Introduction

Fine-scale variations of conserved 16S rRNA sequences are commonly observed in microbial populations (Rocap et al., 2003; Klepac-Ceraj et al., 2004; Hunt et al., 2008). However, the ecological relevance and factors promoting diversification are not well understood. Within these closely related groups of microorganisms, an even greater level of diversity has been observed when analyzing variation among sequences of protein-encoding genes (for example, Whitaker et al., 2003; Hunt et al., 2008). This level of sequence variation is considered the population-scale diversity (within a species) and is called microdiversity. The amount of microdiversity within a population is driven by population genetic forces such as mutation, recombination, migration, selection and genetic drift. Investigating the spatial and temporal distribution of microdiversity within populations is necessary to elucidate its ecological importance, mechanisms of microbial speciation, and diversification patterns (Achtman and Wagner, 2008; Fraser et al., 2009).

Small freshwater lakes are an ideal system to investigate microbial population dynamics because the spatial boundaries are easily identified and seasonal ecosystem dynamics have been well characterized. At the community level, bacteria inhabiting freshwater lakes are highly dynamic (Eiler and Bertilsson, 2007; Shade et al., 2007; Nelson, 2009) and diverse (Zwart et al., 2002; Shade et al., 2007). Differences in community composition are driven in time by seasonal fluctuations in extrinsic and intrinsic variables (Eiler and Bertilsson, 2004; Shade et al., 2007), and in space by regional factors and landscape position (Yannarell and Triplett, 2005). Recently it has been shown that the bacterial community composition in the two thermal layers of the water column in a stratified lake (epilimnion and hypolimnion) experiences different levels of variation and dynamics during one summer season (Shade et al., 2008). The epilimnion is the upper, warmer water layer during summer stratification. The availability of light in this layer initiated primary production and interactions between phytoplankton and the bacterial community (Kent et al., 2004). In contrast to the epilimnion, the bacterial community in the cooler and deeper water layer, the hypolimnion, was largely unaffected by extrinsic drivers (Shade et al., 2008). The apparent stability of the hypolimnion community during the summer months may result from the absence of a strong driver for successional change.

Over the course of the seasons, turnover of the water column can induce variation in the microbial communities of freshwater lakes. The water column in dimictic lakes mixes in spring and fall as a consequence of differential warming and cooling of the epilimnion versus the hypolimnion. Mixing events disrupt established chemical gradients in the water column and combine two otherwise isolated layers of water. Following the mixing event, the water column re-stratifies during summer and winter. Jones et al. (2008) showed repeated re-sets in the temporal development of the bacterial community in the epilimnion of a Taiwanese freshwater lake after a series of typhoons mixed and destratified the lake. Shade et al. (2008) argue that disturbance by mixing may also have an effect on hypolimnion communities. In contrast to dimictic lakes, meromictic lakes are permanently stratified and do not experience seasonal mixing. The variation in the frequency of mixing events between meromictic and dimictic lakes allows us to test the effects of the disturbance introduced by mixing on microbial populations.

In this study, we examine the spatial and temporal dynamics of closely related methanogen populations in a series of humic bog lakes in northern Wisconsin, USA. Methanogens are strictly anaerobic members of Euryarchaeota. We hypothesize that these strict anaerobes would be sensitive to mixing events that introduce oxygen traces into the anaerobic hypolimnion of bog lakes. In methanogens, the functional gene mcrA encodes the α-subunit of methyl-coenzyme M reductase. This methanogen-specific reductase is a key enzyme in methanogenesis (Ermler et al., 1997) and useful for the non-culture-based assessments of diversity. Using the mcrA gene, the distributions of methanogen species have been surveyed in anaerobic digesters (Leclerc et al., 2004), peat bogs (Juottonen et al., 2006; Bräuer et al., 2006b), soils (Lueders et al., 2001; Luton et al., 2002), sediments from freshwater lakes (Banning et al., 2005) and marine environments (Dhillon et al., 2005). Methanogens have also been identified in the water column of freshwater lakes (Earl et al., 2003; Lehours et al., 2005, 2007) where sequences of the orders Methanosarcinales and Methanomicrobiales dominated clone libraries. Galand et al. (2002) characterized a Methanomicrobiales-associated cluster of sequences that appears to be globally significant in nutrient-poor acidic environments like peat bogs and the humic bogs investigated here. Bräuer et al. (2006a) have sequenced the genome of the first isolate from this cluster, candidatus Methanoregula boonei 6A in which the mcrA gene is found in single copy.

Here we test (1) how methanogen populations differ between five humic bog lakes, (2) whether methanogen populations change across seasons of 1 year including mixing events and (3) how the population structure compares between the lakes with different mixing regimes.

Materials and methods

Description of the sampling sites

The methanogen populations in five dystrophic humic bog lakes in Vilas County in northern Wisconsin, USA, were investigated. Two sets of bog lakes separated by 30 km were studied; ‘Mary Lake’ (MA) and ‘Rose Lake’ (RL) from the Adelaide–Yolanda group of lakes (Juday and Birge, 1941) and ‘South Sparkling Bog’ (SSB), ‘North Sparkling Bog’ (NSB) and ‘Trout Bog’ (TB) in the Trout Lake area of the North Temperate Lakes Long Term Ecological Research Station. The five lakes represent two mixing regimes: RL, NSB, SSB and TB are dimictic (mixing twice a year); MA is meromictic (permanently stratified and never mixed). Lake characteristics are specified in Table 1. All lakes contain darkly stained, acidic water with an approximate average Secchi depth of 1 m. The geology of the area is dominated by quaternary sediments deposited during the Wisconsin glaciation (Clayton et al., 2002). The immediate shore vegetation is predominantly Sphagnum spp. and Vaccinium spp. or adjacent coniferous forests.

Table 1 Lake characteristics
Full size table

Sample collection

Samples from different subsets of lakes were collected at four time points between October 2007 and July 2008 (Figure 1). Integrated water samples were collected at the deepest point of the lakes from the epi- and hypolimnion using connected 2 m PVC pipe segments with stop valves at 1 m intervals. To determine the location of the thermocline separating the hypolimnion and epilimnion, temperature and dissolved oxygen measurements were taken at 0.5–1 m intervals throughout the water column using a YSI 550 dissolved oxygen meter (Yellow Springs, OH, USA) (Supplementary Figures S1–S5, Supplementary materials). Samples were collected in ethanol-washed Nalgene containers and immediately transported to the lab for further processing.

Figure 1
figure 1

Timeline indicating sampling times (1–4), ice cover and approximate times for lake mixing (hatched) in 2007–2008. Lake mixis only applies to dimictic lakes. NSB mixed earlier than the other dimictic lakes in fall 2007 before sampling date (1).

Full size image

Particulate matter, including microbial cells in the water samples, was collected from 150–200 ml sample volume by vacuum filtration in ethanol-washed Nalgene filter holder assemblies onto 47 mm 0.2 μm Supor-200 nylon membrane filters (Pall Life Sciences, Ann Arbor, MI, USA). The filters were temporarily stored at −20 °C until transported on ice to storage at −80 °C.

DNA extraction

Two extractions were performed per membrane filter (2 × one half of a filter) and kept separated as replicates. Each nylon membrane was aseptically cut into pieces of an approximate size of less than 10 mm2 per piece and subjected to 10 min bead beating on a flat bed vortex followed by DNA extraction using the column-purification MO BIO UltraClean Fecal DNA kit (Mo Bio Laboratories Inc., Carlsbad, CA, USA) according to the manufacturer's instructions.

PCR, cloning and sequencing

The primer pair 1AF and 1100 AR (Hales et al., 1996) with good coverage of methanogenic archaea (Cadillo-Quiroz et al., 2006) was used for 16S rRNA gene amplification in a PCR containing final concentrations of 4 mM MgCl2, 0.3 mM dNTPs (each), 300 nM primers and 1.25 U of GoTaq flexi polymerase (Promega Corporation, Madison, WI, USA). Initial denaturation was done for 3 min at 94 °C; melting for 1 min at 94 °C, annealing at 54 °C for 1.5 min, extension at 72 °C for 1.5 min and a final extension at 72 °C for 10 min after 29 cycles.

To amplify the mcrA gene, primers mcrF and mcrR (Luton et al., 2002) were used in a PCR containing final concentrations of 1.5 mM MgCl2, 0.2 mM dNTPs, 200 nM primers and 0.6 U of Phusion High Fidelity DNA polymerase F-530 (Finnzymes, distributed by New England Biolabs Inc., Ipswich, MA, USA). Initial denaturation was done for 3 min at 98 °C; melting for 30 s at 98 °C, annealing at 59 °C for 40 s, extension at 72 °C for 1 min and a final extension at 72 °C for 10 min after 30 cycles.

Phusion PCR products were cloned into Invitrogen pCR II-Blunt TOPO vector using the Zero Blunt TOPO PCR Cloning Kit (Invitrogen, Carlsbad, CA, USA) at the maximum recommended incubation time of 30 min. Taq PCR products were cloned into an Invitrogen TOPO TA Cloning Kit (Invitrogen) according to the manufacturer's instruction. All clones were directly submitted for high-throughput sequencing to the WM Keck Center for Comparative and Functional Genomics at UIUC.

qPCR for specific mcrA sequence types

Three primer pairs designated ‘MA high F’ (5′-3′ TATACCAGCTACGGTGTGGACTAC), ‘MA high R’ (5′-3′ AGTCGGGTACTCCTCGTACTGCT); ‘SSB high F’ (5′-3′ TCGACGACTACACCTACTACGG), ‘SSB high R’ (5′-3′ CATAGAGCGTGACTTCAGTTGC); ‘MA.SSB low2.0 F’ (5′-3′ AGCAGCATCAGGTATCTCCTGT), ‘MA.SSB low2.0 R’ (5′-3′ ATACAGCCTTCGTCGGGTCT) were designed to target highly abundant genotypes in the May clone libraries of MA and SSB.

To test the specificity of the qPCR primer pairs, PCR were performed using as DNA templates a selection of clones from our clone libraries with a range of mismatches to the target sequences. PCR mixes with a total volume of 30 μl contained final concentrations of 1 × GoTaq PCR buffer, 2 mM MgCl2, 0.2 mM dNTPs, 0.2 mM of both forward and reverse primers, and 0.03 U GoTaq Flexi DNA polymerase (Promega). Cycling conditions consisted of 3 min of incubation at 94 °C followed by 30 cycles of alternating temperatures of 94 °C for 30 seconds, 63 °C for 30 s, and 72 °C for 1 min followed by a final elongation at 72 °C for 10 min. Amplification was only detected for clones with less than five mismatches near the 5′ end, and less than one mismatch near the 3′ end of either primer. All sequence types in the clone libraries for the May samples of MA and SSB that fell within the above mismatch criteria were considered to be potentially amplified by the primer pairs when estimating sequence abundance based on clone libraries as in Table 3.

qPCRs were carried out using IQ SYBR Green Supermix (Bio-Rad, Hercules, CA, USA) in a Mastercycler ep Realplex (Eppendorf, Westbury, NY, USA). Clones from each representative sequence were used to generate DNA standards for standard curves using PureLink Quick Plasmid Miniprep Kit (Invitrogen) and quantified using a Nanodrop ND-1000 (Thermo Scientific, Wilmington, DE, USA). Ten-fold serial dilutions from 109 copies to 103 copies of uncut plasmid were used for standard curve generation. qPCR was run with the conditions described above for 40 cycles. The subsequent melting curve generation consisted of 94 °C for 30 s followed by a 20 min stepwise increase of temperature from 63 °C to 94 °C. Each sample was run in triplicate wells. All samples were run on the same 96-well plate.

Sequence analysis

Sequences were automatically trimmed of vectors, manually checked for sequencing errors using Sequencher 4.7 (Gene Codes Corporation, Ann Arbor, MI, USA), and manually aligned. We determined the combined error rate for Phusion PCR and sequencing in our hands by re-processing a known clone. After PCR amplification of a single clone with the Phusion enzyme, the product was cloned once more and 80 clones were sequenced. We observed no PCR and sequencing error when re-sequencing 30 kbp. For the total number of 507 kbp sequenced in this study, assuming a conservatively estimated error rate of 1/30 kpb, approximately 17 erroneous bases resulting from PCR and sequencing error can be expected. We observed 25 sequences containing one unique single nucleotide polymorphisms. These individual bases were changed to an ‘N’ in order to avoid counting possible PCR and sequence error as diversity. Sequences containing multiple unique bases were left unchanged because the likelihood of multiple PCR and sequence errors in the same sequence is low. In addition, Bellerophon (Huber, 2004) was used to check for chimeric sequences that could form during PCR. No potential chimeras with 100% identity to either parental sequence were identified in our dataset indicating an absence of detectable PCR based chimeras. The R software environment (R development core team, 2009) with the package APE (Paradis et al., 2004) was used to calculate a distance matrix for all unique sequences based on the proportion of nucleotide sites that differ (p-distance). Gaps and ambiguous positions (N) were deleted in each pairwise comparison.

The R function hclust was used for hierarchical clustering according to the furthest neighbor clustering criteria applied to the distance matrix. We used Arlequin 3.1 (Excoffier et al., 2005) for performing FST to test for differentiation among samples (Wright, 1951; Hudson et al., 1992). DOTUR (Schloss and Handelsman, 2005) was used for calculating rarefaction curves for each lake after calculating p-distance matrices in R.

Construction of phylogeny

ARB (Ludwig et al., 2004) was used for aligning 16S rRNA gene sequences. A maximum likelihood phylogeny was constructed from 74 unique 16S rRNA gene sequences and 9 reference sequences using Garli version 0.96 with the default settings (Zwickl, 2008). The analysis was repeated six times and similar trees were obtained. Bootstrap support was based on more than 1500 replicates.

Distribution of diversity within populations

Subsampling was done to adjust for the difference in available clone sequences per bog lake. Sequences for all time points were pooled for each lake and then replicate subsamples of 100 sequences were drawn without replacement. All pairwise genetic distances of the subsamples with 100 sequences were calculated and binned at intervals of 0.001. Gaps and ambiguous positions (N) were excluded from the overall length in all pairwise comparisons. The frequency of each binned pairwise distance value was calculated as a fraction of the total number of binned pairwise distances. The subsampling was repeated 1000 times, and the mean frequencies and standard deviations are recorded. The distribution of genetic distance within and between clades (A and B in Figure 6) were determined as the maximum and minimum nucleotide identity from all pairwise comparisons within each lake using MEGA 4 (Kumar et al., 2008).

Nucleotide sequences accession numbers

All 263 unique mcrA gene sequence types are available at GenBank under Accession numbers GU084829–GU085091. Abundance data for the sequence types and the sequences in fasta format are available in separate files in the Supplementary Materials.

Results

16S rDNA and mcrA sequences of methanogens from humic bogs

Ninety 16S rDNA clones of an average length of 1041 bp were sequenced from the October hypolimnion samples from MA, TB and SSB. Out of 90 sequences, 74 were unique sequence types. The phylogenetic relationships between the unique types and select reference sequences are shown in Figure 2. Based on this phylogeny, candidatus Methanoregula boonei 6A8 (Bräuer et al., 2006a) is the most closely related cultured representative to the majority of the sequences retrieved from the bog lakes with 96.3%±0.6 (mean±standard deviation) nucleotide identity between it and the clones within its nearest bootstrap-supported clade.

Figure 2
figure 2

Likelihood tree for 16S rDNA reference sequences (accession numbers in parentheses) and environmental sequences from October samples of MA, TB, and SSB. For the environmental sequences, the number of unique sequence types per clade is given in bold followed in parentheses by the total abundance of sequences in that clade. Bootstrap support (%) is added to major nodes when greater than 70%. Scale bar represents 0.05 changes per nucleotide position.

Full size image

To resolve population structure at a finer scale, we sequenced 1222 clones of the protein-coding methanogen gene mcrA that we obtained from 15 water samples (Table 2). The average length of these sequences was 415 nucleotides. Each hypolimnion sample contained partial mcrA sequences indicating the presence of methanogen DNA in the anaerobic water column. Amplification of epilimnion DNA yielded no PCR product with the exception of one MA sample in January when the epilimnion was anaerobic.

Table 2 Characteristics of mcrA clone libraries.
Full size table

Figure 3a shows the sampling curve calculated for each set of lake samples with no binning of sequences. Sampling was incomplete when considering each unique sequence type. However, when sequences were binned at the 97% nucleotide identity level, curves began to become saturated, indicating that methanogen diversity was well sampled at this scale (Figure 3b).

Figure 3
figure 3

Rarefaction curves for pooled samples from each lake, binned at levels of nucleotide identities of 100% (a) 97% (b).

Full size image

Despite the larger information content of the more variable mcrA gene, we could not resolve a stable phylogeny within the candidatus M. boonei clade containing the majority of our sequences. Although many clusters of sequences could be resolved, their relationships to each other could not be determined. Because relationships among clades could not be resolved, we clustered sequences using hierarchical clustering by nucleotide identity. In Figure 4 we show the hierarchical clustering of those 66 of the 263 unique sequence types that were observed more than twice in our clone libraries to qualitatively identify patterns in space and time for the most dominant sequence types. This set of 66 unique sequence types represents 79% of the total 1222 clone sequences sampled. As we observed in the phylogenetic analysis using the 16S rRNA gene, the majority of the mcrA clone sequences (clusters 1, 2 and 3 in Figure 4) were most closely related to the candidatus Methanoregula boonei 6A8 mcrA sequence with an average nucleotide identity of 83.6%±1.3 (mean±standard deviation).

Figure 4
figure 4

Distribution of mcrA sequence types and abundance among clone libraries. The dendrogram to the left hierarchically clusters the sequence types by their genetic distance. The heatmap to the right indicates the relative abundance of each unique sequence type according to the scale. Persisting genotypes are defined as present at all time points in specific lakes and appear boxed in the heatmap. Only clones with an absolute abundance of more than 2 clones are displayed in this figure. The inset table at the bottom of the dendrogram shows the relative frequency of clones belonging to each cluster from each lake in the complete data set. *Indicate target sequences selected for qPCR the assay. Methanosaeta thermophila PT (Mthe_0569), Methanocella sp RC1 (YP_686530), Methanosarcina acetivorans (AAM07885), cand. M. palustris E19c (EU296536), and cand. M. boonei 6A8 (Mboo_0582).

Full size image

Spatial and temporal patterns

As shown in Figure 4, three primary clusters of sequence types were observed that are differently distributed among the five lakes. The majority of sequences from the bog lakes of the Trout area were found within clusters 1 and 2, whereas the majority of MA sequences populate cluster 3. Sequences in cluster 3 were most similar to the mcrA gene sequences of the Methanomicrobiales-associated Methanoregula-cluster in Bräuer et al. (2006a). Because only one sample was analyzed for RL, the majority of unique sequences from this bog (16 of 35) are not shown in Figure 4. These sequence types are accounted for in the inset table in Figure 4 and can be found mostly in cluster 2. Only three sequences were shared by one or more of the Trout area bog lakes and both of the Adelaide–Yolanda lakes.

Each bog lake contained a unique set of 3–8 persisting sequence types that were present at all sampling times (Table 2, Figure 4). The persisting types contained the most frequently sampled sequence types in each lake (Table 2). Only NSB and SSB shared persisting sequence types with another lake (TB) but not with each other. The six persisting sequence types in MA accounted for 49% of all MA clone sequences and were more closely related to each other (95.9% nucleotide identity) than the persisting sequences within the other lakes (78.8% TB, 86.2% NSB, 87.4% SSB). The most abundant persisting type in SSB dominated each clone library by contributing more than 50% of all clone sequences from this lake (Table 2). The persistence of sequence types through time also indicates a spatially homogenous distribution of methanogen sequences within a single lake.

qPCR was used to independently verify that the abundance patterns in our clone libaries did not result from biases in PCR amplification or cloning. We selected two clone libraries (May samples from MA and SSB) in which the highly abundant sequence type in MA was absent in SSB and vice versa. Two primer pairs (‘MAhigh F/R’ and ‘SSBhigh F/R’) target the highly abundant genotypes in MA and SSB, but in addition also amplify a range of closely related sequence types from the respective libraries that occur in low abundance according to the clone libraries. A third primer pair was designed to target sequences that were at low abundance in both MA and SSB (MA.SSBlow). qPCR results confirmed that highly abundant sequences in May samples of MA and SSB were not abundant in the other respective clone library. For both bog lakes, the highly abundant sequence types were present at approximately 5-fold higher concentrations than low abundant sequence types in the same sample (Table 3). This relationship was similar to the relative abundances inferred from the clone libraries (Table 3). Results from the qPCR assay therefore support the relative abundance pattern we observe from the clone libraries.

Table 3 Ratios based on qPCR and clone libraries between high and low abundance sequence types
Full size table

Using FST we tested for differentiation between lakes and among samples taken at different times from the same lake. The matrix of multiple pairwise FST values is displayed as a heatmap in Figure 5. Theoretically, FST values range between 0 (no differentiation) and 1 (complete differentiation). To test the importance of clone abundance for driving differentiation, we calculated FST values based on the presence–absence of unique sequence types (Figure 5 top), in addition to the whole data sets including clone abundance information (Figure 5 bottom). All FST values >0.04 were significant (P<0.05). The numeric data for Figure 5 is available in Supplementary Table S1 in the Supplementary materials. Differentiation was also measured using UniFrac (Lozupone et al., 2006). This analysis agreed with our results from the FST despite a poorly supported phylogeny (data not shown).

Figure 5
figure 5

FST for all multiple pairwise comparisons of 15 samples. The lower triangle of the matrix represents FST values for the analysis of the clone abundance pattern. The upper triangle gives FST data when only presence–absence of unique sequence types was considered. The numeric matrix of the FST values can be found in Supplementary Table S1 in the Supplementary materials. *Three comparisons within SSB and TB, where FST values were relatively high.

Full size image

FST quantitatively confirmed the qualitative patterns observed in Figure 4. MA was differentiated from all dimictic bog lakes including RL with an average FST of 0.32±0.10 when accounting for the abundance distribution of sequence types. Very low FST values were observed among the Trout area bog lakes (0.07±0.06) with differentiation resulting primarily from differences in abundance of individual clones rather than the types of sequence present in the samples (average FST 0.01±0.06 calculated only with presence/absence of sequences). FST values between MA and RL and between RL and the bog lakes from the Trout area were similar and lower than between MA and the Trout area bog lakes indicating that dimictic RL is intermediate in its level of differentiation. This is because RL is dominated by sequences from cluster 2 (Figure 4), which shares members with the Trout area bog lakes while in MA sequences from cluster 3 predominate.

Comparing temporal samples from each bog lake, FST values ranged from 0.0 to 0.07 when accounting for abundance information, indicating no differentiation among samples taken at different times. The exceptions were three comparisons within SSB and TB, where FST values were relatively high (0.11, 0.12 and 0.17, respectively, marked with asterisks in Figure 5 bottom). Considering only composition (presence–absence information, Figure 5 top), the respective comparisons showed average FST values between 0.00 and 0.01 and thus no signal for differentiation. Therefore, changes over time within SSB and TB related to differences in relative sequence abundance but not in observed sequence types.

Distribution of diversity within lake populations

In Figure 6, we show the frequency distribution of nucleotide identity for all pairwise comparisons of sequence types within each bog lake. Mean and standard deviation were estimated using random subsampling of 100 sequences to account for differences in the number of clone sequences per sample. Sequences from different time points were pooled to represent the diversity of the lake through time because of the temporal consistency observed in Figures 4 and 5. From this frequency distribution we observed differences in the structure of diversity among the different lakes. For MA (Figure 6), an arrow in the figure identifies a region of high microdiversity resulting from highly similar but not identical sequences with nucleotide identities of greater than 96%. Several nucleotide changes within this group were non-synonymous resulting in an average amino acid identity of 97%. For MA, 22% of all pairwise comparisons were greater than 96% nucleotide identity as opposed to 8%, 3% and 7% for TB, NSB and SSB, respectively. In the Trout area bog lakes (TB, NSB and SSB; Figures 6), we observed a higher frequency of pairwise genetic distances with lower percent identity resulting from comparisons between clusters 1, 2 and 3 (Figure 4) and with sequences outside of the three clusters (<83% nucleotide identity, B bracket in Figure 6). The dimictic RL from the Adelaide area showed a composite structure with high levels of microdiversity (arrow in Figure 6 18% of comparisons with greater than 96% nucleotide identity) as well as more divergent types. The distribution for SSB was unique with a high frequency of a single dominant clone sequence. The observed microdiversity in MA and RL samples were not due to PCR and sequencing errors because single nucleotide polymorphisms were removed from the analysis, and because PCR and sequencing errors should be consistent among samples from all lakes. The microdiversity observed in Figure 6 did not result from the higher number of clones sequenced for MA because we compensated for sample size by displaying the mean frequencies of pairwise distances in subsamples of 100 clones. The difference in microdiversity between the methanogen populations in the five bog lakes was also prominent in the change between the rarefaction curves for MA and RL when closely related sequences are binned (Figures 3a and b). Even though the observed number of unique genotypes in NSB, MA and RL was similar (Table 2), the structure of the methanogen populations for these three lakes was different (Figure 6).

Figure 6
figure 6

Distribution of pairwise nucleotide identity for the five lakes and a scaled plot for SSB. The frequency of each binned pairwise nucleotide identity value is shown as a fraction of the total number of pairwise distance values. Dark bars indicate the standard deviations of each frequency value of a random subsample of 100 clones over 1000 repetitions. The graph for RL does not contain bars for standard deviations due to a low number of clones sampled from only one time point. The y axis of SSB (scaled) is scaled to a maximum value of 0.1 to better show the smaller peaks. Arrows on two of the peaks in SSB (scaled) indicate values that extend beyond the scaling of the graph. The brackets indicate the range of pairwise identities seen when comparing sequences within each of the three clusters in Figure 4 (‘A’) and between the three clusters (‘B’). The arrows in MA and RL point out the higher level of microdiversity seen in these lakes.

Full size image

The distribution of diversity observed in Figure 6 is reflected in the average nucleotide identity shown in Table 2 and plotted as a function of lake depth in Figure 7. The deepest lakes RL and MA had the highest average nucleotide identity while the shallower dimictic lakes had higher levels of divergence. As shown in Figure 7 there was a strong positive relationship between depth and diversity in these lakes (R2=0.98), where the single-clone-dominated SSB is considered an outlier and disregarded for the linear fit.

Figure 7
figure 7

Average pairwise nucleotide identity within each lake for methanogen mcrA sequences versus the maximum depth. SSB is an outlier because of the high abundance of a single clone sequence. Error bars are standard errors for the estimates of genetic distance. The R2 of 0.98 does not include the point for SSB.

Full size image

Discussion

Spatial structure among bog lakes

Methanogen populations in each of five humic bog lakes are unique, differentiated by persisting genotypes in each population. The observed differentiation may result from selection due to differing environmental conditions in the bog lakes, low migration rates or a combination of these two forces. Despite attempts to sample bog lakes with similar environmental conditions, the environmental characteristics we could measure differentiate bog lakes in the Adelaide area from the Trout area. The Adelaide area bog lakes MA and RL have a higher maximum depth and a higher average pH (Table 1). In addition, the quaternary deposits surrounding the Trout and Adelaide area bog lakes are slightly different (Clayton et al., 2002). These and other potentially important environmental differences, for example in quality and quantity of dissolved organic matter, may drive the differentiation among Adelaide and the Trout Bog area lakes. Consequently, as in many other studies in microbial ecology, it is difficult to quantify the relevant physical and chemical parameters that result in differentiation on the microbial scale. Despite the geographic similarities between RL and MA, the dimictic RL has an intermediate level of differentiation by FST between MA and the Trout area bog lakes caused by the dominance of sequences from cluster 2. In addition, RL contains both microdiverse sequences and a relatively higher number of divergent sequences than MA. This hybrid pattern of diversity in RL suggests that mixing regime, in conjunction with other physical or biogeochemical characteristics (geographic distance, pH and depth), influences the methanogen population that is present.

Even in the absence of selective drivers, differentiation of methanogens between bog lakes may occur simply by genetic drift with low migration. Given the toxicity of oxygen to methanogens, migration rates between the anaerobic hypolimnia of any two bog lakes are likely lower than for epilimnetic aerobic microorganisms as discussed by Jones and McMahon (2009). The difference between lakes could be interpreted as an indicator of random colonization from a larger and diverse metapopulation (Curtis and Sloan, 2004). However, temporal persistence of different genotypes in each of the five bog lakes suggests that colonization is not an ongoing process but instead, low migration rates decouple local bog populations from one another.

Isolation among nearby natural populations from similar environments is found in soil environments as well. Grundmann and Debouzie (2000) showed spatial separation of ammonia and nitrite oxidizing communities on the millimeter scale in soil samples. Chantratita et al. (2008) found in soil samples taken only meters apart from each other uniquely composed populations of Burkholderia pseudomallei differing in composition as well as abundance distribution. Our data in accordance with these studies show that physical separation and low migration may be important drivers for diversification at a local spatial scale.

Temporal persistence of methanogen sequence types in dynamic environments

Persistence, defined as the continuous occurrence of a sequence type despite environmental changes, has also been observed at the community level in dynamic ecosystems. For example, persistent methanogen 16S rDNA sequences were observed in dynamic soil and engineered systems (Lueders and Friedrich, 2000; Zumstein et al., 2000; Leclerc et al., 2001; Collins et al., 2003; Pender et al., 2004). In contrast, several studies have shown population-level changes in response to seasonal variation in conditions such as temperature (Wu and Hahn, 2006; Hunt et al., 2008), viral lysis (Middelboe et al., 2001), and resource availability (Rocap et al., 2003). Our results show stability at the population scale in bog lakes as we find several persisting mcrA sequence types continuously present at all sampling times (Figure 4). Our current sampling interval of approximately 3 months would not have been sufficient to accurately quantify an immediate short-term response of the methanogen populations to a mixing disturbance if it occurs. Small changes in rare sequence types cannot be detected even at a higher temporal resolution with limited sequencing capacity. However, the similar methanogen populations at a seasonal sampling interval show that if disturbances caused by lake mixing occurred, populations are resilient, quickly reverting to the original abundance distributions.

Although it does not affect the validity of the spatial and temporal patterns described here, the activity of methanogen populations in the water column should be considered. With exceptions as described for the meromictic Knaack Lake in central Wisconsin, USA (Winfrey and Zeikus, 1979), methane measurements based on concentration profiles through the water column generally suggest little or no methanogenic activity outside the sediment. Therefore, the presence of mcrA sequences in the hypolimnion waters may be explained by passive transport into the water column for example through ebullition of biogas formed in the sediment body or resuspension of sediment particles. In this case, hypolimnectic methanogen populations may then mirror methanogen populations in the sediments of the bog lakes. The persistence of sequence types in our data would then point towards a temporally stable population structure of methanogens in the sediment. It should be noted that PCRs with methanogen-specific primers yielded no product with aerobic epilimnion samples, which would be expected if the presence of methanogens in the water column was indeed only because of transport phenomena.

If we have sampled an active methanogen population in the water column, the relative stability over time suggests that methanogens in dynamic environments of bog lakes have developed mechanisms to survive the short-term effects of disturbances, such as oxygen spikes in the hypolimnion. Physiological adaptations to low oxygen concentrations have been suggested for various methanogens. For example, the genomes of Methanocella sp RC-1 (Erkel et al., 2006), Methanosarcina barkeri (Maeder et al., 2006), candidatus Methanosphaerula palustris E1-9c (Cadillo-Quiroz et al., 2008) and candidatus Methanoregula boonei 6A8 (Bräuer et al., 2006a) contain catalase and peroxidase genes. The activity and specificity of methanogen populations for the water column is an interesting topic for future research.

Differences in the structure of diversity

Samples from the deeper MA and RL stand out from the Trout area bog lakes, because they contain microdiversity i.e. an abundance of clone sequences differing only in a few base positions. In contrast to the Adelaide area bog lakes, in particular NSB and TB harbor more genetic diversity between clones and few closely related sequences. Based on these data we hypothesize that the isolated and undisturbed water column in the deeper lakes promotes microdiversification resulting from neutral mutations, whereas competition between genotypes may be driving diversification in the dimictic bog lakes of the Trout area. The degree of microdiversity may then be a function of the disturbance frequency (Connell, 1978), which is driven by the depth of the bog lake (Wetzel, 2001), when more shallow lakes experience complete mixis more regularly than deeper dimictic lakes such as RL. Other depth-associated factors such as productivity, habitat volume, microbial community composition and diversity and overall population size may contribute to the correlation between depth and genetic diversity identified here. In contrast, the unique population structure of SSB with one dominant genotype may result from a recent selective sweep, and recolonization by a small population, or overall smaller population. The intriguing correlation between average nucleotide identity within each lake versus the maximum depth of the sampling site in Figure 7 encourages further research to test the hypothesis that with increasing depth, the average nucleotide identity increases because of more extensive microdiversification in contrast with more divergent methanogen sequences.

Conclusions

Methanogenic archaea were detected in the anaerobic hypolimnion of humic bog lakes. Spatial structure exists among lake populations, which may have resulted from migration barriers or selection allowing populations within each lake to diverge independently. Each lake harbored a unique set of abundant sequence types that were present at all sampling times. Fall and spring turnover of the dimictic lakes did not change the composition of the methanogen populations. Based on our data, we hypothesize that differences in the distribution of diversity within populations from meromictic lakes and shallower dimictic lakes may be driven by differences in mixing regime (for example, the frequency and strength of disturbance) or other depth-associated factors.