The influence of habitat heterogeneity on freshwater bacterial community composition and dynamics

A walkthrough based on...

Shade A, Jones SE, McMahon KD (2008) The influence of habitat heterogeneity on freshwater bacterial community composition and dynamics. Environ Microbiol. 10(4): 1057-1067.

The walkthrough below illustrates the central procedures of an investigation conducted by Shade et al. (2008). The authors investigated the response of freshwater bacterial communities to a range of environmental variables. Three lakes with similar physicochemical characteristics but different mixing regimes (polymictic, dimictic, and meromictic) were sampled on a weekly basis throughout the open-water period (late May to early September) and along vertical temperature and dissolved oxygen gradients. Special focus was given to epi- and hypolimnion communities.

Data preparation

Dissolved oxygen and temperature measurements were averaged across both the epi- and hypolimnion of each lake's water column. Mean differences in these parameters were calculated by subtracting the average of the hypolimnion from that of the epilimnion.

ARISA profiles were grouped into OTUs based on profile alignments and normalised peak areas were converted into relative abundance values per profile. Peak areas from replicate samples were averaged prior to analysis.

Characterising habitat heterogeneity

The contextual data gathered during sampling was examined to characterise each lake's environmental variability across the sampling period. The dissolved oxygen and temperature measurements where visualised in two-dimensional maps (similar to a heatmap) ordered by depth and sampling time.

Assessing OTU richness and persistence

OTU richness, here defined as the number of distinct ARISA peaks in a given sampling unit, was compared between lakes and visualised as a Venn diagram. Histograms were used to compare 1) the average number of OTUs observed at each lake across time alongside the total number of OTUs observed and 2) the total number of OTUs in the epilimnion versus that of the hypolimnion per lake.

The authors assessed whether there were significant differences between the OTU richness of thermal layers and lakes with different mixing regimes using multiple univariate t-tests with Bonferroni correction.

The persistence of a given OTU through time was represented as the proportion of samples in which that particular OTU occurred relative to all samples collected within a given lake layer. Only OTUs which occurred in more than ~3% of samples and which passed a minimal ARISA peak area threshold were included. Persistence values were visualised using a heatmap with OTUs clustered by their persistence patterns across lake layers using hierarchical cluster analysis.

Testing for significant differences between bacterial community composition

Analysis of similarity (ANOSIM) was used to test if groups of samples (defined a priori) were significantly different from one another based on the Bray-Curtis dissimilarity measure. OTU relative abundances were standardised prior to analysis. Stratified lakes had distinct bacterial community composition between thermal layers, however, the polymictic lake showed no separation. The authors compared their ANOSIM results to the results of their indirect gradient analysis (below).

The authors used a permuted analysis of multivariate dispersion (PERMDISP2) test (Anderson, 2006), related to principal coordinates analysis (PCoA), to assess whether the dispersion (i.e. multivariate spread) of samples between groups is equivalent. This analysis is useful in detecting differences in dispersion between groups, rather than differences in centroid location. In the analysis under discussion, this is analogous to assessing whether the bacterial community composition between groups is equally variable.

Indirect gradient analysis

The differences in bacterial community composition across lakes and between layers was assessed by correspondence analysis (CA) as well as partial correspondence analysis (pCA). The pCA analyses (performed in CANOCO) sought to remove the influence of interlake differences by partialling out the effect of a dummy "lake" variable from the response data prior to indirect gradient analysis with CA. After conducting (p)CA on the epi- and hypolimnion layers, the authors correlated axes constructed by (p)CA with contextual parameters to attempt to identify explanatory variables that may account for any differences observed. Further, the first CA axis from each lake layer was subject to mean square successive difference analysis, which tests for serial correlation between successive measurements, in order to detect patterns of change across time.

Procrustes analysis was used to compare the first two axes of CAs performed on each lake layer. The authors considered the Procrustes correlation between these ordinations to represent how similar each layer's dynamics were over the sampling period. Highly correlated dynamics were treated as an indication of common drivers, while weakly correlated dynamics suggest different drivers are in operation between layers.

References

Anderson MJ (2006) Distance based tests for homogeneity of multivariate dispersions. Biometrics. 62:245–253.

The illustration below is a simplified representation of the procedures used by Shade et al. Please consult the manuscript for a more complete explanation.