High resolution time series reveals cohesive but short-lived communities in coastal plankton

Abstract

Because microbial plankton in the ocean comprise diverse bacteria, algae, and protists that are subject to environmental forcing on multiple spatial and temporal scales, a fundamental open question is to what extent these organisms form ecologically cohesive communities. Here we show that although all taxa undergo large, near daily fluctuations in abundance, microbial plankton are organized into clearly defined communities whose turnover is rapid and sharp. We analyze a time series of 93 consecutive days of coastal plankton using a technique that allows inference of communities as modular units of interacting taxa by determining positive and negative correlations at different temporal frequencies. This approach shows both coordinated population expansions that demarcate community boundaries and high frequency of positive and negative associations among populations within communities. Our analysis thus highlights that the environmental variability of the coastal ocean is mirrored in sharp transitions of defined but ephemeral communities of organisms.

Fig. 1

Contrasting dynamics of plankton at different levels of taxonomic resolution. Although relative abundances of bacteria (a) and Eukarya (b) appear relatively stable when evaluated at the phylum to family level, they vary extensively and rapidly at fine-scale taxonomic resolution (OTU-level) (c), also evidenced by different rates of decline in taxonomic similarity (1-Jensen–Shannon distance) for phylum- and OTU-level with increasing time lags (d). Legends for phylum and family-level taxa are ordered from most to least abundant. Error bars in d denote standard error of three independent samples per day

Fig. 2

Example of limited period of higher average abundance of OTUs. For the diatom OTU Euk23325 and the Flavobacteria OTU Bac47519, a the 14 day moving average and b daily dynamics are shown. On short time scale (days) both OTUs are negatively correlated, whereas over longer time scales (weeks) they are positively correlated

Fig. 3

Dynamics of predicted communities and metadata across the time series. a Communities predicted by WaveClust analysis as modular units (clusters) of interacting OTUs based on wavelet decomposition to determine correlations at different temporal frequencies followed by clustering (Methods). Shown are the results from positive correlation at both low and high frequency. b Heat map displaying change in physical, biological, and chemical environmental parameters over the time series. Color scale for each environmental parameter varies between maximum and minimum values, which are for each parameter: WWTMP: 20–10 °C; ATMP: 30–10 °C; DWD: 14–4S, WS: 14–4 m/s; WL: 4.5–1.5 m; WH: 3–0.5 m; Press: 1020–995 hPa; TD: 1–0 incoming/outgoing; Malgae: 4–0 relative concentration; CCon: 10–4 µg/L; NH₄: 1.4–0 µM; NO₂-NO₃: 20–0 µM; PO₄: 0.7–0 µM; Silicate: 10–2 µM; Salinity: 36–33 psu. For more details see Supplementary Data 7, where values for each day are listed. Day 247, which marks Hurricane Earl passage, is framed in the heat map. c Granger causalities linking predicted communities to each other and to environmental parameters. Nodes represent communities (colored according to a) and environmental parameters (gray). Legends for b and c: WTMP, water temperature; ATMP, air temperature; DWD, dominant wave period; WS, wind speed; WL, water level; WH, wave height; Press, pressure; TD, tidal direction; Malgae, macroalgae; CCon, chlorophyll concentration; NH₄, ammonium; NO₂–NO₃, nitrite and nitrates; PO₄, phosphate

Fig. 4

The most closely related OTUs tend to peak in different communities suggesting ecological differentiation. Shown are overlapped histograms of pairs of representative dbc-OTUs falling within the same (blue bars) or different (red bars) communities. Green bars represent the sum of blue and red bars. When OTUs are defined with high sequence similarity (>97%), they preferentially occur in different communities, for both bacteria (a) and Eukarya (b). Observed distributions were significantly different from a random community membership assignment (χ² goodness of fit tests: χ²(1) = 3042, p-value < 2.2e−16, for bacteria; and χ²(1) = 14,735, p-value < 2.2e−16, for Eukarya)