Soil characteristics more strongly influence soil bacterial communities than land-use type
A walkthrough based on...
Kuramae EE, Yergeau E, Wong LC, Pijl AS, van Veen JA, Kowalchuk GA (2012) Soil characteristics more strongly influence soil bacterial communities than land-use type. FEMS Microbiol Ecol. 79: 12–24
The walkthrough below illustrates the central procedures of an investigation conducted by Kuramae et al. (2011). These authors investigated the influence of land use and soil characteristics on bacterial communities sampled from soils in the Netherlands. The authors acquired five samples (each comprising five subsamples) from each of 25 fields representing six land use regimes within a two-day period with stable weather. Indirect and direct gradient analyses were used as well as non-parametric hypothesis testing procedures. The authors showed there was no overall effect of land use on the microbial community composition in their study system, however, identified several parameters that did appear to structure the communities.
Data preparation
Denaturing gradient gel electrophoresis (DGGE), PhyloChip analysis, and real-time polymerase chain reaction (RT-PCR) analysis directed towards phylogenetic marker genes were used to detect operational taxonomic units (OTUs). DGGE, a fingerprinting approach, resulted in a presence / absence table representing 66 phylotypes. PhyloChip analysis yielded relative abundance data for 2869 OTUs. In analogy to the classical "sites × species" data table, OTUs serve as the variables in a "samples × OTU" data table.
See the original text for procedures and thresholds for determining OTU presence / absence and relative abundance.
Description of OTU richness
Results from PhyloChip hybridisation were used to assess and describe OTU and taxon richness across samples. Richness was correlated with contextual parameters and significant correlations reported.
Indirect gradient analysis
Non-metric multidimensional scaling (NMDS) was used on distance and dissimilarity matrices calculated from the DGGE presence / absence data and the RT-PCR and PhyloChip relative abundance data. Hellinger distances were calculated for presense / absence data while Bray-Curtis dissimilarities were calculated for relative abundance data. The rationale for using different (dis)similarity measures is described here.
Direct gradient analysis
Canonical correspondence analysis (CCA) was used to ordinate samples based on their relation to the contextual parameters measured. This is a form of constrained or direct gradient analysis wherein only the variation that is 'accounted for' by the explanatory variables included in the analysis is used in ordination. More information and a wizard are available here. Separation of samples was connected to several contextual parameters and models built using forward selection (see "Variable selection in MLR" under multiple linear regression). Significance was tested by permutation.
Hypothesis testing: Are samples with the same land use regime more similar than those with different regimes?
Analysis of similarity (ANOSIM) was used to test whether there was significance separation between samples based on their respective land use regimes. This hypothesis test used the same distance/dissimilarity matrices as described above (Indirect gradient analysis). Significance was tested by permutation.
Examining individual OTU relationships to contextual parameters
Pearson correlations were computed between OTU abundances derived from PhyloChip analysis and the contextual parameters measured. A false discovery correction for multiple testing was applied to control Type I error rates. Individual OTUs and taxa which significantly correlated with one or more contextual parameters were described and notable results discussed.
The illustration below is a simplified representation of the procedures used by Kuramae et al. Please consult the manuscript for a more complete explanation.