Food or just a free ride? A meta-analysis reveals the global diversity of the Plastisphere

Abstract

It is now indisputable that plastics are ubiquitous and problematic in ecosystems globally. Many suggestions have been made about the role that biofilms colonizing plastics in the environment-termed the "Plastisphere"-may play in the transportation and ecological impact of these plastics. By collecting and re-analyzing all raw 16S rRNA gene sequencing and metadata from 2,229 samples within 35 studies, we have performed the first meta-analysis of the Plastisphere in marine, freshwater, other aquatic (e.g., brackish or aquaculture) and terrestrial environments. We show that random forest models can be trained to differentiate between groupings of environmental factors as well as aspects of study design, but-crucially-also between plastics when compared with control biofilms and between different plastic types and community successional stages. Our meta-analysis confirms that potentially biodegrading Plastisphere members, the hydrocarbonoclastic Oceanospirillales and Alteromonadales are consistently more abundant in plastic than control biofilm samples across multiple studies and environments. This indicates the predilection of these organisms for plastics and confirms the urgent need for their ability to biodegrade plastics to be comprehensively tested. We also identified key knowledge gaps that should be addressed by future studies.

Fig. 1. Overview of the studies and samples included in the meta-analysis.

Cumulative number of studies per year (A), study location (B), number of samples (C), relative abundance of sample type (D), and sample incubation time (E) for studies carried out in the marine, freshwater, other aquatic, and terrestrial environments. A, B Show all 50 studies whereas, C, D, and E show only studies/samples that were included. Studies for which data were not provided (neither publicly available nor provided upon request) are shown with transparent colors in A and white marker edges in B. Note that those studies shown for 2020 were already in press and available online by 5 January, 2020. See Supplementary Tables S1 and S2 for full details of all studies and samples included.

Fig. 2. nMDS plots showing uniFrac distance between samples.

nMDS plots showing weighted (A, B) or unweighted (C, D) uniFrac distance (i.e., accounting for taxon phylogeny with or without taxon abundance, respectively) calculated between all samples and shown in A and C as samples colored by environment and B and D as samples colored by study. Results of PERMANOVA and ANOSIM tests for significance between groups are shown on each plot.

Fig. 3. Average similarity between samples within a study versus between studies.

Average similarity (determined by weighted or unweighted uniFrac; i.e., accounting for taxon phylogeny with or without taxon abundance; top or bottom, respectively) between samples within a study versus between studies, with white boxes showing samples grouped by environment. Study names are colored by environment, with green, purple, blue, and orange being for marine, aquatic, freshwater, and terrestrial, respectively (as in Figs. 1 and 2).

Fig. 4. Summary of the composition, diversity and shared ASVs within sample groupings.

Similarity of the composition of microbial communities on different substrata in different environments. Samples are grouped by weighted uniFrac distance using ward linkage (dendrogram) between sample types (colored by environment) and mean community composition at the phylum level is shown. Those phyla that are grouped into ‘Other’ are phyla that are <1% mean relative abundance. Mean relative abundance of several orders that have previously been suggested to be associated with plastics and Simpsons Index of Diversity (showing median and interquartile range) for each group is also shown. The number of ASVs that are shared between different substratum types (that are >1% in relative abundance in these samples) is shown at the bottom.

Fig. 5. Classification accuracy for random forest models.

Classification accuracy (%) for random forest models (classification or regression for discrete or continuous categories, respectively) constructed for (A) all samples grouped within different metadata categories or (B) samples within each environment grouped within plastic type (general) at different taxonomic levels. Random forest models are trained using a subset of 80% of samples (chosen randomly) and classification accuracy is based on testing using the remaining 20% of samples. Figure S4 shows the top most important features at the ASV level across all metadata categories while Supplementary Section 3 shows all taxonomic levels as well as metadata categories.

Fig. 6. Differential abundance of taxa between early and late incubation times.

Heat trees showing differential abundance of taxa between early and late incubation times (up to or above 7 days, respectively) within substratum types (aliphatic, other plastic, and control biofilms) for each of the marine, aquatic and freshwater environments where samples were taken at different time points. Strength of colors indicates differential abundance, with gray indicating no significant difference (p > 0.05; Wilcoxon rank sum tests with holm-bonferroni false discovery rate correction), and strong yellow or red colors indicating that log₂ fold change is at least threefold higher in the early or late samples, respectively. Some key taxa are indicated in the empty, larger tree on the left and all are shown in Supplementary Section 4. Tests between different substrata at the same time point as well as between different substrata at all time points are also shown in Supplementary Sections 4 and 5. All terrestrial samples were collected after an unknown environmental residence time and could therefore not be included.