Automated clustering of heterotrophic bacterioplankton in flow cytometry data

Flow cytometry has become a standard method to analyze bacterioplankton. Analysis of samples by flow cytometry is automatic, but it is followed by manual classification of the bacterioplankton groups in flow cytometry standard (FCS) files. This classification is a time consuming and subjective task performed by manually drawing the limits of the groups present in cytograms, a process referred to as gating. The automation of flow cytometry data processing based on pattern recognition techniques could provide an efficient tool to overcome some of these disadvantages. Here, we propose the use of model-based clustering techniques for the automatic detection of low (LNA) and high (HNA) nucleic acid bacterioplankton groups in FCS files. To validate our method, we compared the automatic classification with a flow cytometry database from a 9 yr time series collected in the central Cantabrian Sea that had been manually analyzed. The correlation between automatic and manual gating methods was >0.9 for cell counts and 0.7 to 0.95 for side scatter values, a proxy of cell size. In addition, no significant differences were found in the mean annual cycle of LNA and HNA cell abundance depicted by both methods. We also quantified the subjectivity of manual gating. The coefficient of variation for heterotrophic bacteria counts obtained by different analysts was around 10 to 20%. Our results suggest that the combination of flow cytometry and automatic gating provides a valuable tool to analyze microbial communities objectively and accurately, allowing us to safely compare bacterioplankton samples from different environments.