Introduction

Eutrophication of an aquatic system is commonly caused by long-term nutrient discharge into the water as a result of human activities. Shallow lakes are especially at risk of becoming turbid, a state characterized by high phytoplankton biomass and poor water quality [34]. The natural and recreational value of the lake consequently deteriorates.

Bacteria play a key role in the microbial community responsible for recycling of the nutrients in the aquatic systems [4].

Numerous studies have been conducted on the regulation of bacterial communities by environmental factors at the biomass level. Recently, the development of molecular biology tools suitable for ecological studies has made it possible to follow changes in the composition of populations at the genetic level and the spatial and temporal effects on the populations as a consequence of environmental changes. DNA-based techniques, such as cloning and sequencing, fluorescent in situ hybridization, and denaturing gradient gel electrophoresis (DGGE), have particularly expanded our understanding of the diversity and identity of microbes in various habitats [1, 25, 43].

Changes in functioning of the bacterial community have been shown to be related to changes in its species composition [7, 18, 31]. Large, controlled mesocosm experiments in seawater have included studies on genotypic changes in bacterioplankton community composition in response to changes in organic substrate or inorganic nutrient concentrations [3, 27, 31, 33]. A few controlled studies have also been conducted in freshwater mesocosms [8, 22, 32, 48]. In addition, studies have been conducted to follow the development of the structure ofaquatic microbial communities in response to changes in environmental conditions in eutrophic lakes [12, 20, 23, 41]. However, in freshwaters, the littoral zone has received little attention, although its habitats are more varied and the interactions are more complex than those in the pelagial zone. Thus, more experimental studies are needed to better understand the effects of increasing nutrient concentrations on the food web, including bacteria, in shallow lakes under threat of eutrophication.

We recently conducted a mesocosm study on the effects of nutrient enrichment and fish manipulation on phyto- and zooplankton and macrophytes in the littoral of mesotrophic Lake Vesijärvi in southern Finland [11]. The mesocosms were set up in two successive summers. In 1998, when macrophytic growth and cladoceran biomasses were low, chlorophyll a concentration increased linearly with phosphorus concentration, and mesocosms became turbid; only three control mesocosms without fish maintained a clear-water state. In 1999, by contrast, the nutrient additions were more moderate, and all the mesocosms maintained a clear-water state. Phytoplankton biomass remained low as a result of grazing by abundant cladocerans and reduced light penetration into water caused by macrophytes, especially freely floating lemnids (Lemna trisulca L.). Weather conditions in spring and early summer appeared to have been largely responsible for the differences in development of macrophytes and zooplankton in the 2 years.

In this article, we report on changes in bacterioplankton community composition (BCC) in the mesocosms. Mesocosms with different nutrient concentrations were selected to represent different levels of eutrophication, from mesotrophic to hypertrophic. To follow the changes in BCC, we used DGGE profiling of partial 16S rRNA gene fragments and nonmetric multidimensional scaling (NMDS). The correlations between the patterns obtained and environmental factors were investigated using canonical correspondence analysis (CCA). In addition, the phylogenetic affiliation of the major DNA bands was examined by DNA sequencing.

Materials and Methods

Description of Study Site and Experimental Design

Lake Vesijärvi in southern Finland, in its present state, is mesotrophic. Total phosphorus concentrations in the littoral zone outside our mesocosms varied between 5 and 50 μg L−1, and total nitrogen (in 1999) varied between 600 and 1400 μg L−1. Two factorial field mesocosm experiments were conducted starting on 4 June 1998 and 29 June 1999. The mesocosms were plastic cylinders of 1-m diameter that reached the lake bottom at 1-m depth. The nutrients, nitrogen as KNO3 and phosphorus as KH2PO4, were added weekly. The concentrations of added nutrients were different in the two experimental years (Table 1). In 1998, the nutrient enrichment levels N2 and N3 were so high that they led to hypereutrophic conditions, which hardly ever occur in Finnish lakes, although possible in central and southern Europe. Therefore, in 1999, the nutrient additions were adjusted to represent mesotrophic to eutrophic conditions more relevant to the Finnish environment. Consequently, N1 in 1998 (100 μg P L−1 and 1 mg NL−1) was comparable with N3 in 1999 (90 μg P L−1 and 0.9mg N L−1). In 1998, roach (Rutilus rutilus L.) and, in 1999, perch (Perca fluviatilis L.; both native species in the lake) were used in the experiments (Table 1). Additional lake water samples were taken just outside the mesocosms. The details of the experimental design have been published elsewhere [11, 38].

Table 1 Sampling dates, corresponding experimental week, and a number of samples from the lake water and mesocosms analyzed using denaturing gradient gel electrophoresis (DGGE)
Full size table

Sample Treatment

Water samples for the bacterial community analyses were taken from the mesocosms on the dates indicated in Table 1. Sampling always took place before nutrient additions. A 1-m-long tube was used to transfer 1 L of water into plastic bottles, which were then transported to the laboratory on ice.

For DNA extraction, 900 mL of sample water was concentrated by centrifugation at 4°C to approximately 500 μL. DNA was extracted using bead beating and CTAB (hexadecylmethylammonium bromide) treatment, purified with a Prep-A-Gene Purification Kit (Bio-Rad), and amplified with eubacterial 16S rRNA gene primers F-968-GC and R-1401, as in [10]. DCode™ Universal Mutation Detection System (Bio-Rad) was used for running the DGGE gels with 6% polyacrylamide and 35–50% formamide–urea gradient, as in [16]. The band intensities from the lake water samples were faint compared with the other samples, which was also reflected in the relatively low bacterial DAPI counts (results not shown). We therefore verified the lake banding patterns by running an additional gel, for which we obtained more DNA by combining several replicate polymerase chain reactions (PCRs).

Strong and well-defined bands were cut from the DGGE gels, sequenced, and compared with databank entries, as in [10].

Statistical Analyses

Denaturing gradient gel electrophoresis images were transferred to GelCompar software (Applied Maths, version 4.1) for alignment of the electrophoresis patterns. The alignment was corrected by visual inspection of the gels. An optical density profile was measured for each DGGE lane (sample), and the band positions and relative (%) intensity of each band were calculated against the whole lane (100%). The data matrix obtained based on the relative DNA band intensities was used for statistical analyses.

NMDS was used to cluster the samples. NMDS analyses were performed with the Community Analysis Package software (version 2.0, Pisces Conservation Ltd., Lymington, UK), which calculated the Sorensen similarity coefficients and visualized the similarities between the samples in two dimensions. CCA, a direct gradient analysis method in the CANOCO 4.5 software package [40], was used to compare the DGGE band matrices with the environmental variables (physicochemical variables and phyto- and zooplankton data). Both DGGE data and environmental variables (except pH, alkalinity, temperature, and total suspended solids) were log(x + 1)-transformed. Forward selection was used to rank the environmental variables in importance for determining the species data, and the Monte Carlo permutation (499 permutations) test was applied to judge which variables contributed significantly to the statistical model. The environmental (explanatory) variables used were fish density, nitrogen concentration measured as NO3, NH4, NO2 (in 1998), and total nitrogen (in 1999), total phosphorus concentration, soluble reactive phosphorus (in 1999), transparency as determined by Secchi depth, pH, alkalinity (in 1998), water temperature, oxygen concentration and saturation, total suspended solids (in 1998), chlorophyll a, biomasses of the phytoplankton groups [(in μg L−1): cyanobacteria, diatoms, dinoflagellates, chloro-, chryso-, conjugato-, crypto-, eugleno-, and tribophytes, phytoflagellates], and biomasses of the zooplankton groups [(in μg C L−1): rotifers, large euplanktonic grazers (>0.5 mm; e.g., Daphnia, Eudiaptomus), small euplanktonic grazers (<0.5 mm; e.g., rotifers, Bosmina, Ceriodaphnia, nauplii), large plant-associated Cladocera, small plant-associated Cladocera, Polyphemus, calanoid copepods (Eudiaptomus), cyclopoid copepods, nauplii, large grazers, small grazers, crustacean grazers, grazers, raptorials, and all crustaceans]. Measurement of the environmental variables has been described previously [11].

Results

Analysis of DGGE Fingerprint Patterns

Denaturing gradient gel electrophoresis fingerprint patterns were obtained for 50 samples from 1998 and for 75 samples from 1999 (Table 1). In general, in 1998, the bands were dispersed across the entire gel gradient, whereas in 1999, they were more concentrated in the middle (Fig. 1). The fingerprints of the lake samples remained relatively constant over the sampling period of both years. The same main bands were present, although their intensities varied and a few unique bands were seen (Fig. 1).

Figure 1
figure 1

Example of denaturing gradient gel electrophoresis (DGGE) fingerprint patterns obtained from the water samples in 1998 and 1999. St: standard; Lake: lake water; N0, N1, N2, and N3: different nutrient levels (Table 1); f: fish. For the numbered bands, a matching sequence is shown in Table 3.

Full size image

NMDS analysis was used to display the relations of the fingerprint patterns in two dimensions (Fig. 2). In 1998, the succession of bacterial communities was affected by the nutrient treatments, even at the lowest nutrient addition level N1, whereas in 1999, the banding patterns in all the treatments remained closer to that of the lake water. In 1998 (Fig. 2A), the samples clustered according to the nutrient treatment; thus, the lake samples and control mesocosms clustered on the left side of the plot, and the general shift of the patterns to the right took place with the increasing nutrient addition levels. In addition, in the mesocosms with high nutrient addition levels, the samples from the early weeks of the experiment were found on the top half, and the samples from the later weeks were found on the bottom half of the plot. In 1998, the so-called bottle effect was detected because the fingerprint patterns of the control mesocosms consistently differed from those of the lake samples. However, the patterns were closer to the ones from the lake water than to the treated mesocosm patterns. The final stress of the first dimension was 0.3606 and of the second dimension 0.245.

Figure 2
figure 2

Clustering of the DGGE fingerprint patterns by nonmetric multidimensional scaling for (A) the 1998 samples and (B) the 1999 samples. Nutrient levels as in Table 1. Numbers next to the symbols show the sampling week.

Full size image

In 1999, the first sampling was performed before thefirst nutrient addition and at the time the patterns from different mesocosms were very similar to each other (Fig. 2B). Samples taken during the experiment clustered according to the sampling week, indicating succession of the populations over time, regardless of the experimental treatments. In the 1999 analysis, the final stress of the first dimension was 0.3863 and of the second dimension 0.2112.

CCA was used to investigate the correlations between DGGE banding patterns and environmental variables, including physicochemical water properties and different groups of phyto- and zooplankton. The CCA plot of samples, species, and significant environmental variables based on the first four axes explained 23.0% in 1998 and 25.9% in 1999 of the variance in species data (Table 2). In 1998, the environmental factors contributing significantly to the statistical model were pH, concentrations of nitrate, chlorophyll a and total suspended solids, and biomasses of cryptophytes, cyanobacteria, small (<0.5 mm) euplanktonic grazers, and cyclopoid copepod nauplii (Fig. 3A). In clustering of the 1998 samples, nutrient treatment had the most pronounced effect; samples from the lake and control mesocosms were placed on the left side of the plot, and samples from the nutrient-enriched mesocosms were placed on the right side together with the vectors indicating increasingly eutrophic conditions. Clustering of the latter samples also changed over time. In 1999, the environmental factors contributing significantly to the model were concentrations of total nitrogen, nitrate, and oxygen, oxygen saturation, temperature, and biomasses of crypto- and chlorophytes, diatoms, crustacean grazers, and zooplankton grazers (Fig. 3C). In 1999, the samples clustered according to the sampling date, whereas the nutrient treatment did not seem to have an effect.

Table 2 Summary of CCA results on bacterial community composition constrained to the significant (p < 0.05) environmental variables
Full size table
Figure 3
figure 3

Canonical correspondence analysis ordination plots of the samples, species (DGGE bands), and environmental variables. Symbols (A, C): open circle, lake water; open square, control mesocosm N0; black square, nutrient levels N1 (1998) and N3 (1999); black diamond, N2 (1998); black star, N3 (1998); down-triangle, N1 (1999); up-triangle, N2 (1999). Numbers next to the symbols show the sampling week. Symbols (B, D): black circle, sequenced band; open circle, band not sequenced. Numbers refer to the matching sequences in Table 3.

Full size image

Analysis of Partial 16S rRNA Gene Sequences

Major bands were excised from the DGGE gels, reamplified, and sequenced. Sequences were obtained for 15 of the 52 bands used in CCA analysis in 1998 and for 14 of the 35 bands in 1999. The sequenced bands comprised 59 and 84% of the total band intensity in 1998 and 1999, respectively. Some strong bands gave mixed sequences because of DNA from several organisms present in close proximity; these were not analyzed further. Altogether, 22 different unambiguous sequences were obtained (Table 3). These sequences represented three divisions: Verrucomicrobia, Actinobacteria, and Proteobacteria. In addition, one sequence from the Cytophaga–Flavobacterium–Bacteroides (CFB) group was obtained.

Table 3 Sequenced DGGE bands from the water samples from years 1998 and 1999 and their closest relatives
Full size table

In 1998, band VE14-98 represented a very common verrucomicrobial phylotype. It had a novel sequence and possessed only 95% similarity to the closest reported relative. At week 4, it was present in lake water and mesocosms, but thereafter only in mesocosms. In several cases, it represented over half of the total band intensity in nutrient-enriched mesocosms. In 1999, VE24-99 was the predominant verrucomicrobial phylotype, present in both lake water and mesocosms. In 1998, Actinobacteria were present mainly in lake water and in control mesocosms; one actinobacterial band could represent as much as one third of the total band intensity of the lane. In 1999, actinobacterial VE44-99 was common in all but the highest nutrient-enriched mesocosms. Thus, for both years, the strong actinobacterial bands were limited to mesocosms having less than 100 μg P L−1 and 1 mg N L−1. In addition to actinobacterial sequences, only one verrucomicrobial sequence and one β-proteobacterial sequence were obtained from the lake water. In the mesocosms, the occurrence of sequenced Proteobacteria was rather sporadic; different phylotypes could prevail in a couple of mesocosms each week. The CFB sequence was detected only during week 4 in 1998.

CCA analysis showed the bands detected in DGGE gels in relation to the environmental variables (Figs. 3B, D). Actinobacteria were correlated with less eutrophic conditions in the CCA, whereas Verrucomicrobia and the only representative of the CFB group were correlated with more eutrophic conditions.

Discussion

Changes in Bacterioplankton Community Composition

In 1998, nutrient treatments affected the diversity and composition of bacterial communities strongly. In addition, probably another unknown factor (or factors) changed over time and affected the bacterial communities. CCA analysis indicated that BCC of nutrient-enriched mesocosms was influenced by the environmental variables related to an increased level of eutrophication (high nutrient, chlorophyll a, and total suspended solids concentrations and pH). Nutrient enrichment clearly increased the biomass of phytoplankton [11]; however, only the concentration of nitrogen (measured as nitrate in 1998) correlated directly with BCC. In Lake Vesijärvi, shortage of inorganic nitrogen has been reported to limit the growth of algae [26], but in the quantities available in our mesocosms (trophic levels from mesotrophic to eutrophic), we did not expect that nitrogen or phosphorus limitation would have affected BCC directly. High biomass of phytoplankton increased photosynthesis rate, which led to elevated pH values. In fact, pH in the high-nutrient mesocosms rose so much (pH 11) that allfish died. A correlation between pH and BCC has also been reported elsewhere [20, 37, 47]. pH may merely reflect changes in other environmental factors, such as phytoplankton, but an increase in pH may also influence BCC directly or indirectly by, for example, changing the solubility of phosphorus [17].

In 1999, the bacterial communities were less diversified and resembled that of the lake water community. However, BCC changed to some extent over time. Seasonal succession of freshwater bacterioplankton has been attributed to such factors as light, temperature, wind-induced currents, and phytoplankton blooms [5, 12, 51]. In our CCA analysis, temperature and temperature-dependent oxygen saturation emerged as significant physical environmental factors. Concentration of nitrogen (measured as total N and nitrate in 1999) again correlated significantly with BCC, but chlorophyll a did not. In 1999, the nutrient levels were generally lower thanin 1998, but still at the highest nutrient level of 1999 (90 μg P L−1 and 0.9 mg N L−1), abundant nutrients were available in the nutrient-enriched mesocosms. However, phytoplankton biomass remained low in 1999 because of grazing by abundant cladocerans. Macrophytes, which dominated the mesocosms in 1999, could be expected to have a direct effect on the attached bacterial communities rather than on planktonic bacterial communities [46].

Certain phytoplankton groups correlated significantly with BCC in both years. In 1998, cyanobacteria and cryptophytes were abundant in the control mesocosms and lake water. Nutrient-enriched (eutrophic to hypertrophic) mesocosms had a high chlorophyll a concentration, mainly because of the great biomass of the chlorophyte Scenedesmus spp. In 1999, when the nutrient concentrations were lower and closer to the actual ones in Northern European lakes, the significant phytoplankton classes were cryptophytes, diatoms, and chlorophytes. These results are consistent with those obtained from mesotrophic lakes in Sweden [21]. Phytoplankton is a major source of autochthonous organic matter in lakes, and algal exudates are highly bioavailable to bacteria compared with other sources of dissolved organic matter (DOM) [36, 45]. The species composition, physiological state, and biomass of phytoplankton seem to regulate the photosynthetic extracellular release, which, in turn, may influence the composition of bacterial communities [24, 28, 30, 31, 42]. When we examined the role of the phytoplankton species more closely, different species from several phytoplankton classes showed significant correlations with BCC (results not shown). However, controlled experiments are needed to verify whether the presence or absence of these species can, in fact, lead to different BCC.

Top-down control, i.e., grazing and viral infections, also plays an important role in the control of microbial communities. In our mesocosms, we did not examine protists, the bacterivores such as heterotrophic nanoflagellates and ciliates, or the effect of viruses. However, biomasses of micro- and mesozooplankton were determined and reported by Hietala et al. [11]. In 1998, small grazers, predominantly rotifers and naupliar stages of cyclopoid copepods, correlated significantly with BCC. The rotifer community was dominated by microphagous species such as Keratella cochlearis and small Trichocerca, which readily feed on bacteria-sized particles [29]. In their mesocosm tracer studies on carbon cycle, Lyche et al. [23] showed that zooplankton community comprised mainly of cyclopoid copepods was weakly coupled to microbial loop. Thus, the correlation between BCC and nauplii might have been a result of other interactions in the complex food web or might have been merely coincidental. In 1999, the general groups of grazers and crustaceans, which predominantly consisted of filter-feeding cladocerans, correlated significantly with BCC. The predominant zooplankton thus seemed to influence BCC, either directly or by selective feeding on protists, which, in turn, influenced BCC, as also reported earlier [12, 14, 15, 19, 20, 33, 48]. In addition, grazers act as producers of DOM through “sloppy feeding”, feces, and other excreta. Grazers release nutrients in organic rather than in inorganic forms, thus supporting heterotrophic rather than autotrophic processes in ecosystems [2]. Also, fish may efficiently return organic and inorganic substrates into the water column through excretory processes and by mixing of sediment [13] and thus enhance bacterial growth. Nevertheless, fish did not affect the clustering of BCC in our study.

In relating the changes in BCC patterns to environmental variables in CCA, we chose to use band intensity data because we felt that they better depict the changes taking place in the bacterial community than absence/presence data, as also stated by Muylaert et al. [24] and Yannarell and Triplett [47]. No doubt PCR amplification can severely distort the relative proportions of different genotypes, but because we compared samples originating from the same environment at different time points, the quantitative results can be assumed to reflect the proportional changes taking place in BCC. In addition, caution is needed when considering the significance of the environmental factors. Transformation of data can have an impact on the results, as shown by Yannarell and Triplett [47], and the significance of a single environmental factor also depends on the other environmental variables included in the run. It is also possible that some of our unsequenced bands may have originated from cyanobacteria or chloroplasts, thus increasing the correlation we detected between BCC and the phytoplankton groups.

Sequence Diversity

The bacterial groups we detected by sequence analysis are among the universal groups found in freshwater environments [49]. Most of our sequences were closely related to previously reported sequences from uncultured freshwater bacteria. Close relatives were typically found from distant geographic locations.

The common occurrence of Verrucomicrobia in our mesocosms suggests that Verrucomicrobia are able to take advantage of a nutrient-rich environment, or that the phylotypes represented by our sequences had some features that made them competitive in a confined environment. Many Verrucomicrobia are prosthecate [50], and this could offer them an advantage in nutrient uptake or make them more resistant to grazing. Our studies on BCC in other Finnish lakes have also indicated that Verrucomicrobia are relatively more prevalent in eutrophic lakes than in oligo- or mesotrophic lakes [10, 16].

Another group of bacteria commonly present in our samples was Actinobacteria. Four of our five sequences belonged to actinobacterial clade acI and one (VEL8-99) to clade acII, as defined by Warnecke et al. [44]. Both of these clades harbor 16S rRNA sequence types from freshwaters and estuaries with varying hydrological and limnological characteristics and from distant geographic locations. Besides mesotrophic Lake Vesijärvi, we have found closely related Actinobacteria in other Finnish lakes of different nutrient and humic content as well as in cyanobacterial blooms [16]. In the CCA analysis, Actinobacteria clustered far from the vectors indicating a high level of eutrophication. This supports our earlier notion that although Actinobacteria are permanent abundant members of bacterioplankton, they do not respond strongly to new nutrient resources [10].

Proteobacteria and the members of the CFB group have, in many studies, been found to be the predominant freshwater bacteria (e.g., [6, 9, 35, 49]). In our mesocosms, their occurrence was rather sporadic. However, because PCR amplification can distort the results [39], and we did not obtain sequences from every strong band in the gels, we cannot draw firm conclusions about the most common bacteria in the mesocosms.

In conclusion, with our mesocosm experiment, we have shown that when a shallow lake shifts from mesotrophic to eutrophic or hypertrophic, and clear-water state shifts to turbid state, BCC also changes considerably. High nutrient levels, especially nitrogen, appeared to induce a shift in BCC. In addition, nutrient loading had indirect effects on BCC through changes in the biomass and composition of phyto- and zooplankton. The 2-year experiment also showed that the results of environmental studies can be greatly influenced by natural variation in weather conditions in different years. Nevertheless, similar types of environmental variables did correlate with BCC for both years. The dominant bacterial divisions we detected remained the same over this period, although the dominant phylotypes were different.

The occurrence of Verrucomicrobia correlated with more eutrophic conditions and Actinobacteria with less eutrophic conditions.