This project will show that metabarcoding targeting bacteria, can successfully be used to replicate morphology-based spatial patterns of change observed around fish-farms.
We set-out to assess the spatial and temporal change in bacterial assemblages around the Dunstaffnage fish-farm. The samples that were analysed were collected over an 18 month period from the cage-edge, and from a 100 and 600 meter distance around the farm. These samples were subjected to next-generation sequencing metabarcoding analysis. We identified over 1000 bacterial taxa and demonstrated, for the first time, clear changes in the bacterial assemblage occurring over time around the fish-farm. We used a machine-learning approach (random forest, RF) to assess the extent to which the bacterial assemblage characterised the Distance from the farm. Our best RF model achieved >80% accuracy, even whilst excluding sampling date, indicating that bacterial assemblages can be used to distinguish Distance regardless of the time the sample was taken. We found that the optimal RF model was based on relatively few taxa and that these were mostly of unidentified genera (reflecting database limitations). The higher-level identification that was possible showed that the key taxa were Gammaproteobacteria and that many are known to be involved in sulphur metabolism and either strictly anaerobic or aerobic.
This feasibility study has shown that bacterial metabarcoding has considerable potential for the accurate classification of marine sediments and potential utilisation as a cost-effective, rapid and highly sensitive tool in the assessment of the seabed surrounding fish-farms.
Scottish Association for Marine Science (SAMS)