Selection of sponge-associated bacteria with high potential for the production of antibacterial compounds

The potential of sponge-associated bacteria for the biosynthesis of natural products with antibacterial activity was evaluated. In a preliminary screening 108 of 835 axenic isolates showed antibacterial activity. Active isolates were identified by 16S rRNA gene sequencing and selection of the most promising strains was done in a championship like approach, which can be done in every lab and field station without expensive equipment. In a competition assay, strains that inhibited most of the other strains were selected. In a second round, the strongest competitors from each host sponge competed against each other. To rule out that the best competitors selected in that way represent similar strains with the same metabolic profile, BOX PCR experiments were performed, and extracts of these strains were analysed using metabolic fingerprinting. This proved that the strains are different and have various metabolic profiles, even though belonging to the same genus, i.e. Bacillus. Furthermore, it was shown that co-culture experiments triggered the production of compounds with antibiotic activity, i.e. surfactins and macrolactin A. Since many members of the genus Bacillus possess the genetic equipment for the biosynthesis of these compounds, a potential synergism was analysed, showing synergistic effects between C14-surfactin and macrolactin A against methicillin-resistant Staphylococcus aureus (MRSA).

Identification of active strains. The active isolates were identified by 16S rRNA gene sequencing.
Genomic DNA of the isolates was extracted using the genomic DNA kit (Analytik Jena). PCR amplification was carried out using the primer pair pA (5′-AGA GTT TGA TCC TGG CTC AG-3′) and pH (5′-AAG GAG GTG ATC CAG CCG CA-3′). The PCR was performed in a total volume of 40 µL including 2 µL of DNA template, 2 U of Taq Polymerase (Promega, Madison, USA), 1× green buffer, 0.2 mM dNTPs, 0.5 µM primer pA, 0.5 µM primer pH, 1.25 mM MgCl 2 and 5% of dimethyl sulphoxide (DMSO). PCR was performed in a Biometra TRIO Thermal Cycler (Analytik Jena) using the amplification conditions as follows: 95 °C 5 min (denaturation), 50 °C 45 s (annealing), 72 °C 1 min (elongation), and 72 °C 5 min (final elongation). Amplified 16S rRNA gene fragments were purified using the Promega SV Gel and PCR Clean Up System kit. Purified PCR products were sequenced (Eurofins Genomics). The obtained forward and reverse sequences were assembled. Then, a BLAST analysis was done to identify closest homologues. Sequences with > 98% sequence similarity (over an average of 1372-1400 bases) to their closest phylogenetic neighbour were assigned to the species level. Sequences with ˂ 98% sequence similarity were identified to the genus level 26 . The 16S rRNA gene sequences of the isolates have been deposited in GenBank with the accession numbers MT314037-MT314061 (Supplementary Table S1). Phylogenetic tree based on the 16S rRNA gene was constructed in MEGA X by using the Maximum Likelihood method (default settings, 1000 bootstraps) based on the Tamura 3-parameter model. The percentage of trees in which the associated taxa clustered together is shown next to the branches. Initial tree(s) for the heuristic search were obtained automatically by applying Neighbor-Joining and BioNJ algorithms to a matrix of pairwise distances estimated using the Maximum Composite Likelihood (MCL) approach 27 .

Scientific Reports
| (2020) 10:19614 | https://doi.org/10.1038/s41598-020-76256-2 www.nature.com/scientificreports/ BOX PCR. Preparation of samples and PCR conditions. The bacterial strains were grown on ISP2 agar plate at 30 °C overnight. Several bacterial colonies were transferred into the collection micro tubes (Cat.No.19560, Qiagen) with 2 Zirconia beads (2.3 mm Carl Roth, Art.: N036.1) and dissolved in 200 µL of sterile water. The mechanical cell disruption was performed by TissueLyser II for 2 × 1 min at 30 Hz. The samples were centrifuged for 2 min at 3220×g. One microliter of supernatant was used as a DNA template for the repetitive element palindromic-polymerase chain reaction (rep-PCR) analysis with BOXA1R oligonucleotide (CTA CGG CAA GGC GAC GCT GACG) 28 . The rep-PCR was carried out in a total volume of 25  Statistical analysis/ Rep-PCR DNA fingerprint analysis. Gel image was normalized, bands were identified and data were statistically analysed by using GelCompare II software version 6.5 (Applied Maths, Belgium). The positions of bands on the gel were normalized using the DNA 5 K ladder (PerkinElmer, Inc.) from 100 to 7000 bp as an external reference standard. Similarity coefficient was calculated by band-based method of Dice, while unweighted pair group method with arithmetic averages (UPGMA) was used for cluster analysis.
Competition assay. Initially, all strains to be tested were cultured on agar and in liquid medium 1-7 days.
Liquid cultures represented the pre-culture for the test strain. Therefore, cells were separated by centrifugation at 6010×g, 5 min, 20 °C; then, inoculated into soft agar medium (10% agar), which was used to overlay the test plates. The tested bacteria were also inoculated and incubated until growth was visible on separate plates. The competition assay started afterwards; therefore, agar-plugs were prepared from the solid medium cultures and placed on the test plates.
Screening culture extracts for antibacterial activity. The five most promising strains from the competition assay were fermented in ten different liquid media: ISP2, ISP2 + NaCl, LB, LB with ASW, marine broth, malt yeast extract, nutrient broth, starch nitrate with ASW, tryptic soy broth (TSB), tryptone and yeast extract medium with ASW. TSB medium consists of casein peptone (17 g/L), K 2 HPO 4 (2.5 g/L), glucose (2.5 g/L), NaCl (5 g/L), soya peptone (3 g/L); tryptone and yeast extract medium with ASW consists of casein peptone (4 g/L) and yeast extract (2.5 g/L); the other media compositions are given in SI. Strains were inoculated from cryoculture into a preculture (20 mL in 50 mL flask), which was incubated at 140 rpm at 30 °C overnight. Then, 1 mL of this preculture was transferred into the main culture (100 mL new medium in 300 mL flask). Fermentation was done at 140 rpm, 30 °C for 2 days. The fermented broths were extracted once using ethyl acetate (1:1). The resulting organic layer was collected and evaporated to dryness under reduced pressure using a rotary evaporator. Therefrom resulting organic extracts were used for antibacterial testing using the agar diffusion method. Extracts were dissolved in methanol (final concentration of 10 mg/mL) and 10 µL were applied onto a 6 mm paper disc that was allowed to dry at room temperature. Plates were pre-incubated for 1 h at 4 °C and subsequently transferred to 30 °C for 24-48 h. Methanol and carbenicillin were used as negative and positive control, respectively.

MS measurements and molecular networking.
To compare the various extracts and to dereplicate the compounds therein, molecular networking was carried out. The sixty-crude extract samples were dissolved in MeOH at the final concentration of 10 mg/mL and subjected to LC-HRMS measurement. Mass spectra were detected on a micrOTOF-QII mass spectrometer (Bruker, Billerica, MA, USA) with ESI-source combined with a HPLC Dionex Ultimate 3000 (Thermo Scientific, Darmstadt, Germany) utilizing an EC10/2 Nucleoshell C18 2.7 μm column (Macherey-Nagel, Düren, Germany) at 25 °C. MS data were obtained in positive mode over a range from 100 to 1000 m/z. For all ions above a threshold of 100, auto MS/MS fragmentation was performed with increasing collision energy (35-50 kV over a gradient from 500 to 2000 m/z) at a frequency of 4 Hz. The injection volume was 2 µL with a concentration of 1 mg/mL. MS/MS data were converted from MassHunter data files (.d) to mzXML file format using MS Convert. The data were uploaded to the Global Natural Products Social (GNPS) molecular networking (https ://gnps.ucsd. edu/). Network files were visualized using the program Cytoscape 3.7.2. Dereplication was done by comparing the MS 2 spectra with the reference spectra in GNPS spectral libraries 25 . Metabolic fingerprinting. Bacteria were inoculated from cryoculture in ISP2 liquid medium supplemented with 2% NaCl and incubated overnight at 30 °C and 140 rpm. Then, ISP2 agar plates (2 per strain) were inoculated each with 50 µL of preculture and incubated at 30 °C for 2 days. The agar plates were cut into small pieces, macerated with ethyl acetate and placed on a shaker overnight. The solution was filtered and the resulting organic extract was dried in a rotary evaporator under vacuum. The concentration of this crude extract was adjusted to 50 mg/mL and subjected to LC-HRMS measurement (procedure done in triplicates).
MS-analysis was performed on 1290 UHPLC system (Agilent, Santa Clara, CA, USA) equipped with DAD, ELSD and maXis II (Bruker, Billerica, MA, USA) ESI-qTOF-UHRMS with the following gradient: A = H 2  www.nature.com/scientificreports/ Data processing was performed with DataAnalysis 4.4 (Bruker, Billerica, MA, USA) using sodium formate for recalibration, RecalculateLinespectra (threshold 10,000) and FindMolecularFeatures (0.5-25 min, S/N = 0). Bucketing was performed using ProfileAnalysis 2.3 (Bruker, Billerica, MA, USA) (30-1080 s, 100-1600 m/z, Advanced Bucketing with 24 s 5 ppm, no transformation, Bucketing basis = H+). The bucket table was subsequently used as input for analysis via R. R (version 3.6.0) 29 with libraries readr 30 , coop 31 , gplots 32 , data.table 33 , parallelDist 34 and devtools 35 were used. For heatmap-generation with several sidebars a variation of heatmap.2 by Griffith 36 was used. Hierarchical clustering in the heatmap is performed with function "hclust" based on "complete linkage" of the cosine similarity results. The complete R-script is deposited here: Cosine-V2.R GitHub repository https ://githu b.com/chris toph-hartw ig-ime-br/cosin e-V2; https ://dx.doi.org/10.5281/zenod o.39329 68. For sample comparison, the cosine similarities (dot product of vectors) between samples were calculated. Samples were sorted according to clustering results and pairwise similarities were used to determine metabolic groups. If the pairwise similarity between to subsequent clustered samples is at the threshold (e.g. 0.9) or higher, they belong to the same metabolic group. Cosine similarity data was also analyzed inside of triplicates and between strains. For outlier in triplicates MS-data were inspected. If the differences could be explained due to large concentration differences, outliers were included in the same metabolic group as the rest of the strain.
Co-cultivation. Selected bacteria (one representative of closely related strains of the 108 active ones with a different species) were cultured in liquid media in a co-cultivation approach. Bacterial strain 1 and bacterial strain 2 were inoculated separately in a 150 mL Erlenmeyer flask as pre-culture. Then, the experiment was conducted as follows: (a) Strain 1 was inoculated as single strain control, (b) strain 2 was inoculated as single strain control, (c) strain 1 was inoculated first and strain 2 was inoculated after 1 day in the same flask, (d) strain 2 was inoculated first and strain 1 was inoculated after 1 day in the same flask, (f) strain 1 and strain 2 were inoculated at the same day in the same flask, (e) medium was used as negative control.
The activity was measured based on the zone of clearance around the paper discs. The co-cultivation experiments that showed the highest activities were selected to perform the large scale cultivation: (1) Bacillus sp. EP6-817 and Lysinibacillus sphaericus EP6-121 inoculated at the different day in NB medium (Peptone 5 g/L, Malt extract 3 g/L, NaCl 5 g/L) and cocultivated for 1 day at 30 °C with shaking at 140 rpm; (2) Verrucosispora sp. EP6-325 and Bacillus sp. EP6-454 inoculated at the same day in M1 medium (Starch 10 g/L, Yeast extract 4 g/L, Peptone 2 g/L) and cultivated for 3 days at 30 °C with shaking at 140 rpm.

Minimal inhibitory concentration and synergistic effect of surfactin and macrolactin.
The minimal inhibitory concentration (MIC) and fractional inhibitory concentration (FIC) were determined as previously reported 37 . In brief, a panel of Gram-negative E. coli ATCC 25922 (wild-type and ΔtolC mutant), Grampositive B. subtilis DSM 10, S. aureus ATCC 25923 (methicillin-sensitive), and S. aureus ATCC 33592 (methicillin-resistant, MRSA) were grown in cation-adjusted Mueller-Hinton II broth (CAMHB, Becton Dickinson) over-night, Listeria monocytogenes DSM 20600 was grown in brain heart infusion (BHI) medium supplemented with 1% (v/v) Tween 80 over 2 days of incubation at 37 °C, until the 30 mL preculture was turbid. Next, the preculture was adjusted to McFarland 0.5-1 turbidity standard (approx. 1.5 -3.0 × 10 8 cfu per mL) and diluted 1:600 in fresh media as seeding cell suspension for the growth inhibition assay. Purified macrolactin A and surfactins were dissolved in DMSO (12.8 mg/mL final concentration), spotted onto 96 well plates and 1:2 dilution series in a volume of 100 µL of the seeding cell suspensions were prepared on 96 well plates to obtain a final compound concentration of 128 -0.0625 µg/mL. Growth and sterility controls were added (DMSO as solvent control), and gentamycin served as positive control. Endpoint MIC values were measured after 18 ± 2 h at 180 rpm shaking speed and 37 °C at 85% relative humidity. Readout was obtained by turbidity absorption measurement at 600 nm and chemiluminescence-based ATP-quantification using BacTiter Glo Microbial Cell Viability Assay (Promega)

Results
Sponge collection and bacteria isolation. During our continuous efforts to identify natural products from marine bioresources, 10 sponge specimens covering 10 different Demospongia species, were collected from Sangihe Island, Indonesia (a description of the specimens with a preliminary genus or species identification is provided in the Supplementary Table S2). From these samples, associated bacteria were isolated using classical agar plate-based methods with various media. In total, 835 marine bacteria were isolated as axenic culture. From most sponge samples, approximately 100 cultures were isolated. However, from two sponge samples, i.e. specimen EP10 (cf. Aaptos suberitoides) and EP15 (cf. Agelas nakamurai), only about 20 isolates were retrieved in this study. Comparing the isolation efficiency of the different media, the highest number of isolates (143) was obtained from SNA medium (consisting of starch and salts); while from NB medium (complex medium) only 23 isolates were retrieved (Fig. S11).
Screening for antibacterial activity. Concerning a biological activity of the isolates, we focussed on antibacterial activity. The 835 isolates were screened for activity against the Gram-positive bacterium M. luteus ATCC 4698 and the Gram-negative bacterium E. coli K12 using a growth inhibition assay on agar plates. This primary screening revealed that 12.9% (108) of the isolates showed antimicrobial activity against at least one of the bacteria tested (Table S3). Among the 108 isolates active in the primary screening, 4.6% inhibited solely Gram-negative E. coli bacteria, 78.7% inhibited Gram-positive M. luteus and 16.7% inhibited both test strains (Fig. 1). Thereby, the highest number of active isolates originated from the sponge EP6 (22 out of 104 isolates showed activity) and the lowest number from sponge EP14 (2 out of 112 isolates showed activity). From sponge sample EP15 no active isolate was retrieved. However, it must be considered that from this sponge only a very low number of bacteria was isolated in total (Fig. 1).
Competition assay. An important step in natural product research is the prioritization of bacterial strains for further investigation. Therefore, a strong dereplication platform, mostly based on MS analyses of the extracts derived from the strains, is key. Another option is the in silico analysis of the strains following genome sequencing. However, both platforms are relatively costly and not available at all laboratories. The goal of this project was to test if a prioritization of strains is possible depending on agar plate-based competition assays. This methodology can be done by microbiologists in virtually every lab, also in remote areas without expensive instrumentation.
The underlying hypothesis is that the strongest competitors should be selected, since these strains have a higher probability of success for the identification of compounds with antibacterial activity. Strong competitors were regarded as the strains that inhibited most of the other strains originating from the same host sponge (Fig. 2). Hence, in a preliminary round, the strains were challenged in bilateral agar plate-based inhibition assays. Strains were incubated as axenic cultures on agar plates during 2-7 days depending on the growth rate of the  Figure 2. Scheme of the competition assay. Bacteria were isolated from the sponge specimens. As example a blue and green coloured sponge are shown. All axenic cultures derived from one sponge were tested against all the others from one sponge. Therefore, bacteria were cultured and agar plugs were transferred to agar plates inoculated with the test strain. Best competitors (inhibiting most of the test strains, which was observed by inhibition zone) were selected and tested against best competitors derived from other sponges. After 24 h of incubation, the inhibition zone was measured. In that way, the strongest inhibitors (in total 25 strains) were selected and qualified themselves for the next round, i.e. the competition assay between the strains selected from the different sponge hosts. In the final round, the number of inhibited strains, inhibition zones and the number of strains inhibiting each strain were documented ( Fig. 3 and Table S4). Bacillus sp. EP1-654 was the winner of this challenge; since this strain inhibited 16 of the 24 other strains (67%). Thereby, it showed inhibitory activity against at least one bacterium from each host sponge, except the bacterial strain derived from sponge EP8, i.e. Bacillus sp. EP8-203. The latter strain was also a high competitor, like Bacillus sp. EP7-200, EP5-815 and EP6-816; it inhibited 12 out of 24 strains of the test set, respectively. Furthermore, Bacillus sp. EP8-203 showed the lowest sensitivity, since it was only inhibited by one competitor. Brevibacterium sp. EP14-508 was the weakest strain in the challenge. It did not inhibit one of the other strains and in turn was inhibited by 16 of the competitors (Fig. 3 and Table S4).
Correlation of marine bacterial taxonomy and chemical fingerprint. As mentioned before, the top five competitors belonged to the genus Bacillus, which was also the most abundant genus in this collection. To get an idea about the phylogenetic relationship and if other genera could also be found among the strong competitors, a phylogenetic tree based on the 16S rRNA gene sequences was built (Fig. 3). The activity and sensitivity pattern were plotted to this tree and showed that the highest competitors belonged to a branch of closely related strains. This trend was observed (i) when all strains were fermented in the same medium (which gives a better comparability) and (ii) when the strains were fermented in the medium they were originally isolated from. Since it is known that variation of the growth conditions, e.g. medium applied, results in a changed metabolome, the isolation medium was also used (Fig. 3). Since this could be an indication that the strains selected by the competition assay would be highly redundant, a BOX PCR was performed (Fig. 4).
A dendrogram based on BOXA1R fingerprint data was constructed by using Dice similarity coefficient and the UPMGA cluster analysis method to determine the phylogenetic relationship of the strains with higher resolution. This revealed that a few strains show a high similarity of the band pattern, but in general, a quite diverse pattern was observed. Taking into account the similarity higher than 75%, the strains are grouped in three major clusters (Fig. 4). The first group includes solely Bacillus strains. The average similarity coefficient among these strains was 93% while the strains EP4-170, EP7-199, EP7-200 and EP8-203 appeared to be identical. The second group represents the most heterogeneous group involving the species from different genera but with the average similarity of 90%. The third group includes four Bacillus strains isolated from EP 1, 10 and 13. They had the same similarity coefficient as group I.
In a next step, the 25 high competitors were subjected to a metabolomics analysis. The taxonomic similarity based on housekeeping genes does not necessarily reflect the similarity of strains in regard to their metabolome, since also closely related strains might carry different biosynthetic gene clusters (BGCs) encoding for antibacterial Activity/sensitivity profile of the strains, which were cultivated in the medium they were originally isolated from. Therefore, different media were used. (c) All strains fermented in ISP2 medium supplemented with NaCl. The bar diagram indicates for each strain how many competitors were inhibited (red) and by how many itself was inhibited (grey). The blue bar indicates the branch of high competitors. Bootstrap values are given at the branches of the phylogenetic tree. www.nature.com/scientificreports/ active natural products. Therefore, the crude extracts of these strains were analysed by chemical fingerprinting. The chemical fingerprint enables to judge the similarity of metabolomes, e.g. visualized in a dendrogram (Fig. 5, complete dataset is shown in Figure S13). Grouping of all extracts using a cosine similarity threshold of 0.9 resulted in 38 groups (cosine 0.8 in 19 groups). Analysis of cosine similarity between strain triplicates was performed and outliers were sorted to the rest of the triplicates if large differences in concentration led to the deviation while the Base Peak Chromatogram pattern was identical (Table S5, metabolic variations are shown in Figure S12). After inspection, 19 metabolic groups (excluding media controls) remained (Table S6, corresponding chromatograms Figure S12). While strain EP14-508 is most dis-similar to all other profiles (Fig. 5), it closely groups to other strains in the DNA analysis. The strains EP7-199 and EP7-200 showed even 100% identity in their 16S rRNA sequence; however, in the chemical fingerprint differences were detectable. This result matches the results of the BOX PCR, that already indicated slight differences, and of the competition assay, since in the latter strain EP7-200 inhibited more strains than EP7-199. Furthermore, the strains of metabolic group 1 are partly from different clades based on the DNA fingerprint (Fig. 4).
Natural products produced.  (Table S7). In addition to the activity tests, the extracts from the five selected strains in ten media, as well as the medium controls were subjected to LC-MS/MS analysis. Thereby, some compounds could be dereplicated based on the GNPS natural products library, e.g. surfactin C12, surfactin C14, surfactin C15 and lichenysin A (Figs. S14, S15, S16).
The production of surfactins was a common feature for many of the here isolated bacteria when cultivated on solid medium. In liquid medium, it was observed that especially in co-culture experiments the production of this compound series was induced. The C14 to C17 surfactins were successfully isolated from a co-culture, in which Verrucosispora sp. EP6-325 and Bacillus sp. EP6-454 were inoculated at the same day in M1 medium (Figs. S7-S10).
The experiments with axenic cultures do not reflect the conditions of the microbiome of a sponge in nature. Therefore, further co-culture experiments were performed, to challenge the strain with the presence of a competitor. A co-culture of Bacillus sp. EP6-817 and Lysinibacillus sphaericus EP6-121 was tested in a way that the www.nature.com/scientificreports/ one strain was incubated for one day before the other strain was added, as well as inoculation on the same day in one flask. A clear increase in the production of one compound was observed when Bacillus sp. EP6-817 was cultured for 1 day, before the culture was inoculated with L. sphaericus EP6-121 (Fig. 6, Fig. S17). This peak was not detected in L. sphaericus EP6-121 extracts and in much lower abundance if Bacillus sp. EP6-817 was cultivated alone. From the co-culture, this peak was purified and its structure elucidated using NMR experiments, proving it to be macrolactin A (m/z 425.2308 [M + Na] + , Figs. S1-S6).

Synergistic effect of surfactin and macrolactin. Surfactin and macrolactin appeared to be compounds
of which the production is increased by the presence of a competitor strain. Furthermore, both corresponding BGCs can be detected in many sequenced Bacillus genomes. Therefore, their antibacterial effects were analysed as single compounds and in combination. However, none of the surfactins were active against these strains (MIC > 128 µg/mL). Instead, macrolactin A was active against the Gram-positive test strains, except B. subtilis DSM 10 (Table S8). In addition, combination effects between macrolactin A and surfactin C14 were assessed by chequerboard assays (Table S9) www.nature.com/scientificreports/ bial effect (FICI ≤ 0.5). As compared to macrolactin A without any supplement, the percentage of growth inhibition is enhanced by factor >10 at 1 µg/mL macrolactin A and surfactin dosing ≥ 0.5 µg/mL, whereas at 2 µg/ mL macrolactin A the percentage of growth inhibition is doubled by surfactin ( Fig. 7 and Fig. S18). At higher macrolactin A concentrations (≥ 4 µg/mL), the synergistic effect cannot be resolved anymore.

Discussion
Sponge-associated bacteria represent an important bioresource for antibiotic compounds. In this study, 12.9% of isolated bacteria exhibited antimicrobial activity against at least one test organism. This number of active isolates is in the range reported for other bioprospecting projects, which was in the range of 2-34% 21,38,39 . E.g., 8 of 400 isolates (2%) from the coastal marine sponges Amphilectus fucorum and Eurypon major revealed activity against E. coli and B. subtilis 40 . In our study, Firmicutes is the phylum with the highest share (77%) on antibacterial activity, thereby especially genus Bacillus contributing to this, which is consistent with other reports 41,42 . However, several of these projects, aiming to isolate bacteria with antibiotic activity started with a lower number of isolates and selected the most promising isolate. For example, from 92 bacteria isolated from red algae, 33% showed antibacterial activity. Of the ten selected isolates, seven were of the genus Bacillus 41 . A total of 158 isolates were isolated from a sponge, thereunder twelve were active isolates, five of them belonging to Firmicutes 42 . In general, Bacillus have to be regarded as proliferative producers of natural products with various biological effects. Together with their sporulation efficiency this can be regarded as a significant advantage for survival in different environments 43 . However, this is also providing a bias in isolation projects, since the spore-formers easily survive different sampling and storage procedures and are fast growers on many standard media. Based on metagenomic approaches, the most dominant groups of sponge holobionts belong to the phylum Proteobacteria [44][45][46] . Therefore, it is obviously hard to compare the cultured communities from marine sponges directly, since variations in media  www.nature.com/scientificreports/ and culture conditions can have a huge impact on the isolation and cultivation success 42 . Furthermore, the great plate count anomaly was also reported for sponge-derived bacteria, e.g. less than 1% of bacteria observed by microscopic analysis in sponge tissues could be cultured using standard medium 47 , and only a low number of bacteria within a sponge grew under laboratory conditions 12 .
In this study, the fraction of isolated bacteria from different sponges varied. However, the low number of 20 isolates originating from sponge EP 15 (Agelas nakamurai) can be explained by the presence of agelasines in this sponge 48 . Agelasines are sponge-derived compounds that show antibacterial, as well as cytotoxic properties. Hence, it might be that either associated bacteria were killed during storage and processing of the samples due to the active compounds present, or that due to the presence of these effective compounds the sponge does not need dedicated antibiotic producers in its microbiome.
To evaluate the approach of prioritizing strains active in a preliminary screening, based on their ability to inhibit other isolates from the same holobiont, the metabolomic fingerprints were analysed. The method was reported to provide a high-resolution strain discrimination, as shown for a group of closely related Streptomycetes 49 . Here, experiments were performed in triplicates, which grouped together; however, a few triplicate samples did not cluster together; they rather formed a group with other strains. This might indicate that these strains, that cannot be clearly differentiated by chemical fingerprinting, are very similar in their metabolome. Furthermore, three strains (strain code EP4-170, EP7-199 and EP7-200) that are highly similar on their metabolic fingerprint showed high identity in the BOX PCR band pattern, and formed a monophyletic clade in the 16S rRNA gene-based phylogenetic tree. However, even these three strains, which are closely related to Bacillus subtilis, did not show an identical activity and sensitivity pattern in the competition assay. Therefore, it can be concluded that analysis of the activity and sensitivity pattern of a strain allows differentiation and therewith prioritization, even between highly similar strains. This result is consistent with another study 50 , which reported significant differences in bioactivity and chemical diversity between strains of the same species. It might even be that strains, which are identical on the genetic level, could show a different reaction in strain interactions. The background of different hosts could have resulted in different fine tuning of the regulation mechanisms that in turn lead to the production of different (levels of) metabolites 20 .
In addition, the activity/sensitivity pattern of the five strongest competitors (all taxonomically related Bacillus species) was different. These bacteria should be prioritized for further analysis, because they can inhibit most of the competitors. The strains were isolated from five different sponges, which might point toward the fact that in each microbiome a high competitor Bacillus strain is selected. It is obvious that isolates from different sponges could produce varying active compounds. At present, most of the activities observed cannot be directly linked to a corresponding compound. However, sponge-derived Firmicutes (particularly the genus Bacillus), are well known for the production of antimicrobial compounds 51 . In this project, the dereplication of the extracts resulted in the identification of 'classic' Bacillus compounds, e.g. C12, C14, C15 surfactins and lichenysin A. Based on the fact that genus Bacillus in general possess many BGCs, which reflects the potential to produce various secondary metabolites, it can be expected that in nature a cocktail of compounds is used to tackle competitors. In terms of biological fitness, the ability to produce relatively low (sub-MIC) amounts of single antimicrobial compounds in combination, while maintaining or even exceeding the growth inhibition effects against competing microorganisms is of major importance. Synergy of at least two different molecules with different mode of action is given, if the effect of the combination at low dosage is larger than the additive effect of both compounds alone. This principle has shaped natural-product producing bacteria in their specialized ecological niches, whereas the classic medicinal approach is currently antibiotic treatment with one specific drug. In a historical retrospective, this has been a dead-end strategy that has assisted the emergence of antibiotic resistance, reducing effective treatment options in the future. The analysis of surfactin, which is sometimes reported as antibiotically active, did not show activity in our MIC tests, but showed synergistic effects combined with macrolactin A. The production of both compounds was clearly triggered by the presence of another strain, since in a bioactivity-guided isolation approach none of the compounds would have been caught from single cultures. The co-culture approach, based on the interaction of microbes can be regarded as an effective approach for the induction of secondary metabolites. Many studies in mixed microbe fermentation cultures have described triggered metabolite production. The co-culture of the marine-derived fungi Emericella sp. (strain CNL-878) and the marine actinomycete Salinispora arenicola (strain CNH-665) revealed markedly enhanced production of emericellamide A and emericellamide B 52 . A number of secondary metabolites were induced by co-culture of two sponge-derived Actinomycetes, Actinokineospora sp. EG49 and Nocardiopsis sp. RV163 53 . Yu et al. 54 detected the chromone derivative 7-methoxy-2,3-dimethylchromone-4-one produced by Streptomyces rochei MB037. Co-culture with the gorgonian-derived fungus Rhinocladiella similis 35 stimulated its production significantly. It can be speculated that increased natural product production is caused by microbial competition for nutrients or space.
Surfactins are cyclic lipopeptides with surfactant-like properties, interfering with many kinds of cell membranes. They were reported to show antibacterial activity in agar diffusion assays against various test strains if used at high concentrations 55,56 . However, these results must be interpreted with caution, since the MICs determined were > 1024 µg/mL 56 , indicating no activity. The debate on the antimicrobial effect of surfactins remains active, but may be accounted to the non-specific mode of action on the cell layer, e.g. cation transport through the bilipid layer 57 , pore-formation on the cell wall 58 , and distortion of the membrane integrity by detergent-like properties of surfactins that induce membrane depolarization events in the host 59 . In consequence the cell's energy balance is affected by compromising the cell wall potential, which results in reduced cell proliferation fitness and inactivation of transport processes into and out of the cell. Furthermore, deficient cell ultrastructure integrity may no longer block diffusion of large molecules, while its efflux efficiency may be severely reduced. A synergistic combination of sub-MIC surfactin and macrolactin A may be explained by this route. Synergistic effects of lipopeptides are described in some studies, i.e. surfactin with iturin , surfactin with fengycin, and iturin Scientific Reports | (2020) 10:19614 | https://doi.org/10.1038/s41598-020-76256-2 www.nature.com/scientificreports/ with fengycin 60 . In another study, synergistic effects between C15 surfactin and ketoconazole were described against Candida albicans SC5314 61 . Macrolactins are macrolide compounds containing a 24-membered lactone ring, which are well known for their broad bioactivity spectrum like antiviral and anticancer properties, as well as activity against multi-resistant and clinically relevant pathogenic bacteria such as MRSA 62,63 . However, information about the common mode of action of this macrolide antibiotic remains scarce. A bacteriostatic mode of action against staphylococci and MRSA could be inferred from morphological changes during cell septum formation, which indicated cell wall synthesis inhibition as in the case of 7-O-malonyl macrolactin A 62 . As known for macrolide antibiotics, macrolactin N targets protein biosynthesis by inhibition of peptide deformylase in S. aureus 63 .

Conclusions
The microbiome of marine sponges is a promising bioresource for natural products with various activities. An agar plate-based competition assay was performed for the selection of antibiotic-producing strains using inexpensive and easily accessible equipment and materials. Using BOX PCR in addition to 16S rRNA gene sequencing for the taxonomic identification, it was shown that diverse strains are selected by the competition assay. Integrating the analysis of the metabolic profile by comparing the cosine similarity of the strain extracts revealed that (i) strains selected from different sponges group together, which might be an indication that a specific metabolome is of ecological relevance. However, (ii) also different strains were selected by the competition assay, pointing towards different options in shaping a microbiome that can contribute to chemical defence. It was shown that molecules with assumed antibacterial effects, e.g. surfactins, are inactive alone (at the concentrations tested), but can have synergistic effects combined with other molecules produced by Bacillus species.