Secreted TAL effectors protect symbiotic bacteria from entrapment within fungal hyphae

The association of the agriculturally significant phytopathogenic fungus Rhizopus microsporus with the bacterial endosymbiont Burkholderia rhizoxinica is a remarkable example of bacteria controlling host physiology and reproduction. Here, we show that a group of transcription activator-like effectors (TALEs) called Burkholderia TALE-like proteins (BATs) from B. rhizoxinica are essential for the establishment of the symbiosis. Mutants lacking BAT proteins are unable to induce host sporulation. Utilising novel microfluidic devices in combination with fluorescence microscopy we observed the accumulation of BAT-deficient mutants in specific fungal side-hyphae with accompanying increased fungal re-infection. High-resolution live imaging revealed septa biogenesis at the base of infected hyphae leading to compartmental trapping of BATdeficient endobacteria. Trapped endosymbionts showed reduced intracellular survival, suggesting a protective response from the fungal host against bacteria lacking specific effectors. These findings underscore the involvement of BAT proteins in maintaining a balance between mutualism and antagonism in bacterial-fungal interactions and provide deeper insights into the dynamic interactions between bacteria and eukaryotes.

example, amplification of a genomic region in a normally beneficial Wolbachia symbiont leads to over-proliferation at the hosts' expense 2 . Various bacteria control their eukaryotic hosts by translocating protein effector molecules into host cells. Therefore, they may use dedicated protein secretion systems (Type 1 to 7SS, Fig. 1a) 3,4 . For example, the type 3 secretion system (T3SS)associated Xanthomonas AvrBs3 effector family (termed transcription activator-like effectors, TALEs) are highly orthologous proteins from bacterial plant pathogens that aid colonisation by mimicking plant transcription factors 5 . In humans, the gut pathogen Helicobacter pylori directly injects virulence factors into host cells. These T4SS-secreted proteins impair intracellular signalling systems, leading to morphological changes and increased inflammation in the host 6 .
An intriguing case of bacteria controlling host physiology and reproduction is the endosymbiosis between the plant-pathogenic zygomycete Rhizopus microsporus and its toxin-producing bacterial endosymbiont Burkholderia rhizoxinica 7,8 (which was recently suggested to be placed in the novel genus Mycetohabitans 9 ). In this agriculturally relevant symbiosis 10 , host reproduction through spores relies exclusively on the presence of endobacteria (Fig. 1b). If R. microsporus is cured of its endosymbionts, the host is unable to reproduce vegetatively 8 . This strict control of sporulation ensures that the symbiosis persists over time, because the endosymbionts are translocated into the fungal spores during host reproduction. Multiple investigations have shed light on various symbiotic factors. For example, bacteria can invade the fungal host by secreting effector proteins via the T2SS 11 , while a linear lipopeptide aids the reinfection process by reducing surface tension 12 . Once inside the fungal hyphae, B. rhizoxinica lives in "stealth mode" due to the presence of a specialised bacterial lipopeptide O-antigen 13 . In addition, a functional T3SS is required for the formation of a stable symbiosis 14 .
While these findings imply that secreted effector molecules produced by bacterial endosymbionts control fungal physiology and development, there is a considerable lack of knowledge about the molecular basis of this interaction and the types of effectors involved.
Here we show that TALEs produced by the symbionts play a key role in controlling fungal reproduction in this phytopathogenic bacterial-fungal interaction (BFI). Furthermore, we present the first real-time snapshot of septa biogenesis in a zygomycete fungus leading to hyphal trapping of endosymbionts incapable of secreting TALEs.

Investigation of a T3SS-associated effector
To identify genes coding for T3 effector proteins we sequenced the genomes of seven endofungal Burkholderia species (Supplementary  Table 1). These seven strains form the B. rhizoxinica complex together with B. rhizoxinica HKI-454, whose genome sequence has been published previously (Supplementary Table 2) 15, 16 . Within each draft genome sequence, we found a gene encoding for a potential T3SS effector, with an alanine-tryptophan-arginine triad (AWR) peptide (Fig. 1c). The awr gene family, present in a number of bacterial pathogens including phytopathogenic Burkholderia and Ralstonia strains, collectively contributes to bacterial virulence 17,18 .
For example, deletion of all five awr genes in Ralstonia solanacearum severely impairs its capacity to multiply in the natural host plant 17 .
To confirm the identified sequences as genuine awr orthologues, a phylogeny was inferred from alignment of the predicted endofungal Burkholderia AWR peptide sequences (Fig.  1d). These sequences were aligned with homologous AWR sequences from Gram-negative plant and animal pathogens and other Burkholderia symbionts ( Supplementary Fig. 1) 17 . The tree was rooted in the Xanthomonas (γ-proteobacteria) sequences, which are more distantly related to the Ralstonia and Burkholderia (βproteobacteria) sequences. According to the inferred phylogeny, the nearest relative of the endofungal AWR proteins was found in the mammalian pathogen Burkholderia pseudomallei (GenBank acc. no. ZP02502372) 19 . In addition, all eight endofungal Burkholderia AWR peptide sequences contained monopartite nuclear localisation sequences (NLS) at the C-terminal end of the protein (Fig. 1e and  Supplementary Table 3). These sequences are consistent with the previously formulated NLS consensus sequence K(K/R)X(K/R) To test whether endobacterial AWRs are involved in fungal reproduction, we generated targeted deletion mutants (Δawr::Kan r , Supplementary Fig. 2) using a double-crossover strategy 21 . AWRdeficient mutants were examined for their reinfection ability, using a previously described sporulation bioassay 21 . When wild-type B.

Fig. 1 Characterisation and functional analysis of an effector protein associated with the type 3 secretion system (T3SS) from endofungal
rhizoxinica and the endosymbiont-free (cured) fungal host are cocultured, mature sporangiophores can be seen after four days (Fig.  1f). The appearance of mature sporangiophores indicates the successful establishment of the symbiosis. Similar numbers of spores were observed upon co-cultivation with B. rhizoxinica Δawr::Kan r leading to complete restoration of the wild-type phenotype ( Fig. 1g and Supplementary Table 2). In addition, the intracellular localisation of the Δawr::Kan r mutant strain was comparable to wild-type B. rhizoxinica (Fig. 1f).

Wide distribution of TALEs in endofungal symbionts
Since we were unable to observe an effect of AWR on fungal reproduction, we continued to search for putative T3 effectors by investigating the genomes of eight endofungal Burkholderia species associated with R. microsporus (Supplementary Table 1) using T3SS prediction tools (see Methods for details). Three putative effectors, encoded by genes with the locus tags RBRH_01844 (2,316 bp), RBRH_01777 (936 bp), and RBRH_01776 (2,994 bp) were previously identified in B. rhizoxinica HKI-454 (Fig. 2a).

Fig. 3 Phenotypic observations of Rhizopus microsporus co-cultivated with
Burkholderia rhizoxinica wild-type or mutant strains in bacterial-fungal interaction (BFI) devices. a, Photograph and simplified illustration showing the microfluidic device in a glass petri dish. The BFI device is made of a patterned poly(dimethylsiloxane) layer bonded to glass-bottomed dish to form microchannels. The microchannels were filled with potato dextrose broth. Scale bar: 5 mm. A simplified two-dimensional representation of the design showing the narrow entry points into the microchannels which limits the number of hyphae that can enter the device 24  The endofungal bat sequences identified in the genomes were confirmed as genuine TALE orthologues using: (i) PCR amplification and Sanger sequencing (Supplementary Table 5) and; (ii) phylogenetic analyses of predicted protein sequences. BAT1 was found in each of the five Burkholderia branches, while BAT2 and BAT3 were present in species belonging to the Eurasian and Pacific branch (Fig. 2d,e). It is conceivable that missing bat sequences evaded detection due to their highly repetitive nature. Using phylogenetic analysis, R. solanacearum TALEs (RipTALs) were identified as the closest relatives of endofungal BAT proteins ( Fig. 2e and Supplementary Fig. 3).
TALEs usually harbour a T3S signal, a NLS, and an activation domain, all of which are required for the transcriptional activation of host plant genes 22 . We searched for NLS motifs within endofungal BAT peptide sequences using the NucPred and cNLS prediction software (see Methods for details). A fraction of BAT1 and BAT2 proteins were predicted to localise to the nucleus (52%) while the remaining fraction localises to the cytoplasm (48%). Localisation of BAT3 could not be predicted (Supplementary Table 7). Although a transcriptional activation domain (AD) seems to be absent in BAT proteins, a BAT1 fusion protein containing a viral NLS and AD was shown to transcriptionally activate a promoter with its target sequence in both human and plant cells 23 . However, their native function in B. rhizoxinica, or in the interaction with R. microsporus, is still unknown.

Impact of BAT mutants on fungal sporulation
In order to identify the role of BAT proteins in the Burkholderia -Rhizopus symbiosis, we performed targeted gene deletion using a double-crossover strategy 21 . Although B. rhizoxinica can be maintained in axenic cultures under laboratory conditions, genetic manipulations are challenging. The long doubling time and the aggregation of cells 32 significantly impair the selection process.
Next, the reinfection potential of the knock-out strains was evaluated using a sporulation assay. The sporulation efficiency was significantly (p<0.05) reduced in all B. rhizoxinica BAT mutant strains as only a limited number of mature sporangia were formed (Fig. 2f  and Supplementary Table 4). In contrast, the wild-type B. rhizoxinica strain readily triggered sporulation after three days of co-incubation. Fluorescence microscopy was used to monitor bacteria (stained with SYTO9) inside the fungal hyphae (counter-stained with calcofluor white) following re-infection. BAT mutants reached an extremely dense cell density within the host cytosol, while wild-type bacteria spread equally among the fungal mycelium (Fig. 2f).

BAT-dependent changes in fungal morphology
To investigate the accumulation of BAT mutant cells in the fungal hyphae, we utilised a tailor-made microfluidic platform that allows live monitoring of BFIs over time using fluorescence microscopy ( Fig.  3a) 24 . First, cured R. microsporus was grown in BFI devices filled with PDB (Fig. 3b). After two days of incubation, the fungus was coincubated with SYTO9-stained wild-type B. rhizoxinica or BAT mutant strains (Δbat1::Apra r , Δbat2::Kan r , and Δbat3::Kan r ). As a control, T3SS-deficient mutants (ΔsctC::Kan r and ΔsctT::Kan r ) 21 or a rhizoxindeficient mutant (ΔrhiG::Kan r ) were also co-incubated. We observed four morphologically distinct types of hyphae ( Fig. 3c and Supplementary Fig. 5), which were classified as follows: (1) vegetative side hyphae; (2) vegetative main hyphae; (3) vegetative empty side hyphae; and (4) abortive sporangiophore. Vegetative main hyphae (morphotype 2) are characterised by their relatively large diameter (app. 10-15 µm). They appear to be tightly packed with organelles (e.g. nuclei, vacuoles, mitochondria, etc.), which are transported through the hyphae at relatively high speed (Video 1). This hyphal classification has been described for members of the Basidiomycetes and Ascomycetes, but rarely in Zygomycetes 33,34 .
We also observed smaller side hyphae, which were either filled with cellular components (morphotype 1) or were completely empty (morphotype 3). Morphotype 4 showed strong green fluorescence indicative of high bacterial cell numbers ( Fig. 3c and Supplementary  Fig. 5). Since B. rhizoxinica accumulates in the hyphal tip prior to the formation of sporangia 8 , this morphotype was termed abortive sporangiophore. Next, we calculated the mycelium area (as a percentage) of all four hyphal morphotypes over time (0, 24, and 48 hours post infection) using brightfield images (Fig. 3d). We monitored changes in the fungal morphological composition by: (i) comparing individual morphotypes at specific time points across B. rhizoxinica wild-type and mutant strains and; (ii) by comparing morphotypes of individual B. rhizoxinica strains between time points. In general, we observed no significant changes in any of the B. rhizoxinica strains, with two exceptions (Supplementary Table 8a,b). After 48 hours, morphotype 1 significantly decreased (p<0.05) and morphotype 2 significantly increased (p<0.05) in fungi co-cultured with BAT1 mutants, while the reverse effect was observed for fungi co-cultured with BAT3 mutants ( Fig. 3d and Supplementary Table 8c). Interestingly, fungal reinfection with wild-type B. rhizoxinica and BAT-deficient mutants resulted in a noticeable increase in morphotype 4 after 48 hours. Although this higher percentage of morphotype 4 was not significant, this effect was virtually absent in the control strains (Fig. 3d).

Monitoring fungal re-infection in real-time
The reinfection process was investigated in more detail by calculating the number of endohyphal bacteria from the integrated density at 485/498 nm (Fig. 4a). One day after infection, BAT1 and BAT3 mutants showed a 30% increase in fungal reinfection, compared to the positive control (sterile R. microsporus ATCC62417/S re-infected with wild-type B. rhizoxinica), while 25% less BAT2 mutant cells were detected inside the fungal hyphae. In a control experiment, endobacteria were virtually absent when T3SS defective mutants were co-cultured with the cured fungus ( Supplementary Fig. 6). In addition, rhizoxin-deficient mutants were able to enter the fungal cells similarly to the positive control ( Supplementary Fig. 7).
The number of fluorescent BAT mutants inside the fungal hyphae after two days of co-incubation was similar to that seen one day after co-incubation ( Fig. 4a and Supplementary Table 9), suggesting that reinfection with BAT mutants takes place within 24 hours of their introduction ( Supplementary Fig. 5). It is possible that a potential increase in bacterial cells after 48 hours was not detected due to bleaching of SYTO9 35 . To minimise the effects of bleaching, light exposure was kept to a minimum by using a low exposure time (100 ms). Surprisingly, the four fungal morphotypes differed across the 48 hour infection period. The number of bacteria observed in morphotype 4 (abortive sporangiophore) after two days of coincubation was significantly higher (p<0.05) compared to one day after infection for all BAT mutants ( Fig. 4b and Supplementary Table  9). Bacterial localisation in morphotype 4 was also observed in microscopic images ( Fig. 4c and Supplementary Fig. 5).
The fungal reinfection and subsequent formation of morphotype 4 was captured in real-time ( Fig. 4c and Video 2). First, BAT mutants penetrate and re-infect selected hyphae of R. microsporus, especially side-hyphae. Following infection, the bacterial cells reach very high densities (strong green-fluorescent signal) and septa are formed at the base of these hyphae.  Table 10).
As a result, the cytoplasmic flow between infected hyphae and the remaining fungal mycelium is abolished, leading to the physical containment of bacteria (Video 2). These observations were surprising, considering that R. microsporus generally does not form septa. In the rare cases where septa can be found, they are formed to either wall off old or injured hyphae or as separators between mycelia and spore-forming structures 33,37 .

BAT mutants induce septation in fungal host
To further investigate the formation of septa, we analysed the total number of septa formed following fungal reinfection with B. rhizoxinica BAT mutants (Δbat1::Apra r , Δbat2::Kan r , and Δbat3::Kan r ) in a BFI device. The total number of septa was smaller or equal to the positive control (≤100% of sterile R. microsporus ATCC62417/S re-infected with wild-type B. rhizoxinica) 24 hours after reinfection, while septa formation increased markedly after 48 hours of co-incubation (Fig.  4d,e and Supplementary Table 10a). Compared to the reinfection process where most bacteria are inside the fungal hyphae after 24 hours (Fig. 4a), the formation of septa appears to take place at a later stage of the reinfection process (48 hours). In addition, septa formation and accumulation of bacteria in side-hyphae correlated with morphotype 4 (Fig. 4c). Using a LIVE/DEAD stain we observed that the majority of BAT-deficient mutants were dead in morphotype 4, while all other morphotypes contained more live bacteria (p<0.05). In contrast, wild-type B. rhizoxinica showed similar ratios of live versus dead bacteria in all four hyphal morphotypes (Fig. 5a,b and Supplementary Table 11). It has to be taken into consideration that LIVE/DEAD staining, utilising SYTO9 and propidium iodide (PI), may underestimate bacterial viability. SYTO9 can stain live and dead cells with different efficiencies 35 , while PI has a strong binding affinity to extracellular nucleic acids, a major component of bacterial biofilms 38 . Since B.
rhizoxinica is a known biofilm producer 32 , we performed LIVE/DEAD stains of axenic cultures prior to co-incubation in BFI devices. The majority of the cells stained green ( Supplementary Fig. 8), indicating that underestimation of bacterial viability due to extracellular nucleic acids is unlikely.

Discussion
Bacteria that live in close association with eukaryotes can control their host by employing specialised tools 39,40 . One prominent example are the TALEs, a T3SS-associated group of highly homologous proteins that allow plant pathogens to modify and exploit their hosts 27,30,41 . In recent years, TALEs have become popular tools in biotechnological applications due to their ability to manipulate DNA in a site-directed manner 42 . As a consequence, TALE-related proteins have been identified in a wide range of bacterial species, including B. rhizoxinica (BAT1, BAT2, and BAT3) 26 . In this study we combined genomic and functional studies to characterise T3SSassociated proteins in the fungal endosymbiont B. rhizoxinica. While fungal sporulation is independent of the T3SS-associated alaninetryptophan-arginine (AWR) protein, targeted gene knockouts resulted in BAT-deficient mutants that are unable to induce fungal sporulation in the host, suggesting that BAT proteins are an important factor for the Rhizopus -Burkholderia interaction. By utilising a tailor-made microfluidic device 24 , we show that the fungal mycelium is composed of four morphologically distinct types of hyphae. These fungal morphotypes are rare in the Zygomycetes 33 , and this is the first time that different cell types have been described in R. microsporus. The vegetative main hyphae of R. microsporus (morphotype 2), is one of the most important type of hyphae. These so called trunk hyphae 43 transport intracellular material across the R. microsporus mycelium at relative high speeds. This allows R. microsporus, and fungi in general, to cope well with heterogeneous distributions of soil nutrients in their natural habitat 34 .
In addition, fast cytoplasmic flow also allows for free migration of B. rhizoxinica across the host mycelium 44 . Hyphae serving as dispersal vectors for motile bacteria are an important feature of BFI ecology 45 and may promote colonisation of new ecological niches 46 .
Microfluidics was used to gain a deeper understanding of the interaction between BAT-deficient endosymbionts and their fungal hosts. We have demonstrated (i) efficient fungal re-infection of BAT mutant strains within one day, and (ii) accumulation and trapping of these strains in specific fungal side-hyphae after two days. Accumulation of mutant endosymbionts is unique to BAT-deficient B. rhizoxinica, as T2SS-and T3SS-deficient mutants are unable to reinfect the sterile fungus efficiently 11,21 . Since total bacterial reinfection reaches its maximum after one day, it is conceivable that BATdeficient bacteria are redistributed inside the fungal mycelium following reinfection and subsequently accumulate in side-hyphae. These infected side-hyphae are then walled-off from the remaining mycelium via the formation of septa, which leads to trapping of BATdeficient bacteria. The occurrence of septa is one of the most surprising results, as Zygomycetes generally lack septate hyphae 33 .
Septation causes the cytosolic exchange to stop and bacteria, which normally move freely within the mycelium 8 , become physically restricted.
Considering that the majority of trapped bacteria are dead, our observations are in agreement with a model where the lack of BAT proteins marks B. rhizoxinica as a pathogen or potentially harmful, thereby eliciting a protective response from the fungus. It should be kept in mind that these hyphae could represent abortive sporangia due to the set-up of the microfluidics devices. R. microsporus is grown submerged in liquid medium under which conditions the fungus struggles to produce sporangia. However, considering that the persistence of the symbiosis is dependent on spores containing endobacteria 8 , we would expect the majority of the trapped endobacteria to be alive (as seen for wild-type bacteria). Therefore, we argue that BAT-deficient bacteria induce a controlled suicide of the host, a mechanism that is well described in plantpathogen interactions. For example, TALEs from Xanthomonas campestris induce a suicide gene in resistant plants, causing programmed cell death of the host and thereby preventing the pathogen from spreading 47 .
Notably, BAT proteins described in this study, are the first TALE-related proteins whose absence causes an increase in host colonisation. In comparison, TALE proteins from plant pathogens aid host colonisation through the induction of susceptibility genes, which leads to a dramatic increase in bacterial cell numbers in the host 31 . It is remarkable that the absence of BAT proteins causes a similar phenotype in fungi, i.e. high numbers of bacterial cells. Thus, BATs might represent DNA-binding proteins with a function different to TALEs 42 . For example, multiple unrelated proteins showing nuclear localisation and interaction with DNA have been identified as mechanisms to aid host colonisation for various pathogens and symbionts 39 . The fact that parasites and mutualists commonly share the same molecular tools (e.g. secretion systems) to control their hosts 1 supports the hypothesis that B. rhizoxinica becomes pathogenic 'in the eyes of the fungus' if BAT proteins are absent. Indeed, multiple investigations have suggested that B. rhizoxinica endosymbionts were initially parasites before switching to mutualists 48 . For example, phylogenetic analyses indicated that B.
rhizoxinica has switched its host multiple times during evolution 16 , while R. microsporus has acquired resistance against a potent phytotoxin (rhizoxin) produced by the endosymbiont 49 . Thus, BAT proteins might represent an important factor for the maintenance of the mutualistic relationship in this BFI and their absence may facilitate a switch from mutualists back to parasites. This challenges the traditional views of symbiotic relationships, which were defined as a mutually beneficial for both partners. Instead, the BFI presented in this paper may be better described as a sliding scale, where symbionts can act as mutualist or pathogen depending on the genetic context 50 .

Conclusion
In summary, we show that BAT-deficient endosymbionts successfully re-colonise the host but become trapped in specific hyphae due to the formation of septa. This represents an unprecedented case where mutant bacteria induce biogenesis of septa in a Zygomycete. In contrast to TALE proteins from phytopathogenic bacteria, a lack of BAT proteins facilitates bacterial colonisation of the host. Finally, microscopic investigations revealed that TALE-deficient mutants could be regarded as pathogens in the fungal cytosol, which would be in line with a plausible antagonism-mutualism shift in the evolution of this microbial symbiosis. The possibility of endobacteria controlling the physiology of a fungal host offers a broader view on the dynamic interactions between bacteria and fungi. In silico predictions and characterisation of Type 3 effectors Potential type 3 secreted effector proteins were predicted using the T3SS PREDICTION server 56 and the EFFECTIVE T3 prediction tool 57 . Nuclear localisation signals (NLS) were predicted using the cNLS domain prediction tool 58 and NucPred 59 . The nucleotide sequences of potential Burkholderia T3 effector genes have been deposited in GenBank under the accession numbers provided in Supplementary Table 3.

Phylogenetic analysis
For phylogenetic analysis, predicted T3 effector protein sequences were aligned using ClustalW 60

Generation of B. rhizoxinica mutant strains
To investigate the role of both AWR and BAT proteins in the symbiosis, four genes (awr: RBRH_03012; bat1: RBRH_01844; bat2: RBRH_01776; bat3: RBRH_01777) were deleted using a double crossover strategy as previously described 21 .
Using a proof-reading polymerase, the upstream and downstream regions of the genes of interest were amplified. Primers were designed to contain 20 bp overlap with the gene of interest and as well as a 20 bp overlap with an antibiotic resistance cassette (bat2, bat3: kanamycin, bat1: apramycin). The kanamycin and apramycin cassettes were amplified from pK19 and pJI773, respectively, using primers carrying the same 20 bp.

Sporulation assay
In a liquid sporulation assay, sterile R. microsporus aerial hypha (app. 0.1 cm 3 ) was grown in 24-well plates containing 750 µL VK medium (5 g/L glycerol, 10 g/L yeast extract, 10 g/L corn starch, 10 g/L corn step solids, 10 g/L CaCO3; pH 6,5). After 15 hours of incubation, 100 µL of overnight wild-type or mutant B. rhizoxinica cultures were added to individual wells in three biological replicates. Co-culture plates were inspected daily for sporulation. The spores were harvested from each well using 500 µL NaCl (0.15 M) and counted using a Thoma Chamber. Experiments were performed three times independently (N = 3) with three technical replicates on each plate. Data are presented as means with SD. Raw data from sporulation experiments were processed with MS Excel. GraphPad Prism 5.03 (GraphPad Software, La Jolla, California, USA, www.graphpad.com) was used for statistical analysis and graphing. Data from spore counts were compared for B. rhizoxinica wild-type and B. rhizoxinica mutant strains using one-way analysis of variance (ANOVA) and Tukey HSD test function in GraphPad. P-values with p<0.05 were considered statistically significant. The Brown-Forsythe test was used to test for equal variance and a p value with < 0.05 was considered significant.

Fluorescence microscopy of co-cultures
One-week old co-culture fungal hyphae were used to visualise the localisation of the B. rhizoxinica mutant strains. The bacterial cells were stained with 5 µM Syto 9 (Invitrogen) and fungal cells were counter-stained with 2 µg/mL calcofluor white (Fluka, Germany) for 5-10 min. Fluorescent microscopy was carried out using a Zeiss LSM 710 confocal laser-scanning microscope, and images were captured using the Zeiss-Zen software.

Inoculation of bacterial-fungal interaction devices with fungus
Bacterial-fungal interaction (BFI) devices were prepared as previously described 24 . Before the BFI devices were inoculated, cured (endosymbiont-free) R. microsporus ATCC62417/S was subcultured on PDA plates for two days at 30 °C. A piece of young, growing mycelium (app. 1 cm 3 ) was cut from the agar plate and placed upside down in front of the microchannels of a BFI device filled with Potato Dextrose Broth (PDB; Becton, Dickinson & Company). The agar plug was positioned with the growth direction of the fungus facing the microchannels. We paid special attention to the size of the agar plug to minimise biases between experiments. After two days of incubation at 30 °C, the hyphae had grown far enough into the microchannels to perform subsequent reinfection experiments.

Bacterial inoculation into bacterial-fungal interaction devices
Bacterial over-night cultures (500 µL) were harvested in an Eppendorf tube and resuspended in 0.5 mL NaCl (0.85%) containing SYTO9 green-fluorescent nucleic acid stain (5 nM final concentration, Invitrogen). Following incubation in the dark for 5 min, stained cells were washed and resuspended in 0.5 mL PDB medium. Cells were counted using a CASY Cell Counter (OMNI Life Science, Bremen, Germany) and then added to the bacteria inlet of the BFI device (app. 50-100 µL), which is connected with the microchannels containing the cured fungus (Fig. 3a,b).

Live-cell imaging of bacterial-fungal interactions
A fluorescence spinning disc microscope (Axio Observer microscope-platform equipped with Cell Observer SD, Zeiss) was used to capture BFIs at 0 h, 24 h, and 48 h after bacterial inoculation. For each time point, 16 images were taken at random position of the microchannels, including both ends of the BFI device (N = 16 biological replicates). Brightfield images were captured using a laser intensity of 7.1 V and an exposure time of 100 ms. Bacterial cells (wild-type B. rhizoxinica, Δbat1, Δbat2, Δbat3, ΔsctT, ΔsctC, and ΔrhiG), stained with SYTO9, were captured at 485/498 nm with an exposure time of 100 ms. As a reference, wild-type R. microsporus ATCC62417, naturally containing endosymbionts, was also analysed, although no additional bacterial cultures were added through the inlet. Each reinfection experiment was performed at least three times independently (N ≥ 3). Images were generated using custom software (Zeiss Zen Blue Software) and analysed using Fiji 36 .
Images, containing spatial calibration meta data, were imported to Fiji and the same spatial scale was applied to all images using the global scale function. Prior to image analyses, all images were converted to 16-bit greyscale, which was propagated to both channels (brightfield and 485/498 nm). The hyphal area and integrated density (ID; product of area and mean grey value) of both channels were measured using the free hand tool and measuring tool implemented in Fiji. According to the brightfield images, four morphologically distinct types of hyphae were defined (Fig. 3d) and the number of septa observed was counted. Subsequent calculations and data visualisation was carried out using MS Excel and GraphPad Prism 5.03. To correct for autofluorescence and variations between experiments, raw ID values at 0 hours post infection (hpi) were averaged from three independent experiments and then subtracted from raw ID values at 24 hpi and 48 hpi. Corrected ID values were divided by the number of bacterial cells (cells/mL) inoculated in the bacterial inlet and then plotted as percent of the positive control (cured R. microsporus ATCC62417/S co-incubated with wild-type B. rhizoxinica HKI-0454). The same approach was applied to correct for the number of septa observed. The corrected number of septa was divided by either the inoculated bacterial cell number (cells/mL) or by the mycelium area (in µm) and then plotted as percent of the positive control.

Imaging of bacterial survival
To test the bacterial survival rate in BFI devices, cells were stained with LIVE/DEAD BacLight fluorescent dyes (Invitrogen) after 72 h of fungal-bacterial co-incubation. To stain cells inside the BFI device, the PDB medium was removed through the bacterial inlet and replaced with 100 µL staining solution (10 nM SYTO9 and 60 nM propidium iodide). Fluid replacement was repeated three times, before the devices were incubated in the dark for 15 min. Following staining, fluorescent dyes were replaced with PDB medium by flushing the devices four times. Fluorescence microscopy and subsequent image analysis was performed as described above.

Statistical analysis
Raw data from re-infection experiments and bacterial survival experiments were processed with MS Excel. GraphPad Prism 5.03 was used for statistical analysis and graphing. One-way analysis of variance (ANOVA) was used to study the relationship between different B. rhizoxinica strains (wild-type and mutant strains) and fungal physiology (e.g. mycelium area, reinfection efficiency, septation rate, bacterial survival) following fungal reinfection using the Tukey HSD test function. P-values with p<0.05 were considered statistically significant. The Brown-Forsythe test was used to test for equal variance and a p value with < 0.05 was considered significant.

Data availability
All data generated or analysed during this study are included in the manuscript and in the supporting files. Source data files have been provided for Supplemental videos.