AME 88:1-18 (2022)  -  DOI: https://doi.org/10.3354/ame01979

ABSTRACT: Increasing awareness of environmental impacts caused by anthropogenic activities highlights the need to determine indicators of environmental status that can be routinely assessed at large spatial and temporal scales. Microbial communities comprise the greatest share of biological diversity on Earth and can rapidly reflect recent environmental changes while providing a record of past events. However, they have rarely been targeted in the search for ecological indicators of habitat types, environmental conditions, or environmental changes. Here, as a proof of principle, we analysed the bacterioplankton community composition of 4 estuaries in North and South America, Europe, and Asia, and looked for indicators of groups of samples defined using partition techniques, according to primary physicochemical variables typically monitored to infer water quality. Indicator value analysis (IndVal) was conducted to identify indicator operational taxonomic units (OTUs; analogous to species in other fields of ecology) in each group. These bacterioplankton-based indicators exhibited a high capacity to predict the group membership of the samples within each estuary and to correctly assign the samples to the appropriate estuary in a combined data set, employing different machine learning techniques. The indicators were composed of OTUs belonging to several bacterial phyla, which responded significantly and differentially to the environmental variables used to define the groups of samples. Moreover, the predictive values of these bacterial indicators were generally higher than those of other biological assemblages commonly used for environmental monitoring. Therefore, this approach appears to be a promising tool to complement existing strategies for monitoring and conservation of aquatic systems worldwide.

KEY WORDS: IndVal · Machine learning · Río de la Plata · Environmental management