Bacterial strains used in this study ( Table S1 ) were mainly isolated from the tissue of the healthy coral P. damicornis, with strains Vibrio mediterranei DSM 13774 and Vibrio coralliilyticus DSM 19607 isolated from the bleached corals Oculina patagonica (Kushmaro et al., 1996) and P. damicornis (Ben-Haim and Rosenberg, 2002), respectively. Strains DSM 13774 and DSM 19607 were provided by DSMZ (German Collection of Microorganisms and Cell Cultures). Antagonistic interactions were tested on marine agar 2216 (MA; BD; Becton, Dickinson and Company) using Burkholder diffusion assays (Burkholder et al., 1966). Each isolate was streaked from the -80°C glycerol stock on MA, and a single colony was then transferred to 6 ml of marine broth 2216 (MB; Becton, Dickinson and Co.) and shaken at 25°C for 24 h. The OD₆₀₀ was measured for each isolate before starting the assay using spectrophotometer (TU-1810, Persee General Instrument co., LTD, Beijing). Ten milliliters of top lawn bacteria (1 ml of bacterial culture added to 100 ml MA with 0.6% [w/v] agar, gently mixed) was poured on a sterile MA (with 1.5% agar) petri dish plate. The lawn was allowed to cool, and then 10 μl of each producer isolate was spotted onto the agar in petri dish plates (12 cm×12 cm). Control producer spots consisted of 10 μl of marine broth. Plates were incubated at 25°C and 32°C for 48 h, respectively, and then imaged. Antagonism was considered to occur when the semidiameter of the zone of inhibition was at least 0.4 mm greater than the semidiameter of the colony formed by the potential producer. This size of inhibition was chosen because it represents a spatial scale relevant to bacterial interactions (Long and Azam, 2001). The semidiameters of the inhibition halo and colony were measured by digital imaging software (ImageJ, NIH), and the decimal places were set to 3 in Set Measurements of ImageJ software. We tested the antagonistic interactions of 32 isolates in pairs, and the experiment was 32×32 (1024) times. Each experiment was carried out in 3 replicates. There are fourteen antagonistic interactions with ambiguous results that were re-verified with 3 replicates.

The phenotypic fingerprint of each isolate, including 71 carbon source utilization assays and 23 chemical sensitivity assays (the detailed assays were presented in Table S2 ), was tested by using BIOLOG GEN III MicroPlate (Biolog Inc.). Isolates were cultured for three days on MA at 25°C, a single fresh colony was then picked and dispersed in the modified inoculating fluid A (Biolog Inc.) with increasing salinity to 30‰ for maintaining the osmotic pressure as the salinity of culture medium MA was approximately 30 ‰, and the turbidity of the suspension was 92-95%T. The cell suspension was inoculated into the GENIII MicroPlate (100 μl per well) and incubated at 25°C and 32°C respectively. Color reactions in the microplate wells were determined using the Biolog Microbial Identification System, MicroStation Reader (Biolog Inc., ELx808BLG, USA) with absorbance measured at 590 nm every 24 h for a total duration of 7 days. The “borderline” value was scored as positive if the value of absorbance was higher than that of the negative control.

For each interaction pair of coral-associated bacteria, one antagonistic pair was defined by the inhibition value in three statuses: x=1 if isolate A inhibited B, y=1 if B inhibited A or if both x and y were 1, which means reciprocal inhibition. If both x and y were zero, which means no inhibition occurred. A phylogenetic tree was constructed based on 16S rRNA gene sequences using the neighbor-joining method with MEGA X (Kumar et al., 2018), and the best DNA model Tamura-Nei was used. Topologies of the phylogenetic trees were evaluated using the bootstrap method with 1000 resamplings. Pairwise phylogenetic distances were calculated using the Tamura-Nei model with MEGA X (Kumar et al., 2018). Phenotypic dissimilarity was quantified by calculating the Jaccard distance according to the results of the GEN III MicroPlate tests. Correlations between antagonism and the phylogenetic and phenotypic distances were evaluated with logistic regression (glm, binomial family). To infer whether the correlation between antagonism and the phenotypic distances was confounded by the correlation between the phylogenetic and phenotypic distances, linear regression for phenotypic distance with phylogenetic distance was performed. The residuals (the variation was not explained by phylogenetic distance) were then used as independent variables in the logistic regressions (Russel et al., 2017). These analyses were performed using R version 4.0.1 (R Core Team, 2020).

The interaction frequency was calculated as the number of antagonistic interactions observed divided by the number of total possible antagonistic interactions determined by the number of strains tested (Vetsigian et al., 2011). Sender-receiver asymmetry, which represents whether the network is more determined by the sender or receiver strains on average, was evaluated by comparing the variance of the sender and receiver degrees (Vetsigian et al., 2011; Perez-Gutierrez et al., 2013). The sender degree of one isolate means the number of isolates it inhibits, and the receiver degree represents the number of isolates that inhibit the isolate (Perez-Gutierrez et al., 2013). The network diameter, the longest path connecting any two strains (Newman, 2003), was calculated using the network analyzer tool implemented in Cytoscape v3.8.0 (Shannon et al., 2003). Nestedness, a property of the interaction network, was calculated as defined by Bascompte and colleagues (Bascompte et al., 2003) using the algorithm implemented in BINMATNEST (binary matrix nestedness temperature calculator) (Rodriguez-Girones and Santamaria, 2006). Antagonistic interaction networks among coral-associated bacteria were visualized with yFiles, a hierarchical layout option of Cytoscape v3.8.0 (Shannon et al., 2003). The “nodes” represent strains in the networks, and the “edges” depict directed antagonistic interactions. The connectivity of a network is the total number of interactions (Landi et al., 2018), represented by the total number of edges in this study. The width of the “edges” was multiple enlarged based on the intensity value (i.e., the size of the inhibition zone) among the antagonism interactions. The size of the “nodes” was enlarged according to the total number of sender and receiver degrees.

Comparing the phenotypic fingerprints of each isolate at 25°C and 32°C, we found that the physiological and biochemical characteristics of most strains varied at different temperatures except Marimonas sp. SCSIO 12655 and Halioglobus sp. SCSIO 12614 ( Figure 1 ). Microbacterium sp. SCSIO 12466 showed the highest number of variations with 20 tested characteristics due to increased temperature ( Figure 1 and Table S1 ). In contrast, Erythrobacter sp. SCSIO 12564 showed only one varied characteristic (i.e. resistance to minocycline), and the tested phenotypical characteristics of strains Halioglobus sp. SCSIO 12614 and Marimonas sp. SCSIO 12655 were not affected by the increasing temperature ( Figure 1 and Table S1 ). According to the total number of positive tests, more kinds of sugar and amino acids could be utilized at the higher temperature ( Figure S1 ). α-D-glucose and L-glutamic acid were commonly utilized at low temperature ( Figure S1 ), while α-D-glucose, L-alanine and L-glutamic acid were commonly utilized at high temperature ( Figure S1 ). In addition, the resistance of five antibiotics (Troleandomycin, Rifamycin SV, Minocycline, Lincomycin, Vancomycin) had no change under the evaluated temperature ( Figure S1 ).

Twenty out of the 32 isolates inhibited at least one isolate ( Table 1 and Figure 2 ). Ten of the twelve strains belonging to Rhodobacteraceae showed antagonistic activities ( Table 1 ). Among them, strain SCSIO 12563 presented the strongest antagonistic activity that inhibited 19 (25°C) or 18 (32°C) strains belonging to various taxonomic groups ( Table 1 and Figure 2 ). Seven isolates (Epibacterium sp. SCSIO 12563, Bacterioplanoides sp. SCSIO 12839, Ruegeria sp. SCSIO 12669, Roseovarius sp. SCSIO 12450, Roseovarius sp. SCSIO 12626, Mycolicibacterium sp. SCSIO 12476, Gordonia sp. SCSIO 12652) were not inhibited by any isolates at either 25°C or 32°C ( Figure 2 ). In addition, strain Microbacterium sp. SCSIO 12466 showed autoinhibition ( Figure 2 ).

V. mediterranei DSM 13774 was inhibited by strain Epibacterium sp. SCSIO 12563 at both 25°C and 32°C ( Tables 1 , S3 and Figure 2 ). The growth of V. coralliilyticus DSM 19607 was inhibited by strain Marimonas sp. SCSIO 12655 at 25°C ( Tables 1 , S4 and Figure 2 ). V. mediterranei DSM 13774 did not inhibit any isolate growth, while V. coralliilyticus DSM 19607 inhibited Marinobacter sp. SCSIO 12456 at both 25°C and 32°C ( Tables 1 , S5 and Figure 2 ).

The total number of antagonistic interactions at 32°C was approximately 20% higher than that at 25°C ( Figure S2 ). Twenty-one antagonistic interactions occurred only at higher temperature (32°C), while 8 occurred only at 25°C ( Figure 2 and Table S4 ). At the level of order or class, the mean number of inhibited isolates increased at higher temperature for producers belonging to Rhodobacterales, Sphingomonadales, Rhizobiales, and Actinobacteria ( Figure S3 ). At the level of strain, Ruegeria sp. SCSIO 12571, Cognatiyoonia sp. SCSIO 12602, Roseovarius sp. SCSIO 12654, Marimonas sp. SCSIO 12655, Altererythrobacter sp. SCSIO 12686, and Janibacter sp. SCSIO 12652 was able to inhibit more isolates with increasing temperature ( Figure 2 ). These results indicated that more antagonistic interactions occurred at the higher temperature. Additionally, among the antagonistic interactions that occurred at both 25 and 32°C ( Tables S3 and S5 ), most of them (62.2%) showed compromised inhibition activities at 32°C, while three strains (Epibacterium sp. SCSIO 12563, Marimonas sp. SCSIO 12655, and Microbacterium sp. SCSIO 12466) showed significantly increased inhibition activities ( Table S5 ).

We further compared the structures of the antagonistic interaction networks at different temperatures ( Figure 3 ), and found that the number of nodes increased by 1, and the number of edges increased by 28% at higher temperature ( Figure 3 and Table S6 ). Additionally, the interaction frequency increased at 32°C (25°C, 0.04 and 32°C, 0.06) ( Table 2 ). The value of sender-receiver asymmetry was negative at 25°C (-0.22) and 32°C (-0.15), indicating that the interaction matrix was sender-determined and temperature independent ( Table 2 ). The values of nestedness, a vital descriptor of ecological network architecture (Landi et al., 2018), showed that the antagonistic network was significantly nested at both temperatures ( Table 2 ; P < 0.00001 both at 25°C and 32°C). To test the relation between sender degree and receiver degree at different temperatures, we depicted their antagonistic interaction using a heatmap ( Figure S4 ). The nested heatmap results showed that the antagonists were highly resistant, while the inhibited strains were unlikely to antagonize other strains ( Figure S4 ).

Among the 496 antagonistic test pairs (pairs of tested strains), we found that the antagonistic probability increased with decreasing 16S rRNA gene phylogenetic distance ( Figures 4A, B ). Furthermore, the correlation between antagonistic probability and phylogenetic distance was stronger at higher temperature (25°C, P = 0.028, pseudo-R² = 0.021; 32°C, P = 0.020, pseudo-R² = 0.021; logistic regression used throughout unless otherwise noted) ( Figures 4A, B ).

To further assess whether the correlation between antagonism and phylogeny was driven by the metabolic niche space of individual isolates, the correlation between antagonism probability and phenotypic distance (calculated on the basis of the dissimilarity of the phenotypic fingerprints) was evaluated. There was no correlation between antagonism and phenotypic distances (25°C, P = 0.885; 32°C, P = 0.891) ( Figures 4C, D ). As the phylogenetic distance correlated with the phenotypic distance at both 25°C and 32°C (25°C, P < 0.01; 32°C, P < 0.05) ( Figure S5 ), we also analyzed the correlation between antagonism probability and phenotypic distance by using the residuals from the linear regression for phenotypic distance with the phylogenetic distances. The results showed that antagonism did not correlate with phenotypic distances at either 25°C or 32°C (25°C, P = 0.510; 32°C, P = 0.908) ( Figure S6 ).

In order to investigate the trade-off between antagonism and resource exploitation, we tested whether the antagonist had a wider metabolic niche space (growth on the 71 carbon sources of the Biolog GENIII MicroPlate) for each antagonistic interaction, and compared it to the non-antagonistic interactions. We found that the antagonists tended to have a wider metabolic niche space, i.e., grew on more carbon sources than the antagonized strains at 25°C (Fisher’s exact test, odds ratio = 2.064, P = 0.028), but not at 32°C (odds ratio = 0.825, P = 0.498) ( Figure 5 ).

The results obtained in this study and reported previously indicated that antagonism was a common phenomenon in the coral-associated bacterial community, and temperature distinctly altered the breadth and intensity of antagonistic interactions (Rypien et al., 2010; Aguirre-von-Wobeser et al., 2014; Tang et al., 2019). The intensity of most antagonistic interactions decreased at higher temperature, which is consistent with previous results (Rypien et al., 2010). In contrast to the findings of Rypien et al. (2010), more antagonistic interactions occurred at higher temperature in this study, suggesting that the breadth of antagonism increased with increasing temperature. Previous studies showed that 6% or 12% of tested isolates were able to inhibit V. mediterranei (Nissimov et al., 2009; Rypien et al., 2010), and 3% or 10% were able to inhibit V. coralliilyticus (Rypien et al., 2010; Miura et al., 2019; Tang et al., 2019). Results of this study showed that strain Epibacterium sp. SCSIO 12563 inhibited the growth of V. mediterranei DSM 13774, and Marimonas sp. SCSIO 12655 inhibited V. coralliilyticus DSM 19607. Those two isolates (SCSIO 12563 and SCSIO 12655) were affiliated to different species in comparison to the antagonists observed in the previous studies (Nissimov et al., 2009; Rypien et al., 2010; Miura et al., 2019; Tang et al., 2019). These results suggest that these two opportunistic vibrion might face a challenging growth environment due to antagonistic interactions in the coral holobiont. Moreover, the growth of V. coralliilyticus DSM 19607 was only inhibited by strain Marimonas sp. SCSIO 12655 at 25°C, while the inhibition was removed at 32°C. The removal of inhibition might contribute to the enrichment of vibrion at higher temperature (Bourne et al., 2008; Frydenborg et al., 2014; Tout et al., 2015; Rajeev et al., 2021) and allow compromising the health of coral host (Gavish et al., 2021).

The antagonistic networks of coral-associated bacteria were sender-determined both at 25 and 32°C, which means that the structures of the networks were more influenced by antagonists than sensitive strains. Furthermore, our results showed that the antagonistic interaction network was significantly nested, with nestedness scores of 0.984 (25°C) and 0.969 (32°C). The highly nested and sender-determined network suggests that the antagonistic interaction network of the coral-associated bacteria community is more determined by antagonists. The nested antagonistic interaction networks were also respectively observed among Gammaproteobacteria in the oligotrophic water environment (Aguirre-von-Wobeser et al., 2014) and Carnobacterium maltaromaticum strains isolated from diverse habitats (e.g., dairy products, diseased fish, and meat products) (Ramia et al., 2020). These results suggested that the nested network is a common feature among bacteria antagonism interactions.

Coral-associated bacterial communities are known to alter in response to increasing seawater temperature (Tout et al., 2015; McDevitt-Irwin et al., 2017; Grottoli et al., 2018; Pootakham et al., 2018; Li et al., 2021). In this study, the antagonistic interaction network was found to be more hierarchical and complex at 32°C than 25°C ( Figure 3 ). The study of Welsh et al. (2016) showed that heat stress decreased the complexity (network edges) of the coral mucus associated bacterial network, and the number of mutual exclusion was decreased in the network at the higher temperature (29-30°C). We should mentioned that the strains used in this study were isolated from coral tissue, while Welsh et al. (2016) investigated the network of bacteria associated with corals mucus. The microbial communities in different compartments of coral holobiont might show distinct networks. Even so, the inconsistent results obtained in this and previous studies highlight the necessity of interpretation of the interactions among coral-associated bacteria on the basis of both wet lab and dry lab results. The interactions among culturable strains could be experimentally investigated, while these results may not be enough for comprehensively understanding the interactions in the whole coral-associated bacterial community. On the other side, although the co-occurrence network constructed based on the relative abundances acquired through culture-independent analyses provides information for the whole bacterial community, it needs more cautions in interpreting the interactions presented in the co-occurrence network (Blanchet et al., 2020).

Competition (including antagonism) is considered one of the factors in community assembly (Perez-Gutierrez et al., 2013; Rocha et al., 2015; Zapien-Campos et al., 2015; Garcia-Bayona and Comstock, 2018). We hypothesize accordingly that the alteration of antagonistic interactions due to increased temperature is one of the processes that governs the assembly of coral-associated bacterial community, and encourage verifying it in combination with studies on natural communities.

Our results showed that closely related bacteria have a higher probability of antagonism than more distantly related bacteria, consistent with previous reports (Russel et al., 2017; Gonzalo et al., 2020). Moreover, we found that the correlation between antagonism and phylogenetic distance was stronger at higher temperature (32°C), implying that phylogenetically similar bacteria are likely to compete more fiercely under heat stress in coral holobionts. In contrast to the results of Russel et al. (2017), antagonism did not significantly correlate with phenotypic distances among coral-associated bacteria in this study, while this was similar to the finding in isolates of Pseudomonas aeruginosa from household drains (Mojesky and Remold, 2020). In the previous studies of both Russel et al. (2017) and Mojesky and Remold (2020), the utilization of 31 carbon sources were tested. These results suggest differential associations between antagonism and physiological and biochemical properties occur in distinct habitats, and phylogenetic similarity might predict the inhibition among bacteria associated with corals. Considering that the limitation in the range of phenotypic characteristics that could be screened using BIOLOG GEN III MicroPlate, expanding the tested index is necessary in the future.

We investigated whether the antagonizing strains are generalists growing on many resources, and found that antagonists were indeed generalists at 25°C ( Figure 5A ), growing on ~20-130% more carbon sources than the inhibited strains. This result supports the hypothesis, i.e., the antagonist is also generalist, proposed by Russel et al. (2017). Meanwhile, the findings obtained at 32°C did not support this hypothesis, as the numbers of carbon sources utilized by antagonizing and antagonized strains were similar ( Figure 5B ). Additionally, more antagonistic interactions occurred due to the increased temperature ( Figure S2 and Table S4 ). These results suggest that temperature significantly affects the trade-off between antagonism and resource exploitation in the antagonistic interactions of coral-associated bacteria. We further speculated that the shift of trade-off is an adaptation to nutrient reduction in bleached coral under thermal stress (Morris et al., 2019; Rädecker et al., 2021).