Enhancing microbial predator-prey detection with network and trait-based analyses

Network analyses are often applied to microbial communities using sequencing survey datasets. However, associations in such networks do not necessarily indicate actual biotic interactions, and even if they do, the nature of the interactions commonly remains unclear. While network analyses are valuable for generating hypotheses, the inferred hypotheses are rarely experimentally confirmed.

We employed cross-kingdom network analyses, applied trait-based functions to the microorganisms, and subsequently experimentally investigated the found putative predator-prey interactions to evaluate whether, and to what extent, correlations indicate actual predator-prey relationships. For this, we investigated algae and their protistan predators in biocrusts of three distinct polar regions, i.e., Svalbard, the Antarctic Peninsula, and Continental Antarctica. Network analyses using FlashWeave indicated that 89, 138, and 51 correlations occurred between predatory protists and algae, respectively. However, trait assignment revealed that only 4.7-9.3% of said correlations link predators to actually suitable prey. We further confirmed these results with HMSC modeling, which resulted in similar numbers of 7.5% and 4.8% linking predators to suitable prey for full co-occurrence and abundance models, respectively. The combination of network analyses and trait assignment increased confidence in the prediction of predator-prey interactions, as we show that 82% of all experimentally investigated correlations could be verified. Furthermore, we found that more vicious predators, i.e., predators with the highest growth rate in co-culture with their prey, exhibit higher stress and betweenness centrality - giving rise to the future possibility of determining important predators from their network statistics.

Our results support the idea of using network analyses for inferring predator-prey interactions, but at the same time call for cautionary consideration of the results, by combining them with trait-based approaches to increase confidence in the prediction of biological interactions. Video Abstract.