Influence of host phylogeny and water physicochemistry on microbial assemblages of the fish skin microbiome

Abstract The skin of fish contains a diverse microbiota that has symbiotic functions with the host, facilitating pathogen exclusion, immune system priming, and nutrient degradation. The composition of fish skin microbiomes varies across species and in response to a variety of stressors, however, there has been no systematic analysis across these studies to evaluate how these factors shape fish skin microbiomes. Here, we examined 1922 fish skin microbiomes from 36 studies that included 98 species and nine rearing conditions to investigate associations between fish skin microbiome, fish species, and water physiochemical factors. Proteobacteria, particularly the class Gammaproteobacteria, were present in all marine and freshwater fish skin microbiomes. Acinetobacter, Aeromonas, Ralstonia, Sphingomonas and Flavobacterium were the most abundant genera within freshwater fish skin microbiomes, and Alteromonas, Photobacterium, Pseudoalteromonas, Psychrobacter and Vibrio were the most abundant in saltwater fish. Our results show that different culturing (rearing) environments have a small but significant effect on the skin bacterial community compositions. Water temperature, pH, dissolved oxygen concentration, and salinity significantly correlated with differences in beta-diversity but not necessarily alpha-diversity. To improve study comparability on fish skin microbiomes, we provide recommendations for approaches to the analyses of sequencing data and improve study reproducibility.


Introduction
Bacteria are ubiquitous in nature and their abundance and comm unity structur e ar e influenced by a v ariety of biological, physical, and chemical factors (Thompson et al. 2017 ).There is a growing interest in characterizing the bacterial communities associated with higher eukaryotic hosts, termed microbiomes (Ursell et al. 2012, Byrd et al. 2018 ).Many of these bacterial communities form symbiotic relationships with their host, conferring benefits including facilitating pathogen exclusion, immune system priming, and nutrient degradation, all of whic h ultimatel y enhance host fitness (Belkaid and Hand 2014, Chiu et al. 2017, Pickard et al. 2017, Hou et al. 2022 ).Gnotobiotic animals grown in sterile laboratory conditions with limited microbiota demonstrate reduced resilience to disease and poorer health outcomes (Williams 2014, Tlaskalo va-Hogeno va et al. 2015 ).T his highlights the importance of r esearc h on micr obiomes and their r elationship with their host organism to optimize animal health.To date, most r esearc h has focused on gut microbiomes in domesticated mammals (47.9%), principally because of the interest in their roles in human health and liv estoc k pr oduction (P ascoe et al. 2017 ), yet mammals constitute less than 10% of total v ertebr ate div ersity (IUCN Red List 2022 ).
Fish encompass 50% of all v ertebr ate div ersity (IUCN Red List 2022 ), and their importance to global food security and ability to live in a wide range of different habitats highlights the need to better c har acterize the micr obial comm unities in fish giv en that their importance for host health.There has been a recent and increasing interest in understanding how fish microbiomes affect growth and health in aquaculture Most of these studies have focused primarily on gut health and gro wth (Talw ar et al. 2018, Huang et al. 2020, Legrand et al. 2020 ), including the effects of feeds (Karlsen et al. 2022 ), and associations with disease and immunological defence (Shi et al. 2022 ).T here ha ve been far fewer studies, howe v er, examining the microbiota in fish skin (Gomez and Primm 2021, Wang et al. 2023, Ber ggr en et al. 2022 ).Understanding the composition of fish skin microbiomes and the dynamics in response to environmental conditions will help establish their associations with health and disease (Ber ggr en et al. 2022, Debnath et al. 2023 ) and in turn, optimize aquacultur e pr actices (P alladino et al. 2021, Sánchez-Cueto et al. 2023 ).To date, there has been relatively little exploration of how skin microbiomes are affected by water physicochemical or biological stressors.
We identified 103 studies on marine and freshwater fish skin microbiomes with publicly available data (as of December 2022).These studies were carried out in a range of aquaculture systems and water physicochemical conditions .T he experimental a ppr oac hes in DNA extr action methods, amplicon sequencing depth, data analysis, and metadata availability also differed across these studies, making microbiome comparisons difficult.Furthermor e, ther e wer e differ ences in data anal ysis pipelines may result in different ASV/O TU count tables , which can influence rates of false positives or false negatives (Olson et al. 2020 ) and the sensitivity and specificity of ASV/OTU detection (Prodan et al. 2020 ).Differ ent indices to assess beta div ersity to measur e community similarities take into consideration different properties of the community composition that are not comparable in cross-study comparisons.For example, a J accar d index considers only the presence/absence of taxa; a Bray-Curtis dissimilarity matrix treats all members as independent units while a Weighted Uni-Frac dissimilarity matrix includes the taxonomic relationship between members (Lozupone and Knight 2005 ).T hus , the chosen dissimilarity matrix influences beta-diversity values and in turn conclusions (Lozupone et al. 2007, 2010, Wong et al. 2016, Fukuyama 2019 ).
In this study, we analysed the microbial community composition of fish skin microbiomes using 16S RNA V4 region sequence data publicly available in the National Center for Biotechnology Information (NCBI) database.We r etrie v ed 36 studies where the sequencing data and associated metadata were sufficiently robust for the pr oposed anal yses.We e v aluated similarities and differences in fish skin bacterial microbiota that correlate with fish taxonomy, water physicoc hemical v ariables, and r earing conditions.We show that closely related fish species host microbiomes dominated by similar bacterial taxa, supporting pr e vious r esults (Brooks et al. 2016 ).We also identify differences in bacterial taxa and community structure associated with marine and freshwater fish and show the features of fish skin microbiomes that conv er ge in differ ent hosts when held under similar housing conditions or water physicochemical conditions.We furthermore emphasize the importance of standardizing, or at least adopting a more consistent approach in sequencing analysis to better enable comparisons in the c har acterization of microbial communities of the fish skin across studies and environmental conditions.

Liter a ture search str a tegy and paper data selection
We applied a systematic approach to select studies for conducting our analysis to investigate the effects of species and environmental parameters on fish skin microbiome bacterial community structur e.Our searc h str ategy consisted of mining the liter atur e for data sets using the following k e ywords: 16S V4 MiSeq ("fish skin" and "microbiome" | "microbiota") published before 2023 and in English from Google Scholar.Our inclusion criteria were: (a) studies had publicly available 16S rRNA amplicon sequencing data, (b) the sequence data included the V4 hypervariable region, and (c) studies were for bony fish skin microbiomes (i.e.excluding ra ys , sharks and lungfish).We used only studies with sequenced V4 region of the 16S rRNA to allow for nucleotide comparisons and the construction of phylogenetic trees.We filtered for studies that used the Illumina MiSeq technology-the most popular method (to maximize the number of studies included) as comparisons acr oss differ ent sequencing c hemistries can r esult in different sequencing biases (Loman et al. 2012, Allali et al. 2017, Bailén et al. 2020 ).Preprint publications were eligible for inclusion.All man uscripts were man ually screened to determine whether they met the inclusion criteria and contained active links to fish skin microbiome datasets.

Data quality assessments and data processing
Datasets were first filtered and validated to be derived from fish skin microbiomes or negative sequencing controls as assessed by the associated metadata.Studies were discarded if they had either low-quality reads that resulted in the loss of the majority of the dataset (ov er all r ead Phr ed scor e < 10; < 90% c hance base pairs ar e corr ectl y sequenced); insufficient metadata to determine the origin of sequences (as occurred for some studies where it was not possible to determine whether the samples were fish skin or water samples), or the study used reads that after quality trimming did not ov erla p r esulting in onl y partial cover a ge of the 16S V4 region.For each of the studies included in this analysis, standard sequencing adaptors and primers available as part of the bbmap bioinformatics toolkit (v38.91) were removed and reads quality trimmed with a minimum overall Phred score of 10 using BBDuk (Bushnell 2013 ).Following quality trimming, the D AD A2 (v1.26.0) pac ka ge in R was used to error corr ect, der eplicate, pseudo pool amplicon sequence variants (ASVs) (which allows for the retention of rare ASVs at low abundances across samples otherwise discarded as spurious due to its low abundance within a single sample) for combined sample inference.Reads were then merged into ASVs, with ASVs shorter than 248 base pairs (V4 region ∼ 254 base pairs) and c himaer as discarded (Callahan et al. 2016 ).Using a novel method, all ASVs were aligned using BLAST (v2.13.0) (Altschul et al. 1990 ) against a dataset containing only the V4 region (McMurtrie et al. 2022 ).ASVs from all studies could then be trimmed to the V4 region only and merged, otherwise not possible due to differences in variable regions sequences (e.g.V3-V4 and V4-V5 become just V4, allowing comparisons).This in turn allo w ed for phylogenetic-based comparisons between studies.Trimmed ASVs were then assigned to taxa using D AD A2 against the SILVA database silva_nr99_v138.1 (Quast et al. 2013 ).ASVs identified from eukaryotes, mitochondria, c hlor oplasts, or not present in at least two samples from the same study were discarded.This r emov ed potentiall y anomalous taxa present in a study only once.All studies were subsequently merged into one phyloseq object for ease of downstream analysis (McMurdie and Holmes 2013 ).ASVs identified in the negative sequencing controls were removed from all studies using the Decontam R pac ka ge (v1.18.0) pr e v alence method at a 0.5 thr eshold (Davis et al. 2018 ).This r emov ed potentiall y contaminating ASVs from the combined phyloseq object ensuring there was no bias, as not all studies published negative sequencing controls.Datasets were further filtered to removed samples with low ASV abundance and low numbers of reads .T his was done as these simple microbiomes when compared with highly complex microbiomes may obscure trends in microbial community structure making cross-comparisons difficult.This was done by retaining samples where the total number of reads was between the 2.5 and 97.5 percentile (k ee ping the mid dle 95%).Ad ditionally, ASVs summarized by abundances falling within the bottom fifth percentile for each fish species across the entire dataset were deemed rare taxa and r emov ed (i.e. the top 95th percentile of most abundant ASVs per fish species were retained only).The best phylogenetic tree substitution model for the 16S phylogenetic tree ASVs from all the merged studies was determined using IQ-TREE (v2.2.0.3) using the Bayesian information criterion (Kalyaanamoorthy et al. 2017 ).Maxim um-likelihood tr ees wer e constructed using IQ-TREE to determine the most likelihood phylogenetic tree (Nguyen et al. 2015 ) and added to the phyloseq object for downstream analysis.Scripts to r epr oduce this process can be found here: https: // github.com/ash-bell/ fish _ 16S _ metastudy _ public .

Amplicon analysis
The effects of different physicochemical factors on ASV abundance wer e anal ysed using R (v4.2.2) (R Core Team 2021 ).Host Fr om left to right, the number of studies that wer e r emov ed for each quality assessment criteria.Some studies passed initial quality control steps that assessed for availability of 16S rRNA V4 fish skin microbiome data but were subsequently discarded downstream for the numbers and reasons shown.linea ges wer e determined by querying TaxIDs from the NCBI taxonomic database (Sc hoc h et al. 2020 ) and stored as a dendr ogr am using the dendextend pac ka ge (v1.17.1) (Galili 2015 ).ASV abundance was normalized as r elativ e abundance within a sample.Phylogenetic distance matrices were determined and ordination plots were constructed using phyloseq (v1.Statistical test applied to calculate the difference between dissimilarity matrixes (PERMANOVA) were performed using the ado-nis2 function in the vegan package.A pairwise PERMANOVA test was performed using the pairwise.adonis2function from the pairwise.adonispac ka ge (Pedr o Martinez Arbizu 2018 ) (v0.4).All r eported tests were performed with PERMANOVA or pairwise PER-MANOVA had an adjusted P -value of < .001.Cultivation groups with only one replicate were not included in pairwise comparisons as the one replicate study became the sole weighting in a group, leading to potential bias.Scripts for this process can be found here: https:// github.com/ash-bell/ fish _ 16S _ metastudy _ public .

Papers chosen for analysis
Our database search for fish skin microbiome studies resulted in 290 manuscripts .T he full documents wer e scr eened for a pplicability to this study, filtering for 16S V4 region fish skin microbiome data that were publicly a vailable , as described abo ve (see the sections "Methods" and "Liter atur e searc h str ategy and pa per data selection").This resulted in 103 manuscripts with original data on fish skin micr obiomes (Fig. 1 ).Onl y studies on bony fish were included with studies on elasmobr anc hs including sharks (three), rays (two), and lungfish (one) excluded due to their distant phylogeny.One teleost study was discarded as it was on eggs only.Another 23 studies were discarded because there was no sequencing data released (this included four studies providing accession numbers but with no associated data).A further 10 studies were discarded as the sequencing was not based on the V4 region, including one study that did not indicate whic h r egion was used.Six studies were also excluded as they did not use MiSeq sequencing technologies (one having a discrepancy between the sequencing instrument used in the methods versus their metadata).Three further studies were excluded as they examined dead fish obtained from markets or food processing plants, thus were likely contaminated with microbes from harvesting, handing and/or pr ocessing.A further fiv e studies had to be r emov ed because the associated metadata did not allow for their identify as fish samples to be confirmed.This resulted in 49 studies that were suitable for inclusion within our meta-analysis (Fig. 1 ).
We then assessed these 49 studies for the sequencing read quality, excluding any studies that did not pass our basic quality control steps (Fig. 1 ).This resulted in the removal of eight of these studies because sequencing reads did not merge (due either to low base pair quality resulting in the truncation of reads, or discrepancies between overlapping regions preventing finding consensus sequences); three studies because there were so few reads per sample that error rates could not be estimated, therefore prev enting err or corr ection; and one further stud y because the n umber of forw ar d and r e v erse r eads differ ed, suggesting an incomplete dataset (Fig. 1 ).Studies were also removed from the analysis if no ASV was present more than twice in samples from the same study, indicating little to no diversity.Two published studies used the same BioProject number and were thus treated as a single study.This reduced the final total to 36 studies for our meta-analysis (Table 1 ).
Some studies used in the meta-analysis had over 100 000 reads per sample but on av er a ge we found 20% of reads did not pass quality control filters, with some studies discarding over 50% of reads due to their low quality.Other studies comprised of samples with around 20 000 reads of which few w ere discar ded during quality control steps.Using Good's cov er a ge of 100% (Good 1953 ), (indicating that no ASV appears in any samples only once) suggests fish skin microbiomes on average (median) reached saturation at 75 unique ASVs (212 unique ASVs, 95th percentile) and (median) 7245 reads (36 460 reads; 95th percentile) indicating that the majority of bacterial species fish skin microbiome are likely ca ptur ed with these numbers.
In total, our phyloseq object contained 8003 bacterial taxa in 1922 samples across 36 different studies, comprising 98 different fish species and sampled from a range of culturing systems and en vironmental conditions .We in vestigated the bacterial composition pr esent acr oss these fish species and assessed for interrelationships between their skin microbiomes and a range of environmental water conditions/parameters.

Bony fish skin have similar bacterial taxa that di v erge at a bacterial order level
Analysis of 98 different fish species indicated that the bacterial phylum Proteobacteria and class Gammaproteobacteria were the most abundant bacterial taxa in fish skin microbiomes .T hese occurred in all fresh and saltwater fish skin microbiomes analysed ( Figures S1 and S2 , Supporting Information ).Fish skin micr obiomes wer e clearl y differ ent in fr esh v ersus saltwater envir onments, and we ther efor e anal ysed fr eshwater fish separ atel y to marine species throughout this manuscript.The bacterial phylum Firmicutes was also highl y pr e v alent in fresh (found in 24/25 species) and saltwater fish (found in 55/57 species) (Fig. 2 ; Figure S3 , Supporting Information ).When comparing the similarities of saltwater fish to other salt water fish and freshwater fish to other freshwater fish, we observed that saltwater fish exhibit greater similarities with each other than freshwater fish do with one another.All saltwater fish skin microbiomes contained the bacterial order Enter obacter ales and family Vibrionaceae, with 68 out of 71 fish species analysed containing the genus Vibrio ( Figure S1 , Supporting Information ).F reshw ater fish skin microbiomes appeared more diverse at both a bac-terial order and family taxonomic le v el compar ed with saltwater fish ( Figure S1 , Supporting Information ).In freshwater fish, Burkholderiales and Enter obacter ales wer e the most abundant bacteria at an order le v el, and Aeromonas the most pr e v alent genus ( Figure S1 , Supporting Information ).Fish skin microbiome community compositions at a genus level were weakly correlated with host lineage (freshwater fish 5% correlation; saltwater fish 6% corr elation; Figur e S4 , Supporting Information ) and sho w ed variations between studies at a fish order le v el ( Figur es S5 , S6 , and S7 , Supporting Information ).

Fish cultured in different environments have different fish skin microbiome communities
Fish skin microbiome community compositions for fish cultivated in different environments had clear differences in their composition ( R 2 = 0.12, fresh and saltwater, P < .001)(Fig. 3 , Tables 2 and 3 ).Fish skin microbiomes were most similar between fish sampled in the wild and those in outdoor aquaculture systems (sea cages or ponds) ( R 2 = 0.01 freshwater, 0.03 saltwater, P < .001).The greatest differ ences wer e seen between fr eshwater fish held in labor atory flow-thr ough systems v ersus all other fr eshwater systems (Tables 2 and 3 ).Skin microbiomes of fish held in recirculating aquaculture system (RAS) tanks were most similar to fish from wild conditions ( R 2 = 0.06 within freshwater, 0.04 within saltwater, P < .001),follo w ed b y other tank systems with unspecified water sources ( R 2 = 0.09 within freshwater, P < .001)and treated flowthrough systems ( R 2 = 0.20 within freshwater, P < .001).Outdoor aquacultur e systems wer e most similar to fish fr om wild conditions, follo w ed b y tank ( R 2 = 0.07 within freshwater, P < .001),and RAS system ( R 2 = 0.14 within freshwater, 0.10 within saltwater, P < .001),and the most different to flow-through systems ( R 2 = 0.17 within freshwater, P < .001).
Both fresh and saltwater fish skin microbiome beta diversity correlated with temperature, salinity, pH, and dO 2 suggesting fish skin microbial community compositions are influenced by these factors (Fig. 4 ).dO 2 concentration was the most strongly correlated physicochemical factor ( R 2 = 0.12 within freshwater, 0.17 within saltwater, P < .001)followed by salinity ( R 2 = 0.12 within freshwater, 0.10 within saltwater, P < .001),pH ( R 2 = 0.06 within freshwater, 0.05 within saltwater, P < .001)and then temper atur e ( R 2 = 0.05 within freshwater, 0.03 within saltwater, P < .001)(Figs 4  and 5 ).Ho w e v er, as onl y a fe w of the studies provided accompan ying physicoc hemical metadata, this meant that onl y a low diversity of fish species was included in the water physicochemical comparisons.In addition, we emphasize that many studies only sampled one fish in the same conditions limiting confidence in these assessments on the effect of these physicochemical factors ( Figure S8 and Table S1 , Supporting Information ).
Table 1.List of fish skin microbiome studies with publicly available sequence data used within this manuscript.Laboratory (tank)-based aquaculture systems include flow-through (with unspecified water tr eatment), tr eated flow-thr ough (with microbial sterilization occurring before water input) or untreated flow-through (with no prior microbial sterilization and the outflow is discarded).A recirculating aquaculture system (RAS) is where the water is reused after passing through a (bio) filter system.Environmental (wild) refers to fish caught in natural freshwater and marine water bodies (e .g. streams , rivers , and seas).Outdoor aquaculture refers to ponds or sea cages exposed to the natural environment.

Variation in the skin microbial composition of Atlantic salmon ( Salmo salar ), a case study
As host factors likely confound attempts to qualify and quantify the influence of water physicochemistry on fish skin microbiomes, we investigated the variation in the skin microbiome for studies conducted on the Atlantic salmon ( Salmo salar ), the fish species for which there was the highest number of independent studies ( n = 5, Table 1 ); one study, ho w e v er, that had to be discarded due a low number of ASVs.Atlantic salmon skin microbial composition differ ed significantl y ( R 2 = 0.34, P < .001) between studies, e v en after accounting for differences in salinity (the only water physicoc hemical factor r ecorded for all studies).This indicates factors other than host species and salinity (such as fish strain, differences in sampling pr ocedur e or DNA extraction) play a role also in determining fish skin microbiome community composition for this species ( Figure S9 , Supporting Information ).
We also investigated whether temperature affected the skin microbiome of Atlantic salmon grown in freshwater (data not recorded for saltwater studies) albeit there were only two studies for this anal ysis ( Figur e S9 , Supporting Information ).Her e, we found the abundance of 109 ASVs comprising 69 genera of bacteria significantl y corr elated negativ el y or positiv el y with temper atur e c hanges (v egan; env.fit;P < .001cutoffs) ( Figur es S10 and S11 , Supporting Information ).

Discussion
Fish skin microbiomes play a crucial role in maintaining fish health, ho w e v er, the bacterial composition of these microbiomes and their interplay in response to environmental parameters are not w ell-understood.Here, w e use a meta-analysis framework to anal yse data fr om publicl y av ailable studies to inv estigate the  commonalities and differences in bacterial taxa in fish skin micr obiomes acr oss a r ange of fr eshwater and marine fish species from both cultivated and wild conditions.We investigated how host phylogeny and the environment influence these microbial assemblages and identify bacterial taxa that may constitute core elements of the fish skin microbiome.We also make suggestions  and recommendations for future studies relating to sequencing and data curation to better enable future comparisons of microbiomes between studies.

Bacterial composition of fish skin microbiomes
All fish skin microbiomes were dominated by the Proteobacteria phylum and Gamma pr oteobacteria class, as has been iden-tified pr e viousl y in the liter atur e (Boutin et al. 2014, Lokesh and Kir on 2016, Kr otman et al. 2020 ).Div er gence of fish skin microbiome community compositions occurred at the level of bacterial order, with no bacterial genus present in all fish species.Saltwater fish skin microbiomes shared greater similarities, compared to that observed within freshwater fish.Supporting this observation, microbiomes of marine damselfish ( Stegastes leucostictus ) separated by 225 km were seen to have similar skin beta-diversities (Xavier et al. 2020 ) and only 3% of the microbiome beta-diversity in cor al r eef fish w as explained b y r eef locations (Chiar ello et al. 2018 ).In contrast the beta diversity of the skin microbiome of fr eshwater Eur opean catfish ( Silurus glanis ) fr om riv er sites separated by 200 km differed significantly (Chiarello et al. 2019 ).These findings may reflect more consistent physiochemical properties across marine environments than in freshwaters.Supporting this, a 1-year time series study on Scomber japonicus skin microbiomes sho w ed little variation in marine water physicochemistry (Minich et al. 2020a ), whereas a study of Amazonian teleost species from freshw ater sites w ere separated by clear hydrochemical gradients (Sylvain et al. 2019 ).These findings support that fish skin microbiomes community compositions are influenced by the physiochemistry of their environment in both marine and freshwater en vironments .Common genera found in the skin microbiomes of freshwater fish included Acinetobacter , Aeromonas , Ralstonia , Sphingomonas (all Proteobacteria), and Flavobacterium (Bacteroidota); and in saltwater fish included Alteromonas , Photobacterium , Pseudoalteromonas , Psychrobacter , and Vibrio (all Gammaproteobacteria class).For some of the fish orders analysed there was only one species repr esentativ e in this meta-analysis (especially for Gadiformes, Istiophoriformes, and Centr arc hiformes), incr easing the likelihood that the presence of any given bacterial genera may be missed and in turn the microbiome bacterial assembla ges mor e univ ersal than suggested.
Gener a suc h as Aeromonas , Acinetobacter , Flavobacterium , and Vibrio contain species with known pathogenicity to fish.Aeromonas hydrophila , e.g. is ubiquitous within freshwater environments and is associated with diseases including bacterial haemorrhagic septicaemia and e pizootic ulcerati ve syndrome (Lategan et al. 2004, Chen et al. 2022 ).Flavobacterium psychrophilum is the primary agent of bacterial cold-water disease and rainbow trout fry syndrome, which is one of the main sources of economic loss of the salmonid industry (Duchaud et al. 2018 ).Acinetobacter spp.contain strains that ar e emer ging as septicaemic disease-causing a gents for a wide variety of freshwater fish (Malick et al. 2020 ), albeit they occur also in the skin microbiomes of nondiseased freshwater fish.Vibrio occurs commonly in the skin microbiomes of healthy saltwater fish but also include species such as Vibrio anguillarum that can cause septicaemia and are one of the most common saltwater fish pathogens in the aquaculture industry (Austin et al. 1995, Weber et al. 2010, Chaudhary et al. 2021 ).Species within the Ralstonia and Sphingomonas genera are poorly categorized especially within fish microbiomes.We cannot specify whether the genera highlighted contain these pathogens, but it is well established they can be commensals or pathobionts and can cause disease in response to host stressors (Bass et al. 2019 ).
The presence of bacteria such as Sphingomonas , Psychrobacter , and Ralstonia in microbiomes has r eceiv ed attention r ecentl y.Sphingomonas spp.synthesize astaxanthin, as part of the carotenoid synthesis pathway responsible for the red pigmentation in salmon and krill (Tam et al. 2021 ) and may act to provide some probiotic protection to fish, e.g. as shown in mortality rates in rohu ( Labeo rohita ) subjected to V. anguillarum infection (Chaudhary et al. 2021 ).Psychrobacter spp.includes opportunistic pathogens in both fish and humans and is the causativ e a gent of disease in Red Sea marine fish (El-Sayed et al. 2023 ).Howe v er, P. nivimaris and P. f aecalis hav e also been documented to hav e pr obiotic qualities, reducing mortality rates in juvenile turbot ( Scophthalmus maximus ) exposed to Tenacibaculum maritimum (see Wuertz et al. 2023 ).Ralstonia spp.include plant and opportunistic human pathogens (Waugh et al. 2010, Said et al. 2020 ), that occur natur all y in water and soil microbiomes (Ryan et al. 2011 ).Ralstonia spp ., along with Photobacterium and Acinetobacter , are dominant members of the fish gut microbiome, and may have a k e y role in dietary function or are transferred from fish skin during predation (Huang et al. 2020 ).The functional roles of these opportunists and commensals, ho w e v er, ar e poorl y understood.For bacteria such as Alteromonas and Pseudoalteromonas in the microbiome of saltwater fish, almost nothing is known about their function.The presence of common bacterial genera across widely differing species indicates some functional commonalities in the fish skin microbiome but gener all y this is still r elativ el y poorl y understood.T his ma y also change in response to environmental stressors and disease.For example, the genus Psychrobacter contains both opportunistic pathogens and symbionts, with thus both a pr otectiv e function but also a disease-causing impact (El-Sayed et al. 2023, Wuertz et al. 2023 ).It is well-established that when fish ar e str essed, their micr obiomes ar e perturbed leading to an incr ease in the abundance of opportunistic pathogens, which may lead to the de v elopment of disease (Minniti et al. 2017 ).A greater understanding of the function of the different bacterial genera and their relationship with the host fish and how they respond to environmental str essors is clearl y needed to better define their r oles in health and disease.

Rela tionship betw een fish phylogen y and skin bacterial community composition (phylosymbiosis)
Pr e vious studies have documented a correlation between the taxonomic relatedness of fish and the similarity in their microbiomes, a concept r eferr ed to as phylosymbiosis (Brooks et al. 2016, Chiarello et al. 2018 ).In our analysis we found a r elativ el y limited link between fish skin bacterial community compositions and host lineage, ho w ever, this may relate to the fact that many fish were represented by a single species.Also, within our analysis, the pr e v alence of bacterial gener a in the skin micr obiome v aried for different fish orders .Illustrating this , for Salmoniformes ther e was lar gel y a consistent bacterial composition and abundance r eported acr oss m ultiple studies, wher eas in the P er ciformes there was a far gr eater v ariability in the most pr e v alent taxa.In the P er ciformes fish or der, only Photobacterium , Pseudoalteromonas , and Vibrio genera were highly prevalent and abundant in all host species analysed.This disparity, ho w e v er, may in part be because the most common P er ciform in this analysis, Dicentrarchus labrax , was cultured in estuarine waters while all other Perciformes species were from saltwater en vironments , lik ely dri ving differences in the skin community composition for the analysis on the P er ciforms.When comparing the skin microbiomes of fish from within the same host orders held within saltwater across multiple studies there was a higher degree of similarity of bacterial genera (see e.g.members from the Pomacentridae fish famil y; Figur e S5 , Supporting Information ).Although some genera were common across different fish species (such as Aeromonas in freshwater and Vibrio in saltwater) some gener a a ppear ed to be limited to specific taxonomic orders, suc h as Chryseobacterium in the Salmonidae ( Figure S7 , Supporting Information ).In studies on coral reef fish where several different fish species wer e anal ysed within the same study stronger corr elations hav e been r eported between host linea ge and skin micr obial comm unity compositions (Chiar ello et al. 2018, Pr atte et al. 2018 ).

Effect of cultiv a tion system on fish skin bacterial community compositions
We hypothesized that the diversity and composition of fish skin bacterial communities are likely to be influenced by fish culturing systems, due to different environmental conditions.Our results indicate that ov er all, 12% of beta diversity was explained by fish culturing systems in fresh and also within saltwater conditions.This supports pr e vious studies indicating that differ ent aquacultur e envir onments can r esult in differ ent fish microbiome comm unity structur es (Minic h et al. 2020b, Lor gen-Ritc hie et al. 2022 ).Skin microbiomes in fish held in flow-through aquaculture systems differed most from all other types of fish cultivation systems (RAS, wild-caught, and outdoor aquaculture), with RAS systems more closely resembling fish captured from the wild or from aquacultur e envir onments, whic h wer e br oadl y similar.These differences in skin microbiota may reflect differences in water supplied for fish in ca ptiv e conditions , e .g. systems sterilized via UV and/or ozone, whereas RAS systems with biofilters only have a rich bacterial flora that will also be also in the water column.

Water physiochemistry factors shaping fish skin microbiomes
For both fresh and saltwater conditions, difference in water physicochemical factors including salinity, temperature, pH, and dO 2 correlated with fish skin bacterial community composition (beta diversity), with the strongest correlations seen with dO 2 concentration, follo w ed b y salinity, pH, and with smallest effect for temper atur e.For fr eshwater fish, temper atur e was positiv el y correlated with alpha diversity.No correlation was observed for pH or dO 2 , and there was a negativ e corr elation with conductivity.For saltwater fish, alpha diversity increased with a more basic pH, higher salinity, and dO 2 concentration but decreased at higher temper atur es.Studies assessing the effect of physicochemical factors on fish skin microbiomes have similarly shown that exposure to high temperatures results in decreases in alpha div ersity ov er time and differences in beta diversity in saltwater gr eater amberjac k ( Seriola dumerili ) (Sánc hez-Cueto et al. 2023 ) and in chum salmon ( Oncorhynchus keta ) (Ghosh et al. 2022 ).Studies on the flag cichlid ( Mesonauta festivus) and black piranha ( Serrasalmus rhombeus) found no effects of pH on skin microbiomes (Sylvain et al. 2019 ), ho w e v er, in Tambaqui ( Colossoma macropomum ) (data included within this study) exposure to a low pH (pH 4.0) compared to control (pH 6.3) resulted in changes in fish skin beta diversity, but not alpha diversity (Sylvain et al. 2016 ).Small differences in salinity (less than 2 ppt) were not found to correlate with fish skin beta diversity in Pacific chub mackerel ( S. japonicu s) (see Minich et al. 2020a ) or for 44 different coral reef fish species (see Chiarello et al. 2018 ) (data included within this study).In the black molly ( Poecilia sphenops ), clear differences were seen for larger salinity shifts (13 ppt), which may relate to changes in fish osmoregulation (Schmidt et al. 2015 ) and in anadromous Arctic Char ( Salvelinus alpinus ) (data included within this study) alpha diversity was lower in freshwater conditions with differences in beta diversity (Hamilton et al. 2019(Hamilton et al. , 2023 ) ).Studies on fish from the River Jordan (data included within this study) found differences in beta diversity in fish skin microbiomes exposed to different temperatures, dO 2 concentrations, conductivity (salinity), and pH.Incr eases in alpha div ersity were also observed with increased temperature, dO 2 concentrations, conductivity (salinity), and a more basic pH.Ho w ever, these physiochemical factors were substantially less correlated with alpha diversity in "corrected" skin communities (when water microbiota at the same abundance in the fish skin were removed).
In contrast, in this study we only found an increasing alpha diversity relationship with increasing temperature, the most correlated physicochemical factor in Krotman et al. ( 2020)'s study.Our alpha div ersity corr elation data illustr ate that salinity, temper atur e, pH, and dO 2 can all influence the fish skin microbiome, but the magnitude and direction of the effect is lik ely de pendent on the magnitude of change in the water physicochemical parameters, and/or by the physiological adaptability of the fish to the environmental conditions.Changes in water physicochemistry may also affect the water microbiome and this in turn could play a part in reshaping the skin microbiome (Krotman et al. 2020, Sylvain et al. 2020 ).
When considering Atlantic salmon onl y, differ ent studies hav e r eported differ ent skin micr obiomes, e v en when accounting for physicoc hemical factors suc h as salinity (Lor gen-Ritc hie et al. 2023 ).Other factors, ther efor e, r elating their maintenance or other aspects of water chemistry must play a role in determining fish skin microbiome community composition.These factors likely include salmon genetics (seed stocks), water microbiota, husbandry practice, life history and/or diet (Minniti et al. 2017, Bledsoe et al. 2022, Lor gen-Ritc hie et al. 2022 ).Our analysis on the effect of water temper atur e on the skin micr obiome in Atlantic salmon was extr emel y limited due to the available data but ne v ertheless the indication from that analysis was for water temper atur e associated changes in the abundance of 69 genera of bacteria in Atlantic salmon skin microbiomes.Further studies are thus much needed to determine what specific bacterial taxa (such as pathobionts and commensals) are enriched or depleted in fish skin microbiomes with features of w ater physiochemistry.This w ould support optimizing aquaculture systems to ensure optimal animal health, and we suggest this is especially true for closed (e.g.RAS) production systems.

Fish skin microbiome concluding remarks
From this study, we show that the dominant taxa in fish skin bacterial microbiomes are the Proteobacteria phylum and Gamma pr oteobacterial class, r egardless of host, envir onment, and water salinity.When comparing different hosts, fish skin micr obiomes div er ge at a bacterial order le v el, with some similarities between closely related species (phylosymbiosis) (Brooks et al. 2016 ).Fish skin microbiome differs significantly between saltwater and freshwater fish and the top genera found in salt and freshwater are completely different, with greater similarities occurring between saltwater fish.These top genera include potential pathobionts, that may cause disease when host microbiomes are perturbed, and commensals which play a crucial role in pathogenic defence and nutrient degradation likely resulting in higher host fitness.Although the type of cultivation environment influences fish skin microbiome composition, the host and other environmental factors appear more influential in determining fish skin micr obiome comm unity compositions.Mor e studies on different fish species maintained in the same environments and across a range of physiochemical conditions are required to better determine specifics on the bacterial taxa enriched by differ ent envir onmental conditions.Meta studies suc h as the one pr esented her ein would notabl y benefit fr om published studies pr oviding mor e compr ehensiv e metadata r elating to the fish (e.g.genetics , strains , and so on) and water physicochemistry exposur e conditions.Encour a ging this will expand the utility of microbiome sequencing data and allow for further investigations beyond the initial, primary, studies.Data integration of amplicon sequencing fr om div er gent studies, r epr oducibility, and futur e

Figure 1 .
Figure 1.Distribution of studies analysed in the current study.103 published studies containing relevant and original data on fish skin microbiomes wer e r etrie v ed and filter ed down to a final 35 unique BioPr ojects fr om 36 published manuscripts.Fr om left to right, the number of studies that wer e r emov ed for each quality assessment criteria.Some studies passed initial quality control steps that assessed for availability of 16S rRNA V4 fish skin microbiome data but were subsequently discarded downstream for the numbers and reasons shown.

Figure 2 .
Figure 2. Relative abundance of fish skin microbiome bacterial community compositions at an ASV level, grouped by: (A) freshwater fish bacterial phylum le v el; (B) fr eshwater fish bacterial class le v el; (C) saltwater fish bacterial phylum le v el; and (D) saltwater fish bacterial class le v el.

Figure 3 .
Figure 3. PCoA of a Weighted UniFrac dissimilarity matrix at the ASV level grouped and colour coded by cultivation system in which the fish were sampled or maintained.Table 2. PERMANOVA and pairwise PERMANOVA post hoc test results comparing different freshwater fish culturing conditions from Fig. 3 .R 2 indicates ho w w ell the v ariables fit the PERMANOVA model.Cultiv ation systems with onl y one study wer e r emov ed in pairwise comparisons due to potentially biased study weighting.All statistical tests have a P (adjusted) value of < .001.

Figure 4 .
Figure 4. Alpha and beta diversity of freshwater fish skin microbiomes associated with features of water physicochemistry.Shannon (alpha) diversity is correlated using a linear regression model against associated physicochemical factors and correlated using Spearman's correlation.Beta diversity is coloured according to associated physiochemical data using a principal coordinate analysis (PCoA) from a Weighted UniFrac dissimilarity matrix at an ASV le v el.(A) and (B) temper atur e, (C) and (D) conductivity, (E) and (F) pH, and (G) and (H) dO 2 concentration.

Figure 5 .
Figure 5. Alpha and beta diversity of saltwater fish skin microbiomes associated with recorded physicochemical factors.Shannon (alpha) diversity is correlated using linear regression to associated physicochemical factors and correlated using Spearman's correlation.Beta diversity is coloured according to associated physiochemical data using a principal coordinate analysis (PCoA) from a Weighted UniFrac dissimilarity matrix at an ASV le v el.(A) and (B) temper atur e. (C) and (D) conductivity, (E) and (F) pH, and (G) and (H) dO 2 concentration.

Table 3 .
PERMANOVA and pairwise PERMANOVA post hoc test results comparing different saltwater fish culturing conditions from Fig. 3 .R 2 indicates ho w w ell the v ariables fit the PERMANOVA model.Cultiv ation systems with onl y one study wer e r emov ed in pairwise comparisons due to potentially biased study weighting.All statistical tests have a P (adjusted) value of < .001.