Gelatin from bovine (Sigma-Aldrich, St. Louis, MO, United States), sodium carboxymethyl cellulose (CMC) (low viscosity, Sigma-Aldrich), and a Tissue-Tek^® optimal cutting temperature (OCT) compound (Sakura Finetek, Alphen aan den Rijn, Netherlands) were used for the sample preparation. Sucrose (Pharmaceutical Secondary Standard, Certified Reference Material, Sigma-Aldrich) was purchased to generate the reference spectra. The organic matter reference sample Suwannee River Fulvic Acid (SRFA II, International Humic Substances Society) was used for quality control during FT-ICR-MS measurement. Ultrapure water (MQW) was generated from a MilliQ Integral five system (Merck, Darmstadt, Germany).

Zea mays L. plants (wild type, B73 cultivar) were grown in field- and laboratory-scale experiments. Metal cylinders (h = 2 cm, d = 0.74–0.91 cm, wall thickness = 400 μm, sharpened) were used to punch out an undisturbed soil sample (1.0 cm height) with roots (undisturbed refers to the soil structure and the spatial relationship between the root surface and soil particle arrangement). Experimental details on the field (Vetterlein et al., 2021) and laboratory experiments can be found in the Supplementary Material to this article (Supplementary Table 1).

The field samples were taken at BBCH 19 (Biologische Bundesanstalt, Bundessortenamt and Chemical industry) following the German coding of the phenological growth stages of maize (Lancashire et al., 1991; Bleiholder et al., 2001; Meier, 2018) from plots established in the Helmholtz Centre for Environmental Research (UFZ) Research Station Bad Lauchstädt, Germany (N°51.390424; E°11.875933), where a hole was excavated and undisturbed soil samples were taken at a depth of 12 and 21 cm. The soil was a haplic Phaeozem (loam) from the vicinity of Schladebach in Saxony Anhalt, Germany (N°51.308725; E°12.104531) (Vetterlein et al., 2021).

To trace the fate of assimilated carbon in the rhizosphere, ¹³C pulse labeling was conducted on the day before sampling (Vetterlein et al., 2021). Gas-tight chambers with a volume of 0.42 m³ were set up and ¹³CO₂ was produced by mixing Na₂¹³CO₃ (99% ¹³C Na₂CO₃ Euriso-Top GmbH, Germany) with a solution of 5 mol L^–1 H₂SO₄ (95–97%, for analysis, EMSURE^® ISO, Merck KGaA, Darmstadt, Germany). Nine plants per chamber and field plot were labeled with two equal portions of 8.14 g of Na₂¹³CO₃ tracer over 4 h with the first pulse given after sunrise and the second pulse 2 h after. After the last application, the chamber was left for another 2 h to allow assimilation of ¹³CO₂. The soil surface was sealed during ¹³C labeling with a plastic foil to prevent direct gas exchange.

The ¹³CO₂ concentration in the gas-tight chamber was determined via gas chromatography (GC-Box, Thermo Fisher Scientific, Bremen, Germany, column: PoraPLOT Q, Agilent J&W) coupled with an isotope ratio mass spectrometer (Delta plus XP, Thermo Fisher Scientific, Bremen, Germany). The ¹³C concentration in the chambers decreased from 29.65 at% ¹³C (30 min after the second pulse) to 1.19 at% ¹³C (right before the end of the 4 h labeling period). The total CO₂ concentration at the end of the labeling period was around 90 ppm. ¹³C measurements were conducted to indicate the successful labeling of the plant biomass. The plant material was analyzed using a mass spectrometer coupled with an elemental analyzer (QMS ESD 100, InProcess Instruments, Bremen, Germany).

The samples from the laboratory experiments were taken after 25 and 31 days of growth in the rhizoboxes (Vienna Scientific Instruments GmbH, Austria, PMMA, H × W × D, 23 × 10 × 3 cm) (Supplementary Table 1). The filling, fertilization, and planting were conducted as described for the soil column experiments (Lohse et al., 2020; Vetterlein et al., 2021).

For the verification of the MSI results, three technical replicates of disturbed BS and RS samples were taken from different positions of the rhizobox experiment. The RS was operationally defined as soil attached to the root for this comparison. The RS was manually separated from the roots by shaking the root. The BS was composed of soil that is not in the direct vicinity of the root. No embedding and cryosectioning were performed for the direct analysis via LDI-FT-ICR-MS to allow the comparison with the MSI workflow. The RS and BS were attached to adhesive tape for LDI measurement without additional sample preparation (Solihat et al., 2019).

SCiLS Lab (Version 21a Core, Bruker Daltonik) in combination with the microscopic images were used to manually select regions of interest (ROI) around the root (Supplementary Table 1) and to define areas with an increasing distance to the root surface to visualize molecular gradients. The peak intensities were normalized to the root mean square (RMS) of the peak intensities in each spectrum, averaged for each region, exported, and further processed with R version 4.0.3 (R Core Team, 2020).

The Box-and-whisker plots denote the 75^th and 25^th percentiles (box), whiskers indicate 1.5 × interquartile range (IQR). For the statistical evaluation of the isotope ratios, one-sample t-tests were conducted (one-sided, non-paired, α = 0.05) to compare the mean isotope ratio (IR) for each distance <150 μm from the root and the root itself with the natural ¹³C₁ abundance of the dihexose (0.1298). As an approximation of the measurement uncertainty of the IR, the relative IQR (IQR/median) of the IR for each distance <150 μm from the root and the root itself was calculated. As an estimation for the measurement uncertainty, the mean of these values was used (field sample: 10.8% and rhizobox: 10.0%).

The RMS normalized data were imported in METASPACE (Palmer et al., 2017) to find colocalized metabolites and to annotate possible metabolite structures by using the Database Chemical Entities of Biological Interest (ChEBI 2018-01) with a maximum 10% false discovery rate.

To preserve the small-scale gradients of the molecules in the undisturbed soil samples, small metal cylinders were used for sampling. These can be applied in laboratory and field experiments to take samples below the soil surface. The size for the undisturbed soil sample of up to 0.91 cm diameter and 1.0 cm height was chosen after we observed the disintegration of the sample and loss of soil for larger undisturbed soil samples especially during the embedding and cryosectioning step.

The sectioning of soil samples required embedding in a mixture of gelatin and CMC. Freeze-drying of the soil sample is not recommended as this may result in shrinkage, a brittle structure, and disintegration during the embedding step. Applying a vacuum (down to 110 mbar) to the bottom of the undisturbed soil sample created sufficient suction to infiltrate the soil with the medium. A simple placement of the undisturbed soil sample in a beaker containing the medium did not result in complete embedding and caused the disintegration of the section during the cryosectioning. After embedding, the soil samples were frozen at −20°C to limit the diffusion and microbial activity in the sample.

The stability of the soil sections was further improved by coating the outside of the embedded undisturbed soil sample with a layer of OCT compound before mounting it into the cryomicrotome. Sections of the loam <80 μm were too brittle and did not keep the soil structure around the root intact. Thicker sections (>100 μm) could not be cut reliably since the undisturbed soil sample would fall off the specimen holder due to the higher friction. To ensure that the storage and measurement in the vacuum did not affect the integrity of the sections, images were taken to evaluate the effect of reduced pressure. A small shrinkage of the soil section after storage in vacuum at 200 to 400 mbar for 80 min was observed (Supplementary Figures 2A,B). No further shrinkage in the vacuum (3 mbar) during the MSI measurement was observed (Supplementary Figures 2C,D).

The sample preparation workflow was applied to recover the spatially resolved information of the root exudates from both field (FP1) and laboratory experiments (RB1-b1). Dihexose (C₁₂H₂₂O₁₁, molecular formula of e.g., sucrose, maltose, lactose, and melibiose) was selected as a target molecule for the MSI since these metabolites have been identified in root exudates (Lugtenberg et al., 1999; Walter et al., 2003; Fan et al., 2012). The highest intensity of the dihexose was detected in the root itself or within less than 100 μm from the root surface (Figure 1). While the intensity decreased gently over 500 μm for the sample taken from the field experiment, the laboratory sample showed a sharp drop within approximately 100 μm from the root surface.

Due to the heterogeneous distribution of signal intensity in the sections, the influence of the shape and size of the ROI on the mean intensities of the dihexose was tested with all soil sections (Supplementary Figures 3, 4 and Supplementary Table 1). The intensity gradient of the dihexose could be recovered in all samples, independent of the size and shape of ROI (small ROI: Figure 1, large ROI: Supplementary Figure 5).

The analysis of the two-dimensional (2D) sections from different depths of the soil sample can be used to reconstruct three-dimensional (3D) distributions of root exudates along the root axis. Four sections of the same root – a few 100 μm apart – showed a comparable intensity decrease for the dihexose mass with increasing distance to the root (FP2-a to FP2-d, Figure 2 and Supplementary Figure 5A).

The masses of coumaric acid, vanillic acid, caffeic acid (aromatic carboxylic acids), and glucose could be detected in the root and the rhizosphere by selecting a CASI window with a smaller center m/z value (m/z 172 ± 10 Da, FP2-d, Figure 3). These compounds have been previously reported in root exudates (Supplementary Table 2). The detection of aromatic compounds is expected because of their selective better ionization with LDI compared with compounds without an aromatic system (Karas et al., 1985). A second LDI measurement of the same section (FP2-d), then, with the higher center m/z value (m/z 341.5 ± 5 Da) also revealed the mass of the dihexose around the same part of the root (Figure 2 and Supplementary Figure 3E).

We, therefore, tested the feasibility to conduct repeated LDI measurements of the same region (RB2) to expand the accessible mass range via selecting different CASI mass windows. Conducting two subsequent measurements of the same region did not alter the mean intensity of the dihexose. Even after the third and fourth measurements of the same section, 60 and 33% of the initial intensity could still be observed (Supplementary Figure 6) with, however, blurred intensity distributions. After four repeated measurements, the measurement region was charred and the section of the root within this region lost its UV-activity (Supplementary Figures 2E,F).

In sample FP1, mass peaks within a 10 Da mass range (m/z 341.5 ± 5 Da) were annotated via METASPACE (Palmer et al., 2017) and 17 metabolites from the ChEBI-Database were found. Seven of these showed a strong localization in the root, among them the dihexose (C₁₂H₂₂O₁₁). The co-located metabolites were C₁₈H₁₈O₇, C₁₉H₂₀O₆, C₁₉H₁₈O₆, and C₁₉H₁₆O₆ (co-localization value with the C₁₂H₂₂O₁₁: 0.59 – 0.76), but their identity as plant metabolites was not further evaluated (Supplementary Table 3). Additionally, C₆H₁₂O₆ and C₆H₁₀O₅ were detected in the spectra (m/z 341.5 ± 5 Da), pointing to the fragmentation of saccharides due to the cleavage of the glycosidic bond and additional loss of water. This fragmentation pattern was confirmed with an LDI-CID-FT-ICR-MS/MS (CID: collision-induced dissociation) experiment of a sucrose standard (Supplementary Figures 7A,B). Without the application of CASI-isolation and collision voltage, only a low intensity of fragmentation was observed, confirming a soft ionization when using LDI (Supplementary Figure 7C).

As an alternative approach to using selected mass windows, a full scan MSI measurement was tested with one section (FP2-d), revealing a high number of signals (Supplementary Figure 8). Only two metabolites could be annotated for the full scan MSI measurement via METASPACE with, however, no spatial correlation to the position of the root (Supplementary Figure 9).

To verify the imaging results with a conventional, destructive approach, disturbed BS and RS samples from a laboratory experiment were generated and analyzed by LDI-FT-ICR-MS. Next to a common background of complex soil OM, molecular formulas unique for BS and RS were detected (Supplementary Figure 10A), resulting in a higher mean O/C-ratio and a lower mean H/C-ratio in the RS sample as compared with the BS (Supplementary Table 4). In particular, this experiment confirmed the enrichment of putative sugars with high O/C and H/C ratios in the RS (Supplementary Figure 10B).

The ultra-high resolution of the FT-ICR-MS allows the detection of multiple Isotopologue signals simultaneously. For the dihexose, the ¹³C₁, ¹⁸O₁, ¹³C₂, ¹³C₁²H₁, and ¹³C₃-Isotopologue were detected localized in the root (Figures 4A–F, exact masses Supplementary Table 5). For both, the field and laboratory experiment, the intensity of the monoisotopic signal and the ¹³C₁-isotopologue followed a linear trend. The slope m of the regression can be used to calculate the ¹³C₁/¹²C isotope ratio (IR). The sample from the field experiment showed a higher slope (FP1, m = 0.17), indicating a possible ¹³C enrichment compared with the non-labeled laboratory experiment (RB1-b1, m = 0.14) and a sucrose standard (m = 0.13, Supplementary Figure 11).

The IR for a sample from the field experiment (FP1) where a ¹³C label was applied decreased with increasing distance to the root. In the root and below 150 μm from the root surface, the IR was significantly (p < 0.05) above the natural abundance (0.1298). The ¹³C uptake by the plant associated with sample FP1 was confirmed by IR mass spectrometry. ¹³C enrichment was detected for the shoot (2.53 ± 0.014 at% ¹³C) and the youngest unfolded leaf (1.44 ± 0.004 at% ¹³C). The mean natural abundance at BBCH 19 for the unlabeled plants was 1.078 ± 0.001 at% ¹³C (n = 24).

Due to the overall decrease in signal intensity with increasing distance to the root (Figure 1A), the coefficient of variation of the IR for each area increased (Figure 4G). Using the relative IQR, we estimated a measurement uncertainty for the spatially and molecular resolved IR of approximately 10%. For the non-labeled laboratory sample (RB1-b1), only inside the root, a significantly higher IR could be detected (p = 0.04), while in the rhizosphere (<150 μm from the root surface) the IR reflects the natural ¹³C₁ abundance (Supplementary Figure 12).

A sample preparation method was developed allowing the spatially resolved analysis of organic molecules at the root-soil interface using LDI coupled with ultra-high resolution mass spectrometry. The primary focus of the sample preparation was to generate thin and stable soil sections. Crucial aspects of the sampling and sample preparation workflow are discussed in the following.

Non-embedded, undisturbed soil samples are not stable during sample preparation and can easily disintegrate during cryosectioning. Freeze-drying of the soil samples resulted in a major volume reduction and a brittle structure of the soil. Conventional resin-embedding requires chemical fixation and the removal of water before the embedding (Mueller et al., 2013; Bandara et al., 2021) and is prone to extract or delocalize small molecules from the soil. Gelatin was previously used for the stabilization of soil samples for microscopic examination of soil structure (Anderson, 1978), and a mixture of gelatin and CMC was recently used for sediment samples (Alfken et al., 2019; Wörmer et al., 2019).

In our workflow, the embedding medium was directly applied to the field moist sample. However, applying the gelatin/CMC mixture onto the top of the undisturbed soil sample resulted in the insufficient embedding/stabilization of the sample, and the obtained sections were brittle and could not be mounted intact onto a glass slide. Therefore, a vacuum was applied to the bottom of the undisturbed soil sample. This supported the penetration of the embedding medium into the soil sample. It should be noted that already a partial filling of the pore volume (as shown in the microscopic images, Supplementary Figure 13) by the embedding medium considerably improved the stability of the sample for subsequent cryosectioning. To minimize the microbial degradation and the diffusion of metabolites, it is recommended to conduct the embedding immediately after sampling. If applicable, non-destructive imaging techniques such as X-ray CT (Gao et al., 2019) could help to pre-select undisturbed soil samples.

The thickness of a section often has a strong effect on the sensitivity of the MSI detection (Alfken et al., 2019). However, for our workflow, the section thickness is limited by the stability of the sections, and sections thinner than 80 μm and thicker than 100 μm tended to be less stable. It should be mentioned that the optimal thickness depends on the substrate texture. For substrates with a higher proportion of sand (i.e., when most of the single particles are larger than 200 μm), the section thickness must be adapted accordingly. To aid identification of ROI containing a root, non-destructive imaging techniques such as optical or fluorescence microscopy can be applied to the sections before subsequent MSI analysis.

As an ionization method, LDI was used. The soil mineral matrix and the aromatic compounds in the root and SOM absorb and distribute the laser energy, even allowing for multiple measurements of the same region (Supplementary Figure 6). A MALDI matrix is often used in MSI and has also been applied for indirect rhizosphere gradients (Veličković et al., 2020b). However, the application of a matrix is an additional sample preparation step that may lead to the diffusion of target molecules on the surface of the section and could also contribute to additional background signals, limiting the ion abundance and hence detection limit of the relevant metabolites.

To achieve the highest sensitivity for observing the gradients of metabolites in the soil despite the SOM background (Supplementary Figure 8), we propose to limit the mass window to a maximum of 20 Da (CASI mode). Although this reduces the number of target molecules detectable in one imaging run, it ensures the sensitive detection of metabolites in soil with high spectral quality.

Targeting other metabolites outside the selected mass window may require multiple MSI measurements with accordingly increased measurement time. E.g., the distribution of [C₉H₈O₃−H]^– (m/z 163.0401, e.g., coumaric acid) and the dihexose [C₁₂H₂₂O₁₁−H]^– (m/z 341.1089, e.g., sucrose) could not be determined in one measurement, but it was possible with two successive measurements from the same section (Supplementary Figure 3E and Figures 2, 3). The inclusion of further mass windows may be limited by the loss of intensity after two consecutive LDI measurements (Supplementary Figure 6).

Without the application of the CASI mode, gradients of metabolites could not be detected around the root (Supplementary Figure 9). However, a large number of signals in the full mass spectrum (Supplementary Figure 8), point to the ionization of SOM by LDI (Solihat et al., 2019), which could be used to analyze the changes in the SOM composition due to rhizosphere processes. The observed chemical complexity further highlights the need for ultra-high mass resolution, as offered by FT-ICR-MS. The further extension of the MSI method to a larger mass range may allow for a comprehensive non-targeted metabolite analysis in the rhizosphere.

Here, we used disturbed soil samples (i.e., via mechanical separation into BS and RS) to screen (no embedding and cryosectioning, Solihat et al., 2019) for target molecules used in subsequent imaging runs (Supplementary Figure 10). Using the LDI measurements of BS and RS samples from a laboratory experiment, the enrichment of molecular formulas of putative sugars with high O/C-ratio was observed in the RS (Supplementary Figure 10B).

The overall MSI workflow was tailored to prevent the delocalization of metabolites. By having short storage times either at low temperature or in a vacuum, and just the necessary sample preparation steps (i.e., no MALDI-matrix application), artifacts were minimized. Since the embedding medium does not infiltrate the complete pore volume (as detected via fluorescence microscopy, see Supplementary Figure 13) and was used sparingly, the bias of the embedding process can be minimized. In any case, we could find the highest intensity close to the root surface and always a good correlation of the position of the root with the metabolite signals, pointing to little or no bias by the sample preparation.

The spatial resolution of LDI-FT-ICR-MSI (25 μm) is sufficient to reveal the distribution of individual plant metabolites in the soil, which cannot be achieved using bulk analysis. A strong decrease in the intensity of the dihexose with increasing distance to the root up to 150 μm could be detected (Figure 1 and Supplementary Figure 5). The low dihexose intensity in regions furthest away from the root indicates a low level of blank contribution by the sample preparation workflow (see also Supplementary Figure 1). The non-linear decrease in intensity can be caused either by microbial utilization (Fischer et al., 2010) or by sorption to minerals or organic matter in the soil (Kaiser and Guggenberger, 2003) but is also expected from diffusive gradients around the roots (Raynaud, 2010; Vetterlein et al., 2020; Landl et al., 2021).

The less pronounced gradient in the laboratory experiments compared with the field experiment may be explained by the higher water content and hence, increased diffusion during the laboratory experiments (Raynaud, 2010; Holz et al., 2018b). Additionally, the higher root length density and different root age in the laboratory compared with the field experiment could lead to changing carbon input by the roots into the soil (Vetterlein et al., 2021). Molecular gradients in a soil section could also be affected by the proximity of the underlying neighboring roots below the 2D-plane analyzed by MSI, which can be identified by X-ray CT. The shape of the observed molecular gradient may also depend on the orientation of the 2D plane analyzed by MSI concerning the root axis (i.e., cross-section or longitudinal section).

As shown in Figure 3, several metabolites could be identified in and around the root. All the compounds assigned to the detected exact masses have been described previously as root exudate components (Supplementary Table 2). However, the origin of the compounds cannot be specified – potential microbial metabolites could contribute to the observed gradient. Structural identification of the detected masses may be achieved by orthogonal analysis using chromatographic techniques coupled with mass spectrometry (van Dam and Bouwmeester, 2016). Differences in the distribution of certain molecular species may, again, depend on differences in the initial concentration or selective adsorption to minerals.

The distribution and decomposition of organic compounds released by the roots are often determined by using isotopic labels such as ¹⁴C, ¹⁵N, or ¹³C (Pausch and Kuzyakov, 2012; Pausch et al., 2013; Vidal et al., 2018). Depending on the experimental setup used, the detectable size of the rhizosphere can vary from a few hundred μm (Holz et al., 2019; Bilyera et al., 2021) up to several mm (Schenck zu Schweinsberg-Mickan et al., 2010; Holz et al., 2018b). The maximum distance from the root where the dihexose could be detected was well below 1 mm. Considering that a single molecular formula was analyzed, the corresponding lower sensitivity as compared with measuring a bulk isotope label could be the reason for the apparent smaller extent of the rhizosphere. Bilyera et al., 2021 reported an observable rhizosphere extent of 120 μm and 880 μm for a mature wild type maize root and the root tip region, respectively based on ¹⁴C imaging, which correlates well with our results (highest intensity of the dihexose within less than 100 μm from the root surface).

Several Isotopologues (¹³C, ¹⁸O, ²H) of the dihexose were detected in the root (Figure 4), indicating a possible application of the presented workflow for the detection of stable isotope-labeled metabolites. It should be mentioned that even without the application of a ¹³C pulse labeling, the natural isotope abundance can be detected (Supplementary Figure 12), highlighting the versatility of the LDI-FT-ICR-MS workflow. Although ultra-high resolution mass spectrometry provides information about the accurate mass of Isotopologues, the precision and accuracy of the IR (as determined from the peak intensities) depends on the absolute magnitude of the detected signal as described for Orbitrap-MS (Khodjaniyazova et al., 2018) and FT-ICR-MS (Cao et al., 2021).

For our LDI-FT-ICR-MS workflow, the ion abundances and accuracy of the ¹³C₁/¹²C IR were affected by the concentration of the analyte and the laser power (Supplementary Figures 14, 15). Such information is crucial to calculate thresholds for a reliable determination of the ¹³C₁/¹²C IR. In the sample from the field experiment for which ¹³C labeling was applied, a dihexose labeling could be detected up to a distance of 150 μm from the root surface, based on the significantly higher IR values (Figure 4) pointing to the enrichment of ¹³C in the rhizosphere. This was confirmed by IR-MS, showing ¹³C enrichment in the plant biomass and a strong depletion of ¹³CO₂ in the atmosphere during the labeling period. We conclude that combining our presented MSI workflow with stable isotopic labeling could therefore provide even more specific insight into in situ root exudation and for the characterization of metabolite pools (Arrivault et al., 2017).