Fish skin microbiota under healthy and diseased conditions

Background: The welfare of farmed sh is inuenced by numerous environmental and management factors. Fish skin is an important site for immunity and a major route by which infections are acquired. The objective of this study was to characterize bacterial composition variability on skin of healthy, diseased and recovered Gilthead Seabream (Sparus aurata) and Barramundi (Lates calcarifer). S. aurata, which are highly sensitive to gram-negative bacteria, were challenged with Vibrio harveyi. In addition, and to provide a wider range of infections, both sh species (S. aurata and L. calcarifer ) were infected with gram-positive Streptococcus iniae, to compare the response of the highly sensitive L. calcarifer to that of the more resistant S. aurata. All experiments also compared microbial communities found on skin of sh reared in UV (a general practice used in aquaculture) and non-UV treated water tanks. Results: Skin swab samples were taken from different areas of the sh (lateral lines, abdomen and gills) prior to controlled infection, and 24, 48 and 72 hours, 5 days, one week and 1 month post-infection. Fish skin microbial communities were determined using Illumina iSeq100 16S rDNA for bacterial sequencing. The results showed that naturally present bacterial composition is similar on all sampled sh skin sites prior to infection, but the controlled infections (T 1 24 h post infection) altered the bacterial communities found on sh skin. Moreover, when the naturally occurring skin microbiome did not quickly recover, sh mortality was common following T 1 (24 h post infection). We further conrmed the differences in bacterial communities found on skin and in the water of sh reared in non-UV and UV treated water under healthy and diseased conditions. Conclusions: Our experimental ndings shed light on the sh skin microbiome in relation to sh survival (in diseased and healthy conditions). The results can be harnessed to provide management tools for commercial sh farmers; predicting and preventing sh diseases can increase sh health and welfare, and enhance commercial sh yields.

from all treatment groups (UV and non-UV) were acclimated to the environment for two weeks in separate tanks and fed 2% of their body weight daily.

Fish tagging and sampling
Fish were anesthetized using clove oil (25 µl/L for 10 min until loss of movement, followed by 12.5 µl/L for continued anesthesia supplemented with aeration). Each sh was assigned a different serial number by injecting a subcutaneous (S.C) P-tag (Trovan). Fish from each treatment group were sampled using a sterile cotton swab (FLOQSwabs in tube ® 553C -COPAN) during the tagging process at the beginning of the experiment (T 0 ) and at each time point as described below.
For the V. harveyi bacterial infection experiments, samples were taken by swabbing a ~ 1 cm 2 area of sh skin mucous layer. One swab was taken from each area: abdomen (A), gills (G, taken from the right laments between the rst and second gill arch), and right side lateral line (L). Samples were collected at the beginning of the experiment (T 0 ), 24 h after stress and exposure to pathogen (T 1 ), after 1 week (T 2 ) and after 3 or 5 weeks (T 3 -week 3 in the winter experiment and week 5 in the summer experiment) ( Table   1).
After sampling, each swab was inserted into a clean, sterile and dry test tube and was kept at -80ºC until analysis. Fish were monitored daily for signs of disease. In addition, 500 ml samples of water from the sh tanks were ltered through a 0.22 µm lter paper (Macherey-Nagel (MN) USA) at the set time points, with an additional sampling point at 60 min after infection (Ts). All samples were kept at -80 ºC until used.

Pathogenic bacterial culture and application
V. harveyi, and S. iniae were obtained from the National Center for Mariculture (NCM) pathology department. Bacterial stocks kept in -80 °C, were inoculated in a laminar ow hood on tryptic soy agar (TSA, DIFCO USA) prepared with 25% sterile seawater, and incubated at 24 ± 1 °C for 48-72 h. After the incubation period, the bacterial isolates were transferred to tryptic soy broth (TSB, ACUMEDIA USA) prepared with 25% sterile seawater and incubated again for another 48-72 h at 24 ± 1 °C. OD values from the bacterial concentration were read at 600 nm using a microplate spectrophotometer (PowerWave TM XS, BioTek, Winooski, USA).

Stress implementation, infection and sh monitoring
After sh tagging (described in Section 2.1), stress was implemented to magnify the impact of the bacterial infection as follows, Fish were netted out of the water for 5 min (handling stress), and then subjected to a needle scratch on their caudal n by a sterile (23 G) needle. Immediately afterwards, sh were immersed in a Vibrio harveyi bacterial suspension (250,000 bacterium / ml) in a reduced water tank volume (5 l). After 60 minutes of immersion, the water tank was gradually re lled to its initial volume of 100 l within one hour. After a 24 h recovery period, rst samples were taken (T 1 ). All bacterial infections were done the same way except for a small modi cation for S. iniae. In the S. iniae trial, sh were transferred to an aerated container with 5 l of seawater containing bacterial suspension at a concentration of 5 x 10 7 CFU bacteria/l for 10 min and returned to their respective tanks. Fish were monitored daily throughout the experiments for signs of disease, sh showing clinical signs were recorded and freshly dead sh were sampled for bacteriological analysis to resolve disease etiology.
Mortality rates from the different treatment groups in all experiments were recorded.

DNA extraction, library preparation and Illumina sequencing
Swab samples taken from different treatments at different time points were individually clipped under sterile conditions and set up for DNA extraction using the MoBio 96-well plate PowerSoil DNA Isolation Kits (MO BIO Laboratories, California, USA), following the manufacturer's protocol. All steps of DNA extraction were carried out in a sterile UV-hood (DNA/RNA UV-cleaner box, UVT-S-AR bioSan, Ornat, Israel) to reduce external contaminations. In every DNA extraction, 200 µl of RNase free water was used as a negative control (Sigma Aldrich, Israel). All samples were placed randomly in the DNA extraction plate to exclude any bias.
For the V. harveyi infection experiment (Table 1), in order to increase phylogenetic resolution and diversity estimates, a multiplex PCR using ve different sets of the 16S rDNA genes was applied to cover about 1000 bp of the 16S rRNA gene [34] (Supplementary Table 1). First PCR (PCR I) reactions were performed in triplicates, where each PCR-I reaction (total 25 µl) contained: a) 12.5 µl of KAPA HiFi HotStart ReadyMix (biosystems, Israel), b) 0.4 µl of equal v/v mixed primers forward and reverse primers, c) 10 µl of molecular graded DDW (Sigma, Israel), and d) 2 µl of (2-100 ng/µl) DNA template. PCR I reactions were performed in Biometra thermal cycler (Biometra, TGradient 48) as follows: initial denaturation at 95 °C for 2 min, followed by 35 cycles of 98 °C for 10 sec, 61 °C for 15 sec, and 72 °C for 7 sec. The PCR I routine ended with a nal extension at 72 °C for 1 min. Upon completion of PCR I, we ran an electrophoresis gel to verify that all samples were successfully ampli ed. Following successful and veri ed ampli cation, triplicate samples were pooled together and cleaned using Agencourt® AMPure XP (Beckman Coulter, Inc, Indianapolis, USA) bead solution following the manufacturer's protocol.
Library preparation was performed using a second PCR (PCR II) to connect the Illumina linker, adapter and unique 8 base pair barcode for each sample [34]. The PCR II reactions were prepared by mixing 21 µl of KAPA HiFi HotStart ReadyMix (biosystems, Israel), 2 µl of mixed primers with Illumina adapter (Supplementary Table 2), 12.6 µl of RNase free water (Sigma, Israel), and 4 µl of each sample from the rst PCR product with 2 µl of barcoded reverse primer. This was placed in Biometra thermal cycler (Biometra, TGradient 48) as follows: initial denaturation 98 °C for 2 min, and then 8 cycles of 98 °C for 10 sec, 64 °C for 15 sec, 72 °C for 25 sec, and a nal extension of 72 °C for 5 min. Then all PCR II products were pooled together and cleaned using Agencourt® AMPure XP (Beckman Coulter, Inc., Indianapolis, USA) bead solution following manufacturer's protocol, where 50 µl of pooled PCR II product were cleaned using 1:1 ratio with the bead solution for more conservative size exclusion of fragments less than 200 bp, and at the nal step, 50 μl of DDW with 10 mM Tris [pH = 8.5] were added to each sample. This was followed by aliquoting 48 µl of the supernatant to sterile PCR tubes and storing in -80 °C, while an additional 15 µl of the nal product was sent to the Hebrew University (Jerusalem, Israel) and sequenced on full lane of 250 bp paired-end reads (to correct for sequencing errors and enhance total read quality) using Illumina MiSeq platform.
For the S. iniae infection experiment (Table 1), we used V4-16S rDNA F515 and R806, [35] and its related Illumina primers (Supplementary Table S3) for PCR I and PCR II using the same aforementioned protocols and procedures, however, DNA samples were sequenced using 150 bp paired-end reads using Illumina iSeq100 platform at our laboratories.

Sequence curation and quality control
First, the V. harveyi infection experiment sequences were ltered for PhiX using Bowtie2 [36], then incomplete, low-quality reads (phred Q threshold 33) and incomplete paired sequences were removed using PEAR software [37]. Following the previous quality control steps, sequences were analyzed using QIIME-2 software [38]. In QIIME-2, sequences were aligned, checked for chimeric sequences and clustered to different OTUs (operational taxonomic unit) based on 99% sequence similarity, then classi ed based on Greengenes database V13.8 [39]. The generated OTU table was also cleared from sequences classi ed as f__mitochondria, c__Chloroplast, k__Archaea and K__Unclassi ed. Both the number of raw sequences and bacterial classi ed sequences were recorded in Supplementary Table 1 and the third primer set (F649-R889) was selected as representative for the microbial community composition (see Supplementary materials and methods Doc1, Data validation).
After the S. iniae infection experiment in 2020, the sequences (F649-R889) from the earlier V. harveyi infection experiment (summer 2015 and winter 2016) were curated and analyzed together as follows: rst, samples were ltered for primer sequences, then sequence errors were cleared with MAX_CONSIST=20 and repeated sequences were removed. Then sequences were clustered using DADA2 [40], and paired-end sequences were merged with minimum overlapping of 20 base pairs. After merging, samples were cleaned from chimeric sequences, the sequences were assigned to taxonomical classi cation using Silva database V138 with 99% sequence similarity [41] and an ASV table was generated. A similar analytic procedure was performed for the S. iniae infection experiment sequences, however, we rst produced the paired-end sequences (to obtain similar fragment length as in the V. harveyi experiment) using PEAR, and then we followed the same protocol.

Data curation and analysis
Data curation: Both generated ASV tables (2015 + 2016 and 2020 experiments) were curated as follows: only sequences classi ed in the kingdom Bacteria were maintained, then sequences classi ed as NA_Phylum, Chloroplastes_Order and Mitocondria_Family were removed from both ASV tables. Then only samples having a total sequence number of over 1,000 sequences were maintained for downstream analysis. Following initial data curation, additional lters were applied to remove noise, for example, we removed low read ASVs (≤ 10 reads) (Supplementary table S4). Afterwards, a rarefactions curve was produced (Supplementary Figure S1).
Data analysis: Non phylogenetic alpha diversities, including (A) Chao1 species' diversity estimate [42], (B) Shannon diversity [43], (C) Simpson diversity index [44] were calculated using the VEGAN package in R [45]. Faith's phylogenetic diversity [46] was calculated from the curated dataset using the PhyloMeasures package in R [47]. After determining alpha diversity, we compared beta diversity among groups and treatments. To investigate the absolute and weighted "abundance" of shared ASVs, we generated different Venn diagrams using the "eulerr" package in R [48]. Then PCoA dissimilarity ordination plots were generated based on weighted unifrac distance matrix explaining beta diversity variations among the different treatments and temporal scales. Signi cance tests were performed for the various treatments using single or pairwise comparisons using permutational multivariate analysis of the variance (adonis) based on Bray-Curtis distance matrix [49]. Taxonomic distribution graphs were generated based on the ASV tables, each phylum was assigned a distinct color and all genera under the same phylum were assigned different shades of the same color.

Effect of UV and non-UV treated water on sh survival
For the V. harveyi infection experiments, sh mortality was recorded daily ( Figure 1). Dead sh were removed from the experiment and subjected to bacteriological analysis to con rm mortality etiology. Figure 1A shows that S. aurata sh reared in non-UV treated water had signi cantly higher survival rates following V. harveyi infection (60% survival in summer and 20% in winter), compared to sh reared in UV treated water (no survivors in either season).
In our second round of experiments, we assessed S. iniae bacterial infection in the less susceptible S. aurata and more susceptible L. calcarifer sh ( Figure 1B). For the S. iniae infection, the survival rate of L. calcarifer increased from 20% in UV treated water to 40% in non-UV treated water. The survival of S. aurata infected with S. iniae in UV treated water was 100% compared to 80% in non-UV treated water.
3.2 Commensal bacterial diversity at different spatial and temporal changes following pathogenic bacterial infection Fish skin bacterial diversity estimates, including non-phylogenetic Chao1 species' diversity, Shannon and Simpson diversity index, and Faith's phylogenetic diversity (Figure 2) all showed a slightly higher diversity for non-UV treated water compared to UV treated water. Interestingly, during the infection (T 1 ), we saw a remarkable decrease in sh skin ora diversity estimates of both V. harveyi and S. iniae pathogens and these diversities returned to their initial level at T 2 and T 3 , corresponding to one week and one month post infection. Figure 2 presents diversity estimates at the different body sites (Abdomen, Gills and Lateral) during V. harveyi infection in the summer 2015 and winter 2016 experiments. At different time points, we noticed a higher similarity in those diversity estimates for both sh abdomen and lateral lines compared to gills which showed slightly higher estimates, however these differences showed to be insigni cant when compared using Tukey's test (Supplementary Figure S2).
To better illustrate these differences and to evaluate related patterns in the bacterial communities, PCoA ordination plots based on weighted unifrac distance matrix were generated for the different experiments.   Table S5) show distinctly unique sh skin bacterial communities in sh from UV and non-UV tanks for S. aurata during both V. harveyi ( Figure  3B) and S. iniae infection ( Figure 3C) but not for L. calcifer ( Figure 3D, Supplementary Table S5). Interestingly, S. aurata only showed a signi cant difference in community composition when comparing the water treatments at T 0 but not at T 1 for the V. harveyi infection. However, during the S. iniae infection, signi cant differences in bacterial communities were evident when comparing the UV and non-UV tanks at all time points. Note, there were no signi cant differences in the bacterial composition of gills at T 0 when comparing UV and non-UV treatments for S. aurata during the V. harveyi infection (Supplementary  Table S5).
When looking at core and unique microbial ASVs, only 21, 31 and 25% of all ASVs are shared between both UV and non-UV treatments for winter, 2016 V. harveyi infection and for S. iniae infection in S. aurata and L. calcifer respectively, these also constitute 93, 93 and 85% of weighted bacterial abundances, respectively (Supplementary Table S6). In contrast, the percent and weighted percentages of the shared and unique microbial communities were similar among the different time points and body sites. Notice, sh skin microbial communities in non-UV treatment had a higher percentage of unique ASVs compared to sh reared in UV treated tanks, this percentage declined during infection (Supplementary Table S6B) and gradually increased post infection. Interestingly, the number of shared ASVs has shown to positively correlate with disease severity and negatively correlate with survival rate (Figure 1).
When comparing the microbiome of different sh body sites (abdomen, gills and lateral line), Figures 3A and B do not show a clear separation. When testing the signi cant differences between sh body sites at different time points in UV and non-UV treatment (Supplementary Table S7), microbial communities do show signi cant differences but only at a few time points. Differences in microbial communities are evident when comparing the microbial communities of the lateral line to gills at T 0 and T 1 in the non-UV treatment in both summer and winter experiments, and once again when abdominal microbial communities were compared at T 1 in the summer experiment (Supplementary Table S7A). The unweighted and weighted percentages of unique ASVs for different body sites at different time points for non-UV treatments (Supplementary Table S7B Table  S8A). Interestingly, the microbial community on skin of surviving sh returned to its original composition two weeks post infection (comparing between T 0 and T 3 , P-values > 0.05). When infecting both S. aurata ( Figure 3C) and L. calcarifer sh species with S. iniae ( Figure 3D), we attempted to monitor changes in the microbial communities at higher temporal resolution to understand their interactions and impacts on sh health. Therefore, additional sampling time points were added at 48 h (T 1-2 ), 72 h (T 1-3 ), 5 days (T 1-5 ), 1 week (T 2 ), and 2 weeks (T 2-2 ) post infection. Differences in the microbial communities between the UV and non-UV treatments were observed ( Figure 3C, Supplementary Table S5), in addition, the PCoA plot presented in Figure 3C also shows interesting temporal patterns. The microbial communities showed a gradual deviation from T 0 downward along axis2 (explaining 19% of variance) for samples taken at T 1 and T 1-2 (24h and 48 h after infection, respectively), while at T 1-3 (72 h after infection) the microbial communities began to return to the original composition, which was similar to T 1 (Supplementary Table S8B). Interestingly, after ve days (T 1-5 ), one week (T 2 ), two week (T 2-2 ) microbial communities gradually moved to cluster with T 3 (one month after infection) which was similar to the original microbial communities (P-value = 0.056 between T 0 and T 3 ). In the UV treatment, there were no signi cant differences at the different time points compared to T 0 (Supplementary Table S8C), yet a signi cant difference was observed comparing different stages of infection (T 1-2 , T 1-3 , T 1-5 , T 2 , T 2-2 and T 3 ), quite similar to differences seen in the non-UV treatment.
During S. iniae infection of L. calcarifer, which is a highly susceptible sh species (see Figure 1B), T 0 did not show a clear separation of the microbial communities or statistical differences between the non-UV and UV treatments (Supplementary Table S5). There were no clear differences at different time points neither before nor after infection in both non-UV and UV treatments (Supplementary Table S8D Figure 4 illustrates the relative bacterial abundance during V. harveyi infection in the summer 2015 experiment, the bar graph shows three main bacterial phyla dominating the total bacterial abundance, Proteobacteria (blue, red and white), Firmicutes (pink) and Actinobacteria (yellow). At T 0 , before infection, Proteobacteria abundance was 63.9 ± 12.2%, followed by Firmicutes (14.8 ± 11.0%) and Actinobacteria (13.1 ± 8.1%). Following infection, these relative bacterial abundances changed at T 1 (24 h after infection) and T 2 (1 week after infection) and T 3 (3 weeks after infection) yet they were still dominant and the nal relative abundances at T 3 were 56.6 ± 6.3% for Proteobacteria, 27.1 ± 19.0% for Firmicutes and 11.0 ± 4.9% for Actinobacteria. To better understand the changes in relative bacterial abundances and their effect on sh health before, during and after infection, we analyzed the relative bacterial abundances to pinpoint their signi cant changes at different sampling points using DeSeq (Supplementary Figure S3). DeSeq analysis showed seven ASVs to signi cantly differentiate at different weeks post infection (T 2 and T 3 ), its relative abundance declined to less than 1% of the total bacterial abundance and was replaced with Del ta ASV, their relative abundances were 11.2 ± 8.2% and 24.2 ± 16.9% respectively ( Figure 4).

Bacterial community compositions
When repeating the same experiment in winter 2016, we analyzed relative bacterial abundances in both UV and non-UV treated tanks ( Figure 5). Figure 5 shows the similar three main bacterial phyla dominating the total bacterial abundance: Proteobacteria, Firmicutes and Actinobacteria. The relative abundances of these bacterial phyla proved to be different in the non-UV and UV treatment tanks. At T 0 , before infection, Proteobacteria abundance in non-UV vs. UV treated water was 57.3 ± 6.5% and 53.5 ± 9.1% respectively, followed by Actinobacteria (18.2 ± 9.6% and 15.6 ± 9.8%) and Firmicutes (15.1 ± 7.1% and 20.1 ± 9.3%). DeSeq analysis shows that the abundance of ten bacterial ASVs signi cantly differentiate in different UV treatments and time points Del ta and Bacillus genus of Proteobacteria and Firmicutes are among the most abundant ASVs (Supplementary Figure S4). Interestingly, Delftia (white) and Bacillus (yellow) showed a signi cantly different distribution following UV treatment at T 0 : the abundance of both Delftia and Bacillus decreased from 33.9 ± 15.3% to 4.9 ± 2.4%, and from 5.6 ± 3.6 to 0.2 ± 0.5% respectively. At T 1 , 24 h after infection with V. harveyi, the bacterial community was dominated by the Vibrio family (red), with a relative abundance of 85.0 ± 10.4% and 80.6 ± 13.7% for non-UV and UV treated water, respectively ( Figure 4). Moreover, Delftia genus also showed a higher abundance in non-UV treated (2.9 ± 5.2%) compared to UV treated water (0.3 ± 0.3%), while Photobacterium (cyan, belonging to Vibrio family) showed a higher abundance in UV treated (1.9 ± 3.9%) compared to non-UV treated water (0.2 ± 0.8%) (Supplementary Figure S4). At T 2 (one week post-infection) and T 3 (three weeks postinfection), all sh from UV treated water perished, in non-UV treated tanks, the relative abundance of vibrio decreased to 58.9 ± 20.7% at T 2 and to 6.6 ± 6.5% at T 3 ( Figure 4). On the other hand, Delftia increased from 6.9 ± 5.0% to 16.6 ± 8.7% at T 2 and T 3 respectively and the nal relative abundances of the main bacterial phyla were 50.4 ± 10.9%, 35.1 ± 14.8% and 9.7 ± 2.9% for Proteobacteria, Actinobacteria and Firmicutes respectively, similar to the initial abundance at T 0 . Interestingly, the Del ta ASV was dominant at T 0 for both summer (33.9 ± 15.3%, Figure 4), and winter experiment (29.5 ± 15.1%, Figure 5), whereas at T 1 , Del ta relative abundances signi cantly declined to 2.1 ± 3.9% and  Figure 6), the relative abundance of S. iniae did not dominate sh skin lateral line (unlike sh infected with V. harveyi) and showed a relative abundance of 9.0 ± 8.6% and 5.9 ± 4.5% for non-UV and UV treatments, respectively. Yet, DeSeq analysis showed Unclassi ed_Vibrionaceae family ASV (brown, Supplementary Figure S5), and Unclassi ed_Gammaproteobacteria Class ASV (black, Supplementary Figure S5), to signi cantly differentiate at different time points and between UV and non-UV treatment. Unclassi ed_Vibrionaceae Family showed an increased abundance from 2.4 ± 1.6% at T 0 to 18.3 ± 6.3% at T 1 for non-UV treatment and from 13.7 ± 20.9% at T 0 to 44.5 ± 27.6% at T 1 for UV treated water. At T 1-2 (72 h after infection), Unclassi ed_Vibrionaceae Family abundance further increased to reach 51.4 ± 7.4% for non-UV treated water. In addition, following the infection, Unclassi ed_Gammaproteobacteria (black) ASV showed an increased abundance at the recovery stage (T 1-2 , T 1-3 , T 1-5 and T 2 ), however its abundance was higher in the UV treatment compared to non-UV treated water (Supplementary Figure S5).
The S. iniae infection experiment was also conducted on L. calcarifer sh (Figure 7). At T 0 , before infection with S. iniae, the main bacteria phyla found on the lateral line of L. calcarifer were Proteobacteria, Firmcutes and Bacteroidota at relative abundances of 70.3 ± 3.6% and 72.8 ± 6.5% for non-UV treatment respectively, and 10.9 ± 2.2% and 8.2 ± 4%, 5.1 ± 2.3% and 7.3 ± 2.9% for UV treated water, respectively. At T 1 (24 h post infection), there was a similar yet higher increase in S.

Microbial diversity, composition and survival rate of sh in UV and non-UV treated water post infections
We compared sh skin microbiome in a controlled environment, in UV and non-UV treated water, in healthy and diseased sh. In most experiments, sh reared in UV treated water showed signi cantly reduced survival following bacterial infection ( Figure 1). As expected, infections by S. iniae, a gram positive bacterial pathogen, caused up to 20% mortality in the less susceptible S. aurata, compared to up to 80% mortality and higher disease severity in the highly sensitive L. calcarifer (Figs. 1A and 1B enhanced larva survival in tanks subjected to recirculating aquaculture system (RSA) without UV compared to those reared in RSA with UV treatment, without introducing any stress or pathogenic infection to lobster larvae [70]. Their results strengthen our ndings of higher diversity indices (both Shannon and Species richness "Chao1") when comparing non-UV and UV treated water ( Figure 2).
The higher sh survival rate in the non-UV treatment following infection may be attributed to the stability of microbial communities [72]. A recent paper [73] Table S5). In fact, when comparing between UV and non-UV-treatments, we noticed a signi cant difference in microbial communities at T 0 (before infection) for both S. aurata and L. calcarifer. When S. aurata was infected with V. harveyi and when L. calcarifer was infected with S. iniae, no signi cant differences were observed in sh from either UV treatment following infection. However, signi cant differences between the two UV treatments were seen when S. aurata was infected with S. iniae at all the sampling points (Supplementary  Table S5, Figure 3). Both V. harveyi and S. iniae bacteria are serious pathogens for wild and cultured sh and can cause a wide range of symptoms and even death [74]. Yet, S. iniae pathogen is less virulent to S. aurata compared to L. calcarifer. Our results show that S. iniae dominated the L. calcarifer skin microbiome after infection with no signi cant difference after the bacterial infection, however, with S. aurata, S. iniae was less dominant on sh skin after bacterial infection. These differences remained throughout the experiment on sh reared in UV-treated water tanks.
According to the ecological theory of R/K-selection (MacArthur and Wilson, 1967), selective pressures (i.e. stress, induction or UV treatment) drive microbial succession either by selection for opportunists (Rselection), or for specialists (K-selection). When sh are subjected to UV treatment, the high resource supply per bacterium favors the fast growing species (R-selection). Therefore, different bacterial communities, which were present at T 0 , were destroyed when a strong pathogen that was introduced managed to dominate sh skin microbiome. However, when the sh reacted to a less virulent pathogen (in the case of S. aurata infected with S. iniae) K-selection strategies were maintained, thus lowering mortality rates. This notion was discussed by Vestrum et al. (2018), who showed that different water treatment systems induced differences in larval microbiota. This observation indicates that non-stress conditions promote K-selection and microbial stability by maintaining a microbial load close to the carrying capacity of the system [71].

Spatial variation in sh skin microbiome from different body sites
When investigating the microbial variation in the different body sites of S. aurata, no signi cant variations were noticed ( Figure 3). However, at T 0 and T 1 , diversity estimates showed the bacterial community of gills to be slightly higher in Shannon, Simpson and Faith's phylogenetic diversity, compared to sh skin abdomen and lateral line areas ( Figure 2). Investigating pairwise signi cance, only the microbial communities of the gills signi cantly differentiated at speci c time points, namely when compared with lateral lines at T 0 and T 1 (Supplementary Table S7A). Looking at the unique ASVs (Supplementary Figure S3 and S4) at T 0 and T 1 , up to 53% of the total microbial ASVs were unique for the gills despite being only 9% of the weighted abundance at most (Supplementary Table S7B). Moreover, the gills are a thin barrier between sh blood and the environment [75]. This sophisticated system has a large surface area and delicate structure that provide an ideal port of entry for molecules, particles, and all kinds of pathogens [76]. As such, gill mucosa contains a fully developed immune system, including commensal bacteria [77,78]. This may indicate the important role of gill microbiome on sh survival. This assumption corresponds to our results, which showed gill microbial communities to change signi cantly, especially after infection in the non-UV treatments (Supplementary  Figure S7). Water microbiome formed a different cluster and was statistically signi cant when compared to the sh microbiome. In addition, non-UV treated water also showed a higher diversity compared to UV treated water samples. These ndings were previously reported by Chiarello

Temporal changes in sh skin microbial community during health, disease and recovery from bacterial infection
During the acute infection stage, the pathogenic bacteria signi cantly dominate sh skin microbial communities compared to the original microbial communities seen at T 0 (pre-infection. During the recovery stage, the microbial communities were observed to gradually return to their initial microbial communities, which were present prior to infection (Figure 4-7). However, some variations in observed microbial communities relate to water treatment, season, sh type, the induced pathogens and temporal . This asymmetry and unique diversity pattern in the gill microbial community is different from our observations. Figure 2 shows the symmetrical decrease in microbial diversity indices of both sh gills and skin (abdomen and lateral line) following infection compared to T 0 , yet following infection at T 1 , gills had a higher microbial diversity compared to sh skin scales for their sampling points during the "potential disease" stage, as the authors note that sh did not exhibit disease symptoms. In our sampling scheme, during the disease stage (24 h after infection, T 1 ), sh clearly exhibited disease symptoms. Therefore, the disease stage indicated by Rosado et al. (2019) may not correspond with ours, and that may explain some of the differences seen in gill microbial communities when comparing our results. Microbial dysbiosis was also documented on skin of Dicentrarchus labrax sh when infected with V. harveyi. The authors related this microbial dysbiosis to various sh immune responses including a decrease in the protease and lysozyme activity in infected sh [88]. Microbial dysbiosis may also be the result of antibiotic treatment. Antibiotics can have either bactericidal or bacteriostatic affect, altering commensals as well as pathogenic bacterial communities.
Pindling et al. (2018) showed that exposure of zebra sh larvae to antibiotic pollution (streptomycin, 10 µg/ml) signi cantly reduced their microbial diversity and increased early mortality among larvae, causing mortality within a few days following antibiotic exposure [89]. In our case, sh mortality was a direct result of the bacterial infection. Another study investigated the effect of broad-spectrum antibiotics on skin and gut microbiome of Gambusia a nis, the results showed that antibiotics lowered skin and gut microbial diversity and community composition [27]. Similar to our ndings (Supplementary Table S8), the authors show that bacterial communities failed to return to their original diversity or composition of pretreatment levels after one week recovery, yet after two weeks of recovery these microbial communities returned "to some extent" to their original state [27]. Moreover, in our results, sh reared in UV treated water, a known water disinfectant, showed to have a signi cantly lower microbial diversity compared to sh reared in non-UV water, we attributed the lower microbial diversity to the increase in sh mortality after infection of that group (Supplementary Table S5).
In our results, no signi cant variations in bacterial communities were seen two weeks into the recovery stage, compared to the original state (T 0 ). Researchers con rm that even if changes in the 16S bacterial composition are noticed during the recovery stage, the biochemical pro le of the microbial community following disruptions goes back to its original state, highlighting that the original microbial composition may not be required in order to restore microbial original functions [28]. This can explain the nonsigni cant differences in the microbial community composition seen in our results at the end of the recovery period, yet we did not perform microbial functional and biochemical pathway analyses. An interesting and important observation was seen in S. artura at T 1 (24 h post infection) infected with V.
harveyi -an increased abundance of Unclassi ed_Gammaproteobacteria ASV in the summer 2015 experiment (Figure 4). Furthermore, a remarkable increase of Photobacterium (belonging to the Vibrio family and closely related to the pathogenic V. harveyi) was observed in the repeated experiment in winter 2016 experiment ( Figure 5). Both the same Unclassi ed_Vibrioceae family ASVs and the Unclassi ed_Gammaproteobacteria ASV signi cantly dominated S. artura skin during infection with S. iniea at the disease stage ( Figure 6). Moreover, these two families, Unclassi ed_Vibrioceae and at a lower percentage, the Unclassi ed_Gammaproteobacteria ASVs abundance were increased on sh skin during S. iniae infection in L. calcarifer (Figure 7). These induced ASVs abundances were only noticed at T 1 (24 h post infection) and at the disease stage (T 1-2 , T 1-3 ), then they gradually decreased as sh progressed to the recovery stages (T 1-5 , T 2 and T 3 , ve days, one week and one month post infection, respectively). This could indicate the importance of these bacterial ASVs and their role during infection. In this context, Photobacterium spp, members of Vibrioceae family are known for their symbiotic relation with different sh species [90]. Some are also considered opportunistic pathogens that adapt R-strategy and take advantage of the reduced sh microbial diversity during the infection and disease stages, when sh start to recover, these opportunistic pathogens' abundances decline [91,92]. On the other hand, in our results, the Unclassi ed_Gammaproteobacteria class ASV showed a signi cantly elevated abundance during infection and disease stage (T 1 , Figure 4), which may have played an important role in increased sh survival. Notice that sh subjected to UV treated water had lower survival rates, except in the third experiment (where S. artura sh were infected with s. iniae), in which sh from the UV treatment showed a higher survival rate compared to the non-UV treatment. Interestingly, Unclassi ed_Gammaproteobacteria class ASV was highly abundant in that experiment at T 1 , on sh skin from the UV treatment compared to the non-UV treatment.
A similar yet different species, Del ta ( Figure 5) dominated S. artura skin microbiome at T 0 , this species signi cantly declined during infection with V. harveyi. Yet, at T 1 (24 h post infection) in the non-UV treatment we saw a higher abundance of Del ta ASV compared to the UV treatment, which may also be attributed to sh survival. While there was no attempt to isolate those species nor perform a metabolic investigation during this study, our results indicate the importance of sh skin and gill microbiome in sh survival following infection. In addition, the results emphasize the need to preserve high bacterial diversity to mitigate sh pathogens, enhance sh health conditions and increase survival rates during infection.

Conclusion
Various aspects of sh skin microbiome during healthy, diseased and recovery conditions were tested by a set of experiments that show how changes in innate and naturally occurring sh skin microbiome and dysbiosis affect sh health. We examined microbial diversity, composition and survival rate of sh in UV and non-UV treated water before and after infection (with Vibrio harveyi and Streptococcus iniae) during the summer and winter seasons. The results demonstrate a higher survival rate of infected sh (S. aurata and L. calcarifer) in the non UV-treated water environment compared to UV-treated water. We noticed that the higher survival rate was attributed to a stable microbial community. When sh were subjected to UV treatment, the high resource supply per bacterium favored the fast growing species (Rselection), therefore, different bacterial communities were destroyed when a strong pathogen was introduced and managed to dominate the sh microbiome. However, when the sh reacted to a less virulent pathogen (in case of S. aurata infected with S. iniae) K-selection strategies were maintained, resulting in lower mortality rates. This observation stresses the need to preserve high bacterial diversity to mitigate sh pathogens, bacterial diversity enhances sh health and increases survival during infection.
When examining the spatial variation in sh skin microbiome from different body sites of S. aurata, no signi cant variations were noticed. However, the microbial communities in gills signi cantly differentiated at speci c time points, when compared with lateral lines site, before infection (T 0 ) and 24 h post infection (T 1 ). Gills host a high percentage of unique ASVs (up to 53% of the total microbial ASVs were unique for the gills), this may indicate that gill microbiome is key to sh survival.
Temporal changes in sh skin microbiome before and after infection, and throughout recovery, showed that microbial communities gradually restore to their initial pre-infection state. However, some variations were observed in the restored microbial communities, these may be related to water treatment, season, sh and pathogen species and temporal resolutions. Yet, temporal changes indicate the importance of certain bacterial species (ASVs) in disease development and sh survival rate, mainly Del ta, Unclassi ed_Gammaproteobacteria and Unclassi ed_Vibrioceae ASVs. The sequences of these bacterial ASV's were patented under patent publication number WO2021038446. The potential to increase sh survival using these microbial species should be further investigated for future development of prophylaxis treatments to reduce the need for antibiotic application and reduce the adverse effects during bacterial infection outbreaks in aquaculture. Further research of metabolic pathways, functional diversity and bacterial isolation using similar experimental setups will increase understanding of disease ecology and shed light on important microbial functional traits and species that enhance sh survival. To the best of our knowledge, this is the rst study that investigated the effects of sh skin microbiome in disease ecology, and compared the effect of UV vs. non-UV treated water on microbiome.         Bar graph illustrating relative abundances of different bacterial phyla (different colors). Each bar represents one (tagged) sh at different time points (T 0 before infection, T 1 24 h, T 1-2 48 h, T 1-3 72 h, T 1-5 ve days, T 2 one week, T 2-1 two weeks and T 3 three weeks post infection), for UV and non-UV treatments during S. iniae infection in L. calcarifer.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.