Use of a Four-Tiered Graph to Parse the Factors Leading to Phenotypic Clustering in Bacteria: A Case Study Based on Samples from the Aletsch Glacier

Miroslav Svercel,Homayoun C Bagheri,Manuela Filippini,Valentina Rossetti,Nicolas Perony

doi:10.1371/journal.pone.0065059

Abstract

An understanding of bacterial diversity and evolution in any environment requires knowledge of phenotypic diversity. In this study, the underlying factors leading to phenotypic clustering were analyzed and interpreted using a novel approach based on a four-tiered graph. Bacterial isolates were organized into equivalence classes based on their phenotypic profile. Likewise, phenotypes were organized in equivalence classes based on the bacteria that manifest them. The linking of these equivalence classes in a four-tiered graph allowed for a quick visual identification of the phenotypic measurements leading to the clustering patterns deduced from principal component analyses. For evaluation of the method, we investigated phenotypic variation in enzyme production and carbon assimilation of members of the genera Pseudomonas and Serratia, isolated from the Aletsch Glacier in Switzerland. The analysis indicates that the genera isolated produce at least six common enzymes and can exploit a wide range of carbon resources, though some specialist species within the pseudomonads were also observed. We further found that pairwise distances between enzyme profiles strongly correlate with distances based on carbon profiles. However, phenotypic distances weakly correlate with phylogenetic distances. The method developed in this study facilitates a more comprehensive understanding of phenotypic clustering than what would be deduced from principal component analysis alone.

Highlights

For validation of phenotypic diversity and physiological functions, cultivation and characterization of single isolates is a necessary and complementary approach to community assays [1,2,3]
We developed a graphical method that summarizes the information given by a Principal Component Analysis (PCA) on a network layout: it represents in a visually-clear form the relation between strain isolates grouped into phenotypic clusters, and the resources that make up their phenotypic profile
When plotting the bacterial equivalence classes along a vertical line according to their coordinate on the PC1 axis, for clarity we introduce a small spacing between the classes that would otherwise overlap, whilst keeping the distances between non-overlapping classes unchanged

Summary

Introduction

For validation of phenotypic diversity and physiological functions, cultivation and characterization of single isolates is a necessary and complementary approach to community assays [1,2,3]. One possible way to obtain phenotypic profiles is through characterization tests such as Biolog PM and API ZYM. These methods have been often used for obtaining the combined metabolic profile of microbial communities [4,5]. The analysis becomes more difficult as the number of strains increases, limiting the size of data sets that can be handled. In order to render the analysis of such data possible, statistical methods that reduce the dimensionality of the data set are often used, such as Principal Component Analysis (PCA) [6]. PCA allows to group correlated variables associated with a set of entities (here, bacterial isolates) together into factors, which are thought to reflect the latent processes from which the correlations arise. PCA provides no easy way to understand what these processes are, and effectively to understand where the grouping between isolates originates

Methods

Results

Conclusion