separate depth layers are indicated by grey scale. The stress value is 17.3.

Full size image

Taxonomic assignments and distinction between epilimnion- and hypolimnion-associated OTUs

The main focus in comparing the communities in different layers was put on differences between epilimnion and hypolimnion as they present highly different environments. After classification with MOTHUR, a considerable proportion of reads from both data sets remained unclassified (Figure 2a). The proportion of unclassified reads was much higher in the hypolimnia than in the epilimnia (paired Wilcoxon's test, P<0.001). With ARB, half of the unclassified reads were annotated to candidate divisions, most of them belonging to candidate division OD1, which constituted about 70% and 40% of unclassified reads in time-series and four lakes data sets, respectively (Figure 2b). In the time-series data set, the most abundant unclassified OTU (OTU 1237) after classification in MOTHUR (including 38.4% of all unclassified reads and 4% of all reads) was affiliated with OD1 using the positions of the sequences in the ARB reference tree. All other abundant OD1 OTUs, such as OTU1357, OTU1422, OTU1526 and OTU1498, could also be annotated as OD1 using their position in the phylogenetic tree after parsimony insertion.

Figure 2
figure 2

The mean proportion of unclassified reads using the naive Bayesian classifier in MOTHUR (a). The mean proportion of reads annotated with candidate divisions after classification with ARB; the proportion of OD1 reads is indicated in white (b). The proportion of reads annotated with typical freshwater tribes as described in Newton et al. (2011) (c). All panels present proportions from all sequences.

Full size image

OTUs were also annotated using a freshwater-specific sequence database to look for the typical freshwater community members in the five lakes (Figure 2c). In the time-series data set, the most abundant freshwater tribes in the epilimnion were acI-B2 (16.3% of all epilimnetic reads), acV-A2 (8.5%) and PnecC (6.9%) and for hypolimnion PnecC (3.0%), acI-B2 (2.3%) and acV-A1 (1.7%). Among lakes, DOC concentration (Figure 3a) was found to have a positive correlation with the proportion of reads affiliated with candidate divisions (Spearman's rho=0.40, P<0.05) and OD1 (Spearman's rho=0.37, P<0.05) and a negative correlation with freshwater tribes (taxa) (Spearman's rho=−0.64, P<0.001). This indicates that generally high DOC lakes are neglected in studies of freshwater microbial communities. OD1 was also positively correlated with methane concentration (Figure 3b; Spearman's rho=0.61, P<0.001) among lakes as well as in time-series data set (Spearman's rho=0.43, P<0.01).

Figure 3
figure 3

Mean DOC concentration (mg l−1) (a) and CH4 concentration (μM) (b) in epilimnion and hypolimnion of Lakes Nimetön, Mekkojärvi, Valkea Kotinen, Alinen Mustajärvi and Valkea Mustajärvi.

Full size image

With combined MOTHUR and ARB classification, the main phyla in the epilimnion of time-series data set were Actinobacteria and Betaproteobacteria, while the main phyla in the hypolimnion were candidate division OD1, Chlorobi and Bacteroidetes (Figure 4a). Even though many of the taxa found from metalimnion were shared with the other layers, the community still had distinct features. For example, Cyanobacteria were mainly found in this layer. OTUs typical for either epilimnion or hypolimnion were identified with Wilcoxon's signed-rank test, using all OTUs with >20 reads. OTUs that had significantly more reads in epilimnion or hypolimnion were defined as epilimnion-associated or hypolimnion-associated. OTUs with different occurrences between epilimnion and hypolimnion contained 43.1% of reads included into the analysis. In the epilimnion by far the most abundant phylum was Actinobacteria, which comprised more than half of all epilimnion-associated reads (Figure 4b). Reads affiliated to Betaproteobacteria were also much more abundant in the epilimnion, constituting almost one quarter of all epilimnion-associated reads. In all, there were only representatives from seven phyla among the epilimnion-associated OTUs. Among the OTUs that were more abundant in the hypolimnion, there were representatives from 20 different phyla (Figure 4c). A third of all hypolimnion-associated reads were affiliated with candidate division OD1, while Chlorobi and Bacteroidetes were also well represented. The only phylum with equal proportion of epilimnion- and hypolimnion-associated reads was Verrucomicrobia.

Figure 4
figure 4

Ternary plot showing the distribution of OTUs in the time-series data from Lake Alinen Mustajärvi (a). Axes represent the three depth layers of the lake (epilimnion, metalimnion and hypolimnion) and the percentage of reads associated with each layer for each OTU. The size of the symbol indicates number of reads associated with each OTU and taxonomic affiliation are indicated by colours. Pie charts indicating the proportion of reads affiliated with OTUs associated with epilimnion (b) and hypolimnion (c) as determined by Wilcoxon's signed-rank test using the time-series data set (false discovery rate q=0.021). Colours represent different phyla as in (a). In (a), OTUs represented by at least 50 reads are included, panels (b) and (c) include all OTUs represented by at least 20 reads in the time-series data set.

Full size image

Since the presence of OD1 in lakes is rarely reported in the literature, the fit of the most common 16S rRNA-targeting primers used for bacterial community analyses to all OD1-related sequences in RDP database (release 10, update 27) was tested with RDP Probe Match (http://rdp.cme.msu.edu/probematch/search.jsp) (Table 1). The primers used in our study (341F and 805r) exactly matched a much higher percentage of OD1 sequences (64% and 60%, respectively) compared with other commonly used primers (from 0% to 22%).

Table 1 Comparison of fit of commonly used primers with sequences affiliated with candidate division OD1
Full size table

Temporal dynamics of OTUs in Lake Alinen Mustajärvi

In the time-series data set, the bacterial communities in epilimnion and metalimnion were more dynamic in composition than in the hypolimnion (average Bray–Curtis distances 0.75 (±0.15), 0.74 (±0.15) and 0.67 (±0.15), respectively), which is also apparent in the non-metric multidimensional scaling plot (Figure 1). Reads affiliated with OD1 constituted a considerable and persistent proportion of hypolimnetic reads, the average representation being 24.8% (Figure 5a). Still, each year the OD1 population apparently crashed upon autumn overturn. The most common OD1, OTU1237, contributed almost half of all reads annotated as OD1.

Figure 5
figure 5

Proportion of all OD1-related sequences and of the 10 most abundant OD1-affiliated OTUs (a). The distribution of the distance from the centroid (b) determined from the temporal dynamics of the bacterial communities in the corresponding layers. The distance from the centroid was estimated by MJ Anderson's permutated analysis of beta-dispersion and represents a statistical measure for the variability of the communities over time.

Full size image

The abundant OTUs in the hypolimnion were less dynamic when compared with rarer OTUs, as suggested by a significant negative correlation between their relative abundances and corresponding standardized variances (Spearman's rank correlations R=−0.41, P<0.001). Results of the permutated analysis of beta-dispersion revealed that there was a significant difference in the temporal variability of taxa over time (F=10.468, P<0.001), with the hypolimnion showing a significant higher homogeneity (lower distance to centroid) than the other two layers (Figure 5b).

Association networks of the hypolimnetic bacterioplankton community

Complete results from time-series data analysed by LSA were visualized in association networks (Supplementary Figure 2; Supplementary Table 3). From all possible interactions, 9.2% were significant at significance level P<0.01 and with a false discovery rate q<0.05. OTUs annotated as OD1 were often significantly positively correlated with each other, most likely indicating functional interdependencies or similar environmental preferences. Further, many OD1-related OTUs were among the most connected OTUs, for example, OTU1357, for which 28% of all possible interactions were significant. Chlorobi OTU1239 (second most abundant OTU in the hypolimnion) was negatively related to the OD1 OTU1357 and OTU1422, suggesting competition or differences in growth constraints among these OTUs. From the complete network, two subnetworks were derived to visualize only the OTUs connected to CH4 (Figure 6a) and to NH4+ and combined NO2− and NO3− (Figure 6b). OTU1357, OTU1237 and OTU1462 annotated as OD1 were positively related to CH4 concentrations, as was OTU727, annotated as Verrucomicrobia. OD1 OTUs were also overrepresented in the number of positive correlations with ammonium including six of the most abundant OD1 OTUs. Only two OTUs, one OD1 and one Betaproteobacteria, were significantly correlated with the combined concentration of nitrite and nitrate.

Figure 6
figure 6

Subnetworks organized around methane (a) and inorganic nitrogen (b) compounds. The subnetworks were extracted from the entire association network (see Supplementary Figure 2) and show all OTUs (nodes) correlating with the respective compound. Circles represent bacterioplankton OTUs where colours represent their taxonomic annotations at the phyla level as in Figure 3a. The size of the circles reflects the number of samples where each OTU could be detected. Blue solid lines (edges) indicate positive correlations and red dashed lines (edges) indicate negative correlations, whereas the width of the line reflects the LSA coefficient. Abbreviations for the environmental variables are translated as following: NH4, ammonium concentrations, NO, combined nitrate and nitrite concentrations and CH4, methane concentrations.

Full size image

Nutrient and DOC concentrations in Alinen Mustajärvi increased from epilimnion to hypolimnion while the oxygen concentration decreased (Table 2). The δ13CH4 values and methane concentrations indicated that the methane was oxidized in the hypolimnion (Supplementary Figure 3). Methane concentration in the epilimnion and metalimnion was lower than in hypolimnion (χ2=89.7556, P<0.001) while the δ13CH4 value was higher (χ2=84.7, P<0.001). There was a drastic decrease in methane concentration between bottom and 4 m depth coinciding with a minor enrichment in δ13C-CH4. The change in δ13C did not seem drastic since the more CH4 there is, the smaller the change in isotope value. However, calculation of CH4 oxidation using fractionation factor determined for methane oxidation in polyhumic boreal lakes (1.037; Kankaala et al., 2007), showed that this change between 4 and 6 m corresponded to half up to almost all of the oxidized CH4 in the water column. Thus, even though we could not determine whether the oxidation was aerobic or anaerobic, a major portion of CH4 was oxidized in suboxic hypolimnion. Several OTUs annotated as OD1, such as OTU28, OTU1237 and OTU1357, were highly abundant in these depth horizons and as mentioned above correlated positively with methane concentrations.

Table 2 Mean δ13CH4 values and mean concentrations of O2, DOC, CH4, NH4+, combined NO3− and NO2− and PO4− for each layer in Alinen Mustajärvi
Full size table

Discussion

A key finding of our study was that the representation of tribes previously described as typical for freshwater (Newton et al., 2011) was low in the hypolimnia while the proportion of candidate divisions and other poorly described phyla was very high. At the same time, these hypolimnetic microbial communities were more stable compared with the more dynamic communities in the epilimnion, corroborating the previously findings from other types of lakes (Shade et al., 2008; Barberan and Casamayor, 2011; Berdjeb et al., 2011; Pollet et al., 2011). This may simply reflect more pronounced seasonal shifts in temperature (range 0.9–18 °C in the epilimnion compared with 3.6–7.5 °C in the hypolimnion), light and nutrient concentrations in the epilimnion compared with the more stable deeper strata. Other possible reasons for higher variability in the epilimnetic community over time may be selective feeding by heteronanoflagellates (Fu et al., 2003; Massana et al., 2009) or pulsed inputs of organic compounds from allochthonous production that are more labile in the surface layers, while quenched before reaching the deeper waters via sedimentation. Additional explanations for the low variability in hypolimnetic community may arise from the fact that anaerobic metabolic pathways have a lower energy yield compared with aerobic metabolism and that the mineralization of organic compounds in anaerobic systems is a multistep process that must involve complex communities (Megonigal et al., 2005). Lower energy yield and the need for a complex community may also facilitate the higher diversity in the bacterioplankton communities of the hypolimnion compared with the epilimnion.

There was a clear shift in bacterial community composition from epilimnion to hypolimnion with metalimnion sharing features with both neighbouring layers. Low proportion of typical freshwater tribes in the hypolimnion emphasizes that the microbial communities in this type of waters are generally underexplored. The proportion of freshwater taxa was also decreasing while the proportion of candidate divisions increased with increasing DOC concentration in both the epilimnion and hypolimnion. The epilimnion-associated OTUs were mainly annotated as acI-B2, acV-A2, xip-A1, PnecC and Novosphingobium (Novo-A2), which have previously been described as abundant tribes in the epilimnion of humic lakes (Newton et al., 2011). In the hypolimnion, the most abundant OTU (OTU1237) was annotated as OD1. In previous studies, Chlorobi (Taipale et al., 2009; Barberan and Casamayor, 2011) and Bacteroidetes (Barberan and Casamayor, 2011) have been found to be major groups in the anoxic hypolimnion, but the dominance of OD1-related sequences has been reported only once (Briée et al., 2007) in a suboxic pond, where it comprised 25% of the sequences from a clone library. We suggest that OD1 may well have been overlooked and neglected as a result of insufficient coverage in previous bacterial PCR surveys. Most commonly used primers for such bacterial community surveys do not target this group, whereas the bacterial primers used in our study (341f and 805r) have a much higher coverage to known sequences of candidate division OD1. We found the proportion of OD1 sequences in Lake Mekkojärvi to be 7.0%, whereas Taipale et al. (2009) did not detect them in the same lake using the bacterial 16S rRNA primer set 27f-518r. Similar underestimation by most commonly used bacterial primers was recently observed also for another poorly characterized phylum, Verrucomicrobia (Bergmann et al., 2011).

Other abundant OTUs in the hypolimnion were annotated as Proteobacteria such as the epsilonbacterial genus Sulfurimonas (OTU1271) and the gammaproteobacterial genus Methylobacter (OTU1103). Members of the genus Methylobacter are group I methanotrophs (Bowman et al., 1993) and they were likely contributing to the methane oxidation as indicated by the stable isotope data. In addition, anaerobic denitrification is suggested to be a common process in the studied lakes as indicated by the presence of Sulfurimonas and the oversaturation of N2 in the suboxic hypolimnia (Tiirola et al., 2011). Sulfurimonas isolates represent sulphur-oxidizing chemolithoautotrophs (for example, Inagaki et al., 2003; Takai et al., 2006) that can perform autotrophic denitrification. This chemolithotrophic process couples denitrification with the oxidation of reduced inorganic sulphur compounds and members of the genus Sulfurimonas have been identified as the main denitrifiers in chemoclines of the Baltic Sea (Glaubitz et al., 2010). OTUs identified as Chlorobi (OTU1239 and OTU1268) were also abundant; isolates belonging to the phylum Chlorobi have been described as anaerobic phototrophs that require sulphur and can photoassimilate acetate or pyruvate directly into the amino acids leading to cell growth (Hoare and Gibson, 1964; Hell et al., 2008). OTU727 correlating positively with methane concentration was affiliated with phylum Verrucomicrobia. Recently, Verrucomicrobia have been implicated with methanotrophic metabolism in extremely acidic environment (Dunfield et al., 2007; Pol et al., 2007). These organisms harbour a novel methylotrophic pathway, with PmoA genes distinct from Proteobacteria, which were considered to be the only phylum capable of methanotrophy before the discovery of these new methanotrophs.

Metabolic pathways in candidate division OD1 have not been specified as the members of this candidate phylum have never been cultivated and only a partial genome has been recovered (Elshahed et al., 2005). Nevertheless, even the scarce data available for this group may still give a hint about the ecology and metabolic capabilities of organisms within this division. OD1 has mainly been detected in sulphur-rich, anoxic environments (for example, Elshahed et al., 2005; Briée et al., 2007; Barberan and Casamayor, 2011), and the partial genome revealed genes known only from anaerobic or facultatively anaerobic microorganisms presenting a similar genomic organization as methanotrophic Archaea (Elshahed et al., 2005) found in anoxic marine sediments (Hallam et al., 2004). Network analysis revealed mostly significant positive correlations with methane and ammonium concentrations, and stable isotope data were indicating methane oxidation in the same depth where OD1 OTUs were highly abundant. Furthermore, the distribution of OD1-related reads in the studied lakes suggests preference towards environments with high DOC and CH4 concentrations. Together, these findings suggest that members of the division are very likely anaerobic, and might have a role in anaerobic methane oxidation. In addition, OTUs within the OD1 candidate division were mainly positively correlated with each other, suggesting that these OTUs share many ecological properties with considerable niche overlap. Furthermore, OD1 OTUs represent abundant and persistent populations with low proportional variability over the ice-free season. Together with network theory this suggests that several OD1 OTUs occupy key roles in the environment. OTUs with a high number of associations (such as OTU1357 which had the highest number of significant correlations) can be regarded as a ‘hub’ in the complex network (see Montoya et al., 2006) that anaerobic lake bacterioplankton communities form in the hypolimnion of boreal lakes.

Still, there is clearly a need to verify these hypotheses on metabolic processes and the identity of the involved key players by functional metagenomic and single cell approaches. Such approaches may support the hypotheses postulated from our field survey on the metabolic capabilities of the diverse and abundant candidate division OD1 and other poorly known groups in boreal lakes. In conclusion, this study shows that the previously described freshwater-related tribes are not abundant in suboxic hypolimnia, which reflects the current focus on aerobic freshwater microbiology. Further, the persistent dominance and high network connectivity of candidate division OD1 suggests that these previously overlooked microbes have an important ecological role in these mostly anoxic environments. Overall, our results for bacterioplankton in boreal lakes revealed a higher diversity and more persistent community composition in the hypolimnion compared with epilimnion and metalimnion. Together with high abundance of poorly characterized bacteria this points to distinct ecological properties of these suboxic communities.