The energy–diversity relationship of complex bacterial communities in Arctic deep-sea sediments

A walkthrough based on...

Bienhold C, Boetius A, Ramette A (2012) The energy–diversity relationship of complex bacterial communities in Arctic deep-sea sediments. ISME J. 6: 724–732.

The walkthrough below illustrates the central procedures of an investigation conducted by Bienhold et al. (2012). The authors investigated the response of structure and diversity of bacterial communities in relation to metabolic energy availability in deep-sea sediments from the Siberian continental margin. Other contextual parameters were measured and examined in the analysis. The sampling design availed of the natural depth gradient provided by the region's continental slope. Communities were detected using automated ribosomal intergenic spacer analysis (ARISA) and 454 massively parallel tag sequencing (MPTS).

Data preparation

Prior to statistical analysis, ARISA and 454 sequences were processed and operational taxonomic units (OTUs) identified. Taxonomic annotation of 454 tags was performed and poorly annotated tags were removed. A Mantel test using Spearman's rank-correlation coefficient was used to compare the original and truncated 454 data sets. The authors reported high and significant correlations and concluded that the truncation did not distort the overall pattern in the community data.

OTU abundance data was Hellinger transformed and several contextual parameters were log10 or square root transformed prior to analysis with linear methods (e.g. redundancy analysis). Space was modeled by subjecting spatial coordinates to transformation by third order polynomials. This created several spatial variables and some were retained following forward selection (see "Variable selection in MLR" under multiple linear regression). 

Assessing OTU richness as a function of energy availability

Simple scatter plots relating OTU richness to pigment concentrations observed at each station allowed the authors to describe links between these two variables. To overcome differences in the size of 454 data sets, Chao1 richness estimates were calculated from resampled data sets of the same size as the smallest available data set. When dealing with ARISA data, the total number of OTUs was used as a measure of richness and plotted against pigment concentrations.

Assessing beta diversity as a function of enzyme activity and energy availability

Mantel tests, using Spearman's rank-correlation coefficient, on dissimilarity and distance matrices were used to assess if differences between communities are correlated to differences in 1) pigment concentrations and 2) enzyme activities observed across samples.

Disentangling the effects of factors influencing community structure

Redundancy analysis (RDA), a form of constrained analysis, was used to assess the influence of contextual explanatory variables on the OTU abundance data. A stepwise forward selection procedure (see "Variable selection in MLR" under multiple linear regression) was used to select variables which "best" explained variation in the abundance data, with significance assessed by permutation. The final model, including only selected explanatory contextual variables, was the final result of this stage of the analysis. Given this reduced model, variation partitioning (VP) using partial RDA was then used to determine how much variation in the OTU abundance data was attributable to each selected contextual parameter, controlling for the influence of the others.

Path analysis was employed to detect plausible directional relationships between selected variables. An initial model, informed by the results of variation partitioning (above), was subject to iterative refinement. Bacterial community structure, as determined by ARISA, and enzyme activity were treated as response variables and correlation with contextual variables related to energy availability, space, and sediment depth was determined using the RV coefficient. Several statistics were used to test each model evaluated through path analysis. Among these, the Bayesian information criterion was iteratively minimised to arrive at an optimal model.

Examining individual OTU relationships with contextual parameters

Pairwise univariate Spearman correlations were conducted between OTU abundances, both individually and aggregated at different taxonomic levels, and pigment concentration (selected through forward selection). Significance was determined by permutation. This procedure allows insight on the response of individual OTUs and taxa to pigment concentration, an indicator of energy availability.

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