For the entire experiment, 304 samples were processed for DNA extraction including 19 controls, 45 “BE” samples, 92 gill swabs, 92 skin swabs, and 56 digesta swabs .After swabbing the BE and fish mucosal sites, individual swab heads were broken off into a 2 ml PowerSoil tube and then stored at −20°C for 2 weeks until DNA extraction to preserve microbiome integrity . All molecular processing was done according to the standard Earth Microbiome Project protocols . Lysis in single tubes were used to minimize noise from well-to-well contamination . A serial dilution of a positive control, Escherichia coli isolate , along with negative control blanks were included to estimate the limit of detection of the assay . By using the Katharoseq method, we empirically calculated the read count used to exclude samples . For library preparation, DNA samples of equal volume were processed using the EMP 16S rRNA 515F /806R primers with 12 bp golay barcodes at a miniaturized PCR reaction volume of 5 μl reactions in triplicate . After PCR, equal volumes of each library were pooled and processed through the MinElute PCR purification kit followed by a 1× Ampure cleanup. The final library was sequenced using a MiSeq 2 × 250 bp kit . Sequences were uploaded, demultiplexed, and processed in Qiita , using the Qiime2 commands . Specifically, sequences from the first read were trimmed to 150 bp following the EMP protocol, and processed through the deblur pipeline and SEPP to generate Amplicon Sequence Variants “ASVs” . ASVs were rarified to 5,000 reads per sample. General Alpha and Beta diversity measures were generated in Qiita. Microbial Alpha diversity comparisons were calculated for richness, Shannon diversity ,pot drying and Faith’s Phylogenetic Diversity . For statistical analysis, grouped comparisons were compared using Kruskal–Wallis test with Benjamini Hochberg FDR 0.05 .
To compare the age of fish with alpha diversity metrics, both linear regression and Spearman correlation were used using PRISM 9.0 . Beta diversity measures were calculated using both Unweighted UniFrac and Weighted normalized UniFrac . Categorical group comparisons of beta diversity were calculated using PERMANOVA tests . Lastly, to quantify the effects or sources of microbes from the BE onto the fish mucus, we applied the microbial source tracking software SourceTracker2 . Prior to SourceTracker2 analysis, ASVs which had less than 100 total counts across the dataset were removed to reduce sparsity and improve performance of the microbial source tracking. Skin samples did not differ in microbial diversity based on rearing type. In the BE, water generally was highest in microbial diversity, while both air stones and air diffusers had the lowest diversity across all sample types. When comparing the water communities of the FT and RAS tanks, the richness and phylogenetic diversity trended higher in RAS . Interestingly, the inlet pipe biofilms were highly variable across the FT and RAS systems with the FT tank having a very high microbial diversity compared to RAS systems. The tank side biofilms were generally higher in microbial diversity in the RAS tanks as compared to the FT tank. When comparing beta diversity, the largest compositional differences were due to the feed vs all other sample types, with most feed pellet communities highly differentiated from the BE and fish mucus with the exception of live rotifer feeds. Many chloroplasts ASVs were present in the pellet feeds, likely from plant ingredients, which likely drove this separation. Upon chloroplast removal, read counts for feed samples drop to levels which would largely exclude them from analysis thus suggesting that feed samples have very low proportions of microbes. The second largest driver in microbial community composition was the fish body sites for both Weighted and Unweighted UniFrac .
For individual body sites, the tank systems also had a moderate impact with gill samples being more differentiated across tank systems . Specifically, for gill samples, the tank rearing system had an impact on the microbial community for both Unweighted Unifrac distance and Weighted normalized Unifrac distances . Pairwise comparisons of Unweighted Unifrac distances revealed that gill microbiomes of RAS reared fish were also differentiated but in general less differentiated as compared to the FT reared fish . Pairwise comparisons of Weighted normalized Unifrac distances revealed the same pattern, with fish reared in different RAS systems having a differentiated community but more even more differentiated when compared to fish reared in FT systems . Skin microbial communities were only influenced by the rearing method when comparing Unweighted Unifrac but not with Weighted normalized Unifrac. When comparing YTK from the same age and genetic cohort reared in three different conditions, gill microbial communities were more influenced by the environmental conditions than the skin, while microbial communities of the BE were highly variable across tank systems.After quantifying the variation which existed across tank systems at a single age of fish, we next wanted to evaluate the extent by which mucosal microbiomes varied with fish age. Specifically, we sought to investigate factors governing the randomness vs. deterministic mechanisms for microbial colonization in marine fish over time. Fish were sampled at three age points including 43, 137, and 430 dph. At 430 dph, fish were either collected from an offshore sea pen or from the indoor environment. The indoor fish at 430 dph had been in the sea pen but were transferred back to the indoor environment to be used as broodstock . These fish were in the indoor tanks for 79 days before sampling. Fish from 43 to 137 dph were always reared in cannabis indoor systems. At each body site: gill , skin , and digesta , microbial diversity was compared across fish ages. Additionally, fish from 430 dph were separated by either indoor or ocean net pen. When comparing richness measures, all three body sites were influenced by age with the gill being most influenced followed by digesta and then skin . A similar pattern was observed for Faith’s PD, which takes into account microbial phylogenetic diversity with all three body sites being influenced by age.
The gill was most influenced followed by digesta and lastly skin . Shannon diversity had the same pattern with gill , digesta , and skin all being influenced by fish age in the same order of impact. When comparing only samples at 430 dph, gill diversity was larger for fish which were transferred from the ocean net pen back into the indoor environment as compared to ocean net pen reared fish. This effect was also seen in the skin, but to a much smaller degree. To model age and microbial diversity across the body sites, we performed a regression and Spearman correlation for each diversity measure. For this analysis, we excluded ocean net pen reared fish from 430 dph to compare only indoor fish . For richness, both gill and skin samples were positively associated with fish age while digesta samples were negatively associated with fish age . For Faith’s PD, both gill and skin again were positively associated with fish age . Lastly for Shannon diversity, skin was positively associated with fish age while digesta was negatively associated with fish age . These cumulative results suggest a general mechanism for alpha diversity changes in the marine fish YTK, S. lalandi, whereby alpha diversity may continue to increase over time in the gill and skin surfaces while digesta samples start highly diverse but then adapt or reduce in complexity over time. Next we wanted to understand how the composition of microbial diversity changed over time and to also determine if there was evidence for succession. To determine if age was associated with microbial niche differentiation across body sites, we compared the fish body site microbiome independently at each of the four ages or conditions including 43 dph , 137 dph , 430 dph “indoor tank” , and 430 dph “seapen” . Body sites at each age group, even as early as 43 dph, had unique microbial communities measured using Unweighted and Weighted normalized Unifrac distance metrics . For Weighted normalized Unifrac, based on the F-statistic, body site microbial communities were most differentiated at 430 dph, especially in the open sea pens. This result suggests that body site microbial communities continue to differentiate throughout the lifetime of the fish. We then sought to answer the question if certain body sites are more influenced by age. To do this, we compared microbiome differences of age and tank type within each body site independently . For both Unweighted and Weighted normalized Unifrac distance comparisons, the gill microbiome samples were more differentiated across ages as compared to the skin and digesta . Furthermore, when observing the gill samples, the 430 dph fish reared in the indoor tank and ocean net pen were divergent on the PCoA . In addition, fish at 43 dph were also differentiated. Next, we evaluated if overall fish mucosal microbiome similarity to the BE changed with age and if it did, which BE or water sample types were most influential . For indoor reared fish at 43, 137, and 430 dph, we compared the microbiome of the gill, skin, and gut to various hatchery components including tank side, water from the tank, the inlet pipe into the tank, air stones, air diffusers, and feed. For feed, we evaluated 12 different feed types that were used throughout the production schedule ranging from days 1–12 until harvest. The first feed type consistently had a more similar microbial community to the gill, skin, and digesta samples across the different ages thus we used these samples for the feed comparison in the broader BE comparison.
When including all possible BE sample types, a noticeable trend emerged where at the earliest age , the microbial communities across all body sites were generally more similar to the BE . Whereas at later ages, the microbiome of the gill and skin communities generally become more dissimilar from the inlet pipe and feeds, but became more similar to the air diffuser. The digesta samples , however, consistently became more differentiated from the BE samples over time suggesting a stronger niche differentiation in the gut. To quantify this, we included only BE sample comparisons which were consistent in all ages – water, inlet pipe, and first feeds – and compared how the mucosal microbiomes of the fish disperse or converge toward the BE. For both gill and skin samples, the total differentiation of fish mucosal site to the three BE samples was least at 43 dph but increased with age . The gill and skin samples were both more similar to the inlet pipe at 43 dph and became more divergent from the inlet pipe over time . Digesta samples became more differentiated from all BE surfaces equally over time . To estimate the total impact of these differences, we calculated the effect size . For the gill, the dissimilarity differences across the BE samples explained 34.5% of the variation at 43 dph but then increased to 68.8% of the variation explained at 137 dph. For the skin, the largest jump in effect size occurred between 137 dph and 430 dph . These results indicate that niche differentiation occurs at varying rates depending on body site and that some BE microbial sources continue to have an influence on the fish mucosal microbiome throughout the lifespan of the fish, whereas other environmental sources may only be influential during early ontogeny. To identify the extent by which the BE contributes to the mucosal microbiome of the fish, we applied the popular microbial source tracking program SourceTracker2 which uses Bayesian statistics to estimate contributions of features from various sources to sink communities. SourceTrackr2 determined that contributions of the BE varied widely depending on both the body site and the age of the fish. At 43 dph, the tank side biofilm and air stones were the biggest sources of microbes to the gill and skin of the fish larvae, while the majority of microbes in digesta samples were from unknown or unsampled sources . Rotifer feeds also contributed to the gill, skin, and gut microbiomes, but to a lesser extent compared to airstone and tank side . At 137 dph, gill was again influenced by the airstone and air diffusers in the BE, while higher frequencies of skin and digesta samples were colonized by microbes from feeds . However, microbes from unknown sources had the largest overall contribution at 137 dph across all body sites .