Antibiotic-producing symbionts dynamically transition between plant pathogenicity and insect-defensive mutualism

Pathogenic and mutualistic bacteria associated with eukaryotic hosts often lack distinctive genomic features, suggesting regular transitions between these lifestyles. Here we present evidence supporting a dynamic transition from plant pathogenicity to insect-defensive mutualism in symbiotic Burkholderia gladioli bacteria. In a group of herbivorous beetles, these symbionts protect the vulnerable egg stage against detrimental microbes. The production of a blend of antibiotics by B. gladioli, including toxoflavin, caryoynencin and two new antimicrobial compounds, the macrolide lagriene and the isothiocyanate sinapigladioside, likely mediate this defensive role. In addition to vertical transmission, these insect symbionts can be exchanged via the host plant and retain the ability to initiate systemic plant infection at the expense of the plant's fitness. Our findings provide a paradigm for the transition between pathogenic and mutualistic lifestyles and shed light on the evolution and chemical ecology of this defensive mutualism.

S ymbiosis is ubiquitous in nature and constitutes a major source of evolutionary innovation, playing a fundamental role in the origin and diversification of eukaryotic life on Earth 1 . Microbial symbionts influence virtually all aspects of eukaryote biology 2 , and their impact on host fitness ranges from detrimental to beneficial, occasionally shifting along this continuum 3 . Although such shifts have important implications for the ecological and evolutionary dynamics of symbiosis, observations of recent or dynamic transitions between parasitic and mutualistic lifestyles are scarce, and reports on their occurrence rely largely on phylogenetic evidence 4 . Furthermore, genomic analyses across bacterial groups reveal that general signatures distinguishing pathogenic and mutualistic microbes are often lacking [4][5][6] .
Here, we report on a transition between pathogenicity and mutualism in Burkholderia bacteria associated with a widespread group of herbivorous beetles, the Lagriinae (Coleoptera: Tenebrionidae). We show that these insect-defensive mutualists likely evolved from plant pathogenic bacteria, and that they can still infect a plant host, proposing an ecological context in which the lifestyle transition may have occurred. Furthermore, we elucidate symbiont-produced compounds that inhibit the growth of relevant microbial antagonists of the insect.

Results
Burkholderia as vertically transmitted symbionts in Lagriinae. Lagriinae beetles harbour extracellular bacteria in a pair of accessory glands connected to the female reproductive system 7 (Fig. 1a,d) that are transmitted vertically via the egg surface 7 (Fig. 1b,e). Shortly before hatching, some bacterial cells enter the egg and colonize invaginations of the cuticle located dorsally in the embryo that later close to form three compartments in the larva 7 (Fig. 1c,f). To identify the bacterial symbionts associated with the invasive South American soybean pest Lagria villosa 8 and to confirm their vertical transmission route, we characterized the bacterial community in the symbiont-bearing structures of field-collected adult females and eggs laid by these females, as well as in larvae of laboratory cultures using 454 sequencing of bacterial 16S ribosomal RNA (rRNA) amplicons and quantitative PCR. We identified a closely related cluster of bacteria from the genus Burkholderia as the most prevalent taxon in female accessory glands and eggs (65-86% and 30-71% of reads per individual gland or egg clutch, respectively) ( Supplementary  Fig. 1a,b), and we found the same bacteria in mean abundances of 1.39 Â 10 6 , 1.83 Â 10 7 and 1.59 Â 10 8 16S rRNA gene copies per egg, larva and adult female gland, respectively ( Supplementary  Fig. 1c). Longer 16S rRNA reads (1.1-1.3 kb) obtained by Sanger sequencing revealed that the symbionts are most similar to Burkholderia gladioli, a well-known plant pathogen, and that at least three highly similar strains coinfect L. villosa beetles (Supplementary Fig. 2 and Supplementary Table 1). As supported by findings in the congeneric beetle species Lagria hirta, the coexistence of multiple symbiotic B. gladioli strains within individual beetles might be a common feature in Lagriinae beetles (Supplementary Fig. 2 and Supplementary Table 1). One of the strains could be successfully isolated from L. villosa and cultured in vitro (B. gladioli Lv-StA). Additionally, Burkholderiaspecific fluorescence in situ hybridization (FISH) confirmed symbiont localization in the adult female reproductive glands, on the egg surface and in the unusual dorsal organs of larvae ( Fig. 1).
Egg antifungal protection by B. gladioli symbionts. The specialized localization of the symbionts in the larval and adult stage and the vertical transmission route suggested an important functional role of Burkholderia in the insect host. We therefore generated symbiont-free (aposymbiotic) beetles by egg-surface The Burkholderia symbionts were localized by FISH in a cross-section of an L. villosa larva (inset). FISH pictures show Burkholderia-specific staining in red (Burk16S_Cy3), general eubacterial staining in green (EUB338_Cy5), the overlap of these two in yellow and host cell nuclei in blue (DAPI). Scale bars, 20 mm (d) and 50 mm (e,f). sterilization to evaluate potential differences to their untreated symbiotic counterparts. Notably, aposymbiotic eggs suffered more frequently from fungal infestation, pointing to a protective role of the symbionts. To test this hypothesis, we isolated spores of the most frequently encountered fungal antagonist of L. villosa eggs under laboratory conditions, Purpureocillium lilacinum (formerly Paecilomyces lilacinus), that has been previously reported as an egg entomopathogen 9 and as a natural enemy of L. villosa adults and larvae 10 . Upon exposure to the fungal pathogen, surface-sterilized eggs experienced fungal growth significantly more often and at higher levels than control eggs (Fig. 2a,d and Supplementary Fig. 3). Importantly, reinfection of surface-sterilized eggs with Burkholderia symbionts from egg washes or cultured B. gladioli Lv-StA significantly reduced fungal infestation, confirming that the absence of the symbionts rather than the surface-sterilization procedure itself was responsible for increased susceptibility to fungal growth ( Fig. 2a and Supplementary Fig. 3). The symbionts' protective effect was further corroborated by the significantly higher probability of fungal growth on untreated eggs from aposymbiotic as compared to symbiotic mothers ( Supplementary Fig. 4). In addition to P. lilacinum, the symbionts also inhibited the growth of the fast-growing soil fungus Trichoderma harzianum and the entomopathogen Beauveria bassiana in vivo, revealing a generalized antifungal protection by the symbionts (Fig. 2b,c). Although eggs suffering from P. lilacinum infection hatched at similar rates as those without fungus ( Supplementary Fig. 5a), larvae hatching from infected eggs had significantly lower chances of surviving the first instars, demonstrating that fungal inhibition by the symbionts confers a benefit to the host ( Fig. 2e and Supplementary Fig. 5b). Furthermore, the impact of fungal growth on survival varied among the different treatments and was most pronounced for aposymbiotic individuals, suggesting that there may be costs of the symbiosis in the absence of fungal infection ( Supplementary Fig. 5b).
Antibiotic production by B. gladioli Lv-StA. Given that the cultured bacterial symbiont (B. gladioli Lv-StA) showed strong antifungal activity in vitro (Fig. 2f) and its application in vivo fully restored the naturally observed egg protection (Fig. 2a-c), we used this strain to investigate the chemical nature of the symbiont-conferred protection based on whole-genome sequencing, chemical analyses and bioassays. Bioinformatic mining revealed several secondary metabolite biosynthesis gene clusters including those coding for the previously described bioactive compounds toxoflavin 11 (Supplementary Table 2) and caryoynencin 12 (Supplementary Table 3), as well as an orphan gene cluster coding for a complex polyketide, homologous to the etnangien biosynthetic assembly line characterized in Sorangium cellulosum 13 (Fig. 3a and Supplementary Tables 4 and 5). High-performance liquid chromatography mass spectrometry (HPLC-MS)-based metabolic profiling of B. gladioli Lv-StA culture extracts confirmed the production of the azapteridine toxoflavin (1) and the polyyne caryoynencin (2), which represent known bioactive metabolites [14][15][16]20 , as well as a polyketide ARTICLE structurally related to etnangien, which we named lagriene (3) ( Fig. 3b and Supplementary Fig. 6). The structure of lagriene was elucidated using MS and nuclear magnetic resonance (NMR) ( Supplementary Figs 7-14). Additionally, an aromatic glycoside was identified in the extracts ( Supplementary Fig. 6b), and its structure was fully resolved by MS and NMR (4) ( Fig. 3b and Supplementary Figs 7,[15][16][17][18][19][20]. The latter compound, which we named sinapigladioside, contains an isothiocyanate moiety, a rare structural feature among bacterial metabolites 17 . Similar sulfurcontaining structures are typically found in metabolites of plants (mustard oils), mainly of the order Brassicales, where they serve as chemical defence against herbivores and pathogens 18,19 . The production of these four compounds by B. gladioli Lv-StA suggested them as candidates for the protective effect on the eggs. To evaluate the in vivo production of the identified metabolites, we analysed extracts from eggs that were surface sterilized, reinfected with B. gladioli Lv-StA and exposed (or not) to P. lilacinum spores. Using HPLC-MS, we confirmed the presence of toxoflavin, lagriene and sinapigladioside, implying that the symbionts also produce these antimicrobial agents in vivo, without requiring a trigger from the antagonistic fungus (Fig. 3c). Unsurprisingly, caryoynencin was not detected in the extracts, since this highly reactive compound is known to degrade rapidly 12,20 . To shed further light on the functional role of the bacterial metabolites, we assessed the activity of the compounds against a number of entomopathogenic microorganisms in vitro. Whereas toxoflavin and lagriene possess antibacterial activity (for example, against Brevibacillus laterosporus and Bacillus thuringiensis, respectively) (Supplementary Note 1), caryoynencin and sinapigladioside displayed antifungal activities against P. lilacinum (Supplementary Note 1 and Supplementary Fig. 21). Thus, the beetle symbionts provide a blend of chemical defence compounds against fungal pathogens and potential bacterial antagonists.
Ancestral plant-associated lifestyle of the symbionts. The finding that B. gladioli protects an insect was unexpected given that many B. gladioli strains are well-known plant pathogens 21 .
Hence, we set out to investigate (1) whether the beetle symbionts evolved from plant-pathogenic ancestors, and (2) whether they retained the ability to successfully infect host plants.
Characterization of the bacterial symbionts in Lagriinae beetles from Europe (Lagria hirta), Brazil (L. villosa), Japan and in vivo, that is, on L. villosa eggs reinfected with B. gladioli Lv-StA (bottom). Peak intensity represents relative abundance within each chromatogram, but is not comparable among the green, black and blue profiles. In the egg extracts, toxoflavin was detected in one of the three replicates, whereas lagriene and sinapigladioside were detected in all three replicates. Production of the compounds was independent of the exposure to P. lilacinum, and was not detected in the aposymbiotic controls in any of the replicates.  Table 1). Thus, considering also the morphological description of symbiont-bearing organs in numerous other Lagriinae species 7 , the association with B. gladioli is likely ancient and widespread in this beetle subfamily. A 16S rRNA-based phylogeny revealed that the symbionts are interspersed within the monophyletic clade of plant-pathogenic B. gladioli, supporting the symbionts' plant pathogenic ancestry (Fig. 4d). The lack of symbiont monophyly indicates that-despite a considerable degree of symbiont specificity and a characterized route for vertical transmission-Lagriinae beetles at least occasionally exchange B. gladioli strains with their environment.
Interaction of the B. gladioli symbionts with soybean plants.
Experimental exposure of soybean plants to female beetles and subsequent screening for Burkholderia indeed revealed the transfer of symbionts to the plant tissue (Fig. 4a). Furthermore, artificial infection of soybean with B. gladioli Lv-StA resulted in systemic infection (Supplementary Fig. 22) and reduced seed production compared with water-treated controls (Fig. 4b).
Concordantly, cotyledon assays revealed that soybean mounts a defence response against B. gladioli Lv-StA ( Fig. 4c and Supplementary Fig. 23), showing that the plant recognizes the beetle symbiont as a pathogen. It is conceivable that toxoflavin production by Lv-StA is involved in plant pathogenicity, as has been previously demonstrated for B. gladioli and B. glumae [22][23] . Thus, the beetle is capable of transmitting its symbiotic B. gladioli to soybean plants, where the bacteria can reproduce, spread systemically and ultimately impact plant fitness, confirming that the bacteria maintain the potential to interact pathogenically with a plant host, as well as mutualistically with the beetle.

Discussion
The defensive symbiosis with Burkholderia constitutes a potential key innovation in Lagriinae beetles for the protection of the vulnerable and immobile egg stage exposed to the soil environment. Considering that egg-surface contamination is a widespread route for symbiont transmission in insects 24 , this mechanism may have originally evolved for protection or at least may have been reinforced by the additional protective benefit. Symbiont-mediated egg defence is known for some marine crustaceans 25 , and immobile larval stages are symbiotically protected against pathogenic fungi in beewolves 26 and leaf-rolling weevils 27 , indicating that the protection of immature animals through mutualistic microbes may be a common phenomenon. In this context, the prolific production of a broad spectrum of secondary metabolites 28 In (a, b), the centre value of the boxplots represents the median, the boxes denote the interquartile range, and the whiskers represent minimum and maximum values. Statistically significant differences: **Po0.01.
armory. In addition, we provide evidence that secondary metabolite production is important for beetle protection and potentially also for plant pathogenicity. However, the symbiontmediated production of the four identified compounds and their fitness consequences for the beetle under field conditions remain to be assessed, and at present we cannot completely rule out the possibility that competitive exclusion or additional bioactive metabolites contribute to the symbionts' protective effect on beetle eggs. Our observation that a protective beetle symbiont evolved from plant pathogenic ancestors and retains the ability to infect the beetle's host plant provides a plausible scenario for the evolutionary switch from pathogenicity to a dual lifestyle. There are many examples of insect-vectored plant pathogens 30 , and some of these indirectly benefit their insect vector by altering plant physiology or suppressing plant defences 30 . However, the Lagriinae symbiosis is exceptional in that the symbionts are not only consistently associated with their insect host, but also provide a direct benefit to the insect that is independent of its plant pathogenic effects. Our findings provide evidence for a transition from a plant pathogenic to an insect mutualistic lifestyle, and describe an ecological setting in which it likely occurred. They also support that such shifts are not necessarily unidirectional, but can be dynamic on both ecological and evolutionary timescales. In addition, our findings highlight arthropod-associated microbes as promising sources for novel bioactive compounds.
Three-dimensional reconstruction of a beetle larva. An L. hirta larva was fixated in Bouin solution at 4°C, dehydrated in a graded ethanol series and in isopropanol and embedded using the Epoxy Embedding Medium kit (Sigma-Aldrich, Germany) following the manufacturer's instructions. Hardened epoxy blocks were cut in a rotation microtom (Mikrom HM355S, Thermo-Scientific, Germany) to sections 2 mm thick and stained with Toluidine blue-pyrimidine solution, humidified with xylol and covered with Entellan (Merck, Germany). Section images were acquired in an Axioimager Z1 Microscope (Carl Zeiss, Germany), and the reconstruction was carried out using Amira 5.4.1. software.
Symbiont cultivation for bioassays and genome sequencing. Live L. villosa female adults were placed at -20°C for 20 min and subsequently surface sterilized by rinsing in 70% ethanol. The paired glandular structures associated to the ovipositor were dissected in sterile phosphate-buffered saline (PBS), and one of these was stored at -80°C for nucleic acid extraction. The second one was homogenized in 100 ml of sterile PBS and diluted to a factor of 10 À 3 , 10 À 4 and 10 À 5 . Then, 100 ml of each dilution were plated on Nutrient Agar, R2A Agar (Carl Roth GmbH, Germany) and Actinomycete Isolation Agar (Sigma-Aldrich) and incubated at 30°C. After 3 days, colonies with distinct morphologies were selected, and part of their biomass was transferred into a lysis solution (67 mM Tris-HCl (pH 8.8), 16.6 mM (NH 4 ) 2 SO 4 , 5 mM b-mercaptoethanol, 6.7 mM MgCl 2 , 6.7 mM EDTA (pH 8.0) and 1.7 mM SDS) and kept at 90°C for 5 min. This suspension containing free DNA was used for a diagnostic PCR with primers specific to the 16S rRNA gene of Burkholderia, BKH1434Rw (3 0 -TGCGGTTAGRCTASCYA CT-5 0 ) 31 and Burk3fwd (3 0 -CGGCGAAAGCCGGAT-5 0 ; modified from ref. 34). Pure cultures of colonies corresponding to B. gladioli Lv-StA were kept as glycerol stocks until further use. Genomic DNA isolation was performed with the QIAGEN Genomic-tip 100/G kit (Qiagen, Hilden, Germany) following the manufacturer's instructions. Genome sequencing was carried out using SMRT (single molecule, real-time) technology provided by Eurofins Genomics, Germany. To identify candidate biosynthesis gene clusters, antiSmash 2.0 (refs 35,36) and the Artemis genome browser and annotation tools 37 were used.
Cultivation and extraction of B. gladioli Lv-StA for metabolic profiling. Bacteria were grown in either MGY liquid medium consisting of yeast extract (1.25 g l À 1 ) and M9 salts (50 Â , part A: 350 g l À 1 K 2 HPO 4 ; 100 g l À 1 KH 2 PO 4 ; part B: 29.4 g l À 1 tri-Na-citrate-dihydrate; 50 g l À 1 (NH 4 ) 2 SO 4 ; 5 g l À 1 MgSO 4 ) and glycerol (10 g l À 1 ) (for toxoflavin/lagriene production) or in PDB (Difco) (for caryoynencin/sinapigladioside production) at 30°C and 110 r.p.m. for 5 or 1 days, respectively. The cultures were extracted with ethyl acetate, dried with sodium sulfate and concentrated under reduced pressure. For LC-MS measurements the extracts were dissolved in 500 ml methanol. For lagriene production, 50 l of MGY medium were inoculated with a 1-day-old bacterial pre-culture (1.5 l in MGY) and incubated at 30°C for 26 h followed by incubation at 28°C for 76 h. For sinapigladioside isolation, 3 l of PDB was inoculated and incubated at 30°C and 110 r.p.m. for 24 h. The extraction was performed as described above.
Fungal inhibition on eggs and survival assays. A layer of vermiculite substrate was added to 96-well plates and moistened with sterile water. Filter paper discs were then added individually to each well, excluding outermost rows and columns to avoid heterogeneous humidity conditions. A total of 50 fungal (P. lilacinum, T. harzianum or B. bassiana) spores suspended in water were inoculated into each well (Supplementary Table 6).
For the assay with P. lilacinum, we initially used 12 different clutches laid by field-collected females. Only the six clutches that hatched were included in the analysis, that is, a total of 720 L. villosa eggs (120 eggs per clutch). Eggs from each clutch were divided into four groups of 30 eggs and randomly assigned to four different treatments. For the assays with T. harzianum and B. bassiana, 80 eggs from the same clutch were used (20 eggs per treatment), respectively. All eggs were placed individually and distributed randomly in relation to treatment in the 96-well plates containing the fungal spores. The first group remained untreated as a control. The three remaining groups were washed in PBS and then surface sterilized by submerging them for 5 min in 90% ethanol, followed by 30 s in 12% NaClO, and a final rinse with sterile water. From the three surface sterilized groups, one (reinfected culture) was reinfected with a PBS suspension (2.5 ml per egg) of symbiotic B. gladioli Lv-StA (isolated from L. villosa) previously grown in King B medium and adjusted to a concentration of 2 Â 10 6 cells per ml (to achieve a cell number comparable to naturally infected L. villosa eggs). The second (reinfected natural) was reinfected with the PBS suspension (2.5 ml per egg) recovered from the egg-washing step previous to sterilization that contained B. gladioli and possibly other microbes naturally present on the eggs. Then, 2.5 ml of PBS was added to each egg of the final group (Apo). Plates were stored in closed boxes at 25°C and monitored daily for visible growth of fungal mycelia on the egg surface. Corresponding treatments were not labelled during monitoring (blind assessment). Hatching rate and survival during the first larval instar and early days of the second instar were also assessed in the assay using P. lilacinum.
Metabolic profiling of B. gladioli Lv-StA on L. villosa eggs. Half of an L. villosa egg clutch was used for each replicate within 36 h after being laid. The eggs were surface sterilized as described in the previous section, and three groups of equal numbers of eggs were assigned each to an individual treatment. The first two groups were reinfected with 2 Â 10 6 B. gladioli Lv-StA cells per egg that were previously recovered by centrifuging a pure liquid culture and resuspending in sterile PBS. Eggs from one of these two groups were exposed to P. lilacinum spores (250 spores per Petri-dish). The third group remained aposymbiotic as a control. The eggs were kept in a Petri-dish with a layer of moist vermiculite and filter paper at 25°C. After 3 days, each group was extracted in methanol, and the crude extracts were analysed using liquid chromatography-high-resolution electrospray ionization-mass spectrometry (LC-HRESI-MS) as described below ('General analytical chemistry procedures' section). The experiment was carried out on three independent L. villosa egg clutches.
Nucleic acid extraction for amplification and sequencing. The accessory glands dissected from adult Lagriinae beetles as described above, whole larvae and eggs (previously subjected to chemical extraction in methanol) were used for nucleic acid isolation. Tissue samples were homogenized in liquid nitrogen and subjected to DNA extraction using the MasterPure complete DNA and RNA isolation Kit (Epicentre). Before protein precipitation, samples were incubated at 37°C with 4 ml lysozyme (100 mg ml À 1 ). The rest of the procedure was carried out following the manufacturer's instructions. Isolated nucleic acids were resuspended in Low TE buffer and stored at -20°C. The 16S rRNA gene fragment was amplified using general bacterial primers fD1 (5 0 -AGAGTTTGATCCTGGCTCAG-3 0 ) and rP2 (3 0 -ACGGCTACCTTGTTACGACTT-5 0 ) 38 and Burkholderia-specific primers BKH1434Rw (3 0 -TGCGGTTAGRCTASCYACT-5 0 ) 31 and Burk16S_1F (3 0 -GTTGGCCGATGGCTGATT-5 0 ). The PCR conditions were 3 min at 94°C, followed by 32 cycles (bacterial primers) or 42 cycles (Burkholderia primers) of 40 s at 94°C, 60 s at 65°C (bacterial primers) or 62°C (Burkholderia primers) and 60 s at 72°C, and a final extension step of 4 min at 72°C. Purified PCR products were either sequenced directly, or first cloned into E.coli StrataClone SoloPack Competent cells (Agilent Technologies, Frankfurt, Germany) using the TOPO TA Cloning Kit (Invitrogen, Darmstadt, Germany) for the cloning reaction. 80 mL of each transformation mixture were then plated on LB-ampicillin plates previously spread with 40 mL of 2% X-gal (Zymo Research, Freiburg, Germany) and incubated at 37°C overnight. White colonies were used for plasmid insert sequencing using the M13 primer pair from the Strata Clone kit. PCR products were sequenced bidirectionally on an ABI 3730xl capillary DNA sequencer (Applied Biosystems, Foster City, CA, USA). For symbiont quantification in L. villosa eggs, larvae and female accessory glands, primers Burk16S_1F (3 0 -GTTGGCCGATGGCTGATT-5 0 ) and Burk16S_1R (3 0 -AAGTGCTTTACAAC-CCGAAGG-5 0 ), which amplify a 172 bp region of Burkholderia 16S rRNA, were used for quantitative PCR in a RotorgeneQ cycler (Qiagen) following the protocol described for the Rotor-Gene SYBR Green PCR Kit. PCR conditions were as follows: 95°C for 10 min, followed by 45 cycles of 95°C for 10 s and 65°C for 30 s. A melting curve was subsequently performed with a temperature ramp from 60 to 99°C within 4.25 min.
Phylogenetic analyses. Burkholderia sequences obtained from accessory glands from females of the six different Lagriinae species were curated manually in Geneious 6.0.5 (ref. 39) and aligned using the SINA alignment software 40 . The phylogenetic reconstruction including a representative or the single available symbiont sequence for each investigated Lagriinae species, and Burkholderia references, was based on an approximately maximum likelihood algorithm in Microbial community analysis. DNA samples from 16 L. villosa egg clutches and single accessory glands from 5 adult females were used individually for bacterial community characterization. The 454 pyrosequencing and sequence processing and analyses were carried out as described previously 43 , with minor modifications. Briefly, sequences were obtained from MR DNA (Shallowater, TX, USA) by bacterial tag-encoded FLX amplicon pyroseqencing (bTEFAP) using 16S rRNA primers Gray28F (5 0 -GAGTTTGATCNTGGCTCA-3 0 ) and Gray519R (5 0 -GTNTTACNGCGGCKGCTG-3 0 ) 44 and subsequently analysed in Qiime 45 . After quality filtering, between 10,168 and 26,597 high-quality reads per sample were available for analysis. Sequences were clustered into operational taxonomic units (OTUs) using 97% similarity cutoffs, and one representative sequence per OTU was extracted for taxonomy assignment using the uclust consensus taxonomy assigner. For graphical representation, OTUs corresponding to the same genus were combined, and the percentage of reads in each genus relative to the total reads per sample was plotted.
Insect-mediated transmission of Burkholderia to soybean plants. To test for transmission of B. gladioli from the insect host to soybean plants (Glycine max, cv. Alligator, N.L. Chrestensen Erfurter Samen-und Pflanzenzucht GmbH, Germany), we confined 18 individual L. villosa adults to single leaves and later used quantitative PCR to assess the presence and abundance of live Burkholderia in the leaf tissue. In a first group of nine plants, we attached magnetic cages containing a single beetle to two independent leaves per plant. As a control, a second group of nine plants had an empty magnetic cage attached to one leaf. After 3 days, all beetles were removed, and plants were maintained at 21-23°C with a 16 h light regime for 12 more days until leaves were removed and stored at -80°C for nucleic acid extraction.
Stored leaves were weighed individually, ground in liquid nitrogen and homogenized. For each sample, a fraction of known weight was separated for RNA extraction and processed using the MasterPure complete DNA and RNA isolation Kit (Epicentre) following the manufacturer's instructions. Reverse transcription was carried out using the Quantitect Reverse Transcription Kit (Qiagen) with Burkholderia-specific primers Burk16S_1F (3 0 -GTTGGCCGATGGCTGATT-5 0 ) and Burk16S_1R (3 0 -AAGTGCTTTACAACCCGAAGG-5 0 ). The resulting complementary DNA (cDNA) was used for quantitative PCR as described above ('Nucleic acid extraction, amplification and sequencing' section). Beetles recovered from the experiment were dissected to confirm sex, revealing that 5 out of the original 18 individuals were males. Since adult males lack symbiotic B. gladioli, only the 13 replicates involving female beetles were included in the analysis. To test for statistically significant differences in B. gladioli titres between L. villosa-exposed and control plant leaves, Mann-Whitney U-test was carried out in SPSS 17.0.
Plant fitness effect upon B. gladioli infection. A total of 36 soybean plants (G. max, cv. 29-I, Semillas Panorama, Colombia) were grown for 28 days before treatment. A single leaflet of the first trifoliate leaf on each plant was wounded in a circular area (0.5 cm diameter) using a robotic device that mimics herbivory damage (MecWorm 46 ). Symbiotic B. gladioli Lv-StA previously isolated from L.villosa as described above were cultured overnight in King B liquid medium at 30°C and constant shaking (200 r.p.m.) and resuspended in sterile water at a concentration of 10 5 cells per ml. Half of the plants (N ¼ 18) were inoculated with 10 ml of the bacterial suspension on the wounded area, and the second half were treated with the same volume of sterile water as a control. Plants were kept at room temperature with a 16 h light regime for 38 days after inoculation. Total seed number was determined for all plants, and tissue samples were recovered from three regions on each plant: (1) the wounded area, (2) a different area on the same leaflet and (3) a leaflet of a younger leaf (not wounded). RNA was extracted from the recovered tissues, and quantitative PCR specific for the 16S rRNA gene of Burkholderia was carried out on the corresponding cDNA on a 167 bp fragment using primers Burk16S_StAG_F (5 0 -CTGAGGGCTAATATCCTTCGGGG-3 0 ) and Burk 3.1_R (5 0 -TRCCATACTCTAGCTTGC-3 0 ) as described for the horizontal transmission experiment.
Cotyledon assay. Symbiotic B. gladioli Lv-StA from L. villosa and Escherichia coli K-12 (Agilent Technologies, USA) were cultured overnight in King B liquid medium at 30°C and continuous shaking (200 r.p.m.). A fraction of the cultured B. gladioli cells were killed in 70% ethanol for 5 min. All cultures were centrifuged and resuspended in sterile water. The cotyledon bioassay procedure was based on a previously described protocol 47 with minor modifications. Briefly, 150 cotyledons from 5-day-old G. max seedlings were washed in distilled water, placed in 10% NaClO and submerged in distilled water. Groups of 10 cotyledons, using 3 replicates per treatment, were cut and placed on moist filter paper. Then, 50 ml of bacterial suspension containing 10 6 cells, sterile water (negative control) or b-glucan (200 mg mg À 1 ) (positive control elicitor from the cell wall of the phytopathogen Phytophtora sojae) were applied on the wounded area of each cotyledon. After a 24 h incubation period, only cotyledons that retained the liquid (eight cotyledons per treatment) were individually washed in Millipore water, and the content of mixed glyceolin isomers was determined by measuring absorbance at 285 nm.
General analytical chemistry procedures. Analytical HPLC was performed on a Shimadzu LC-10Avp series HPLC system consisting of an autosampler, highpressure pumps, column oven and photodiode array detector. HPLC conditions were as follows: C18 column (Eurospher 100-5, 250 Â 4.6 mm) and gradient elution (MeCN/0.1% (v/v) trifluoroacetic acid (TFA) 0.5/99.5 in 30 min to MeCN/0.1% (v/v) TFA 100/0, MeCN 100% for 10 min), flow rate 1 ml min À 1 . Preparative HPLC was performed on a Shimadzu LC-8a series HPLC system with photodiode array detector. LC-MS measurements were performed using an Exactive Orbitrap High Performance Benchtop LC-MS with an electrospray ion source and an Accela HPLC system (Thermo Fisher Scientific, Bremen). HPLC conditions were as follows: C18 column (Betasil C18 3 mm 150 Â 2.1 mm) and gradient elution (MeCN/0.1% (v/v) HCOOH (H 2 O) 5/95 for 1 min, going up to 98/2 in 15 min, then 98/2 for another 3 min; flow rate 0.2 ml min À 1 ). For tandem mass spectrometry measurements, a Q Exactive Orbitrap mass spectrometer with an electrospray ion source (Thermo Fisher Scientific) was used. NMR spectra were recorded on a Bruker AVANCE III 600 MHz instrument equipped with a Bruker cryo platform. Spectra were normalized to the residual solvent signals. The infrared spectra were recorded on a JASCO FT/IR-4100 type A.
Isolation of bioactive compounds. The crude extract was defatted with hexane and fractionated by size exclusion chromatography with Sephadex LH20 using MeOH as eluent. Final purification of compound 3 (lagriene) was achieved by preparative HPLC using a Phenomenex Synergi 4 mm Fusion-RP80A column (250 Â 21.2 mm) with a flow rate of 10 ml min À 1 and a gradient method Structure elucidation of lagriene and sinapigladioside. For compound 3 (lagriene), a molecular formula of C 44 H 74 O 11 was deduced from HRESI-MS measurements. 13 C and DEPT135 spectra revealed the presence of three quaternary, 22 methine and 13 methylene and 6 methyl carbon atoms. The proton and carbon NMR data indicated the structural relatedness to etnangien 48 . Analysis of the H,H-correlation spectroscopy (H,H-COSY) and the heteronuclear multiple bond correlation (HMBC) couplings identified the backbone of 3 ( Supplementary  Fig. 7). A coupling constant of J H,H ¼ 11 Hz for the protons H-26/H-27 disclosed the Z configuration of the respective double bond, whereas all other double bonds were found to be in E configuration (J H,H ¼ 15 Hz). HMBC coupling of H-16 and C-38 indicated the position of cyclization. For compound 4 (sinapigladioside), a molecular mass of m/z 468.1334 amu (M-H) À and a molecular formula of C 21 H 26 NO 9 S (calcd. 468.1334) was determined by HRESI-MS. The number of carbon atoms was corroborated by 13 C NMR analysis and the multiplicity was assigned by DEPT135 measurements. Proton and carbon NMR data revealed the presence of a para-substituted aromatic compound. HMBC couplings of H-8 and C-4 and H-7 and C-3, C-4 and C-6 disclosed the connection of the vinyl moiety. A coupling constant of J ¼ 13.9 Hz for H-7/H-8 pointed to the E configuration of the double bond. HRESI-MS data and a characteristic band in the infrared spectrum (ṽ 2104 cm À 1 ) suggested the presence of an isothiocyanate residue. This assumption was corroborated by an HMBC coupling of H-8 with the quaternary carbon C-9. Analysis of the H,H-COSY spectra revealed the spin systems of the sugar moieties and HMBC couplings of the anomeric protons H-1 0 to C-1 and H-1 00 to C-4 0 , respectively, established their connectivity. An HMBC coupling of the O-methyl proton H-3 00 -OMe with C-3 00 revealed the position of the methyl group. The nuclear Overhauser effect measurements finally elucidated the sugar moieties as rhamnose and O-methyl xylose, respectively. The NMR data are also in good agreement with literature data 49 .
Antimicrobial bioassays. The antifungal activity of the metabolites was studied by agar diffusion tests. A total of 50 ml of a solution of the respective compound (1 mg ml À 1 in methanol as a stock solution and respective dilutions) were filled in agar holes of 9 mm diameter (potato dextrose agar, seeded with a spore suspension). After incubation at 30°C for 24 h, the inhibition zone was measured. Antibacterial activity was tested as described before 50 . Compounds were tested against Bacillus thuringiensis, Brevibacillus laterosporus, P. lilacinum and (if enough material was available) Bacillus subtilis, Staphylococcus aureus, Escherichia coli, Pseudomonas aeruginosa, Staphylococcus aureus, Enterococcus faecalis, Mycobacterium vaccae, Sporobolomyces salmonicolor, Candida albicans, Penicillium notatum and Aspergillus fumigatus (see Supplementary Table 6 for information on the strains used). Given the high instability of caryoynencin 12,20 , antimicrobial assays for this compound were carried out with an authentic caryoynencin fraction derived from a previous study 12 .
Statistical analyses. Cox mixed effects models with a random intercept per clutch were used to analyse the effect of treatment on P. lilacinum growth on eggs, as well as the effect of treatment and fungal growth at the egg stage on the survival of early instar larvae, assuming a Gaussian distribution for the random effects. These statistical analyses were carried out in R 2.14.1. using the coxme package 51 . For T. harzianum and B. bassiana, the effect of treatment on fungal growth on the eggs was assessed using Mantel-Cox log rank tests in SPSS 17.0. Growth probability of all fungi and larval survival probability (P. lilacinum assay) were plotted based on Kaplan-Meier models using the rms package in R 2.14.1 (ref. 52). Generalized linear mixed models with a Poisson distribution were used to analyse fungal growth level and hatching rate using the lme4 package in R 2.14.1 (ref. 53), and overdispersion was calculated manually to test for the validity of the distribution family.
Burkholderia abundance in plants upon exposure to the insects, and seed output following Burkholderia infection, were analysed using the Mann-Whitney U-test in SPSS 17.0. Similar distribution for seed output data was confirmed using a two-sample Kolmogorov-Smirnov test. The quantitative PCR data on Burkholderia abundance in different plant locations upon infection were analysed in an analysis of variance and Tukey's post hoc tests after confirming the normal distribution of the data.
For the cotyledon assay, the absorbance data were not normally distributed, and thus a Kruskall-Wallis test with Dunn's post hoc test was carried out on the in SPSS 17.0. Two-sided tests were used in all cases.
Data availability. The 16S rDNA nucleotide sequences of Burkholderia symbionts have been deposited in the GenBank nucleotide database with accession numbers KT888026 to KT888030, and KU358660 to KU358661. Nucleotide sequences for the toxoflavin, caryoynencin and lagriene gene clusters have been deposited in the European nucleotide archive with accession numbers LT797825 to LT797838. Raw 454 sequencing data corresponding to bacterial 16S rDNA amplicons from L. villosa female accessory glands and egg clutches have been deposited in the NCBI Sequence Read Archive with accession codes SAMN04364624 to SAMN04364628, and SAMN04510296 to SAMN04510311, respectively (within BioProject PRJNA306502). The authors declare that all other data supporting the findings of this study are available within the paper and its Supplementary Information files, or from the corresponding authors on request.