P. repens infructescences (Figure 1) were collected at Pringle Bay and at the J.S. Marais Botanical gardens near Stellenbosch, and P. neriifolia infructescences from Pringle Bay and Franschhoek, in the Western Cape Province of South Africa. The infructescences were aseptically opened and parts from the individual flowers, outer involucral bracts, inner involucral bracts, and involucral receptacle were removed. Tissue samples from the different floral parts were suspended in a solution of 6% yeast extract and 0.05% SDS (Sodium Dodecyl Sulphate) and incubated at 40°C for 20 min (Hayakawa and Nonomura, 1989) followed by dilution to 10⁻⁶ to remove the antimicrobial detergent effect of SDS (Hayakawa and Nonomura, 1989). These suspensions were then plated onto three types of artificial media: YEME as described by Cafaro and Currie (2005), glycerol-asparagine agar (Pridham and Lyons, 1961) and starch casein agar (Kuster and Williams, 1964).

Plates were incubated at 28°C for approximately 14 days, whilst being continuously monitored. Colonies provisionally identified as actinomycetes based on morphology were obtained in pure culture by repeated isolation on YEME (Cafaro and Currie, 2005). Isolates were then grouped based on similar colony morphology. Representatives were selected from each group for further study. Care was taken to include isolates from each host species and location sampled.

Genomic DNA was isolated by adding mycelium from each single pure colony into Prepman Ultra™ solution (Applied Biosystems). For identification of isolates, the 16S ribosomal RNA gene region was selected for sequencing. For further phylogenetic placement of isolates, the trpB, rpoB, and gyrB gene regions were amplified. The 16S rRNA gene region was amplified using primers pA and pH designed by Edwards et al. (1989). The methods from Guo et al. (2008) were used for the trpB and rpoB PCR reactions and those from Rong et al. (2009) for the gyrB gene. All reactions were performed using SuperTherm taq polymerase (Southern Cross Biotechnology, Cape Town, South Africa) on an Applied Biosystems Veriti Thermal Cyler. The 16S rRNA (Edwards et al., 1989), gyrB (Rong et al., 2009), rpoB and trpB (Guo et al., 2008) amplicons were sequenced using the methods previously described. Purified PCR products were sequenced using the ABI BigDye system (Applied Biosystems) and purified using sodium acetate precipitation. Sequences were analyzed on an ABI 3130 sequence analyzer (Applied Biosystems).

Isolates obtained were initially identified by comparing their 16S rRNA sequences to all type strain sequences available in the RDP database using the Seqmatch program hosted by the Ribosomal Database Project II (RDP-II) (Maidak et al., 2001). The closest matching sequences of validly described species were downloaded and included in a single data set with our sequences. Mycobacterium tuberculosis was used as an outgroup.

The sequences for the 16S rRNA, gyrB, rpoB, and trpB (Supplementary Table S1) were concatenated along with those from selected reference strains using SequenceMatrix software (Vaidya et al., 2010). Sequence ends were trimmed using Bioedit (Hall, 1999) and the sequences aligned using MAFFT (Katoh et al., 2005). The best-fitting nucleotide substitution model was determined using JModeltest version 0.1 (Posada, 2008) and selected using the Akaike Information Criteria (Posada and Buckley, 2004). A maximum-likelihood phylogeny was produced with PhyML version 3.0 (Guindon et al., 2009) using the TIM3 nucleotide substitution model for 16S rRNA and the TVM+I+G nucleotide substitution model for the concatenated data set (Guindon et al., 2009; Darriba et al., 2012). Tree confidence was measured using a bootstrap search with 1000 replicates (Felsenstein, 1985). The tree was visualized using Mega 4 (Tamura et al., 2007).

Antifungal bioassays were performed using the method described by Cafaro and Currie (2005). Bacterial isolates used in bioassays were randomly selected, with at least two isolates chosen from each group (Table 1). The fungal isolates selected were representative of fungi occurring in infructescences of Protea spp. and ubiquitous saprophytic fungi. Knoxdaviesia proteae and Sporothrix splendens are the most common fungi in the infructescences of P. repens, and S. phasma is common in P. neriifolia (Roets et al., 2009). Fusarium and Penicillium spp. are ubiquitous saprophytic fungi and have occasionally been reported from the infructescences of Protea spp. (Lee et al., 2005; Visagie et al., 2014). Candida albicans is sensitive to many antifungal compounds and is a common isolate used in antifungal assays which could serve as a control for antifungal activities observed. All fungi were retrieved from the CMW culture collections (Forestry and Agriculture Biotechnology Institute, University of Pretoria, South Africa; Table 2), except for C. albicans (retrieved from the ATCC) and the Penicillium sp., which was isolated from a P. repens infructescence (data not shown). The clear zones between fungal growth and putative antibiotic producing bacterial isolate were measured. This is referred to as the zone of inhibition and is used as an indication of the potency of antibiotic and the level of susceptibility of the test organism (Cafaro and Currie, 2005).

All the Streptomyces strains were cultivated in 200 mL YEME (4 g yeast extract, 10 g malt extract, and 4 g glucose in 1 L distilled water) medium in 500 mL Erlenmeyer flasks. The profiles of secondary metabolites of the bacterial strains were analyzed by LC/MS every 2 days. When the cultures were monitored, 10 mL aliquot of each culture was extracted with 10 mL of ethyl acetate (EtOAc). The extracts were concentrated in vacuo and resuspended in 1 mL of methanol (MeOH). After filtering the methanol-soluble extracts, 10 μL from each concentrated extract in MeOH was injected to LC/MS (Agilent 1200 Series HPLC coupled with Agilent 6130 single quad mass spectrometer). Gradient solvent conditions were consistently used (0~20 min: 10~100% CH₃CN in H₂O, 20~25 min: 100% CH₃CN/flow rate: 0.7 mL/min, 0.1% formic acid in CH₃CN and H₂O). A diode array UV detector scanned from 200 to 700 nm. The mass spectrometric analyses were performed on both positive and negative ion modes. The acquired LC/MS profiles were analyzed with Agilent Chemstation 4.3 software.

The UV spectra of the major common secondary metabolites detected in the LC/MS chemical profiling were compared with the house-built UV spectral library interlinked with Agilent Chemstation 4.3 software. For further analysis, the major secondary metabolite of Streptomyces Group 1 was produced by large-scale (18 L) cultivation of a representative strain (PrRe2I22). The whole culture was extracted with 27 L ethyl acetate to yield 5 g of dry material. The extract was fractionated by a reversed-phase (C₁₈) chromatographic column (20, 40, 60, 80, and 100% MeOH in water). The major metabolite eluted in the 60 and 80% MeOH fractions. These fractions were subjected to further purification by HPLC (Kromasil 100-5-C₁₈ 250 × 10 mm, flow rate 2 mL/min, UV 360 nm detection, with 60% aqueous MeOH, 0.1% formic acid) to isolate the major compound. The major compound eluted at 35 min to yield 15 mg as yellow soild. The structure of the major metabolite of the strain PrRe2I22 was identified by NMR spectroscopic analysis including ¹H, ¹³C, COSY (correlation spectroscopy), HSQC (heteronuclear single quantum correlation), and HMBC (heteronuclear multiple bond correlation) experiments as well as literature comparison (Noguchi et al., 1988). The major metabolite of Group 2 was compared with that of Group 1 because their spectroscopic and chromatographic properties were identical. The major compound produced by Group 3 was identified by the UV and MS data acquired by LC/MS analysis and ¹H and ¹³C NNR spectra.

Colonies of the fungal strains were inoculated onto the center of 90 mm petri dishes containing YEME agar and incubated at 28°C. After 4 days, four 6 mm diameter sterile paper disks were placed on each inoculated agar petri dish. Sterile paper disks were imbued with three concentrations (10, 20, 50 μL) of fungichromin (10 mg/mL) dissolved in DMSO, and for the negative control 10 μL DMSO was used. Every petri dish was monitored for 28 days.

An infructescence specimen was dissected into the involucral bracts and florets. Then each part was soaked in MeOH and organic compounds were extracted with MeOH for 3 h. The MeOH extracts were analyzed by LC/MS using the following gradient conditions; 0~20 min: 10~100% CH₃CN in H₂O, 20~25 min: 100% CH₃CN/flow rate: 0.7 mL/min, 0.1% formic acid in CH₃CN and H₂O. A diode array UV detector scanned from 200 to 700 nm. The mass spectrometric analyses were performed on both positive and negative ion modes. Matrix-assisted laser desorption/ionization (MALDI) MS was also performed against pure fungichromin and the infructescence specimen. For MALDI MS of pure fungichromin, 0.2 mg/mL fungichromin in MeOH (0.5 μL) was first spotted and dried on the 384-well MALDI target plate (ASTA Inc., Suwon, Korea) followed by 0.5 μL of the matrix solution. The matrix solution was 10 mg/mL 2, 5-dihydroxybenzoic acid (DHB) in acetonitrile/water (50:50 v/v). For direct analysis of infructescence specimen, dissected florets were first attached onto the MALDI target plate by using a conductive double-sided tape and gently dried under moderate vacuum (~50 Torr). After drying, the DHB matrix solution was sprayed on a floret surface by using an airbrush. MALDI mass spectra were recorded in the positive reflectron mode by using an ABI 4800 Plus MALDI-TOF/TOF analyzer (Applied Biosystems, Foster City, CA, USA).

A total of 83 putative actinomycetes were obtained. All of these isolates could be classified in one of three morphological groups. Isolates residing in Streptomyces sp. Group 1 had soft black colonies, produced gray spores and a yellow substrate pigment was observed in surrounding growth medium. Members belonging to Streptomyces sp. Group 2 produced yellow translucent colonies with gray spores. A red substrate pigment was produced by all isolates in this group. Streptomyces sp. Group 3 had white colonies and produced light gray spores while a red substrate pigment was observed. From the total number of isolates, 20 representing the different morphological groups were selected (Table 1) for further detailed identification.

Isolates of Streptomyces sp. Group 1 were from P. repens in Pringle Bay and Stellenbosch, and P. neriifolia in Franschhoek. Streptomyces sp. Group 2 isolates were from both hosts at all locations sampled, while Streptomyces sp. Group 3 isolates were from P. neriifolia in Pringle Bay and P. repens in Pringle Bay and Stellenbosch (Table 1).

The aligned 16S rRNA data set consisting of sequences obtained in the present study and those of type strains from RDP database consisted of approximately 650 characters from the species-specific α-region of the 16S rRNA gene sequence (Kataoka et al., 1997). Maximum likelihood analysis confirmed that isolates having a similar morphology also grouped together based on 16S rRNA sequences (Figure 2). Furthermore, all the isolates collected in this study grouped among species of Streptomyces.

According to the 16S rRNA maximum likelihood phylogeny, isolates belonging to Streptomyces sp. Group 1 formed a well-supported, distinct clade with the type strains of S. griseofuscus, S. murinus, S. phaegriseochromogenes, S. costaricanus, and an undescribed Streptomyces sp. NRRL 30562 from snakevine in Australia (Castillo et al., 2002, 2006, Figure 2). The ML phylogenetic tree based on the four combined genes was able to separate our isolates in this group into two clades; one which included isolates from P. repens (PrRe2I4, PrRe2I22, PrRe3I4) with 90% bootstrap support, while PrNe0I20 from P. neriifolia formed a sister clade and grouped with the type strain of S. murinus with 77% bootstrap support (Figure 3). These two clades formed a larger group which was separated (93% bootstrap support) from the S. costaricanus, S. phaeogriseochromogenes, and S. griseofuscus (Figure 3). Isolates from this group could belong to S. murinus or the proposed species S. padanus (Xiong et al., 2012).

In the 16S rRNA phylogenetic tree, isolates in Streptomyces sp. Group 2 formed its own clade which did not include any type strain sequences (Figure 2). The closest type strains were those from S. misionensis, S. levis, and S. phaeoluteichromatogenes (Figure 2). In the MLSA phylogenetic tree, these isolates also clustered separately with considerable distance to all type strains and with 100% bootstrap support and thus most likely represents an undescribed species. Interestingly, this clade formed a subgroup containing isolates from both P. repens and P. neriifolia.

Members of Streptomyces sp. Group 3 formed a distinct cluster with 93% bootstrap support with an unidentified strain (MG3) from soil in Germany (Hoster et al., 2005). According to our MLSA, these isolates remained in a distinct cluster, with 100% bootstrap support and were separated from the type strains of S. atratus and S. sanglieri (Figure 3).

Seven bacterial isolates were included in the assays and these represented the three different groups identified from Protea infructescences (Table 2, Figure 4). Isolates residing in Streptomyces Group 1 generally had strong inhibitory activities against all tested fungi and the antifungal activity against Sporothrix spp. and K. proteae was generally higher than against saprophytic test fungi (Table 2, Figure 4).

Isolates belonging to Streptomyces Group 2 was able to inhibit the growth of all tested fungi, although this was generally limited and much lower than observed for members of Group 1 and only limited inhibition of the two saprophyte test fungi was recorded (Table 2, Figure 4). Streptomyces sp. Group 3 had noticeable inhibition against Penicillium and Fusarium isolates, similar to the inhibition observed for Group 1. Interestingly, isolates from this group had lower levels of inhibition against the Sporothrix spp. tested fungi, while K. proteae was generally inhibited to levels similar to Penicillium and Fusarium. The isolates from Group 3 also had much lower inhibitory activity against C. albicans.

To identify secondary metabolites from the Streptomyces spp. strains responsible for the growth inhibition of fungi, we cultivated Groups 1 to 3 Streptomyces spp. strains in liquid culture medium. Based on the LC/MS (liquid chromatography and mass spectrometry) chemical profiling of the EtOAc extracts of the cultures, a common major secondary metabolite in Group 1 and 2 strains was distinctively detected. This compound displayed a typical UV spectral feature (λ_max at 325, 340, and 362 nm), indicating the existence of five conjugated double bonds, and the low resolution molecular ion [M+Na]⁺ at m/z 693. Comparison of the UV spectrum of the compound with our house-built UV library interlinked with the LC/MS software clearly indicated that this metabolite was similar to pentamycin, which bears a pentaene moiety. For definite structure identification, a large culture (18 L) of the Streptomyces sp. strain (PrRe2I22) was prepared. Subsequently, we analyzed ¹H (Supplementary Figure S1), ¹³C (Supplementary Figure S2), and two-dimensional NMR spectra of the compound including ¹H-¹H correlation (COSY; Supplementary Figure S3), heteronuclear single quantum coherence (HSQC; Supplementary Figure S4), and heteronuclear multiple bond correlation (HMBC; Supplementary Figure S5). The spectroscopic analysis and the comparison of the spectroscopic data with the literature (Noguchi et al., 1988) unequivocally identified the compound of Groups 1 and 2 as fungichromin, which is also called pentamycin, a polyene polyketide secondary metabolite as we expected (Figure 5). Therefore, the major secondary metabolite of Groups 1 and 2 Streptomyces strains was elucidated as fungichromin. We confirmed that fungichromin is responsible for the antifungal activity observed using an antifungal assay (Figure 5). Petri dishes were observed daily for 28 days (Figure 5). Fungichromin inhibited the growth of the fungi in a dose-dependent manner (Figure 5).

A common major secondary metabolite in Group 3 strains was also detected. This compound showed a distinct UV spectral feature (λ_max at 220, 260, and 350 nm) and the low resolution molecular ion [M+H]⁺ at m/z 276. Compared with our house-built UV library, the molecular ion detected in MS, and ¹H and ¹³C NMR spectra (Supplementary Figures S6, S7), this metabolite was identified as actiphenol (Highet and Prelog, 1959; Sun et al., 2014), which belongs to the cycloheximide class (Figure 6).

The LC/MS analysis of the MeOH extraction of the infructescence specimen clearly detected fungichromin in the involucral bracts and florets. The exact ion [M+Na]⁺ at m/z 693 was observed at the identical retention time to that of pure fungichromin (Figure 7). We also attempted to detect fungichromin directly from the infructescence specimen by MALDI MS. Figure 8 shows MALDI mass spectra of pure fungichromin and the floret part of the infructescence specimen. For the pure fungichromin, sodium and potassium adduct ions of fungichromin ([M + Na]⁺ and [M + K]⁺) were clearly detected at m/z 693.50 and 709.48, respectively (Figure 8A). In addition to these ions, another intense signal was observed at m/z 1363.50 and this signal was found to be sodium adduct ions of fungichromin dimer ([2M + Na]⁺) by MALDI tandem MS (MS/MS). After confirming that fungichromin is readily detected by MALDI MS, we performed direct analysis of the infructescence specimen by MALDI MS. Figure 8B shows a chemical fingerprint obtained directly from this specimen. From the floret part of the infructescence specimen, both sodium adduct ions of fungichromin and its dimer were detected by MALDI MS. Although the signal corresponding to [M + Na]⁺ at m/z 693.50 was crowded by other metabolite ion signals, the signal corresponding to [2M + Na]⁺ at m/z 1363.50 was distinctly visible (Figure 8B).

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.