Introduction

A major challenge in aquatic microbial ecology is to determine temporal and spatial patterns in community structure of pelagic microorganisms. While the spatio-temporal distribution of bacterioplankton communities in marine systems has been well studied in the past [1–3], little is known about seasonal and multi-annual dynamics in bacterial community composition (BCC) of pelagic freshwater bacteria [4, 5]. The majority of studies published so far only focus on a small number of sampling dates, stations, or depths and, thus, miss long-term seasonal patterns, which are crucial for our systematic understanding of the role of bacteria in freshwater ecosystems. Only a handful of studies have addressed multi-annual and spatial variations in bacterioplankton assemblages to date [1, 3, 5].

Attentive and assiduous data acquisition and conscientious appraisal are the basis for reliable ecosystem models, especially when linking microbial diversity to ecosystem functioning. Therefore, examining relationships between bacteria and other organisms or environmental parameters over multiple seasons are of great importance. For example, by combining network analyses with long-term data on BCC and environmental variables, Fuhrman and Steele [6] identified a repeated co-occurrence of several bacterial phylotypes over time. Other studies assessing seasonal patterns of bacterioplankton communities in lakes revealed that BCC dynamics are directly correlated to seasonal changes in environmental conditions [7, 8]. Thereby, occurrence and abundance of specific bacterial subgroups, e.g. Actinobacteria and Betaproteobacteria [9–13], were shown to be directly impacted by seasonal variability of certain environmental factors. However, all studies were performed within a short and defined time frame and, therefore, could hardly be used to draw general conclusions about multi-annual BCC dynamics, for which long-term data are required.

In this unique study, we examined seasonal patterns of BCC over multi-annual cycles in Lake Tiefwaren, NE Germany, for the first time. A distinctive feature of this lake is that its internal phosphorus load has been reduced substantially in 2001 by lake restoration [14, 15]. To better account for differences in bacterial lifestyle, we distinguished between free-living (FL) and particle-associated (PA) bacteria. We hypothesised that community compositions of FL and PA bacteria vary with seasons and that these variations are a consequence of temporal changes in environmental conditions (e.g. nutrients, phyto- or zooplankton). In the present study, we used denaturing gradient gel electrophoresis (DGGE) to profile bacterial communities and to identify seasonal changes in their structure. Relationships between BCC and environmental variables were tested by using different statistical approaches. Annual reassembly of BCC in winter and summer was studied separately. This comprehensive approach allows for a more reliable generalization of seasonal patterns in BCC and its linkage to specific environmental variables.

Materials and Methods

Study Site and Lake History

Lake Tiefwaren (maximum depth 23 m, lake surface 1.41 km2, total water volume 13.8 × 109 m3) is a dimictic meso-eutrophic hardwater lake situated in the Mecklenburg Lake District ca. 120 km north of Berlin, Germany (geographical position 53°31'N, 12°41'E, WGS84). Due to massive external nutrient loads in the late 1980s, Lake Tiefwaren was eutrophicated with high P-concentrations (TP and SRP). To reduce phosphorus in the water column and to terminate P-release from the sediment into the water column, a new restoration technique was developed, and a pilot-equipment for lake restoration was engineered, which combined the hypolimnetic injection of sodium aluminate and calcium hydroxide. For this treatment, ca. 137-g Al3+ and 154-g Ca2+ per square meter of profundal sediment (below >10 m) were added stepwise into the hyplimnic water from 2001 to 2005 (for more details see [14–16]). P-concentrations in the pelagic water decreased immediately after the treatment in 2001. Due to the formation of a new sediment boundary layer with a high capacity for P-retention, the P-flux from the sediment into the water was completely suppressed. Today, concentrations of TP and SRP in the lake are <0.02 and <0.01 mg L-1, respectively, making Lake Tiefwaren a phosphorus limited freshwater ecosystem.

Sampling Procedure and Measurements of Chemical Variables

Lake Tiefwaren was sampled monthly from 2003 to 2009 by taking a mixed epilimnic water sample (0–5 m depth), but during lake mixis, the upper 0–10 m was sampled. Temperature, oxygen, pH and conductivity were measured (WTW, Weilheim, Germany) to monitor thermal stratification.

Chemical variables were measured by standard chemical procedures [17, 18]: Subsamples for soluble reactive phosphorus (SRP) were determined directly after filtration through 0.6-μm filters, and unfiltered samples for total phosphorus (TP) were incubated with 5 % (w/v) K2S2O8 for 30 min at 134 °C in a steam autoclave prior to analysis. Phosphor concentrations were measured photometrically (FIA compact analyser MLE, Dresden, Germany) using the molybdenum-blue method [19]. Nitrogen fractions were oxidised with Oxisolv® (Merck, Darmstadt, Germany) at 120 °C for 45 min in an autoclave and measured with a Foss FIAstar 5010 analyser (Rellingen, Germany). Chl a concentrations were quantified photometrically after filtration of appropriate aliquots (200-mL) through membrane filters (pore size 1.2-μm, cellulose mixed ester). Extraction was performed by adding 99.8 % acetone; absorbances of the processed samples were recorded at three different wavelengths (630, 665 and 750 nm) for calculating Chl a concentrations.

Measurements of Biological Variables

Phytoplankton composition and biomass were determined based on morphology in Lugol-fixed samples giving a counting accuracy of ±10 % for total phytoplankton. Phytoplankton biomass (mg fresh weight L-1) was estimated by geometrical approximations using the computerised plankton counter Opticount [20]. Crustacean plankton was collected by vertical hauls with a zooplankton net (mesh size 90-μm). Zooplankton and invertebrates were preserved in a 4 % formalin–sugar solution [21], and subsamples containing the dominant species were counted microscopically.

Primary production (PP) and bacterial protein production (BPP) were determined by H14CO –3 uptake and [14C]-leucine incorporation in triplicates by in situ incubation for 4 h (PP) and 1 h (BPP) according to Simon and Azam [22] and Kirchman [23]. Separation of particle-associated (PA) and free-living bacteria (FL) was done by sequential filtration through 5.0- and 0.2-μm cellulose-nitrate membranes, respectively [9].

Cell numbers of the total bacterial community were counted for all samples from 2005 to 2009 using an epifluorescence microscope (at × 1,000 magnification) after filtration of 2-mL of lake water onto black membrane filters (0.2-μm pore size, Nuclepore polycarbonate membrane, Whatman, Maidstone, UK). No pre-filtration step was applied for solely counting bacterial cells attached to particles, and due to methodological reasons, counts from 2003 and 2004 were omitted. Bacterial cells were stained with 4',6'-diamidino-2-phenylindole (DAPI, [24]). Proportions of different microbial taxa were determined by fluorescence in situ hybridization (FISH) with horseradish peroxidase-labelled oligonucleotide probes (catalysed reporter deposition, CARD-FISH) according to the protocol of Pernthaler et al. [25] and Sekar et al. [26]. We used the EUB338 oligonucleotide probe mix (EUBI–III) for the detection of Bacteria [27, 28] and probes that are specific for Actinobacteria (HGC69 [29]) and the actinobacteria freshwater cluster acI (AcI-852 [30, 31]).

DNA Extraction

PA and FL bacteria were separated by sequential filtration of 200-mL lake water through filters with 5.0- and 0.2-μm pore size (47 mm diameter, Nuclepore Track-Etch polycarbonate membrane, Whatman, Maidstone, UK), respectively. Filters were stored at −20 °C until DNA extraction.

Due to protocol optimizations over time, two different but comparable DNA extraction methods were used for filters from 2003 to 2007 and from 2008 to 2009. Both extraction procedures yielded quantitatively and qualitatively similar results.

DNA extraction of samples from 2003 to 2007 was done using a protocol with sodium-phosphate-buffer (120-mM, pH 8), sodium dodecyl sulphate (SDS) and zirconium beads. Filters were mixed with zirconium beads (0.2-g) and the buffer solution for 10 to 15 min on a horizontal Vortex mixer at maximum speed. After centrifugation (17,000 g, 10 min), the supernatant was incubated with additional sodium-phosphate-buffer and lysozyme (10-mg/mL in TE-buffer) for 30 min at 37 °C to digest bacterial cell walls and the peptidogylcan layer. SDS and proteinase K were added and incubated overnight at 55 °C. After addition of 7.5-M ammonium-acetate, DNA was precipitated with 0.7 volumes of ice cold EtOH at −20 °C for several hours. DNA was washed two times with 70 % EtOH and resuspended in 20–50-μL nuclease-free water.

DNA of samples from 2008 to 2009 was extracted using a standard phenol-chloroform method. Zirconium beads, TENP-buffer (containing 50-mM Tris, 20-mM EDTA, 100-mM NaCl, 20-mg/mL polyvinylpyrrolidone) and 500-μL phenol-chloroform-isoamylalcohol (25:24:1) were added to the filters and mixed for 15 min on a horizontal Vortex mixer. Afterwards, the samples were heated to 60 °C for 10 min, cooled down on ice and centrifuged at 4 °C (17,000 g). Remaining proteins were removed by repeating the phenol-chloroform extraction step. After addition of sodium acetate (3-M, pH 5.2) and 1-mL EtOH, DNA was precipitated at −20 °C overnight. DNA was washed two times with EtOH and resuspended in 20–50-μL nuclease-free water. The DNA content was measured with a UV/Vis spectrophotometer (NanoPhotometer™, Implen). DNA was stored at −20 °C until further analysis.

PCR Amplification of 16S rRNA Gene Fragments

16S rDNA fragments of ~560 bp were amplified for DGGE using the universal primer pair 341f and 907r (5'–CCT ACG GGA GGC AGC AG–3' and 5'–CCG TCA ATT CMT TTG AGT TT–3' [32]) and the Actinobacteria primer pair HGC236f and HGC664r (5'–GCG GCC TAT CAG CTT GTT–3' and 5'–AGG AAT TCC AGT CTC CCC–3' [30]). An additional GC rich nucleotide sequence of 40 bp (GC clamp) was attached to the 5'-end of the forward primers to stabilise migration of DNA fragments in the DGGE gel. PCR was performed as previously described [33], with minor modifications (1 % dimetylsulfoxid (DMSO) and Crimson Taq™ DNA Polymerase, New England BioLabs Inc.). DNA was quantified with a UV/Vis spectrophotometer (see above) and diluted to obtain equal amounts of community DNA in each PCR reaction. For each reaction, 1–3-μL DNA was used as template. PCR started with an initial denaturation step of 3 min at 95 °C, followed by 32 cycles of denaturation (95 °C, 30 s), annealing (55 °C, 1 min) and elongation (68 °C, 2 min). A final elongation of 15 min at 68 °C completed the reaction.

DGGE Analysis of PCR Products

PCR products were analysed by using denaturing gradient gel electrophoresis (DGGE, [32]) with the Ingeny PhorU DGGE-System (Ingeny International BV, GP Goes, Netherlands). DGGE was performed in a 7 % (v/v) polyacrylamide gel with a denaturing gradient of 40 % to 70 % urea and formamide for the universal primers and from 55 % to 65 % for the Actinobacteria primer pair. PCR product (~500 ng) of each sample was loaded onto the DGGE gels. All gels were run in 1× TAE electrophoresis buffer (40-mM Tris–HCl, 20-mM acetic acid, 1-mM EDTA, pH 8.3) for 20 h at a constant voltage of 100 V and 60 °C. After electrophoresis, DNA bands were stained with 1× SYBR® Gold (Invitrogen). Selected bands were excised, and the DNA was eluted with 20–50-μL of DNase free water. DNA fragments were reamplified and sequenced as previously described [9] using universal primers 341f and 907r or HGC236f and HGC664r, respectively.

Cluster Analysis of DGGE Banding Patterns

Firstly, DGGE analysis was performed for monthly samples taken between 2003 and 2009 on separate gels for each year, but for ultimate statistical analyses, we have applied a minimum of DGGE gels and combined several selected samples from the years 2003, 2004, 2006, 2008 and 2009 to avoid huge disruptions when comparing completely different DGGE gels. Each season was included to cover a wide temporal range (two consecutive years on the beginning of our study period, one year in the middle, and two consecutive years at the end of our study period). DGGE gels and cluster analyses were performed for all years and were compared for distinct differences between years (which were not detectable). Hence, it can be assumed that the selected samples were representative for all monthly samples.

Cluster analyses of DGGE banding patterns were performed based on the presence/absence matrix of distinct bands using the GelCompare II software package (version 5.0, Applied Maths, http://www.applied-maths.com). Banding patterns were compared band based using Dice correlation as similarity coefficient and UPGMA (Unweighted Pair Group Method with Arithmetic Mean) with the position tolerance option of the software to fit the curves. Boolean character sets (1 or 0) corresponding to the present or absent phylotypes were extracted, and cluster analyses were calculated. Additionally, relative intensities of DNA bands on the DGGE gels were calculated and compared to the binary presence/absence table. Analyses with quantitative data based on standardised band intensities yielded similar results in nonmetric multi-dimensional scaling (NMS) compared to the presence/absence data of DGGE banding patterns but with a better resolution. Therefore, we have subsequently calculated all statistical analyses (e.g. NMS and Mantel tests) with relative and standardised band intensities. These semi-quantitative data were used for NMS analyses. NMS is an ordination method that applies an iterative optimization procedure to plot samples such that distances between the points indicate the degree of similarity among DGGE banding patterns. NMS ordinations avoid distortions originating from the nonnormal distribution of phylotype data obtained from DGGE profiles and environmental data. NMS analyses were done with 99 runs according to the Bray–Curtis similarity algorithm.

Phylogenetic Tree Construction

All 16S rRNA gene sequences were quality checked and aligned using the Greengenes NAST aligner [34]. Aligned sequences were subsequently imported into the Greengenes ARB database (version 236469) and corrected manually for alignment errors. All sequences from this study were added to the reference tree provided in the database containing only high-quality full-length small-subunit (SSU) rRNA sequences using the quick-add tool provided in ARB [35]. Following this step, more than 1,000 closely related full-length sequences from the reference database were selected to calculate a stable maximum-likelihood backbone phylogenetic tree using FastTree. Short sequences from the DGGE bands were added to this tree afterwards also using the quick-add tool. Phylogenetic classification was done according to the Linnaean Taxonomy. Finally, all full-length reference sequences were removed from the tree without altering the topology, leaving a phylogenetic tree. All sequences identified as Actinobacteria were further classified using the classification tool provided in Greengenes [36]. We chose this method for classification because the sequences obtained from the DGGE bands varied in length and gene region depending on the primers (341f/907r or HGC236f/HGC664r).

All partial sequences of 16S rRNA genes obtained in this study were deposited in GenBank with accession numbers JQ657827 to JQ658297.

Statistical Analyses

Long-term data of physicochemical and biological variables were tested with time series analysis of annual means to depict considerable trends of environmental changes over multiple years. For identification of stringent seasonal patterns and specific environmental variables accounting for changes in BCC, limnetic data of Lake Tiefwaren were also analysed by MDS. Distance matrices for the environmental dataset were computed considering Euclidean algorithms. DGGE profiles were compared with selected measured limnetic parameters and combinations of them by performing Mantel tests [37] with 999 randomised runs using Spearman's rank correlation method and the parameters described above. Due to multiple comparisons, the significance level of our Mantel tests had to be adjusted using the Bonferroni correction. Accordingly, Mantel tests with p <0.008 are significant with 95 % confidence. Statistical analyses were conducted using the PASW statistics software package 17.0.2 (IBM SPSS Statistics, http://www.spss.com), OriginPro 8 (OriginLab Corporation, http://www.originlab.com) and the software Primer 6 (PRIMER-E Ltd, http://www.primer-e.com).

Results

Environmental Variables

Epilimnic water temperature in Lake Tiefwaren oscillated regularly and periodically between ca. 2 °C and 25 °C in each year without any conspicuous inter-annual trend throughout the observation period (mean 13.31 ± 6.67 °C, min. 1.8 °C, max. 24.6 °C; data not shown). In addition, pH, oxygen concentration, conductivity, Chl a and NO3 concentration showed pronounced seasonal variations but no overall trend over the long-time period of several years. Particularly, pH and dissolved oxygen strongly increased in summer due to photosynthetic activities (pH 8.0–9.0 (annual mean 8.41 ± 0.23), O2 9–12 mg L-1 (annual mean 10.49 ± 1.67) and Chl a 1–10 μg L-1 (annual mean 4.1 ± 1.7)). Due to the previous lake restoration, phosphorus concentrations (total phosphorus (TP) and soluble reactive phosphorus (SRP)) were low and constant ranging from 0.01–0.04 mg TP L-1 and 0.002–0.02 mg SRP L-1 throughout the whole study period (Fig. S1A).

Phytoplankton Development

Although annual averages of Chl a concentrations did not change much over the study period, Chl a maxima decreased over the years, whereby the high maxima (>10 μg L-1 Chl a) recorded in summer 2003, 2004 and 2005 did not reoccur in the later years. From 2006 on, Chl a peaks were reduced considerably (≤5 μg L-1, data not shown) accompanied by a decrease in phytoplankton biomass. Phytoplankton community composition greatly varied over the years and did not show any uniform pattern. Considering the long period between 2003 and 2009 of our study, each year was characterised by different dominant algal groups with an exiguous seasonal succession of the phytoplankton community. However, a consistent decline of Dinophyceae and Bacillariophyceae was observed, whereas blooms of specific autotrophs, e.g. Cyanobacteria and Chlorophyceae, alternated in several years (bloom events with biomass >1.0 mg L-1). Lake restoration started in summer 2001 and reduced P concentrations in the lake water almost immediately, and phytoplankton biomass decreased thereafter. Nevertheless, this severe disturbance did not influence our analyses several years later.

In parallel to Chl a, maxima in primary production (PP) dropped from 2003 to 2006 (maximum of 100 μg C L-1 day-1), but stabilised in the following years on a higher level (150–200 μg C L-1 day-1) (Fig. S1B). Winter values were always low (<10 μg C L-1 day-1), indicating a reduced algal productivity at low light conditions. Chl a and PP showed similar patterns and the same seasonal as well as long-term trend.

Bacterial Production

Similar to temperature and phytoplankton, bacterial protein production (BPP), measured as [14C]-leucine uptake by epilimnic bacteria in triplicates (standard deviation <15 % between replicates), showed a pronounced seasonal dynamic: BPP was always highest during summer, reaching values up to 110 μg L-1 day-1 (Fig. S1C). In parallel to PP (Fig. S1B), BPP was low in 2004 and 2005 but not in 2006. During summer stratification, the major part of BPP was linked to PA bacteria, whereas FL bacteria accounted for only 36.5 % ± 9.1 of total BPP. During the study period, total and particle-associated BPP were positively correlated to water temperature (correlation coefficient 0.615, p < 0.01, n = 68). Cell specific BPP increased in summer by a factor of 3–5, indicating high bacterial activities.

Bacterial Abundance and Community Composition

Total bacterial numbers, counted for the years 2005–2009, showed a similar seasonal trend as BPP and reached maximum values of 2.5 × 106 cells mL-1 during the summer (mean of all counts was 1.22 ± 0.45 × 106 cells mL-1, n = 51, Fig. S2A). Abundance of total bacteria (including PA and FL fractions) was significantly higher in summer compared to the rest of the year (two-sample t-test, p < 0.05, df = 49). Time series analyses revealed a slight decrease of bacterial abundance over several years (data not shown).

CARD-FISH showed that bacteria detected with the oligonucleotide probe EUBI–III accounted for 40 % to 80 % of all DAPI stained cells (Fig. S2B). In average, cells that were targeted by the Actinobacteria specific FISH probe HGC69 accounted for 20 % to 40 % of total DAPI counts, whereby members of the acI cluster contributed between 50 % and 65 % of all actinobacterial cells. In contrast to total bacteria, Actinobacteria did not show a clear seasonal trend. Abundances of Actinobacteria and acI-Actinobacteria were relatively constant (ca. 0.4 and 0.2 × 106 cells mL-1, respectively) over time.

Differences in bacterial community composition (BCC) were analysed by DGGE and by sequencing of selected bands. In total, 405 (universal bacterial primer) and 288 (Actinobacteria specific primers) 16S rDNA sequences were obtained during this study. Roughly, 55 % of all sequences were derived from the FL fraction. Analyses of sequences obtained by using the universal primer set (341f and 907r) revealed that ca. 70 % of all FL but only 13 % of all PA microorganisms were affiliated with the phylum Actinobacteria. Equal parts of both fractions (each ca. 15 %) belonged to Betaproteobacteria, whereas Alphaproteobacteria were mainly attached to particles (13 %). Due to their larger cell size and their ability to form colonies, Cyanobacteria mainly occurred in the PA fraction (43 %) and rarely appeared free-living (Fig. S3). Our results are in accordance with those reviewed by Newton et al. [38] and previous FISH-based studies indicating that Actinobacteria and Betaproteobacteria are the most abundant bacterial phyla in freshwater systems.

Almost all Actinobacteria sequences (98 %) belonged to the group Actinomycetales. A detailed analysis of this cluster indicates that in average, 30 % of these sequences were affiliated with the acI-A lineage of ‘Candidatus Planktophila limnetica’ (42 % of FL and 15 % of PA) and 43 % belonged to the Corynebacterineae (33 % of FL and 51 % of PA) (Table 1). Within the Corynebacterineae, ca. 70 % belonged to the Mycobateriaceae, 25 % to the Nocardiaceae and 4 % to the Williamsiaceae (only PA) (Table 1).

Table 1 Proportion of different phylogenetic groups of Actinomycetales and Corynebacterineae. Sequences were separated into total, free-living (FL) and particle-associated (PA) bacteria
Full size table

Comparison of Free-Living vs. Particle-Associated Bacteria

DGGE banding patterns of FL and PA bacteria showed distinct differences regarding their seasonal dynamics. Both groups differed significantly (p < 0.01, n = 48), as indicated by the formation of specific clusters in nonmetric multi-dimensional scaling (NMS) analyses of the DGGE profiles (Fig. 1a). Seasonal dynamics were more pronounced for PA than for FL bacteria. Similarity between PA and FL bacteria clusters in the DGGE dendrogram was 63 %. Within the PA and FL clusters, the similarity was ≈ 70 % and ≈ 90 %, respectively (Fig. 1b).

Figure 1
figure 1

a Nonmetric multi-dimensional scaling (NMS) plot representing differences between free-living (FL) and particle-associated (PA) bacterial communities calculated from DGGE banding patterns. b Cluster analysis of DGGE banding patterns. Dice similarity within PA and FL communities is 70 % and 90 %, respectively, and between PA and FL, it is 63 %. Sampling dates are indicated by year/month

Full size image

Free-Living (FL) Bacteria and Actinobacteria

Cluster analysis of FL bacteria resulted in highly similar DGGE banding patterns of tested samples (Fig. 2a, n = 20). The similarity index S ≈ 90 % denotes no separation between different years or seasons. Some samples even showed analogous or almost identical DGGE banding patterns (e.g. 2004/07 and 2008/11, 2008/05 and 2009/05, 2004/11 and 2009/03). DGGE analysis of FL Actinobacteria (Fig. 2b, n = 20) showed similar results and no distinguishable differentiation between years or seasons. Dominant bands of DNA sequences affiliated with the acI-A lineage of ‘Candidatus Planktophila limnetica’ did not disappear at all and were present throughout the whole study period.

Figure 2
figure 2

UPGMA dendrogram for FL bacterioplankton communities calculated based on DGGE banding patterns of bacterial 16S rRNA gens amplified from water samples of Lake Tiefwaren between 2003 and 2009. a Cluster analysis of FL bacteria. b Cluster analysis of FL Actinobacteria. Similarity within both groups ≈90 %

Full size image

Particle-Associated (PA) Bacteria and Actinobacteria

Throughout the entire sampling period of 7 years, no consistent gradual trend emerged over the years. In our comparison, community composition of PA bacteria differed between winter and summer (Fig. 3a, n = 25) and was quite stable during winter. In summer, however, numbers of DGGE-bands increased slightly, and microbial diversity was much higher and directly related to temperature and phytoplankton dynamics. Since DGGE only detects the most abundant phylotypes, we might have missed a substantial proportion of the bacterioplankton diversity, especially the rare phylotypes. The detected abundant bacteria, however, revealed two clearly separated clusters: one comprising almost all late spring and summer samples (May until September, named as summer cluster) and a second one with nearly all samples from late autumn, winter and early spring (November until May, named as autumn/winter cluster). The similarity index between these two clusters was around 65 %. The within-similarity of the identified clusters was slightly higher in the autumn/winter (>75 %) than in the summer (>68 %). Some DGGE banding patterns of single samples, especially of the autumn/winter cluster, appear to be identical, e.g. 2004/04, 2009/03 and 2008/11, as well as 2003/11 and 2004/02 or 2006/03 and 2008/05. In the summer cluster, numbers of DGGE-bands of PA bacteria were much higher in parallel to the more pronounced phytoplankton dynamics.

Figure 3
figure 3

a Cluster analysis of DGGE banding patterns of PA bacteria indicating a summer and autumn/winter cluster. The similarity index between these two clusters is ≈65 %. b DGGE clusters of PA Actinobacteria

Full size image

For PA Actinobacteria, two main clusters (summer and autumn/winter) were identified (n = 25). In contrast to PA bacteria, community composition of PA Actinobacteria was more diverse in the autumn/winter than in the summer cluster (Fig. 3b).

Relationship Between Bacterial Community Composition and Environmental Parameters

NMS plot for environmental variables of Lake Tiefwaren exposed a distinct segregation into different seasonal patterns and separate seasonal clusters (spring, summer, winter, autumn, Fig. S4). Hence, chemical data of summer samples are located in-between spring and autumn and are widely dispersed; samples collected in winter stayed close together and formed a very narrow cluster. Ice coverage during winter did not occur regularly in Lake Tiefwaren. Our definition of seasonal clusters was based on limnetic variables, mainly on lake stratification/mixing and water temperature.

Whereas time series analyses of annual means did not show any particular trend of limnetic parameters over several years (data not shown), temperature is one of the most important seasonal drivers for annual shifts in BCC. We found very strong correlations of PA BCC to various environmental factors, especially temperature, nitrogen and zooplankton (Table 2). The results of all statistical comparisons are given in Table 2. There, we have differentiated between summer and winter (including autumn and spring) seasons and distinguished between FL and PA bacteria as well as Actinobacteria, respectively. Our Mantel tests show correlations between PA BCC and nutrients, especially nitrogen (NO2, NO3, NH4 and TN) during summer. Particularly, BCC of PA bacteria—due to the characteristic of being associated to particles—was tightly linked to the abundance of Dinophyceae (Manterl test: r = 0.375, p = 0.008) and zooplankton. In contrast, BCC of PA Actinobacteria was significantly correlated to total phosphorus (Table 2) and to Cryptophyceae (Mantel test: r = 0.303, p = 0.002). We identified more significant correlations between BCC of PA bacteria and environmental factors than for PA Actinobacteria communities (Table 2), but we did not find any significant correlations with both communities of FL bacteria and FL Actinobacteria.

Table 2 Statistical comparison (Mantel tests) of bacterial and Actinobacteria DGGE profiles and environmental variables separated into all years and seasons, summer and winter samples (indicated as other seasons, including autumn, winter and spring). Values printed bold indicate significant (p ≤ 0.05) correlations, according to multiple comparisons on not independent similarity matrices and Bonferroni corrections, significance level  ≤ 0.008 (six tests). Grouping of parameters was done for nitrogen (NO2, NO3, NH4 and TN), phytoplankton (biomasses of the following phytoplankton groups: Chlorophyceae, Bacillariophyceae, Cryptophyceae, Chrysophyceae, Dinophyceae, Chroococcales, Hormogonales, Hormogonales and Cyanophyceae) and zooplankton (Daphnia, other Cladocerans, Calanoida, and Cyclopoida)
Full size table

Discussion

Seasonal Variability

Composition of bacterial communities in aquatic ecosystems is considerably influenced by environmental factors, which change over seasons (e.g. temperature and pH). Our long-term data revealed clear seasonal patterns for a variety of environmental parameters, which were also reflected in BCC dynamics. BPP was always maximal in summer (Fig. S1C), but not necessarily correlated to PP (Fig. S1B) or bacterial abundances as determined by CARD-FISH (Fig. S2B). Seasonal changes in freshwater bacterioplankton abundance and activity have been investigated in previous studies that show similar results [3, 33]. Our statistical analyses between BCC and environmental variables indicate that temperature was a strong driver for the observed seasonal dynamics in BCC of Lake Tiefwaren. Particularly, PA bacteria that were—disregarding Cyanobacteria—mainly dominated by Alpha- and Betaproteobacteria showed a strong and significant correlation to temperature. Furthermore, BCC of PA bacteria was positively correlated to nutrients, including different nitrogen species (NO2, NO3, NH4 and TN; Table 2), especially in summer when nutrient concentrations were generally low. In contrast, BCC of PA Actinobacteria was significantly correlated to total phosphorus, but only when taking samples from all seasons into account. Limitation of bacterioplankton growth by nutrients has been experimentally tested [39], and it has been shown that changes in nutrient concentrations can greatly affect BCC in the past [40] by favouring specific microorganisms with the ability to rapidly respond to certain nutrients [41]. In our study, solely BCC of PA bacteria and Actinobacteria was significantly correlated to nutrients, which may indicate that the high nutrient requirement of the relatively large bacteria associated to particulate organic matter leads to a higher importance of nutrient availability in structuring BCC of PA bacteria in the long run.

Seasonal changes in freshwater microbial community structure have been shown by previous studies—although monitored only for a single year [7, 42, 43]. Exemplarily, similar seasonal patterns were also found in three temperate lakes of contrasting trophic status [8] and in a comparative study of a natural lake and a water reservoir [44]. These studies concluded that the observed variations in BCC were caused by changes in limnetic conditions, such as mixing and stratification, and that distinct bacterial communities can reoccur from year to year based on seasons. Since the observation period of the aforementioned studies was rather short, they may not allow for any systematic conclusion for which long-term data are needed. Surprisingly, most of the long-term studies on microbial communities were done in marine ecosystems so far [1, 2, 45]. Solely, Shade et al. [5] reported on BCC dynamics over six consecutive years in eutrophic Lake Mendota and found a reoccurring and hence predictable dynamic of BCC. A regular phenology was repeated across years, implying that freshwater bacterial communities were more predictable in their dynamics than previously thought. Additionally, most of the other studies focused only on total bacterial communities and did not account for different bacterial lifestyles. To better distinguish between different size fractions, we analysed PA and FL bacteria separately. This size fractioning is only feasible in oligo- to mesotrophic ecosystems. In more eutrophic as well as in turbid systems (very shallow lakes or rivers, streams and estuary creeks), this differentiation between different bacterial size fractions may be almost impossible due to numerous particles, including filamentous organisms or suspended sediments. A few short-term studies [33, 46, 47], which also differentiated between two size fractions, revealed pronounced differences between PA and FL bacteria, including their response to changes in environmental conditions. Consequently, this is the first study to distinguish between PA and FL bacterial fractions in the long term.

Comparison of Particle-Associated (PA) vs. Free-Living Bacteria (FL)

Traditional sampling for BCC analyses often removes a substantial part of PA bacteria by pre-filtering water samples through gauzes of different sizes to exclude larger organisms such as phyto- and zooplankton [48]. However, separation of pelagic bacteria into PA and FL fractions via serial filtration reveals a much better resolution in BCC dynamics and its relationship to specific environmental parameters [33, 49] than studies that collected the whole bacterial community on 0.2-μm filters [5, 40]. For separation of PA and FL bacteria, filters with different pore sizes can be used, affecting the results accordingly. For example, Hollibaugh et al. [50] and Beman et al. [51] used 1.0–1.2-μm A/E glass fibre filters to separate PA from the FL fraction. Others collected FL bacteria after pre-filtration through 3.0-μm filters [52]. However, in most studies on BCC including the present one, PA and FL bacteria were defined as fractions >5.0 μm (PA) and <5.0 >0.2 μm (FL), respectively (e.g. [33, 46, 47, 49, 53]). Our choice for using 5.0-μm filters to separate between PA and FL bacteria is based on numerous microscopical observations of various freshwater samples, including samples from mesotrophic Lake Tiefwaren. Nevertheless, for better comparing studies on BCC in various aquatic systems, a more defined and restricted use of filter pore sizes for separation between different bacterial fractions is advisable.

To study microbial community structure, we applied the DGGE approach, a robust and reliable method to characterise abundant members of a given bacterial community [32, 54]. Several of our DGGE analyses suggest in a reproducible manner that FL bacteria are more similar between samples than PA communities (Fig. 1a, b). BCC of abundant FL bacteria showed highly similar DGGE banding patterns and, thus, formed a distinct and narrow cluster in our NMS analysis. Communities of FL bacteria were relatively stable throughout the season and were dominated by Actinobacteria. In contrast, PA bacteria (mostly affiliated to Cyanobacteria and Alphaproteobacteria) were more diverse and showed pronounced seasonal variations in BCC, reflecting changes in particle quality over time [49, 55, 56]. Cyanobacteria that often have a size larger than 5.0 μm in diameter and tend to form colonies, mainly contributed to the PA fraction, but only little to the FL fraction. In accordance to the seasonal dynamics of phytoplankton, community composition of PA bacteria was much more diverse in summer than in winter (Fig. 3a). Particles mostly consist of detrital and algal aggregates and/or zooplankton, densely colonised by bacteria, fungi and protozoa [56]. Quality and abundance of particles change with seasons, e.g. pollen grains represent a microbial habitat that is rich in organic matter, and hence, such particles are rapidly colonised and decomposed by pelagic microorganisms (heterotrophic bacteria and aquatic fungi) within a few days [57].

Actinobacterial-related DGGE bands were dominated by Actinomycetales (98 %) and had a slightly higher appearance and diversity in winter than in summer (Fig. 3b), suggesting actinobacterial growth also at winter conditions. Observed high occurrence of FL Actinobacteria in the winter season is supported by a previous survey [9], indicating that total Actinobacteria, including members of the acI cluster, reach their maximal absolute as well as relative abundance in fall/winter, but also in late spring. A comparable seasonal pattern in Actinobacteria occurrence has been observed in an oligo-mesotrophic alpine lake [10]. The recently described lineage ‘Candidatus Planktophila limnetica’ [58] formed a consistent group of the acI-A cluster of FL Actinobacteria. This phylum was highly abundant throughout the whole study period, indicating its universal occurrence in Lake Tiefwaren. Most likely, this group—due to its small cell size and unique cell wall—is more resistant against grazing than others [35]. The reduced variability observed in Actinobacteria diversity might be caused by the limited power of the detection technique in our study. DGGE tends to preferentially reproduce the same fragments when applying it as a very specific fingerprint technique, often resulting in an ‘apparent community stability’.

The separation between PA and FL bacteria and Actinobacteria, however, revealed a much more complex relationship between the different bacterial fractions and available nutrients (Table 2). Particularly, BCC of PA bacteria was coupled to the dynamics of both phyto- and zooplankton. For example, PA bacteria strongly correlated to Dinophyceae, whereas PA Actinobacteria positively correlated to Cryptophyceae. Parveen et al. [46] observed that attached Actinobacteria were highly correlated to the biomass of Chrysophyceae, indicating ecosystem-related adaptation of Actinobacteria to certain microalgae. PA bacteria of Lake Tiefwaren were dominated by Cyanobacteria, Alpha- and Betaproteobacteria and showed close linkage to zooplankton biomass in summer. However, FL bacteria as well as ultramicrobacteria, including Actinobacteria, did not appear to be affected by zooplankton—presumably due to their small cell size that may relieve them from grazing pressure.

Seasonal Succession and General Patterns of Bacterial Dynamics

Based on bacterial abundances, a periodical seasonal trend with high bacterial cell numbers in summer and decreased values in winter can be assumed (Fig. S2). Cell specific BPP increased in summer by a factor of 3–5, indicating high bacterial activities presumably when top down control may be high, but unfortunately, we could not determine protozoan grazing rates. BCC of PA bacteria was more diverse and variable over time than those of FL communities. Most pronounced changes in PA bacterial communities occurred during the transition from winter to spring and later to summer. In November, communities shifted towards a rather stable winter community, which reoccurred every year. Temperature is the major driving force for variations of BCC in Lake Tiefwaren. Hence, thermal stratification and thermal stability of a lake (including spring snowmelt, ice-off and fall-overturn) can control the emergence of specific bacterial communities and often lead to increased heterogeneity of BCC [59]. In rivers, where BCC is strongly correlated with temperature [4, 60], seasonal variation of microbial assemblages is relatively well predictable [61]. Our results show that seasonal dynamics of BCC is influenced by abiotic as well as biotic drivers (Table 2). However, only a few studies have compared BCC of rivers and lakes in respect to seasonality. In oceans, yearly reoccurring patterns in BCC can be relatively well predicted in long-term studies, e.g. patterns in distribution and abundance of bacterioplankton could be related to the variation in selected abiotic and biotic environmental variables [1, 45]. In freshwater, there is a great need for similar long-term studies on seasonal reassembly and stability of BCC. Our study shows that freshwater bacteria behave in a similar manner as their marine counterparts, and especially, separation into PA and FL bacteria reveals distinct seasonal differences in annual BCC dynamics depending on bacterial lifestyle.

Conclusions

This is the first long-term study that analysed BCC in Lake Tiefwaren and revealed consistent, reoccurring seasonal patterns. It also demonstrates that separation between FL and PA bacteria and Actinobacteria can achieve better correlations with abiotic and biotic parameters, especially when considering different seasons. DGGE banding patterns strongly correlate with several abiotic variables, particularly with temperature and nutrient availability. In particular, community composition of PA bacteria is significantly correlated to nitrogen compounds, whereas that of PA Actinobacteria is significantly correlated to total phosphorus. Abiotic factors strongly influence seasonal patterns in BCC of FL and PA bacteria, but additionally, changes in phytoplankton and zooplankton succession affect these bacterial groups in a more complex manner. In winter, when biotic factors are less important, a stable microbial community reoccurs every year. Separation between FL and PA bacteria and Actinobacteria increases the resolution of such seasonal patterns and shows that both bacterial fractions differ in lifestyles and occupy different ecological niches. A future challenge in microbial ecology will be to better understand and predict seasonal patterns of distinct functional groups of bacteria even in the long run.