Environmental DNA metabarcoding read numbers and their variability predict species abundance, but weakly in non-dominant species

Metabarcoding of environmental DNA (eDNA) provides more comprehensive, efficient, and non-invasive sampling of biological communities than conventional methods. However, limitations of metabarcoding include taxon-specific biases in amplification and sequencing that distort assessments of community composition. Further, hyper-abundant species may disproportionately affect community composition assessments and impair the detection of rare species (i.e., "species masking"). In this study, we examine methodological approaches to improve eDNA metabarcoding assessments of community structure using fish community diversity in a pond in south Florida using MiFish primers modified to improve cichlid detection. Mitochondrial 12S eDNA amplicon sequencing via Illumina NovaSeq was analyzed using the DADA2 model-based exact sequence inference. The fish species and abundances in the system were recorded during piscicide treatment and subsequent native species restocking. Our results demonstrate that (1) ultra-high-throughput sequencing on the newer NovaSeq patterned flow cell provided reliable detection of very rare taxa-with detections of a single individual. (2) Read numbers were significantly correlated to the total surface area of the fish population, and numerical abundance to a lesser degree; however, dominant taxa largely drove those correlations, and simulations showed that biases in the most abundant taxa will have disproportionate effects on the strength of the correlation. (3) The read number coefficient of variation for each species across spatially separated replicate samples may provide less biased abundance estimates compared with estimates based on average read counts. Finally, (4) exact sequence inference detected multiple haplotypes and population genetic diversity within a species. Our results demonstrate the real-world metabarcoding capacity to reveal community structure and reliably detect rare species and unique haplotypes and shows that read numbers can, to a limited degree, be used to infer the size of fish populations. Careful examination of detection biases among dominant taxa and spatial variation among samples are required for rigorous eDNA-based estimates of community structure. Our results demonstrate the capacity of NovaSeq metabarcoding to reveal freshwater fish community structure and reliably detect rare species and unique haplotypes. Metabarcoding read numbers were significantly correlated to the total surface area of the fish species' populations, allowing for conditional inferences of population sizes. However, dominant taxa largely drove those correlations, and simulations indicated that biases toward the most abundant taxa will have disproportionate effects on the strength of the correlation.