Antimicrobial activity, antibiotic susceptibility and virulence factors of Lactic Acid Bacteria of aquatic origin intended for use as probiotics in aquaculture

Background The microorganisms intended for use as probiotics in aquaculture should exert antimicrobial activity and be regarded as safe not only for the aquatic hosts but also for their surrounding environments and humans. The objective of this work was to investigate the antimicrobial/bacteriocin activity against fish pathogens, the antibiotic susceptibility, and the prevalence of virulence factors and detrimental enzymatic activities in 99 Lactic Acid Bacteria (LAB) (59 enterococci and 40 non-enterococci) isolated from aquatic animals regarded as human food. Results These LAB displayed a broad antimicrobial/bacteriocin activity against the main Gram-positive and Gram-negative fish pathogens. However, particular safety concerns based on antibiotic resistance and virulence factors were identified in the genus Enterococcus (86%) (Enterococcus faecalis, 100%; E. faecium, 79%). Antibiotic resistance was also found in the genera Weissella (60%), Pediococcus (44%), Lactobacillus (33%), but not in leuconostocs and lactococci. Antibiotic resistance genes were found in 7.5% of the non-enterococci, including the genera Pediococcus (12.5%) and Weissella (6.7%). One strain of both Pediococcus pentosaceus and Weissella cibaria carried the erythromycin resistance gene mef(A/E), and another two P. pentosaceus strains harboured lnu(A) conferring resistance to lincosamides. Gelatinase activity was found in E. faecalis and E. faecium (71 and 11%, respectively), while a low number of E. faecalis (5%) and none E. faecium exerted hemolytic activity. None enterococci and non-enterococci showed bile deconjugation and mucin degradation abilities, or other detrimental enzymatic activities. Conclusions To our knowledge, this is the first description of mef(A/E) in the genera Pediococcus and Weissella, and lnu(A) in the genus Pediococcus. The in vitro subtractive screening presented in this work constitutes a valuable strategy for the large-scale preliminary selection of putatively safe LAB intended for use as probiotics in aquaculture.


Background
Aquaculture has the potential to make a significant contribution to the increasing demand for aquatic food in most world regions; however, in order to achieve this goal, the sector will have to face significant challenges, including the production intensification, the disease control and the prevention of the environmental deterioration [1]. In fish farming, the widespread use of antibiotics as prophylactic and therapeutic agents to control bacterial diseases has been associated with the emergence of antibiotic resistance in bacterial pathogens and with the alteration of the microbiota of the aquaculture environment [2,3]. This resulted in the ban of antibiotic usage as animal growth promoters in Europe and stringent worldwide regulations on therapeutical antibiotic applications. This scenario has led to an evergrowing interest in the search and development of alternative strategies for disease control, within the frame of good husbandry practices, including adequate hygiene conditions, vaccination programmes and the use of probiotics, prebiotics and immunostimulants [4][5][6]. Recently, novel strategies to control bacterial infections in aquaculture have emerged, such as specific killing of pathogenic bacteria by bacteriophages, growth inhibition of pathogen by short-chain fatty acids and polyhydroxyalkanoates, and interference with the regulation of virulence genes (quorum sensing disruption), which have been reviewed by Defoirdt et al. [7]. With regard to probiotics, they are defined as live microbial adjuncts which have a beneficial effect on the host by: (i) modifying the host-associated or ambient microbial community; (ii) improving feed use or enhancing its nutritional value; (iii) enhancing the host response towards disease; and/or (iv) improving its environment [8]. To date, most probiotics proposed as biocontrollers and bioremediation agents for aquaculture belong to the LAB group (mainly to the genera Lactobacillus, Lactococcus, Leuconostoc, Enterococcus and Carnobacterium), to the genera Vibrio, Bacillus, and Pseudomonas or to the species Saccharomyces cerevisiae [8,9]. Recently, a probiotic culture (Bactocell W , Pediococcus acidilactici CNCM MA18/ 5 M) has been authorized for the first time for use in aquaculture in the European Union.
According to the FAO/WHO [10], the development of commercial probiotics requires their unequivocal taxonomic identification, as well as their in vitro and in vivo functional characterization and safety assessment. In Europe, the European Food Safety Agency (EFSA) proposed a system for a pre-market safety assessment of selected groups of microorganisms used in food/feed and the production of food/feed additives leading to a Qualified Presumption of Safety (QPS) status [11][12][13]. The QPS approach propose that the safety assessment of a defined taxonomic group could be made based on establishing taxonomic identity, body of knowledge, possible pathogenicity and commercial end use. According to the EFSA approach [13], most LAB species are included in the QPS list and, therefore, demonstration of their safety only requires confirmation of the absence of determinants of resistance to antibiotics of human and veterinary clinical significance. However, in the case of enterococci, a more thorough, strain-specific evaluation is required to assess the risk associated to their intentional use in the food chain. In this work, we present the antimicrobial activity against fish pathogens and the in vitro safety assessment beyond the QPS approach of a collection of 99 LAB belonging to the genera Enterococcus, Lactobacillus, Lactococcus, Leuconostoc, Pediococcus and Weissella, previously isolated from aquatic animals regarded as human food [14] and intended for use as probiotics in aquaculture.

Direct antimicrobial activity of the 99 LAB of aquatic origin
The 99 LAB strains isolated from fish, seafood and fish products displayed direct antimicrobial activity against, at least, four of the eight tested indicator microorganisms ( Table 1). The most sensitive indicators were Listonella anguillarum CECT4344, Ls. anguillarum CECT7199 and Aeromonas hydrophila CECT5734, followed by Lactococcus garvieae JIP29-99, Streptococcus iniae LMG14521 and Streptococcus agalactiae CF01173. On the contrary, Photobacterium damselae CECT626 and Vibrio alginolyticus CECT521 were the less sensitive indicator microorganisms.

Extracellular antimicrobial activity of the 49 pre-selected LAB
The antimicrobial activity of supernatants from the 49 pre-selected LAB (9 E. faecium selected based on their preliminary safety assessment and 40 non-enterococcal strains) with direct antimicrobial activity against fish pathogens was assayed against three indicator microorganisms by an ADT (Table 3). In this regard, 24 (49%) and 10 (20%) strains displayed extracellular antimicrobial activity in their supernatants and/or 20-fold concentrated supernatants against Pediococcus damnosus CECT4797 and L. garvieae JIP 29-99, respectively, but none of the strains inhibited the Gram-negative strain A. hydrophila CECT5734. Interestingly, the antimicrobial activity of the respective supernatants was sensitive to proteinase K treatment, but was not affected by the heat treatment, revealing the proteinaceous nature and heat stability of the secreted antimicrobial compounds (i.e., heat-stable bacteriocins). The 24 LAB strains secreting bacteriocins into the liquid growth medium belong to the species P. pentosaceus (15 strains), E. faecium (8 strains), and Lb. curvatus (1 strain).
In vitro safety assessment of the 49 pre-selected LAB The 49 pre-selected LAB were further submitted to a comprehensive safety assessment by different in vitro tests.

Hemolysin production, bile salts deconjugation and mucin degradation
None of the non-enterococcal strains showed hemolytic activity, similarly as found for the 9 enterococci. Moreover, bile salts deconjugation and mucin degradation abilities were not found in any of the tested strains.

Enzymatic activities
The results of the analysis of enzymatic activity profiles of the tested LAB are shown in Table 4. None of the strains showed lipolytic activity, except E. faecium LPP29, TPM76, SMA7, and SMF8 which produced esterase (C4) and esterase lipase (C8). Moreover, none of the LAB strains showed protease activity (trypsin and α-chymotrypsin). Nevertheless, peptidase activity (leucine, valine or cystine arylamidase) was found in all the species. All strains showed acid phosphatase (except E. faecium TPM76 and Lc. cremoris) and naphthol-AS-BI-phosphohydrolase activities, but none displayed alkaline phosphatase activity. β-Galactosidase was found in most species (but not in all strains) except Lb. curvatus and L. cremoris. However, α-glucosidase was only found in the three Lc. cremoris strains. β-Glucosidase and Nacetyl-β-glucosaminidase activities were observed in most E. faecium, Lactobacillus spp., L. cremoris, and P. pentosaceus strains, but only in two W. cibaria strains, while the three Lc. cremoris strains showed β-glucosidase but lacked N-acetyl-β-glucosaminidase activity. On the other hand, α-galactosidase, β-glucuronidase, α-mannosidase, and α-fucosidase activities were not detected in any of the tested LAB strains.

Antibiotic susceptibility determined by the broth microdilution test
The distribution of MICs of the tested antibiotics is summarized in Tables 5 and 6. Microbiological breakpoints for ampicillin, vancomycin, gentamicin, kanamycin, streptomycin, erythromycin, clindamycin, tetracycline, and chloramphenicol reported by the FEEDAP document on the assessment of bacterial products used as feed additives in relation to antibiotic resistance [15] were used to categorise the 49 LAB as susceptible or resistant strains. In this document, the genus Weissella, which is considered a group of heterofermentative Leuconostoc-like LAB [16], is not included. For this reason, the respective MICs were interpreted by using the breakpoints given for the genus Leuconostoc. Besides, due to the lack of microbiological breakpoints for penicillin and linezolid on the FEEDAP document, we interpreted our results on these antibiotics according to the cut-off levels proposed by Klare et al. [17] for pediococci, namely 1 and 2 mg/L for penicillin and linezolid, respectively. According to our results, the percentages of strains showing antibiotic resistance in the genera Weissella, Pediococcus, Lactobacillus and Enterococcus were 60, 44, 33 and 11%, respectively, while none of the leuconostocs and lactococci showed this phenotype. In summary, 97.5% of the 40 non-enterococal

Discussion
In this work, the antimicrobial activity against fish pathogens and the in vitro safety of 99 LAB previously isolated from fish, seafood and fish products [14] have been assayed by using microbiological, biochemical and genetic assays in order to identify and select the most suitable candidates to be further evaluated as probiotics for a sustainable aquaculture. LAB are widely known for their ability to inhibit bacterial pathogens by the production of antimicrobial compounds such as organic acids, oxygen peroxide and ribosomally-synthesized peptides referred to as bacteriocins, which constitutes a desirable property for probiotics and a sustainable alternative to antibiotics [9,18]. In this respect, most of the LAB of aquatic origin tested in this work displayed a broad antimicrobial spectrum against the main Gram-positive and Table 5 MICs distribution of 10 antibiotics for the 9 enterococcal strains   Gram-negative fish pathogens, being remarkable that a high number of strains (24 out of 49 strains, 49%) were identified as potential bacteriocin producers. Recently, bacteriocin production ability has been proposed as a key property for selection of probiotic LAB to be used in aquaculture as an alternative to antibiotics to fight against fish pathogen infections [19], similarly as proposed for human and farm animal probiotics [20][21][22]. In aquaculture farming, lactococcosis produced by the zoonotic agent L. garvieae, causing hemorrhagic septicaemia and meningoencephalitis, is one of the most serious diseases affecting several marine and fresh water fish species [23]. With regard to this, our work shows that putative bacteriocinogenic LAB active against this relevant fish pathogen are common amongst the microbiota isolated from aquatic animals (10 strains, 20%). The application of probiotics in aquaculture may modify the microbial ecology of the aquatic hosts and their surrounding environment, and thus the assessment of their safety to the target aquatic species, the environment and humans constitutes an essential issue [24]. To date, several studies describing the screening and evaluation of LAB as probiotic candidates for aquaculture have been reported [25][26][27][28]; however, the safety assessment of the strains is generally limited to in vivo challenge tests and rearing trials in order to confirm their lack of toxicity to the aquatic hosts [24,25,[28][29][30][31]. Strikingly, in vitro safety assessment studies have not been generally addressed, despite they have lower economic and ethic costs and result very effective to evaluate the safety of a high number of candidate probiotic strains not only for the host species, but also for humans and the environment. According to EFSA [13], most of the LAB species tested in this work (P. pentosaceus, Lb. curvatus, L. lactis, Lc. mesenteroides) are included in the QPS list and, therefore, demonstration of their safety only requires confirmation of the absence of determinants of resistance to antibiotics of human and veterinary clinical significance. However, in the case of enterococci, a more thorough, strain-specific evaluation is required to assess the risk associated to their intentional use in the food chain, while no guidelines are given for the safety assessment of the species W. cibaria [13].
Our results show that enterococcal virulence factors were more frequently found in E. faecalis than in E. faecium, which is in concordance with previous reports [32][33][34]. In this respect, most of the E. faecalis (95%) and a large percentage of the E. faecium (53%) strains evaluated in this work showed, at least, one virulence factor, being efaAfs, gelE and agg the most frequently detected genes. With regard to gelE, which encodes for an extracellular zinc endopeptidase that hydrolyzes gelatin, collagen, hemoglobin, and other bioactive compounds, this gene was detected at high frequency in E. faecalis, with all the gelE + strains showing gelatinase activity. However, five out of nine E. faecium strains harbouring gelE were unable to degrade gelatin, suggesting the carriage of a non-functional gene, as previously reported [32,33]. Likewise, in the case of E. faecium P68 and E. faecium GM29 harbouring cylL L -cylL S , the lack of hemolytic activity may be explained by the absence of cylM, whose product is involved in the posttranslational modification of cytolysin. On the other hand, esp and hyl, which encode a cell wall-associated protein involved in immune evasion and an hyaluronidase enzyme, respectively, were not found in any of the tested LAB. Previous studies have reported that esp and hyl are more common in ampicillin-resistant/vancomycin-resistant E. faecium (VREF) than in ampicillin-susceptible/VREF  strains [35]. In this context, the increase in the incidence of VREF at hospital settings has been attributed mainly to the spread of ampicillin-resistant VREF exhibiting esp and/or hyl [36,37]. Therefore, the fact that the E. faecium strains evaluated in this work lack these genes might be related with their non-clinical origin and absence of ampicillin resistance. The use and frequent overuse of antibiotics, including those used in human medicine, in fish farming has resulted in the emergence and spread of antibioticresistant bacteria in the aquaculture environment. This possesses a threat to human and animal health due to the increase of acquired antibiotic resistance in fish pathogens, the transfer of their genetic determinants to bacteria of terrestrial animals and to human pathogens, and the alterations of the bacterial microbiota of the aquatic environment [11,29]. In our study, the percentage of enterococcal strains showing acquired antibiotic resistance was 68%. Interestingly, the results found in E. faecium (71%) and E. faecalis (62%) were similar, however, higher percentages of resistance to ciprofloxacin and/or norfloxacin, rifampicin, and glycopeptides were observed in E. faecalis. Nevertheless, the occurrence of erythromycin and tetracycline resistance was frequently detected amongst E. faecium (45%) but only in one E. faecalis strain (5%). In spite of the high prevalence of acquired antibiotic resistance found in enterococci of aquatic origin, they showed low incidence or absence of resistance to the clinically relevant antibiotics vancomycin (8.5%) and ampicillin, penicillin and gentamicin, respectively, which is in agreement with previous studies [33,38]. Moreover, the percentages of strains showing antibiotic resistance in the genera Weissella, Pediococcus and Lactobacillus were 60, 44 and 33%, respectively, while none of the leuconostocs and lactococci showed this phenotype. In this regard, our results indicate that the LAB susceptibility patterns of MIC values to clinically relevant antibiotics are species-dependent, similarly as previously described by other authors [39,40]. Moreover, multiple antibiotic resistance was commonly found in strains within the genus Enterococcus (37%), mainly in E. faecalis, while being very infrequent in the non-enterococcal strains (5%).
According to EFSA [29], the determination of MICs above the established breakpoint levels, for one or more antibiotic, requires further investigation to make the distinction between added genes (genes acquired by the bacteria via gain of exogenous DNA) or to the mutation of indigenous genes. According to our results, acquired antibiotic resistance likely due to added genes is not a common feature amongst the non-enterococcal LAB of aquatic origin (7.5%). In this respect, this genotype was only found in the genera Pediococcus (12.5%) and Weissella (6.7%). Although P. pentosaceus LPV57 and LPM78 showed resistance to kanamycin (MIC of 128 mg/L), the respective resistance gene [aac(6´)-Ie-aph(2´´)-Ia] was not found in these strains. Similarly, P. pentosaceus TPP3 and SMF120 were phenotypically resistant to tetracycline (MIC of 16 mg/L), but did not contain tet (K), tet(L) or tet (M). In this respect, Ammor et al. [41] reported that pediococci are intrinsically resistant to the latter two antibiotics, as well as to glycopeptides (vancomycin and teicoplanin), streptomycin, ciprofloxacin and trimethoprim-sulphamethoxazole. Other authors proposed a MIC for tetracycline in pediococci ranging between 8 and 16 mg/L [42], or of 32 mg/L for oxytetracycline in P. pentosaceus [17]. The tetracycline breakpoints suggested for pediococci by EFSA are lower than the MICs observed in our work and others [17,42]. On the other hand, the only antibiotic resistance detected in Leuconostoc strains was for vancomycin, which is an intrinsic property of this genus. It has been previously reported that Leuconostoc strains display poor, if any, resistance to antibiotics of clinical interest [38]. With regard to lactococci, the three L. cremoris strains evaluated were susceptible to all the antibiotics; however, relatively high MICs for rifampicin (16-32 mg/L) and trimethoprim (≥ 64 mg/L) were detected. In fact, most lactococcal species are resistant to trimethoprim [41]. As expected, all strains of heterofermentative Lactobacillus spp. were resistant to vancomycin but susceptible to the rest of the assayed antibiotics, except Lb. carnosus B43, which showed the highest MIC for ampicillin and penicillin (MICs of 8 and 4 mg/L, respectively). In this context, the presence of modifications in the low affinity penicillin-binding protein (PBP) that confers resistance to penicillin and β-lactams in E. faecium and Streptococcus pneumoniae, has been reported [43,44]. Moreover, nine PBPs have been described in Lb. casei ATCC 393 [45], which leads us to suggest that a similar mechanism may be also responsible for the ampicillin and penicillin resistance found in Lb. carnosus B43. The resistance to vancomycin detected in Pediococcus, Leuconostoc and Lactobacillus species in this study might be due to the presence of D-Ala-D-Lactate in their peptidoglycan rather than D-Ala-D-Ala dipeptide [46]. In this context, all tested W. cibaria strains showed MICs ≥ 128 mg/L for vancomycin, suggesting that vancomycin resistance is an intrinsic property of this species. In relation to Weissella spp., studies on antibiotic resistance profiles are very limited [47] and breakpoints have not been defined by EFSA [15]. In our study, most W. cibaria strains showed low MIC values; however W. cibaria BCS50 showed relatively high MICs for penicillin (8 mg/L) and kanamycin (64 mg/L), and W. cibaria SMA25 showed MICs of 128 mg/L for kanamycin, 8 mg/L for gentamicin, erythromycin and neomycin, and 2 mg/L for clindamycin. Therefore, these two strains were discarded of this study, while W. cibaria P50, P61, P64, P73, SMA14, SDM381 and SDM389 were not included in the final selection due to their MICs for kanamycin (32-64 mg/L). According to these results, as a rule of thumb, we propose for W. cibaria the breakpoints assigned to Leuconostoc spp. by EFSA [15], until further studies establish the wild-type MIC ranges within this species. In spite of that, different MICs for rifampicin and trimethoprim for W. cibaria and Lc. cremoris were found in this study. The reduced susceptibility of W. cibaria towards trimethoprim could indicate an intrinsic resistance to this antibiotic [48]. In our work, the only antibiotic resistance genes found were mef(A/E), which encodes a drug efflux pump conferring a low to moderate level of resistance to 14 (erythromycin and clarithromycin)-and 15 (azithromycin)-membered macrolides but not to lincosamide or streptogramin B antibiotics [49], and lnu(A), encoding the lincosamide O-nucleotidyltransferase that inactivates lincomycin and clindamycin [50]. In this respect, P. pentosaceus LPM78 and W. cibaria SMA25, displaying erythromycin resistance (MIC = 8 and ≥ 8 mg/L, respectively), carried the gene mef(A/E), which can be found in a variety of Gram-positive bacteria, including corynebacteria, enterococci, micrococci, and several streptococcal species [51,52]. On the other hand, two pediococci (P. pentosaceus LPM78 and LPM83) that showed resistance to clindamycin (MIC = 4 and 2 mg/L, respectively) carried the gene lnu(A), which had been only previously found in staphylococci, streptococci, enterococci and lactobacilli of animal origin and in staphylococci isolated from humans [50,53]. Strikingly, the clindamycin resistant strains P. pentosaceus LPP32 and B5 and W. cibaria SMA25 (MIC = 4 and 2 mg/L, respectively) did not harbour this gene nor lnu(B). To our knowledge, this is the first description of mef(A/E) in the genera Pediococcus and Weissella, and lnu(A) in the genus Pediococcus. The detection of resistance genes for macrolide and lincosamide in non-enterococcal strains suggests a wider distribution of this group of genes than previously anticipated.
The in vitro subtractive screening proposed in this work also include the assessment of bile salts deconjugation, mucin degradation, biogenic amine production and other potentially detrimental enzymatic activities such as the β-glucuronidase activity, which should be absent in probiotic candidates [54][55][56]. Excessive deconjugation of bile salts may be unfavourable in animal production since unconjugated bile acids are less efficient than their conjugated counterparts in the emulsification of dietary lipids. In addition, the formation of micelles, lipid digestion and absorption of fatty acids and monoglycerides could be impaired by deconjugated bile salts [57]. Similarly, excessive degradation of mucin may be harmful as it may facilitate the translocation of bacteria to extraintestinal tissues [55]. In this respect, it is worthy to note that none of the 49 tested LAB deconjugated bile salts nor exhibited mucinolytic activity, the latter indicating their low invasive and toxigenic potential at the mucosal barrier. These results are in accordance with previous findings showing that LAB do not degrade mucin in vitro [58,59]. Moreover, β-glucuronidase activity has been associated with the generation of potential carcinogenic metabolites [56]; however, none of the LAB tested in our study displayed this harmful enzymatic activity. In a previous work [60], we demonstrated that none of the 40 non-enterococcal strains evaluated herein produced histamine, tyramine or putrescine. With regard to enterococci, the nine E. faecium strains only produced tyramine, being E. faecium CV1 a low producer of this biogenic amine. Although the lack of biogenic amine production by probiotic strains is a desirable trait, it should be borne in mind that tyramine production by enterococci is a very frequent trait [60,61]. Finally, several studies have suggested that probiotic microorganisms might exert a beneficial effect in the digestion process of fish due to the production of extracellular enzymes [62][63][64][65]. In our work, the LAB strains of aquatic origin within the genera Pediococcus, Enterococcus and Lactobacillus showed a higher number of enzymatic activities than Lactococcus, Leuconostoc and Weissella, being the enzymatic profiles similar amongst strains within the same genus. In this respect, nearly all the strains produced phosphatases, which might be involved in nutrient absorption [64], and peptidases and glucosidases that breakdown peptides and carbohydrates, respectively. However, the tested LAB showed weak lipolytic activity and no proteolytic activity.

Conclusions
This work shows that antimicrobial/bacteriocin activity against fish pathogens is a widespread probiotic property amongst LAB isolated from aquatic animals regarded as human food. However, particular safety concerns based on antibiotic resistances and virulence factors were dominant within E. faecalis (100%) and E. faecium (79%), and acquired antibiotic resistance genes were not commonly found (7.5%; erythromycin and clindamycin) amongst the non-enterococcal isolates of aquatic origin. To our knowledge, this is the first large-scale study describing the antimicrobial activity against fish pathogens and the safety assessment beyond the QPS approach of LAB isolated from aquatic animals. The in vitro subtractive screening presented herein, which allowed the selection of 33 strains (8 E. faecium, 11 P. pentosaceus, 1 Lb. carnosus, 1 Lb. curvatus, 3 L. cremoris, 3 Lc. cremoris and 6 W. cibaria) out of 99 LAB isolates of aquatic origin, constitutes a valuable strategy for the large-scale preliminary selection of putatively safe LAB intended for use as probiotics in aquaculture and to avoid the spreading of bacterial cultures with harmful traits into the aquatic environment. Nevertheless, a comprehensive in vivo assessment of their lack of toxicity and undesirable effects must be also carried out using cell lines, live food and, ultimately, aquatic animals before their unequivocal consideration as safe probiotics for a sustainable aquaculture.

Bacterial strains and growth conditions
A total of 99 LAB (59 enterococci and 40 non-enterococci) of aquatic origin with antimicrobial activity against spoilage and food-borne pathogenic bacteria of concern for the fish industry, previously isolated and identified by our group from fish, seafood and fish products [14], were used in this study ( Table 1). The LAB strains were isolated on non-supplemented MRS (Oxoid, Ltd., Basingstoke, United Kingdom) or KAA (Oxoid) agar (1,5%, w/v) at 25°C, and taxonomically identified [14] by sequencing of the genes encoding 16S rRNA (16S rDNA) [66] and/or superoxide dismutase (sodA) [67]. Unless otherwise stated, LAB were grown aerobically in MRS broth at 32°C.

Direct antimicrobial activity assay
The antimicrobial activity of the 99 LAB against the main Gram-positive and Gram-negative fish pathogens was assayed by a qualitative stab-on-agar test (SOAT) as previously described by Cintas et al. [68]. Briefly, pure cultures were stabbed onto MRS or Tryptone Soya Agar (TSA) (Oxoid) plates supplemented with glucose (2%, w/v) and incubated at 32°C for 5 h, and then 40 ml of the corresponding soft agar (0.8%, w/v) medium containing about 1 × 10 5 CFU/ml of the indicator strain was poured over the plates. After incubation at 28-37°C for 16-24 h depending on the indicator strain, the plates were checked for inhibition zones (absence of visible microbial growth around the stabbed cultures), and only inhibition halos with diameters >3 mm were considered positive. L. garvieae JIP29-99 was grown aerobically in Tryptone Soya Broth (TSB) (Oxoid) at 37°C. S. agalactiae CF01173 and S. iniae LMG14521 were grown aerobically in Brain Heart Infusion (BHI) broth (Oxoid) at 37°C. A. hydrophila CECT5734, Ls. anguillarum CECT4344, Ls. anguillarum CECT7199, and Ph. damselae CECT626 strains were grown aerobically in TSB at 28°C. V. alginolyticus CECT521 was grown aerobically in TSB supplemented with NaCl (1%, w/v; Panreac Química S.A.U, Barcelona, Spain) at 28°C.

Extracellular antimicrobial activity assay
The antimicrobial activity of supernatants from LAB cultures grown in MRS broth at 32°C for 16 h was determined by an agar well-diffusion test (ADT) as previously described by Cintas et al. [68]. Supernatants were obtained by centrifugation of cultures at 10,000 × g at 4°C for 10 min, adjusted to pH 6.2 with 1 M NaOH, filtersterilized through 0.22 μm-pore-size filters (Millipore Corp., Bedford, Massachussets, USA) and stored at −20°C until use. Fifty-μl aliquots of cell-free culture supernatants were placed into wells (6-mm diameter) cut in cooled MRS or TSB agar (0.8%, wt/vol) plates previously seeded (1 × 10 5 CFU/ml) with the indicator microorganisms Pediococcus damnosus CECT4797, L. garvieae JIP29-99 or A. hydrophila CECT5734. After 2 h at 4°C, the plates were incubated under the same conditions mentioned above to allow for the growth of the target microorganisms and then analyzed for the presence of inhibition zones around the wells. To determine the proteinaceous nature of the antimicrobial compounds, supernatants showing antimicrobial activity were subjected to proteinase K treatment (10 mg/ml) (AppliChem GmbH, Germany) at 37°C for 2 h. After proteinase K inactivation by heat treatment (100°C, 10 min), samples were assayed for residual antimicrobial activity by an ADT as described above using P. damnosus CECT4797 as indicator microorganism. Supernatants with no added enzyme were treated as indicated above and used as controls. For further characterization of the antimicrobial compounds, 7 ml of supernatants from an overnight culture of LAB were subjected to peptide concentration by ammonium sulphate precipitation. Ammonium sulphate was gradually added to the supernatants to achieve 50% saturation. Samples were kept at 4°C with stirring for 3 h, and then centrifuged at 10,000 × g at 4°C for 30 min. Pellets and floating solid material were combined and solubilized in 350 μl of 20 mM sodium phosphate (pH 6.0), and antimicrobial activity of the resulting 20-fold concentrated supernatants was determined by an ADT as described above.

Production of gelatinase by enterococci
Gelatinase production was determined using the method previously described by Eaton and Gasson [32]. Briefly, enterococci were grown in MRS broth overnight at 32°C, and streaked onto Todd-Hewitt (Oxoid) agar plates (1.5%, w/v) containing 30 g of gelatine per litre. After incubation overnight incubation at 37°C, the plates were placed at 4°C for 5 h before examination for zones of turbidity (protein hydrolysis) around the colonies. E. faecalis P4 was used as positive control.

Production of hemolysin
To investigate hemolysin production by the 99 LAB, the strains grown in MRS broth were streaked onto layered fresh horse blood agar plates (BioMérieux, Marcy l'Étoile, France) and grown at 37°C for 1-2 days [32]. β-hemolysis was revealed by the formation of clear zones surrounding the colonies on blood agar plates. E. faecalis P4 was used as positive control.

Deconjugation of bile salts
The ability of the 49 pre-selected LAB to deconjugate primary and secondary bile salts was determined according to Noriega et al. [73]. Bile salt plates were prepared by adding 0.5% (w/v) sodium salts of taurocholate (TC) and taurodeoxycholate (TDC) (Sigma-Aldrich Corporation, St. Louis, Missouri, USA) to MRS agar (1.5%, w/v) supplemented with 0.05% (w/v) L-cysteine (Merck KGaA, Darmstadt, Germany). Overnight liquid cultures of strains (10 μl) were spotted onto agar plates and incubated under anaerobic conditions (Anaerogen, Oxoid) at 37°C for 72 h. The presence of precipitated bile acid around the colonies (opaque halo) was considered as a positive result. A fresh fecal slurry of a healthy adult horse was used as positive control for bile salts deconjugating activities.

Degradation of mucin
The capacity of the 49 pre-selected LAB to degrade gastric mucin was determined as described by Zhou et al. [58]. Mucin from porcine stomach type III (Sigma-Aldrich Corp.) and agar were added to medium B without glucose at concentrations of 0.5% (w/v) and 1.5% (w/v), respectively. A volume of 10 μl of 24 h viable bacterial cultures was inoculated onto the surface of medium B. The plates were incubated anaerobically at 37°C for 72 h, subsequently stained with 0.1% (w/v) amido black (Merck KGaA) in 3.5 M acetic acid for 30 min, and then washed with 1.2 M acetic acid (Merck KGaA). A discoloured zone around the colony was considered as a positive result. A fresh fecal slurry of a healthy adult horse was used as positive control for mucin degradation ability.

Determination of enzymatic activities
The APIZYM test (BioMérieux, Montallieu Vercieu, France) was used for determination of enzymatic activities of the 49 pre-selected LAB. Cells from cultures grown at 32°C overnight were harvested by centrifugation at 12,000 g for 2 min, resuspended in 2 ml of API Suspension Medium (BioMérieux) and adjusted to a turbidity of 5-6 in the McFarland scale (approx. 1.5-1.9 × 10 9 CFU/ml). Aliquots of 65 μl of the suspensions were added to each of the 20 reaction cupules in the APIZYM strip. The strips were incubated at 37°C for 4.5 h and the reactions were developed by addition of one drop each of the APIZYM reagents A and B. Enzymatic activities were graded from 0 to 5, and converted to nanomoles as indicated by the manufacturer´s instructions.

PCR detection of antibiotic resistance genes
The presence of genetic determinants conferring resistance to aminoglycosides except streptomycin [aac