We selected 146 bacterial isolates from organic-poor, sandy soils under vegetation patches of sand sedge (Carex arenaria L.) growing in natural field sites (De Ridder-Duine et al., 2005) (Table S1). The bacterial isolates were pre-cultured from −80°C glycerol stocks on 1/10 TSBA (5.0 gL⁻¹ NaCl, 1.0 gL⁻¹ KH₂PO₄; 3 gL⁻¹ Oxoid Tryptic Soy Broth; 20 gL⁻¹ Merck Agar, pH 6.5) (Garbeva and De Boer, 2009) and incubated for 5–7 days at 20°C prior to screening.

Reference strains that produce known antibiotics in monoculture were obtained from the DSMZ strain collection (Leibniz Institute DSMZ-German Collection of Microorganisms and Cell Cultures, Braunschweig, Germany). These reference strains were: Streptomyces kanamyceticus (DSM 40500), producer of kanamycin, Streptomyces rimosus (DSM 40260), producer of oxytetracycline and Streptomyces nodosus (DSM 40109) producer of amphotericin A and B. These strains were pre-cultured from −80°C glycerol stocks on GYM agar plates (4.0 gL⁻¹ Glucose, 4.0 gL⁻¹ BACTO™ Yeast extract, 10.0 gL⁻¹Malt extract, 2.0 gL⁻¹ CaCO₃, 20 gL⁻¹ Merck Agar, pH 7.2) and incubated for 7 days at 28°C before inoculation into 96-well source plates (see below). In the agar-overlay assay, two bacterial strains were selected to act as model organisms for human pathogenic bacteria: Escherichia coli WA321 (DSM 4509) as Gram-negative target organism and Staphylococcus aureus 533R4 Serovar 3 (DSM 20231) as Gram-positive target organism. The target strains were pre-cultured from −80°C glycerol stocks on Luria Bertani (LB) agar plates (10.0 gL⁻¹ NaCl, 10 gL⁻¹ Bacto™ Tryptone, 5 gL⁻¹ Bacto™ Yeast extract, 20 gL⁻¹ Merck Agar) Sambrook and Russell (2001) and incubated at 37°C for 24 h before inoculation in the antimicrobial screening assay. Characteristics of the target and the control strains are listed in Table S3.

For the high-throughput interaction assay polystyrene Nunc™ OmniTray™—plates (size 128 × 86 mm; cap. 90 mL; Nunc™, Nalge Nunc International, Rochester, NY, USA Cat # 82-264728) were used. Each OmniTray™ plate was filled with 45 mL of 1/10 TSBA (2%) agar. Plates were kept in the laminar flow cabinet until the agar was completely solidified.

96-well Microtiter plates (Greiner bio-one B.V., Alphen a/d Rijn, The Netherlands, Cat# 655180) were prepared to inoculate the selected bacterial isolates and the reference strains. Each well was filled with 150 μl liquid LB broth. Bacterial isolates were inoculated in 10 rows containing quadruplicates of each strain, the 11th row was kept empty and the 12th row was used as positive control by inoculating known antibiotic-producing Streptomyces strains in duplicate with one free well between each strain (Figure 1). Inoculation was done by picking cells from a single colony of each bacterial strain with a disposable inoculation loop (VWR international B.V., Amsterdam, The Netherlands Cat# 50806-404) and transferring to the designated well in the 96-well source plates. The plates were incubated for 2 days at 24°C, after which the plates were prepared for long-term storage (−80°C freezer) by adding 50 μl of 50% (v/v) glycerol to achieve a final concentration of 12.5% (v/v). In total, 15 Microtiter plates (source plates A–O) containing different compositions of monocultures of bacterial isolates were prepared for the high-throughput interaction assay.

A Genetix QPix 2 colony picking robot (Molecular Devices, UK Limited, Wokingham, United Kingdom) was used for the high-throughput interaction assay. The Genetix QPix 2 robot was mounted with a bacterial 96-pin picking head and programmed to replicate the source plates (96-well Microtiter plates) into the OmniTray™ plates (Figure 1). The source plates were replicated two times, one set of inoculated plates was removed from the robot and was used as control to estimate growth and antimicrobial activity of the monocultures. The remaining plates in the robot were used for the interaction assay by inoculating a second set of source-plates in various combinations. The second set of bacterial isolates was inoculated at the same position as the first set of bacteria, in this way the bacterial isolates had physical cell contact and could interact in one-to-one interactions (in quadruplicates). The inoculated OmniTray™ plates (monocultures and interaction plates) were incubated for 4 days at 24°C. In total, 146 bacterial isolates were combined with each other in various arrangements and tested in 2798 unique interactions for the production of antimicrobial compounds.

For detection of antimicrobial activity, an agar overlay assay was performed on the 4th day of incubation (Nkanga and Hagedorn, 1978). The two target organisms E. coli WA321 and S. aureus 533R4 were grown overnight in liquid LB broth at 37°C, 220 rpm. Fresh LB- agar (1.5% Merck Agar) was prepared, cooled down to ~45°C and the target organisms were added to a final OD₆₀₀ of 0.002 corresponding to approximately 6 × 10^∧5 CFU/mL (E. coli WA321) or 4 × 10^∧5 CFU/mL (S. aureus 533R4) and mixed well. A volume of 15 mL liquid LB-agar containing the target organisms was poured over the OmniTray™ plates with the empty 11th row as the start position for pouring. After solidification of the overlay agar, the OmniTray™ plates were incubated overnight at 37°C. The next day (5th day), plates were examined for visible zones of inhibition (ZOI). Monocultures or mixed-cultures of the soil bacterial isolates were scored as positive for antibiotic production if at least two out of four replicates produced zones of inhibition (Figure 1A). The majority of activity reported (>55%) involved ≥3 out of 4 replicates. For confirmation of the high-throughput screening results, several of the antibiotic-triggering/suppressing interactions were tested outside the HTS setup (Figures S7, S8).

For identification of the bacterial isolates, PCRs were performed directly on colonies or with extracted genomic DNA. For genomic DNA extraction the QIAGEN QIAmp DNA Mini Kit (QIAGEN Benelux B.V., Venlo, The Netherlands cat# 51 304) was applied according to the manufacturer's manual. For the colony PCRs, a few colonies of each bacterial isolate were scraped from the plate with a disposable inoculation loop (VWR international B.V., Amsterdam, The Netherlands Cat# 50806-404) and re-suspended in 250 μl sterile MQ-water. The re-suspended bacterial cells were pulse vortexed and heated to 95°C for 5 min. Tubes were centrifuged for 3 min at 12,000×g and 1 μl supernatant from each bacterial isolate was applied in a 50 μl PCR- master mix (Promega Corp. Madison, USA cat# M7505). For 16S rRNA gene amplification, one of the two primer combinations was used: (1) forward primer pA (5′- AGA GTT TGA TCC TGG CTC AG -3′), reverse primer 1492r (5′- GRT ACC TTG TTA CGA CTT -3′), amplifying ~1492 bp from the 16S rRNA gene or (2) forward primer 27f (5′- AGA GTTT GAT CMT GGC TCAG -3′), reverse primer 1492r amplifying ~1465 bp from the 16S rRNA gene (Edwards et al., 1989; Lane, 1991) (modified). All PCR reactions were performed on a MJ Research Peltier thermal cycler 200 PCR machine (Harlow Scientific, Arlington, USA) with the following settings: initial cycle 95°C for 5 min. and 30 cycles of 94°C for 30 sec., 55°C for 30 sec. and 72°C for 1 min. After amplification, a volume of 5 μl of each PCR reaction was loaded on a 1.25 % (w/v) agarose gel and checked after electrophoresis for presence of PCR fragment. The PCR products were sent to MACROGEN (MACROGEN Europe, Amsterdam, The Netherlands) for sequencing.

Obtained sequence chromatograms of the 16S rRNA gene were examined for quality and trimmed to approximately the same size (~650 bp) using 4 PEAKS V1.7.2 for MAC OS X (www.nucleobytes.com) © 2006 Mek&Tosj.com and Clustal W. The aligned 16S rRNA gene sequences were compared against those available in the NCBI database by BLASTN (blast.ncbi.nlm.nih.gov) (Altschul et al., 1997). The sequences obtained during this study are deposited in NCBI GenBank under accession numbers KJ685218–KJ685361. For two isolates, the 16S rRNA sequences were available from previous work: P. fluorescens (strain AD21): DQ778036, Pedobacter sp. (strain V48): DQ778037 (De Boer et al., 2007a).

The bacterial interaction pairs that triggered or suppressed antimicrobial activity against the target organisms were visualized with Cytoscape 3.0.2 (www.cytoscape.org) for MAC OS X (Shannon et al., 2003). Interaction visualizations were performed with the following parameters: each phylogenetic class was visualized as a single node with different symbols for each phylogenetic class, the interactions between the phylogenetic classes (nodes) were visualized by links (edges) connecting each interacting phylogenetic class. Node colors were scaled to the number of interactions between the different phylogenetic classes (see Figure legends). For visualization, self-loops (interactions within the same phylogenetic class) and edges (interactions between phylogenetic classes) were bundled to single links between the respective phylogenetic classes (the darker the line the higher the number of interactions between the phylogenetic classes).

Statistical analyses on frequencies for induction and/or suppression of antimicrobial compound production between the different Gram-groups were performed with http://math.hws.edu/javamath/ryan/ChiSquare.html using online chi square tests. Results of the chi-square test are shown in Tables S5, S6.

16S rRNA gene sequence analysis revealed that the 146 bacterial isolates tested in this study belonged to 4 phyla covering 7 classes and 9 genera: Proteobacteria (14 alpha-Proteobacteria, 65 beta-Proteobacteria, 29 gamma-Proteobacteria), Bacteroidetes (19 Flavobacteria, 1 Sphingobacteria), Actinobacteria (11 Actinobacteria) and Firmicutes (7 Bacilli) (Table 1 and Table S1).

We developed a high-throughput assay to screen for production of antimicrobial compounds by interacting bacteria (Figure 1). In total 146 isolates were screened in monocultures and in 2798 random one-to-one interactions. For 17 isolates (11%), no activity against E. coli and S. aureus was detected not in monocultures nor in mixed cultures (Table S1 and Figure 2A). For 20 isolates (14%) antibacterial activity was observed in both monoculture and mixed cultures. For 48 isolates (33%), this was restricted to monocultures only and for 61 isolates (42%) antibacterial activity was only apparent during interactions (Figures 2A, 3). The number of isolates (110) involved in activity against the Gram-positive target strain S. aureus 533R4 was more than twice the number of isolates (45) with activity against the Gram-negative target strain E. coli WA321 (Table 1, Table S1). Despite the high number of bacterial isolates involved in antimicrobial activity in interactions, the frequency of interaction-mediated induction of antimicrobial activity was low ~6% (154 interactions out of 2798). This implies that interaction-mediated induction was only occurring in a limited number of combinations (Tables 2, 3). Most interactions (72%) did not have an effect on antimicrobial activity (induction or suppression) and about 22% of the interactions suppressed antimicrobial activity in isolates that revealed activity in monoculture (Figure 2B).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.