Previous work found yearly recurrence in microbial community composition at the BBMO13,28,34, and at the Bay of Banyuls14, both in the NW Mediterranean Sea. Our approach focused in the connectivity of microorganisms and how they organize themselves from a network perspective. Similar to previous studies13,14,28,34, our temporal network displayed seasonality with annual periodicity for most global network metrics. In general, our measured global network metrics are within previous work range22–25, 35–37 (Table 2 for edge density, transitivity, and average path length). Contrary to early works reporting biological networks generally being disassortative (negative assortativity based on degree)38, our single static network and monthly subnetworks were assortative. Microorganisms had more and stronger connections and a tighter clustering in colder than in warmer waters. Seasonal bacterial freshwater networks36 also showed higher clustering in fall and winter than spring and summer, but in contrast to our work, networks were biggest in summer and smallest in winter. In agreement with our results, Chaffron et al. reported higher association strength, edge density, and transitivity in polar regions (colder) compared to other regions (warmer) of the global ocean37. Colder waters in the Mediterranean Sea are milder than polar waters, but together, these results suggest that either microorganisms interact more in colder environments, or that their recurrence is higher due to higher environmental selection exerted by low temperatures and therefore, they tend to co-occur. Alternatively, lack of resources (mostly nutrients) in summer or in the tropical and subtropical ocean may prevent the establishment of several microbial interactions. In any case, temperature may not be the only driver of network architecture.
Table 2
Global network metrics of previously described microbial association networks
Edge density
|
Transitivity
|
Average path length
|
Sampling
|
Location
|
Domains
|
Notes
|
Reference
|
0.04
|
0.26
|
3.05
|
Monthly samples August 2000 - March 2004
|
Subsurface deep chlorophyll maximum depth off the southern California coast (SPOT)
|
Archaea, bacteria, and eukaryotes
|
Edge density for microbial network including environmental factors. Transitivity and average path length for microbial network.
|
22
|
0.14
|
0.33
|
1.94
|
Monthly samples August 2000 - January 2011
|
Two depths at SPOT
|
Free-living bacteria and some picoeukaryotes
|
Metrics from surface layer network.
|
23
|
0.02
|
0.24
|
|
Monthly samples March 2008 - January 2011
|
surface ocean (0-5m) at SPOT
|
Free-living eukaryotes (0.7–20 µm), bacteria (0.22–1 µm) and viruses (30 kDa–0.22 µm)
|
|
24
|
0.04
|
0.28
|
2.07
|
Monthly samples August 2003 - January 2011
|
Five depths at SPOT
|
Free-living bacteria
|
Metrics from 5 m layer network.
|
25
|
(0.023) W:0.033, Sp:0.032, S:0.036, F:0.029
|
(0.472) W:0.518, Sp:0.480, S:0.475, F:0.573
|
(4.84) W:2.16, Sp:5.03 S:7.26, F:3.04
|
Spatial samples
|
52 samples from freshwater lakes (surface) in China
|
Bacteria
|
Metrics for (whole network) and seasonal networks: W: winter, Sp: spring, S: summer, and F: fall
|
36
|
0.005,
0.003,
0.008
|
0.2,
0.0,
0.43
|
3.05,
3.02,
2.56
|
Spatial sampling
|
68 stations from the Tara Oceans expeditions (TARA) at two depths across eight oceanic provinces
|
Organisms from seven size fractions spanning from viruses to small metazoans
|
Metrics from surface networks including eukaryotes only, eukaryotes and prokaryotes (0.5-5 µm), and prokaryotes only (0.2–1.6 µm)
|
35
|
0.002
|
0.036
|
|
Spatial sampling
|
Samples from 115 stations from the TARA at two depths covering all major oceanic provinces from pole to pole
|
Bacteria, archaea, and eukaryotes from six size fractions.
|
Metrics represent the means of sample-specific subnetworks.
|
37
|
The effects of environmental variables on network metrics are unclear39, yet, our approach allowed identifying potential environmental drivers of network architecture. Correlation analyses pointed to the usual suspects that have been already found to influence microbial abundances. For instance, our results indicated that temperature and day length, key variables driving microbial assemblages in seasonal time-series12–14, and to a lesser extent inorganic nutrients, were the main factors influencing global network metrics. This is also in agreement with earlier works indicating that phosphorus and nitrogen are the primary limiting nutrients in the Western Mediterranean Sea40,41. Altogether, our correlation analysis is a step forward to elucidate the effects of environmental variables on network metrics, although we did not consider several other variables that could affect networks (e.g. organic matter).
Our preliminary network (significant associations derived with eLSA) contained 18% negative edges compared to 0.9% in the single static network (after applying EnDED and Jaccard index). Thus, our filtering strategy removed proportionally more negative edges. Associations may represent positive or negative interactions, but they can also indicate high niche overlap (positive association) or divergent niches (negative association) between microorganisms42. We hypothesize that most of the removed negative edges represented associations between microorganisms from divergent niches, most likely corresponding to colder or warmer months.
We found more highly prevalent associations within specific months, than when considering all ten-years of data. Furthermore, our results indicate a potentially low number of core interactions and a vast number of non-core ones. Usually, core microorganisms are defined based on sequence abundances, as microorganisms (or taxonomical groups) appearing in all samples or habitats being under investigation43. Shade & Handelsman43 suggested other parameters, including connectivity, will create a more complex portrait of the core microbiome and advance our understanding of the role of key microorganisms and functions within and across ecosystems43. Using a temporal network we identified core associations based on recurrence, which contributes to our understanding of key interactions underpinning microbial ecosystem function. Considering associations within each month, we found more highly-prevalent associations in colder than in warmer months. Our results indicated microbial connectivity is more repeatable (indicating higher predictability) in colder than in warmer waters. On one hand, the microbial community in colder waters being more recurrent13 may explain our observations indicating a more robust connectivity. On the other hand, it may be the stronger connectivity that leads to more similar communities in colder waters in BBMO. Last but not least, the interplay of both species dynamics and interactions may determine community turnover in the studied ecosystem. From a technical viewpoint, the overall single static network may have missed to capture summer associations resulting in smaller monthly subnetworks. For instance, a previous work in freshwater lakes constructed season specific networks and found more associations in summer than winter with Cyanobacteria dominating in summer, which may be due to strong co-occurrence patterns and suitable living conditions36.
Several network-based analysis have been used to study Cyanobacteria associations. For example, Chow et al.24 determined for 12 Cyanobacteria (Prochlorococcus and Synechoccus) 44 potential relationships with two potential eukaryote grazers (a ciliate and a dinoflagellate), 39 to other bacteria and three between Cyanobacteria, which were all positive. Similarly, all cyanobacterial ASVs in our study connected primarily to other bacterial ASVs, and exerted mainly positive associations. In agreement, Cyanobacteria also displayed primarily positive associations in a networks determined for the global ocean35.
Identifying different potential association partners of closely related Cyanobacteria, may indicate adaptations to different niches. A recent study found distinct seasonal patterns of closely related taxa indicating niche partitioning at the BBMO, including Synechococcus ASVs34. Our approach can complement and further characterize “sub”-niches by providing association partners for different ASVs. Moreover, in contrast to a single static network, temporal networks allow identifying associated partners in time (Supplementary Fig. 6). An increase in abundance of a microorganism may promote the growth of associated partners and a decrease may hinder the growth of partners or cause predators to prey on other microorganisms. Moreover, given the majority of association partners being other bacteria, the growth of Cyanobacteria may affect other bacteria and their growth, which is why it is necessary to explore potential interaction partners36.
From a technical perspective, our approach allowed us to see what the single static network captured since all our temporal network observations are linked to it. Thus, future studies with higher sampling frequency may be able to construct networks within a month. However, our approach is a good starting point that allows us to move forward, but still, it has limitations, suggesting caution when making biological interpretations from the temporal network. Another limitation is that we disregarded local network patterns by using global network metrics. Future work could use the local-topological metric based on graphlets44. Counting the number of graphlets a node is part of quantifies their local connection patterns, which allows to infer seasonal microorganisms through recurring connection patterns in a temporal network. Such a network-based approach would complement the detection of the seasonal microorganisms based on sequence abundances13.