Two different sources of leaves were used in this study: greenhouse grown A. thaliana L. plants cvs. Columbia (Col-0) (Lehle Seeds, USA), naturally colonized by environmental bacteria (see Maignien et al., 2014 for details) and in vitro grown plants (0.5 X MS with vitamins at pH 5.8 with 0.7% agar, Figure S1A) grown from surface sterilized seeds. Leaves excised from axenic plants incubated overnight with defined mixtures of bacteria were imaged to showcase the suitability of Leaf-FISH for taxa identification (see section Visualization of Multiple Bacterial Taxa in Arabidopsis Leaves for full description of incubation methods). Leaves from plants grown from seeds inoculated with Methylobacterium adhaesivum B5A (isolated in this study, see Methods on Supplementary Material) or the reference strain Methylobacterium extorquens PA1 (Knief et al., 2010a) were used to explore leaf colonization preferences under in vitro conditions (see sections Monoisolate Inoculation of Arabidopsis Seeds and Characterization of Methylobacterium strains for colonization of Arabidopsis phyllosphere for full description of inoculation methods).

Eelgrass (Zostera marina) was collected in Woods Hole, MA (US) in June 2013.

Fresh leaves (10–20 in vitro leaves/leaf fragments per tube) (Figure S1B) were incubated at room temperature on an orbital shaker at low speed (90 rpm) for 1 h in 10 ml of 1 X PBS (Fisher Scientific, USA) supplemented with 2% paraformaldehyde (EMS, USA). We tested other technical approaches to optimize the fixation step, such as variations of the agar embedding technique described in Brandl et al. (2001). Specifically, we embedded leaf sections in 0.8% low melt agarose using small hybridization chambers (e.g., Secure-Seal, EMS, USA). However, the use of agarose was extremely labor intensive and agarose coated-leaves were labile and difficult to manage. We combined agarose coating with a light vacuum to increase the effectiveness of the fixation step but this resulted in in severe alteration of the distribution of the bacterial cells, many of which were removed from the plant tissues and found in the agarose coat. Processing of “naked” leaves, defined here as tissues not embedded or treated with any coating agents prior to fixation, required less intensive manipulation and reduced preparation time and costs, as it allowed pooling leaves during fixation and pigment removal steps. The lack of protective coating agent did not result in significant losses of bacteria from leaf surfaces (see Supplementary Material).

Leaves were incubated in 10 ml of freshly prepared 50% ethanol. Sequentially, the percentage of ethanol in the mixture was increased (70, 80, 85, 90, and 95%) and, in each step, leaves were incubated for at least 30 min. Typically, in vitro material was incubated twice in 95% ethanol while greenhouse material required three to five replacements to achieve complete whitening of the tissues (Figures S1C, S2A,B). Plant material was stored at −20°C in 10 ml of 95% ethanol for 2 days prior to further optimize pigment removal. Long-term storage of fixed tissues was performed at −20°C in 50% ethanol/1 X PBS.

Tissues were rehydrated prior to any hybridization steps in sterile distilled water or 1 X PBS. Two to four leaves were transferred to each 0.7 ml Eppendorf tube containing 400 μl of pre-warmed hybridization mix and incubated at 46°C for 1.5 h. Hybridization buffer (0.01% SDS, 0.02 M Tris-HCl pH 7.5, 0.2 M/0.1 M NaCl and 20%/30% “Hi-Di” Formamide) was prepared fresh. Appropriated probe(s) combinations were added to each hybridization mix for a final concentration of 5 ng μl⁻¹ each. The probes used in this study were the general eubacteria set EUB338 I, II, and III (Amann et al., 1990; Daims et al., 1999), Methylobacterium-specific mybm-1388 (Pirttilä et al., 2000) and Pseudomonas-specific PSE227 (Watt et al., 2006). Leaves were transferred to new tubes with 1 ml of pre-warmed wash buffer (0.01% SDS, 0.02 M Tris-HCl pH 8.0. 0.22/0.11 M NaCl; 5 mM EDTA if stringency ≥20%) and incubated at 48°C for 10 min. Leaves were rinsed with sterile distilled water, placed on a microscopy slide and tissues were dried in the dark. ProLongGold with or without DAPI (Life Technologies, USA) was added to each leaf before covering with a glass cover. Samples were stored overnight in the dark at room temperature. For long-term storage, slides were kept at 4°C.

The hybridization specificity of all probes was tested by evaluating their performance in pure and mixed cultures of bacteria. An appropriate volume of bacterial culture was filtered through a 0.22 μm filter (GTTP, Millipore, USA). After fixation in 1 X PBS and 1% paraformaldehyde for 1 h at room temperature, sections of the filters were subjected to a standard in situ hybridization protocol using a mixture of probe EUB338 I, II, and III conjugated with Rhodamine Red, mybm-1388 conjugated with Alexa-488 and PSE227 conjugated with Alexa-647 (Life Technologies, USA). Total number of cells present per field were evaluated by counting in a standard epifluorescence microscope (Zeiss Axioskop2) using DAPI staining.

Images were acquired with a Zeiss LSM 780 (Carl Zeiss Microscopy GmbH, Germany) laser scanning confocal microscope equipped with 32-channel GaAsP detector and 8 laser lines using a 100 × 1.4 NA objective lens. Each field of view was imaged using sequential excitation with the appropriate laser combinations for each fluorochrome combination, including 633- (Alexa-647, emission ~670 nm), 561- (Rhodamine Red, emission ~591 nm), and 488- (Alexa-488, emission ~520 nm). To maximize the signal-noise ratio each image was obtained as the average of four images, at a resolution of 12 or 16-bit (4,096–65,536 gray levels) and frame size of 1024 X 1024. The resulting size of the images was 84.94 μm × 84.94 μm. 3-D images were generated using Z-stacks in multitrack mode. The number of tracks was optimized using the multitracking option to minimize cross-talk and fluorophores bleaching. To generate 3-D images typically 5–16 images were taken per field, each at 0.5–1 μm across the Z-axis. Starting from the leaf surface, our ability to image the leaf inner tissues was determined by the residual background autofluorescence emitted by the leaf cells. Subcuticle and substomata areas were easily imaged (approx. 10–15 μm depth), even in mature leaves from greenhouse plants. In young leaves from in vitro plants, the low autofluorescence after pigment removal made possible to image across the blade. Larger composite images, up to 400 μm × 400 μm, were generated by tiling smaller ones generated as described above. Capture in lambda mode was used to generate crosstalk-free multi-fluorescence images by parallel or sequential acquisition of images with spectral information of each pixel.

Spectral images were unmixed using the linear unmixing algorithm implemented in the ZEN Software package (Carl Zeiss) following manufacturer protocols. Linear unmixing was applied using appropriate reference spectra library acquired defining region of interest (ROI) in sets of reference images. For identification of background plant-associated autofluorescence multiple samples of plant tissues (surface cuticle areas, trichomes, chloroplasts, and certain areas of stomata) were included in the reference spectra libraries. ROI were drawn on images corresponding different Arabidopsis leaves and included samples collected from different structures and depths. In most cases a combination of three plant associated references were enough to unmix images and minimize the signal in the residual component. Reference spectra also included unlabeled cells grown as monocultures of Methylobacterium, Sphingomonas, and Pseudomonas for detection of possible autofluorescence as that shown for Pseudomonas in Video S2; cells in culture of Sphingomonas labeled with EUB338 (Rhodamine Red), Methylobacterium labeled EUB338 (Rhodamine Red) and mybm-1388 (Alexa-488) and Pseudomonas labeled with EUB338 (Rhodamine Red) and PSE227 (Alexa-647) and acquired with each excitation wave length. For multicolored Leaf-FISH image presentation, the image of each unmixed fluorophore channel was assigned a different pseudocolor. As recommended in Valm et al. (2012), autofluorescence layers were treated as other spectral character to be identified after unmixing. To generate the multicolor Leaf-FISH, images generated with each unmixed fluorochrome plus those generated with the unmixed autofluorescence of the plant tissues were false colored. All resultant images were merged, and brightness was independently adjusted to generate the final image. Additional imaging processing was performed in Fiji (Schindelin et al., 2012).

Arabidopsis sterile seeds (10% bleach 20 min, rinsed in sterile distilled water 10 times) were incubated for 20 min at room temperature in 1.5 ml Eppendorf tubes containing 0.5 ml of 10⁴ cell ml⁻¹ of the appropriate bacterium isolate. Each bacterial isolate used for seed inoculation were cultured at 30°C and 180 rpm for 48 h in liquid medium, LB (0.5 g l⁻¹ NaCl) or MM medium. Prior inoculation, cell density and viability of the culture was assessed using Live/Dead BacLight L7012 (Live Technologies, USA). Sterile and inoculated seeds were transferred to respective rectangular transparent polypropylene microboxes (Combiness Microbox, Belgium) (10–12 seeds per microbox, four microboxes per treatment per experiment). Plants were grown in 200 ml of autoclaved 0.5 X minimal MS medium with vitamins, at pH 5.8, 0.7% agar and filtered lids (+XXL). Previous experiments concluded that high gas exchange (81.35 GE day⁻¹) contributed to general macroscopic features of the leaves and more vigorous growth in carbon source-free MS medium (data not shown). Sowed seeds were stored at 4°C in the dark for a period of 48–72 h. Germination was carried out under low light, achieved by covering the microboxes with a shadow mesh in a Precision Plant Growth Refrigerated Incubator 818 (Thermo Scientific, United States) under 8 h light (23°C) 16 h dark (21°C) cycle. After 2 weeks, seedlings were exposed to full light conditions (120–180 μmol m⁻² s⁻¹). Microboxes were rotated each 2–3 days to minimize positional effects. Adult stage was reached after 6 weeks in culture (bolting observed in less than 25% of plants).

For this seed inoculation protocol, we defined as “successful colonizer” any isolate that consistently demonstrated abundant colonization of Arabidopsis leaves without interfering with the development of the host plant. In addition, we consider as a desirable trait low growth on the MS medium plant medium. Following the criteria described above, we selected for colonization of Arabidopsis phyllosphere under in vitro condition studies two strains of Methylobacterium; M. adhaesivum (B5A) isolated in this study and a closely-related laboratory-evolved isolate of M. extorquens PA1. In vitro plants grown from inoculated seeds were then used for Leaf-FISH experiments. Sterile plants were grown as reference and used to evaluate any potential alterations of plant morphology or development caused by the seed inoculation. To assure the repeatability of the colonization patterns observed, two cohorts of plants were independently seed-inoculated for each isolate of Methylobacterium.

Indolic acid (IAA) production of Methylobacterium adhaesivum B5A and the reference strain Methylobacterium extorquens PA1 was measured as defined in Glickmann and Dessaux (1995). Phytohormone production was assessed by triplicate on cultures grown in liquid LB and MM at 30°C and 180 rpm for a period of 10 days. The second, seventh and tenth day in culture, optical density at 600 nm and IAA concentration (Salkowski's reagent R2, 4.5 g of FeCl₃ per liter in 10.8 M of H₂SO₄) at 535 nm were measured. Indol concentration was determined using a five point concentration curve (0-10 mg ml⁻¹) of synthetic 3-indolacetic acid (Sigma-Aldrich, USA) freshly prepared in each respective medium. The in vivo effect of IAA production of each Methylobacterium strain (B5A, PA1) was evaluated by quantification of their ability to trigger an adventitious organogenic response of Arabidopsis leaf tissues. Leaves freshly excised from several 6-week-old in vitro Arabidopsis axenic plants were pooled and incubated at room temperature for 20 min in saline double distilled water (0.85% NaCl) or saline water supplemented with an aliquot of Methylobacterium (strains B5A or PA1) grown 48 h in MM. Rinsed leaves (8–13) were placed in petri dishes (eight plates per strain) with 10 ml of MS with vitamins pH 5.8, 0.7% agar supplemented with 30 g l⁻¹ of sugar and maintained under an 8 h light (120–180 μmol m⁻² s⁻¹) 16 h dark cycle. Control leaves (axenic) were placed in MS medium (negative controls, triplicate) or MS medium supplemented with 1.5 mg l⁻¹ IAA (positive controls, triplicate). The presence of adventitious roots or calli was determined every 2 weeks by visual inspection over a 6-week period.

Background signal caused by the autofluorescence of Arabidopsis leaves completely masked any signal emitted by the fluorophores of the bacterial probes (Figures S2C,D) using a routine microscope for transmitted light and incident light fluorescence (Zeiss Axioskop2). CLSM (Zeiss LSM780) followed by intensive processing of confocal images post-capture led to slightly better results in leaf margins (data not shown) but we were unable to confirm bacterial visualization of other leaf areas. In order to reduce the background fluorescence emitted by the plant tissues, we incorporated in our protocol a pretreatment step consisting of a series of dehydrating steps in ethanol. The use of ethanol to gradually dehydrate Arabidopsis leaves successfully removes most of the pigments from the cells (Figure S1C). Under microscope, treated leaves show a translucent (bleached) appearance (Figure S2B) and lower levels of autofluorescence. Background signal reduction permitted clear identification of the specific fluorescence emission of the bacterial probes and visualization of the bacterial cells present on leaves (Figures S2E–H) even using routine microscopes (Zeiss Axioskop2).

In images generated through spectral imaging confocal microscopy, plant residual autofluorescence was incorporated into confocal images after unmixing as layer(s) that could be modified during post-imaging processing. Background signal(s) generated detailed images of the plant surface and subcuticular areas, allowing simultaneous visualization of the leaf microstructures and associated bacteria (Figures 1A–C). Tridimensional images show the continuous distribution of some bacteria on the leaf surface into subcuticle tissues (Figure 1A). Each of the probe fluorophores and plant autofluorescence were incorporated to the image as an individual layer that can be shown simultaneously (Figure 1B) or independently (Figure 1C) as 2-D or 3-D images. As expected, bacterial cells are irregularly distributed on peak and valleys of the leaf surfaces (Figures 1D,E) and are frequently found associated with cuticle cracks, stomatas, or close to veins.

We selected eelgrass (Zostera marina) to test the transferability of the Leaf-FISH method to other plant systems. Eelgrass's adaptations to oceanic lifestyle affect crucial mechanisms that terrestrial plants use to influence the structure of their associated leaf microbiomes. Eelgrass lacks stomata (a common entry point for pathogens in land plants), lost genes for the production of volatile terpenes, and presents a modified cell wall with low-methylated pectins and sulfated galactans, important for ion homoeostasis, nutrient uptake and O₂/CO₂ in saline conditions (Olsen et al., 2016). Land and aquatic plants also differ in leaf microbiome core composition and dynamics. While leaf microbiome is distinct from air inoculum in Arabidopsis (Maignien et al., 2014) in Zostera its composition is highly variable, mirroring their adjacent coastal seawater in eelgrass (Fahimipour et al., 2017). Despite the deep modifications of the leaf physiology and morphology, our Leaf-FISH method successfully generated images of the natural occurring microbial community on eelgrass leaves. Bacteria, labeled with a mixture of probe EUB338 I, II, and III conjugated with Rhodamine Red, were easily observable in the orthoprojections or 3-D images generated of the leaf surface (Figure 2). As expected, bacterial density in this submerged marine plant was much higher than in the terrestrial Arabidopsis.

To assess the specificity of our protocol we generated a simplified bacterial community by inoculating axenic leaves of in vitro grown Arabidopsis with a known mixture of some of the most frequently reported phyllosphere bacterial taxa (Sphingomonas, Pseudomonas, and Methylobacterium) (Vorholt, 2012). Probe testing was performed on pure cultures. Cells (visualized with the DNA stain DAPI) successfully hybridized with the general probe EUB338 I, II, and III (Figures S3A,B). Efficient simultaneous hybridization on filters using general probe EUB338 I, II, and III and genus-specific probes was also achieved for Methylobacterium and Pseudomonas cultures (mybm-1388 and PSE227, respectively). FISH experiments performed on mixed cultures on filters and on leaves certified the None of our taxon-specific probes showed significant cross hybridization to unintended targets. High hybridization efficiency (Figures S3A,B) and specificity of the selected FISH probes was observed in filters and in planta (Figures S3B–D). Spectral unmixing analysis was needed whenever Pseudomonas was present, as the specific strain used in this study was characterized by high autofluorescence (see Video S2 in Supplementary Material).

Our method successfully generates taxonomically informative, tridimensional images for visualization of bacteria inhabiting plant surface and subcuticle spaces (Figure S4) or associated with leaf microstructures (Figure 3). Clear specific fluorescence emission associated with each of the fluorophores conjugated with the set of probes targeting cells of Methylobacterium, Pseudomonas, and Sphingomonas (Figures 3A–C) was observed even in dense bacterial aggregates (Figures 3D–F). Background-free images (Figure 3A) and/or images showcasing specific taxa (Figure S5) can be generated by removing plant autoflourescence signals during post-capture image processing. Under our experimental conditions, P. putida W619 was abundant on the leaf surfaces but also it was frequently found inside stomatal cavities and subcuticle spaces (see Figure S4, Videos S2, S3). So, the ability to generate tridimensional images was essential to gain access to the complete distribution of P. putida W619 in the Arabidopsis phyllosphere (Figures 3A–C).

The ability to manipulate the background signal associated with plant autofluorescence and the strong hybridization signal of taxa-specific probes opens the door to multiple imaging approaches, such as generation of composite images across the irregular surface of leaves. We have generated tridimensional tiled-images to capture bacterial distribution across large areas of the leaf surface, for example 4 × 4 tiles for a total imaged area of 340 × 340 μm (Figure 4A), while simultaneously providing enough resolution to single out rare taxa in a structurally informed context even in a “mock” microbial communities overwhelmingly dominated by a single species (Figure 4B). Post-capture image processing removing plant associated background signal (Figure 4C) generates images suitable for automatic image processing such as standard particle counting analysis as implemented in Fiji (Schindelin et al., 2012; Figure 4D). No discrepancies were observed between cell density estimates calculated by visual counting or by automatic cell counting programs. Additional taxa-specific automatized analysis can be easily performed using different combinations of the layers generated by each probe emission patterns (as those shown in Supplementary Figure S5).