Determinants of community structure in the global plankton interactome

TLDR: It is found that environmental factors are incomplete predictors of community structure and associations across plankton functional types and phylogenetic groups to be nonrandomly distributed on the network and driven by both local and global patterns.

Abstract: Species interaction networks are shaped by abiotic and biotic factors. Here, as part of the Tara Oceans project, we studied the photic zone interactome using environmental factors and organismal abundance profiles and found that environmental factors are incomplete predictors of community structure. We found associations across plankton functional types and phylogenetic groups to be nonrandomly distributed on the network and driven by both local and global patterns. We identified interactions among grazers, primary producers, viruses, and (mainly parasitic) symbionts and validated network-generated hypotheses using microscopy to confirm symbiotic relationships. We have thus provided a resource to support further research on ocean food webs and integrating biological components into ocean models.

Full text

Title: Top-down determinants of community structure in the global
plankton interactome
Authors: Gipsi Lima-Mendez
1,2,3,†
, Karoline Faust
1,2,3,†
, Nicolas Henry
4,5,†
, Johan
Decelle
4,5
, Sébastien Colin
4,5,6
, Fabrizio Carcillo
2,3,7
, Samuel Chaffron
1,2,3
, J. Cesar
Ignacio-Espinosa
8
, Simon Roux
8
, Flora Vincent
2,6
, Lucie Bittner
4,5,6
, Youssef Darzi
2,3
,
Jun Wang
1,2
, Stéphane Audic
4,5
, Léo Berline
9,10
, Ana M. Cabello
11
, Laurent
Coppola
9,10
, Francisco M. Cornejo-Castillo
11
, Francesco d'Ovidio
12
, Luc De
Meester
13
, Isabel Ferrera
11
, Marie-José Garet-Delmas
4,5
, Lionel Guidi
9,10
, Elena
Lara
11
, Stéphane Pesant
14,15
, Marta Royo-Lonch
11
, Guillem Salazar
11
, Pablo
Sánchez
11
, Marta Sebastian
11
, Caroline Souffreau
13
, Céline Dimier
4,5,6
, Marc
Picheral
9,10
, Sarah Searson
9,10
, Stefanie Kandels-Lewis
16
, Tara Oceans coordinators
‡
,
Gabriel Gorsky
9,10
, Fabrice Not
4,5
, Hiroyuki Ogata
17
, Sabrina Speich
18,19
, Jean
Weissenbach
20,21,22
, Patrick Wincker
20,21,22
, Gianluca Bontempi
7
, Silvia G. Acinas
11
,
Shinichi Sunagawa
16
, Peer Bork
16
, Matthew B. Sullivan
8
, Chris Bowler
6,*
, Eric
Karsenti
6,16,*
, Colomban de Vargas
4,5,*
and Jeroen Raes
1,2,3,*
.
Affiliations:
1
Department of Microbiology and Immunology, Rega Institute KU Leuven, Herestraat 49, 3000 Leuven, Belgium.
2
VIB Center for the Biology of Disease, VIB, Herestraat 49, 3000 Leuven, Belgium.
3
Laboratory of Microbiology, Vrije Universiteit Brussel, Pleinlaan 2, 1050 Brussels, Belgium.
4
CNRS, UMR 7144, Station Biologique de Roscoff, Place Georges Teissier, 29680 Roscoff, France.
5
Sorbonne Universités, UPMC Univ Paris 06, UMR 7144, Station Biologique de Roscoff, Place Georges Teissier, 29680
Roscoff, France.
6
Ecole Normale Supérieure, Institut de Biologie de l’ENS (IBENS), and Inserm U1024, and CNRS UMR 8197, Paris, F-75005
France.
7
Interuniversity Institute of Bioinformatics in Brussels (IB)
2
, ULB Machine Learning Group, Computer Science Department,
Université Libre de Bruxelles.
8
Department of Ecology and Evolutionary Biology, University of Arizona, Tucson, AZ, 85721, USA.
9
CNRS, UMR 7093, LOV, Observatoire océanologique, 06230, Villefranche/mer, France.
10
Sorbonne Universités, UPMC Univ Paris 06, UMR 7093, LOV, Observatoire océanologique, 06230, Villefranche/mer, France.
11
Department of Marine Biology and Oceanography, Institute of Marine Science (ICM)-CSIC, Pg. Marítim de la Barceloneta,
37-49, Barcelona E08003, Spain.
12
Sorbonne Universités, UPMC, Univ Paris 06, CNRS-IRD-MNHN, LOCEAN Laboratory, 4 Place Jussieu, 75005, Paris,
France.
13
KU Leuven, Laboratory of Aquatic Ecology, Evolution and Conservation, Charles Deberiotstraat 32, 3000 Leuven.
14
PANGAEA, Data Publisher for Earth and Environmental Science, University of Bremen, Hochschulring 18, 28359 Bremen,
Germany.
15
MARUM, Center for Marine Environmental Sciences, University of Bremen, Hochschulring 18, 28359 Bremen, Germany.
16
Structural and Computational Biology, European Molecular Biology Laboratory, Meyerhofstr. 1, 69117 Heidelberg, Germany.
17
Institute for Chemical Research, Kyoto University, Gokasho, Uji, 611-0011 Kyoto, Japan.
18
Department of Geosciences, Laboratoire de Météorologie Dynamique (LMD), Ecole Normale Supérieure, 24 rue Lhomond,
75231 Paris Cedex 05, France.
19
Laboratoire de Physique des Océan, UBO-IUEM, Palce Copernic, 29820 Polouzané, France.
20
CEA, Genoscope, 2 rue Gaston Crémieux, 91000 Evry France.
21
CNRS, UMR 8030, 2 rue Gaston Crémieux, 91000 Evry, France.
22
Université d'Evry, UMR 8030, CP5706 Evry, France.
‡
Tara Oceans coordinators and affiliations are listed below.
†
These authors contributed equally to this work
*Correspondence to: jeroen.raes@vib-kuleuven.be, vargas@sb-roscoff.fr,
cbowler@biologie.ens.fr, karsenti@embl.de

Abstract: Reconstructing global species interaction networks and identifying the
abiotic and biotic factors that shape them are fundamental yet unsolved goals in
ecology. Here, we integrate multi-kingdom organismal abundances and rich
environmental measures from Tara Oceans and find that environmental factors are
incomplete predictors of community structure. To study biotic effects, we
reconstructed the first global photic-zone co-occurrence network. Interactions are
non-randomly distributed across plankton functional types and phylogenetic groups,
and show both local and global patterns. Known and novel interactions were
identified among grazers, primary producers, viruses and (mainly parasitic)
symbionts. We show how network-generated hypotheses guide confocal microscopy
analyses towards discovery of symbiotic relationships. Together, this effort provides a
foundational resource for ocean food web research and integrating biological
components into ocean models.
One Sentence Summary: A species interaction network from the global ocean shows
novel insights in top-down effects on community structure.
Introduction
The structure of oceanic ecosystems result from the complex interplay between
resident organisms and their physico-chemical environment. In the world’s largest
ecosystem, oceanic plankton (composed of viruses, prokaryotes, microbial
eukaryotes, and zooplankton) form intricate and dynamic trophic and symbiotic
interaction networks (1-4) that are also influenced by environmental conditions.
Ecosystem structure and composition are governed by abiotic as well as biotic
control. The former includes environmental conditions and nutrient availability (5),
while the latter encompasses grazing, pathogenicity and parasitism (6, 7). Like in
terrestrial and intertidal ecosystems, determining the relative importance of both
processes represents a grand challenge in ecology (5), but overall, abiotic effects have
historically been considered to be the factor most strongly determining community
structure (8). The challenge is to establish a quantitative understanding of biotic and
abiotic interactions in natural systems where the organisms are taxonomically and

trophically diverse (9). Although experimental methods were developed to detect
interactions encompassing virus-host associations (10-13) and competition and
cooperation among bacteria (14), they are not sufficiently high-throughput yet to be
applied community-wide in natural systems. However, sequencing technologies are
now enabling community profiling across trophic levels, organismal sizes, and
geographic ranges, providing the opportunity to predict organismal interactions across
entire biomes based on co-occurrence patterns (15). Previous bioinformatics efforts
addressing these issues have provided insights on the structure (16, 17) and dynamics
of microbial communities at specific locations or organismal domains (18-20).
Here we analyze data from 313 plankton samples the Tara Oceans expedition (21)
derived from 7 size-fractions covering collectively 68 stations at 2 depths across 8
oceanic provinces (Table S1), spanning organisms from viruses to small metazoans.
For these samples, viral (13), prokaryotic and eukaryotic abundance profiles were
derived from clusters of metagenomic contigs,
mi
tags (22) and 18S rDNA V9
metabarcodes, respectively (9, 23, 24) (Table S1). In addition, rich environmental
data from on-site and satellite measurements were collected (21, 25, 26). On this
dataset, network inference methods and machine learning techniques are leveraged to
disentangle biotic and abiotic signals shaping ocean plankton communities, and to
construct a global-ocean cross-kingdom species interaction network (interactome).
The interactome is then used to explore top-down relationships in the photic zone and
validated using microscopic investigation of host/symbiont pairs and in silico analysis
of phage-host pairings.
Evaluating the effect of abiotic and biotic factors on community structure
Given the breadth of the dataset we first re-assessed the effects of environment and
geography on community structure. Using variation partitioning (27) we found that on
average, the percentage of variation in community composition explained by
environment alone was 18%, by environment combined with geography 13%, and by
geography alone only 3% (28);(29). In addition, we built random forest-based models
(30) to predict abundance profiles of the Operational Taxonomic Units (OTU) using
a) OTUs alone, b) environmental variables alone, c) OTUs and environmental
variables combined, and tested for each OTU whether one of the three approaches
outcompeted the other (see Methods). These analyses revealed that 95% of the OTU-

only models are more accurate in predicting OTU abundances than environmental
variable models, while combined models were no better than the OTU-only models
(31);(32). This suggests that, unlike previously assumed (8), abiotic factors have a
limited effect on community structure.
To study the role of biotic interactions, we developed a method to identify robust
species associations in the context of environmental conditions. Twenty-three taxon-
taxon and taxon-environment co-occurrence networks were constructed based on
9,292 taxa, representing the combinations of two depths, seven organismal size ranges
and four organismal domains (Bacteria, Archaea, Eukarya, viruses) (33). To reduce
noise and thus false positive predictions, we restricted our analysis to taxa present in
at least 20% of the samples and used conservative statistical cutoffs (see Methods). A
global network was obtained by performing the union of the individual networks. This
network features a total of 127,995 unique edges, of which 92,633 are taxon-taxon
edges and 35,362 are taxon-environment edges (Table 1). Node degree does not
depend on the abundance of the node (OTU) (33). As such this network represents a
novel, extensive resource to examine species associations in the global oceans (33-
36).
Next, we assessed how many of the observed taxon links were indirect associations
representing ‘niche effects’ driven by geographic or environmental parameters (i.e.,
associations between taxa that are only due to a common response to an
environmental condition (15)). Motifs consisting of two correlated taxa that also
correlate with at least one common environmental parameter (“environmental
triplets”) were examined using three approaches (interaction information, sign pattern
analysis, and network deconvolution (37)) to identify associations that were driven by
environment (32, 34); 27,868 such taxon-taxon-environment associations (30% of
total) were detected. Among environmental factors, we found that PO
4
, temperature,
NO
2
and mixed layer depth were frequent drivers of network connections (Figure
1A). Notably, while the three methodologies pinpoint indirect associations, only
interaction information directly identifies synergistic effects in these biotic-abiotic
triplets. Exploiting this property, we disentangled the 27,868 environment-affected
associations into 8,961 edges driven solely by abiotic factors (38) (excluded from the
network for the remainder of the study) and 18,907 edges whose dependencies result

from biotic-abiotic synergistic effects. This revealed that a minority of associations
can be partly or completely explained by an environmental factor.
Evaluation of predicted interactions
Because co-occurrence techniques were thus far applied principally to bacteria, we
assessed the sensitivity of the approach for detecting eukaryotic interactions based on
V9 rDNA metabarcodes. We created, through extensive literature searches, a list of
573 known symbiotic interactions sensu lato (i.e., parasitism and mutualism) in
marine eukaryotic plankton (36, 39). We extracted 42 genus level interactions for
which both partners (OTUs) were present in the abundance pre-processed input
matrices, and found that 40.5 % of these were predicted, and up to 49 % when only
parasitism was considered – a considerable number, given the fact that the list is
based on interactions that are from other locations and potentially transitive or
facultative. The probability of having found each of these interactions by chance
alone was <0.01 (Fisher exact test, average pval = 4e-3, median pval = 5e-7). Most of
the false negative interactions were due to the strict filtering rules we determined to
avoid false positives. Based on this sensitivity and a false discovery rate averaging to
9% (computed from null models; see Methods), we estimate the lower and upper
limits for the number of interactions among eukaryotes present in our filtered input
matrices to be 55,000 and 150,000.
Biotic interactions within and across kingdoms
We next focused on the integrated network containing 83,672 predicted biotic
interactions (31) (36) that were non-randomly distributed within and between size
fractions (Figure 1B, C) (40). Copresences (positive associations) outnumbered
mutual exclusions (anticorrelations; 73% versus 27%), and a non-random edge
distribution with regard to phylogeny was observed (Figure 2A), with most
copresences derived from syndiniales and other dinoflagellates, and exclusions
involving arthropods. On higher taxonomic ranks (e.g., Order), we found that
although taxonomically related groups do co-occur (2,500 associations within the
same order; (15, 16)), 32% (1,157) was found across different orders (38, 41). Certain
combinations of phylogenetic groups are over-represented. For instance, a clade of
syndiniales (the MALV-II Clade 1 belonging to Amoebophrya (3)) shows a
significant enrichment in positive associations with tintinnids (P = 2e-4), amongst the

Frequently asked questions

Q1. What contributions have the authors mentioned in the paper "Title: top-down determinants of community structure in the global plankton interactome" ?

The authors show how network-generated hypotheses guide confocal microscopy analyses towards discovery of symbiotic relationships. Together, this effort provides a foundational resource for ocean food web research and integrating biological components into ocean models. 

Q2. What are the future works in "Title: top-down determinants of community structure in the global plankton interactome" ?

The analyses presented place new emphasis on the role of top-down biotic interactions in the epipelagic zone, and present myriad hypotheses that will guide future research to understand how symbionts, pathogens, predators and parasites interact with their target organisms, and ultimately help elucidate the structure of the global food webs that drive nutrient and energy flow in the ocean. 

Q3. How did the authors reduce noise and false positive predictions?

To reduce noise and thus false positive predictions, the authors restricted their analysis to taxa present in at least 20% of the samples and used conservative statistical cutoffs (see Methods). 

Q4. What is the important factor in determining the structure of oceanic ecosystems?

In the world’s largest ecosystem, oceanic plankton (composed of viruses, prokaryotes, microbial eukaryotes, and zooplankton) form intricate and dynamic trophic and symbiotic interaction networks (1-4) that are also influenced by environmental conditions. 

Q5. What are the common local associations between the two groups?

Approximately two thirds of local associations occur in MS (8,371) followed by SPO (1,119), while the rest are contributed by IO (946), with SO (901), SAO (123) and RS (891), and NAO (60) (Figures 2C-G). 

Q6. What factors are the frequent drivers of network connections?

Among environmental factors, the authors found that PO4 , temperature, NO2 and mixed layer depth were frequent drivers of network connections (Figure 1A). 

Q7. What are the key symbioses in marine ecosystems?

These results demonstrate that the combination of molecular ecology, microscopy and bioinformatics provide a powerful toolkit to unveil key symbioses in marine ecosystems. 

Q8. What is the role of biotic interactions in the ocean?

The analyses presented place new emphasis on the role of top-down biotic interactions in the epipelagic zone, and present myriad hypotheses that will guide future research to understand how symbionts, pathogens, predators and parasites interact with their target organisms, and ultimately help elucidate the structure of the global food webs that drive nutrient and energy flow in the ocean. 

Q9. How did the authors find the percentage of variation in community composition explained by environment alone?

Using variation partitioning (27) the authors found that on average, the percentage of variation in community composition explained by environment alone was 18%, by environment combined with geography 13%, and by geography alone only 3% (28);(29). 

Q10. What is the way to predict OTU abundances?

These analyses revealed that 95% of the OTU-only models are more accurate in predicting OTU abundances than environmental variable models, while combined models were no better than the OTU-only models (31);(32). 

Q11. How many interactions were found by chance?

The probability of having found each of these interactions by chance alone was <0.01 (Fisher exact test, average pval = 4e-3, median pval = 5e-7). 

Q12. What is the effect of PFTs on biogeochemical processes?

Their approach being particularly suitable for predicting parasitic interactions, the authors assessed their potential impact on biogeochemical processes by exploring a functional sub-network (22,223 edges) of known and novel plankton parasites (9) together with classical ‘plankton functional types’ (PFTs (56)). 

Q13. What is the role of alveolate parasitoids in the zooplankt?

This emphasizes the important role of alveolate parasitoids as top-down affectors of zooplankton and microphytoplankton population structure and functioning (3) - although the latter group is also affected by grazing (1).