A biomarker-based approach to predicting gill health in Atlantic salmon (AquaGill)

BACKGROUND

Gill health has emerged as an area of increasing focus for the sustainability and productivity of Atlantic salmon aquaculture in Scotland and around the world. Increasing mortality linked to complex gill disease (CGD), amoebic gill disease (AGD), and environmental stressors such as harmful phytoplankton or micro-jellyfish blooms has highlighted the limitations of current diagnostic approaches. This situation is exacerbated by climate-driven changes, such as warming waters and associated shifts in harmful planktonic organisms.

Traditional assessment of gill health relies heavily on histopathology, which is time-consuming, destructive, and costly, and is therefore difficult to scale for routine monitoring. Previous collaborative research between the Scottish Fish Immunology Research Centre, Scottish Sea Farms, and BioMar established a large gill tissue biobank and identified candidate gene expression biomarkers associated with gill pathology. These studies demonstrated that molecular tools have the potential to provide a rapid and scalable method for detecting and quantifying gill damage. Access to such a tool would support farmers, fish health professionals, and product development teams evaluating novel functional feeds or new genetic lines, enabling earlier interventions, improved fish welfare, and greater industry sustainability.

AIMS

The AquaGill project was a multistakeholder project with the goal of developing a scalable, molecular tool for assessing gill health in Atlantic salmon. The work focused on a concept that could replace or complement traditional methods with a biomarker approach for predicting gill health status.

The specific project objectives were to design and validate a predictive biomarker panel using gene expression data and machine learning. The project also aimed to test the panel’s robustness under real-world conditions, including micro-jellyfish exposure and nutritional interventions. Finally, it sought to evaluate whether non-destructive sampling, particularly gill swabs, could provide a reliable alternative to tissue sampling for molecular assessment, benchmarked against histopathology.

OVERVIEW

The project leveraged an existing Atlantic salmon gill biobank established through earlier SAIC-supported studies, from Scotland and Tasmania. Each sample was linked to detailed histopathology and supported by matched biological material (RNA, microbiome swabs, proteomic material, and plasma), creating a dataset that represents a range of farm conditions and gill health states.

Work Package 1 delivered a high-throughput gene-expression assay and a targeted biomarker panel. Targets were selected from prior studies and designed to reflect key biological processes relevant to gill health (immune response, tissue repair, osmoregulation, and structural integrity), with primer design and validation used to ensure specificity in the duplicated Atlantic salmon genome.

Gene expression data from over 360 Scottish fish were used to train and test machine learning models, linking biomarker profiles to gill health status. Following quality control, 72 high-quality genes were retained for model development, and performance was assessed using an independent test set. The primary modelling approach (Random Forest) was used to optimise features and tuning, with alternative algorithms explored as comparators. Results were further supported through cross-platform checks against matched RNA-seq datasets.

Work Package 2 strengthened commercial relevance by testing the panel during a natural environmental challenge involving the micro-jellyfish Muggiaea atlantica. Weekly sampling over a ten-week period captured changes in gill condition over time, and these data were analysed alongside the biobank to evaluate robustness.

Work Package 3 assessed the feasibility of non-destructive sampling for routine monitoring by comparing RNA extracted from gill swabs with matched gill tissue from the same fish. Swab extraction protocols were optimised to improve yield and quality, and a subset of biomarker genes and pathogen markers was measured by qPCR to assess how closely swab-based profiles tracked tissue-based results.

RESULTS

The Gill Biobank dataset captured a broad spectrum of gill health, with histopathology scores ranging from no detectable damage to severe damage. Modelling these data demonstrated that a threshold HistoScore of 3 on a 0–12 scale  provided the most biologically meaningful distinction between healthy and damaged gills. This supports a clear threshold for action for fish health professionals.

The Random Forest classification model achieved a peak accuracy of 87.7% using the 40 most informative genes, with a sensitivity of 90.6% for detecting damaged fish and a specificity of 80.0% for identifying healthy individuals. The model showed substantial agreement with histopathological classification, as indicated by Cohen’s kappa of 0.695.

The predictive gene signature reflected a coherent progression of gill pathology. Key features included strong activation of inflammatory pathways, markers of epithelial damage, tissue remodelling, and osmoregulatory changes. Principal Component Analysis revealed a continuous molecular gradient aligned with histopathological severity, demonstrating that gene expression profiles captured the transition from healthy to damaged tissue states.

In Work Package 2, analysis of fish exposed to micro-jellyfish demonstrated that the biomarker panel maintained its discriminatory power under environmental stress. Principal Component Analysis again showed clear separation of samples along a gradient of tissue damage, with the first principal component explaining a substantial proportion of the variance. Integration with the full biobank dataset confirmed that the core gene signature remained stable, capturing jellyfish-induced irritation without compromising predictive accuracy. Most samples from this exposure fell within the lower range of histopathological scores, consistent with microscopy findings.

Work Package 3 demonstrated that gill swabs could yield RNA of sufficient quality for transcriptomic analysis, although quantities were low and variability was higher than in tissue-derived samples. Despite these limitations, amplification of housekeeping genes and pathogen markers showed strong consistency between swab and tissue samples. A subset of biomarker genes could also be detected in swab-derived RNA, indicating that non-destructive sampling is technically feasible, though further optimisation is required to improve reliability and sensitivity.

IMPACT

The AquaGill project and its partners delivered a biomarker panel and analysis pipeline for assessing gill health in Atlantic salmon, providing a scalable complement to histopathology. The development of a predictive biomarker panel, coupled with a machine learning framework, enables early detection of gill pathology with high accuracy, supporting timelier and more effective interventions.

Validation of the biomarker panel across different geographies, environmental conditions, including micro-jellyfish exposure and geographically distinct datasets, supports its potential industry wide-use.

The project established the feasibility of non-invasive monitoring through gill swabbing, which, with further refinement, could enable repeated, welfare-friendly assessment of fish health in commercial settings. Despite current technical limitations, this represents a significant step-change in enabling longitudinal monitoring without the need for destructive sampling.

From an industry perspective, the outcomes support improved fish health and welfare management. The tool can be used for the development of novel functional feeds to support non-medicinal health management strategies. In practical terms, this enables faster go/no-go decisions in feed R&D and stronger evidence packages for customers.

Overall, the project strengthens the application of molecular diagnostics in aquaculture, providing tools that may help reduce mortality, improve fish welfare, and enhance the resilience of salmon farming systems in the face of environmental and disease-related challenges.