MULTI-METHOD KEYSTONE SPECIES IDENTIFICATION REVEALS DISTRIBUTED ECOLOGICAL RESILIENCE IN AQUAPONIC MICROBIAL COMMUNITIES

Introduction

Aquaponics integrates recirculating aquaculture with soilless plant production in a shared water circuit, relying on microbial communities to mediate nutrient cycling underpinning system stability. Whether individual keystone species - taxa whose removal would disproportionately restructure community composition relative to their abundance - exist in aquaponic systems remains unknown (Garza et al., 2026; Rottjers & Faust, 2023). Identifying such taxa, or confirming their absence, would clarify the ecological architecture of these systems and inform microbiome-based monitoring and management strategies. Prior work has combined presence-impact statistics, co-occurrence network topology, and machine-learning approaches in other natural aquatic ecosystems, but not in recirculating aquaculture or aquaponics.

Methods

Eighty-eight 16S rRNA V4 amplicon libraries were collected from three compartments (clarifier, fish tank, grow bed) across six sampling dates (May–November 2024) at a research-scale facility (Auburn University, USA) operated under coupled and decoupled hydraulic configurations. After DADA2 processing with cutadapt primer trimming-which reduced chimera rates from 85.7% to acceptable levels-825 ASVs were retained after prevalence filtering (≥20% of samples; 61,787 total ASVs). Three independent keystone screens were applied: (i) leave-one-out PERMANOVA presence-impact analysis, in which each focal taxon was removed before Bray–Curtis distance calculation to avoid circularity, with compartment and hydraulic coupling as covariates and false discovery rate (FDR) correction (150 candidate taxa tested); (ii) Spearman co-occurrence network hub identification among the 50 most CLR-variable taxa (n = 60 samples; |ρ| > 0.6, padj < 0.001); and (iii) LASSO regression predicting Bray–Curtis dissimilarity between consecutive samples at each location.

Results

A distributed ecological resilience framework, rather than dependence on a few keystone taxa, appears to best describe the aquaponic environment. Compartment was the dominant driver of community structure (PERMANOVA R² = 0.256), with smaller contributions from hydraulic coupling (R² = 0.069), plant growing season (R² = 0.055), and illumination regime (R² = 0.031). Grow bed samples showed the highest alpha diversity; fish tank samples showed the lowest. Leave-one-out PERMANOVA identified 67 taxa at FDR < 0.05 and R² > 0.02. Cetobacterium sp. (Fusobacteriota) ranked highest, with three ASVs each explaining 8.2–11.0% of community variance-a signal driven by its restriction to fish tank samples. Other high-impact taxa included Dysgonomonadaceae (R² = 0.067), Thermomonas sp. (R² = 0.054), and Ferruginibacter sp. (R² = 0.045), spanning four phyla. The co-occurrence network yielded 50 nodes and 335 edges (density = 0.27); four biofilm-associated hub taxa were identified across Proteobacteria, Planctomycetota, and Armatimonadota. LASSO retained zero individual taxa as predictors of temporal turnover. Critically, no taxon appeared in more than one detection method: presence-impact candidates and network hubs were entirely non-overlapping (Figure 1).

Figure 1. Non-overlapping keystone candidates across three detection methods. UpSet plot showing set membership of the 71 unique candidates recovered from presence-impact PERMANOVA (67 taxa), co-occurrence network hubs (4 taxa), and LASSO temporal prediction (0 taxa). No taxon was supported by more than one method.

Discussion

The absence of cross-method consensus is the key finding: aquaponic systems do not appear to harbor classical keystone taxa as in natural aquatic environments. The pattern is consistent with distributed ecological resilience, where community-structuring function is spread across many taxa rather than concentrated in a few irreplaceable ones (Allison & Martiny, 2008). One genus, Cetobacterium, may serve as a compositional indicator in fish tanks, but not as a system-wide keystone; its strong presence-impact signal appears to result from compartment specificity. Network hub taxa reflect co-response to shared environmental drivers rather than direct metabolic control, as shown by the inability of graphical LASSO to recover a sparse network even at maximum regularization.

Operationally, these findings support compartment-level community monitoring over surveillance of universal keystone targets.

References

Allison, S.D. & Martiny, J.B.H. (2008). Resistance, resilience, and redundancy in microbial communities. Proceedings of the National Academy of Sciences, 105(Suppl. 1), 11512–11519.

Garza, D.R. et al. (2026). Keystone taxa in microbial communities: mechanisms, detection, and ecological implications. [in press]

Rottjers, L. & Faust, K. (2023). From hairballs to hypotheses: biological insights from microbial networks. FEMS Microbiology Reviews, 10.1093/femsre/fuac005.

Tsuchiya, C. et al. (2008). Distribution of Cetobacterium somerae, an anaerobic bacterium in the family Fusobacteriaceae, in freshwater fish with different feeding habits. Letters in Applied Microbiology, 46, 38–42.