# Gilthead sea bream gut bacteriome as a valuable tool for seafood provenance analysis

## TL;DR

This study uses gut bacteria in gilthead sea bream to track where they were caught, helping prevent seafood fraud and improve traceability.

## Contribution

The study introduces a novel method combining gut bacteriome profiling with machine learning to identify regional seafood provenance biomarkers.

## Key findings

- Gut bacteriome diversity differed significantly between fishing areas, with distinct bacterial phyla and classes in the Center-South region.
- A machine learning model achieved 100% accuracy in identifying the Center-South region and 71.1% for the South region.
- Five key bacterial biomarkers were identified, influenced by human activities and environmental factors.

## Abstract

The increasing demand for high-quality seafood underscores the significant challenges posed by rampant seafood fraud. This study aimed to identify regional capture biomarkers by using the gut bacteriome of Sparus aurata specimens through state-of-the-art long-read sequencing allied to machine learning tools. The gut bacteriomes of animals from four different fishing areas on the Portuguese coast were sequenced. The alpha and beta diversity analyses were shown to enable Center-South gut bacteriome differentiation from other fishing areas due to higher abundance of species of the phyla Pseudomonadota, Bacteroidota, and Bacillota and classes such as Alphaproteobacteria, Betaproteobacteria, and Bacilli. The gradient boosting machine (GBM) model selected by the H2O automatic machine learning pipeline presented a high global accuracy during training and validation phases, identifying Center-South and South sample provenance with 100% and 71.1% accuracy, respectively. By integrating the most important OTUs to the GBM model with the regional biomarkers identified through point biserial correlation analysis (indicspecies packages), a reduced set of five provenance biomarkers was identified, belonging to Gammaproteobacteria, Betaproteobacteria, and Bacilli classes, possibly highlighting the anthropogenic activities surrounding the fishing areas and local environmental abiotic factors. This study highlights the extensive and valuable information obtained by long-read sequencing and couples it with the potential of machine learning algorithms to ultimately demonstrate its efficiency in providing efficient and highly accurate seafood provenance biomarkers. This study also reports the likely influence of industrial and recreational activities, population density, and water management facilities on the gut bacteriome of S. aurata.

This study significantly contributes to a topic of utmost importance—seafood provenance analysis and seafood fraud—by leveraging gut bacteriome profiling. Through the application of long-read sequencing and machine learning, it identifies reliable biomarkers that distinguish gilthead sea bream from different fishing areas. These findings enhance traceability methods by providing a robust tool to combat seafood fraud and ensure food authenticity, thereby protecting the supply chain, the consumer, and the environment. Additionally, this study explores the possible interactions between the gut bacteriome and the industrial, recreational, and environmental factors that could influence the identified biomarkers of regional provenance while also offering insights into the composition of the marine ecosystems surrounding the fishing areas. This approach has broader implications for fishery management, sustainable sourcing, and regulatory enforcement.