A walkthrough based on...
Halliday E, McLellan SL, Amaral-Zettler LA, Sogin ML, Gast RJ (2014) Comparison of Bacterial Communities in Sands and Water at Beaches with Bacterial Water Quality Violations. PLoS ONE 9(3): e90815.
The walkthrough below illustrates the central procedures of an investigation conducted by Halliday et al. (2014). The authors sampled water, wet sand, and dry sand from two beaches during the summer of 2007, particularly during water quality violations. The authors examined the abundances of taxa specifically linked with sewage discharge, such as the genus Enterococcus, using the EPA 1600 method and qPCR targetting the 23S rRNA gene. Guided by these counts, and the occurrences of water quality violations, Halliday et al. pyrosequenced the hypervariable regions of the 16S ribosomal DNA in order to survey bacterial communities in selected samples.
Data preparation
For most of their analyses, Halliday et al. standardised their read abundance data on a per sample basis to arrive at relative abundance data. Prior to estimating the diversity at each site, the authors subsampled their data set (see below). Guided by the GAST taxonomy assignments, the authors also created subsets of their data, creating a data set which included reads exclusive to taxa that are potential indicators of water quality violations.
Assessing changes in local diversity (alpha diversity)
The authors randomly subsampled 8,050 sequence tags - the lowest number of reads recovered in a single sample - from each sample's sequence pool to mitigate the effect of differing sequencing depth across their samples prior to calculating Shannon-Weaver diversity indices for each sample.
Examining and testing differences in between-sample diversity (beta diversity)
Halliday et al. employed non-metric multidimensional scaling (NMDS) using the Bray-Curtis dissimilarity measure on square root transformed data to visualise sample dissimilarities. This was complemented by ANOSIM performed on objects grouped a priori by differences in site, sample type and water quality violation events. They also employed the similarity percentage breakdown (SIMPER) procedure to identify OTUs that contributed most to the dissimilarity observed.
Identifying influential environmental factors
Halliday et al. used the BIOENV procedure to identify what could be the 'best' set of explanatory variables, defined by the combination of variables whose Euclidean distance matrix maximally correlates with the Bray-Curtis dissimilarity matrix derived from their response data. The results of this procedure for water, wet sand, and dry sand samples were reported and sample temperature and tidal range identified as the most influential variables in water samples, with a smaller degree of relevance to sand environments.
Comparison to existing data sets
With reference to a previous pyrosequencing study targetting sewage and wastewater samples, the authors selected sequence tags classified into the orders Clostridiales, Bifidobacteriales, and Bacteroidales, which were treated as indicators of sewage contamination. The resulting data set was compared to the sewage data to identify any associations. Halliday et al. performed another ANOSIM on this subset, with samples grouped by water quality and site, with mixed results. The authors performed a SIMPER analysis and concluded that the differences observed were primarily driven by tags classified as Bacteroidales or Clostridiales. The authors observed that a repetition of the BIOENV procedure on this data set of indicator taxa resulted in stronger correlations.
The illustration below is a simplified representation of the procedures used by Halliday et al. Please consult the manuscript for a more complete explanation.