Although the leaf cuticle is notably ablated by IR-MALDESI, its hydrophobic nature prevents deposition of a consistent exogenous ice layer, which is often incorporated for cryosectioned mammalian tissues. Since it cannot be evenly distributed across the leaf surface, ice formation should be prevented to avoid matrix effects. Figure 2 includes camera images across critical points of the freezing and analysis process for the first frozen leaf sample (Fig. 2A) and second frozen leaf sample (Fig. 2B). The CMC gel was clear in the initial images after the samples were mounted and the enclosure was purged, but it became opaque in the next set of images, indicating that it froze when the stage was cooled. The primary veins of the leaflets also became more opaque, indicating that the leaf tissue froze successfully. Prior to analysis of the region-of-interest (ROI) on the first leaflet (ROI 1), there was no visual evidence of ice deposition on the leaves, but after analysis of the first leaflet in each pair, scattered ice crystals became visible only on the first leaflet. Similarly, ice crystals only became visible on the second leaflet (ROI 2) after it was analyzed. Ablation through the cuticle likely allowed remaining water vapor to crystallize onto newly-hydrophilic areas where the cuticle had been ablated. Ice was not visible on the second leaflet of each pair until after both leaflets were analyzed, so matrix effects during MSI analysis should have been minimal.

Camera images throughout IR-MALDESI analysis of frozen tissue for A the first pair of frozen clover leaflets, and B the second pair of frozen clover leaflets. From left to right, images were taken of the samples after the enclosure was purged and they were mounted on the CMC, after they were frozen via the Peltier-cooled stage, after ROI 1 (first leaflet, outlined in green) was analyzed, after ROI 2 (second leaflet) was analyzed, and once the sample was thawed after analysis

Although scattered ice crystals may form over the course of analysis and result in matrix effects, leaves also continuously lose water over the course of IR-MALDESI analysis of fresh tissue—water can be lost through severed vascular tissue and can freely evaporate from surfaces after the laser penetrates the cuticle. Since ablation relies on endogenous water, water loss would similarly result in matrix effects. Additionally, the resulting tissue shrinkage can pull the sample surface out of focus of the laser since AzC are determined for the whole sample prior to analysis; thus, water loss during fresh tissue analysis can also result in inconsistent ablation. Considering that both methods are predicted to result in minor matrix effects, frozen tissue analysis may be preferable since (1) it preserves the original distribution of water as the endogenous matrix, (2) it maintains effective AzC, and (3) randomly scattered ice crystals should result in less systematic bias than consistent water loss over time.

To enable increased spatial resolution while maintaining undersampling, it is crucial to reduce the diameter of ablation spots. In conventional IR-MALDESI analysis of fresh leaf tissue, high water content and tissue heterogeneity result in inconsistent ablation and large ablation spots, reducing both biological accuracy and spatial resolution. Applying a reflective objective can significantly reduce ablation spot diameter [36], and AzC can improve the overall consistency of ablation [19], but there remains room for improvement. In IR-MALDESI MSI of mammalian tissue, the reflective objective originally reduced ablation spot diameter from ~ 150 to ~ 55 µm [41] and then to 20 µm with a novel 2.94-µm laser [46]; however, the average ablation spot diameter of fresh leaf tissue has only reached ~ 146 µm for IR-MALDESI [36] and ~ 32 µm for UV-MALDESI [37]. In this study, the tissues were intentionally undersampled to ensure that the ablation spot diameters could be accurately measured under the microscope after analysis. In Fig. 3A, the ablation spots from the frozen analysis were visibly smaller than those in Fig. 3B from fresh analysis. Additionally, higher magnification revealed less thermal damage around the edges of the ablation craters in Fig. 3C for frozen analysis compared to Fig. 3D for fresh analysis. As shown in Fig. 3E, the frozen tissue analysis method significantly reduced the ablation spot diameter (p < 2.2 × 10−16 at 95% CI). When analyzing fresh tissue mounted on tape, the average diameter was 65.4 ± 19.5 µm, but the average diameter for frozen tissue on CMC was 43.9 ± 23.3 µm. Interestingly, this finding differs from the original study on mammalian tissue by Joignant et al., which found no significant difference between spot diameters on thawed and frozen tissue with the reflective objective [41]. This discrepancy is attributed to the differences in tissue composition between leaf and mammalian samples.

Images compare ablation spots at 5X magnification for A frozen analysis and B fresh analysis, and 10 × magnification for C frozen and D fresh analysis. E Differences in ablation spot diameter in violin plots and overlaid boxplots for the fresh (left) and frozen (right) methods. The Welch’s paired t-test gave p < 2.2 × 10−16 at the 95% confidence level

Frozen analysis had minimal impact on the classes of metabolites detected. In Fig. 4A, which investigates the known T. repens metabolites detected by each method in relation to metabolic pathways, both methods detected the same number of alkaloids and cyanogenic glycosides, but fresh tissue analysis detected a slightly higher total number of terpenoids, shikimates/phenylpropanoids, and amino acids. Detection of terpenoids appeared most impacted by the method of analysis, and the class of terpenoid further impacted detection; in Fig. 4B, frozen analysis detected slightly more monoterpenoids and gibberellins, while fresh analysis detected more oleanane triterpenoids. Within the shikimate and phenylpropanoid pathway in Fig. 4C, the only superclasses with differences in detection were flavonoids and phenylpropanoids. Phenylpropanoids were only detected in fresh analysis, and fresh analysis also detected slightly more flavonoids, particularly flavanols and flavan-3-ols (Fig. 4D). Isoflavonoids were the most consistently detected superclass, and detection was consistent between methods and across all classes (Fig. 4E). Overall, frozen analysis appears to be beneficial for detecting monoterpenoids and gibberellins but less effective for detecting oleanane triterpenoids. This result warrants further investigation into differences in the character of ablation plumes between fresh and frozen analysis.

Number of T. repens metabolites detected by fresh (green) and frozen (blue) analysis. A Analytes are classified by major pathways, and highly detected categories including B terpenoids and C shikimates and phenylpropanoids are further subdivided. The two largest classes of shikimates and phenylpropanoids, D flavonoids and E isoflavonoids, are further subdivided. Metabolites were compiled from the KNApSAcK database using the search “Trifolium repens” and were considered detected if at least one adduct ([M + H]+, [M + Na]+, and/or [M + K]+) was detected with little-to-no off-tissue signal for at least one sample. Images with spotty spatial localization were omitted

Although the total number of metabolites detected was similar across all classes, the frozen analysis was overall slightly less sensitive than the fresh analysis as shown in the volcano plot in Fig. 5. This plot evaluates ion abundance normalized to the average calculated volume of ablation spots for each method. Of the metabolites demonstrating a significant fold change in abundance between methods, 13/16 were less abundant in the frozen analysis. While frozen IR-MALDESI analysis is known to enhance detection of metabolites in cryosectioned mammalian tissue [47], the increased abundance is a product of the external ice matrix, which is avoided in this method due to potential matrix effects from inconsistent deposition across the cuticle. As such, the decrease in sensitivity for frozen leaf tissue is likely an artifact of the decreased diameter of the ablation spots. The inverse relationship between ion abundance and spatial resolution is well-documented in MSI analysis—increasing spatial resolution by decreasing the diameter of ablation spots results in decreased abundance due to the smaller amount of material sampled from each spot [41, 47 ]. As with any MSI method, advantages and limitations must be weighed for individual experiments; in this case, frozen tissue analysis enables increased spatial resolution but slightly decreases overall ion abundance.

Volcano plot of the abundance of T. repens metabolites normalized to average volume ablated for frozen (blue) vs. fresh (green) methods. Features include the [M + H]+, [M + Na]+, and [M + K]+ adducts of metabolites from the KNApSAcK database search for “Trifolium repens,” but ion images were generated and screened for each feature to remove those with high off-tissue abundance (particularly background from tape and/or CMC), spotty spatial localization, and/or low abundance across all samples. In this conservative estimate, 71/198 features remain. Data and tentative identifications for significant features are provided in Table SI. For practical applications, the volcano plot and data table for raw abundance values are also provided in Fig. S3 and Table SII, respectively

The trend toward decreased ion abundance for frozen analysis could be explained by the decreased ablation diameter; however, it may also be influenced by the intentional quenching of metabolic processes. Interestingly, most features that demonstrated a significant reduction in ion abundance were sodiated (7/13), and one feature, tentatively identified as soyasapogenol C, was significantly more abundant in the fresh method in its sodiated form but significantly more abundant in the frozen method in its protonated form. This discrepancy suggests that results may be influenced by inconsistent salt concentrations; thus, metabolites with a significant fold change in the same direction for at least two adducts were considered with greater confidence. Only one feature, tentatively identified as kaempferol 3-O-galactoside, met this criterion. Kaempferol 3-O-galactoside has a documented relationship to stress response in plants [48], as do many of the other tentatively identified features [49,50,51,52,53]. For many stress-induced defense metabolites, significant accumulation is only documented to occur on a timescale far exceeding that of these experiments, but most time studies analyze bulk extractions and study a limited number of metabolites and species using a limited number of time points. As such, it is likely that the trend toward decreased ion abundance results from a combination of both quenched metabolic processes and decreased ablation diameter, but it is difficult to ascribe this trend to a single cause. To more directly explore metabolic differences resulting from stress, the effects on several well-studied metabolites crucial to early stress responses were evaluated in Fig. 6 by comparing ion abundances in leaflets analyzed chronologically. For each sample, differences in ion abundances between the leaflets analyzed first and second could indicate a stress response; each leaflet that was analyzed second had approximately 30 min to respond to stress while the first leaflet was being analyzed.

A In an early response to wounding, plants activate the phenylpropanoid pathway to produce lignin and/or secondary metabolites for defense. This pathway begins by converting phenylalanine to cinnamic acid and then converting cinnamic acid to p-coumaric acid. B The average abundances of the protonated adducts of phenylalanine (m/z 166.0863), cinnamic acid (m/z 149.0597), and p-coumaric acid (m/z 165.0546) in leaflet one (analyzed first) and leaflet two (analyzed second) of each sample were plotted for fresh (left, green) and frozen (right, blue) analyses. C For fresh (left, green) and frozen (right, blue) analyses, log2FC values were plotted for the change in average normalized abundance between leaflets one and two of each sample. Normalization was conducted voxel-by-voxel for (1) cinnamic acid normalized to phenylalanine (dark), (2) p-coumaric acid normalized to phenylalanine (medium), and (3) p-coumaric acid normalized to cinnamic acid (light). This normalization highlights areas of increased metabolism and/or storage of early phenylpropanoid metabolites

Mechanical wounding of leaves initiates stress signaling responses via reactive oxygen species (ROS) and activation of the jasmonate (JA) signaling pathway [54]. While neither of these responses could be measured directly (ROS are outside of the m/z range and JA signaling metabolites overlapped with background isomers/isobars), most metabolomics studies focus on specialized secondary metabolites such as flavonoids. Flavonoids are synthesized through the phenylpropanoid pathway, and JA signaling is known to upregulate genes involved in this pathway, sometimes in a matter of minutes [32, 55 ]. Since activation of the phenylpropanoid pathway results from mechanical wounding and affects metabolites commonly evaluated in plant metabolomics studies, effects on this pathway were considered most relevant. The phenylpropanoid pathway begins with the deamination of phenylalanine to cinnamic acid and subsequent hydroxylation of cinnamic acid to p-coumaric acid; the plant then uses these precursors to synthesize monolignols and/or specialized secondary metabolites for defense [30] as summarized in Fig. 6A. In Fig. 6B, the average abundances of phenylalanine, cinnamic acid, and p-coumaric acid were calculated for each leaflet of each sample analyzed by each method; representative spectra are available in Fig. S4. Differences between leaflet one and leaflet two were not consistent across the two samples analyzed by the fresh method, but the average abundance consistently decreased for all three metabolites in both samples analyzed by the frozen method. Freezing quenches metabolic processes, so this is most likely an artifact of the method itself and may be related to scattered deposition of ice crystals over the course of analysis; in IR-MALDESI analysis of mammalian tissue with the reflective objective, frozen and thawed tissue had comparable ablation diameters, but frozen tissue with an ice matrix had a smaller ablation diameter [41]. As such, scattered ice crystals building up over the course of analysis may have led to decreased ablation diameter in the affected areas, resulting in decreased ion abundance.

To more directly evaluate the conversion of phenylalanine to cinnamic acid and p-coumaric acid, the abundances of cinnamic acid and p-coumaric acid were normalized to the abundance of phenylalanine at each location. Changes in the average normalized abundances between leaflets are summarized for all samples in a fold change plot in Fig. 6C. The log2FC was positive for both fresh samples for both cinnamic acid (0.15, 0.29) and p-coumaric acid (0.49, 1.00), while the log2FC was inconsistent between frozen samples for both cinnamic acid (0.05, − 0.09) and p-coumaric acid (− 0.24, 0.05), suggesting that more phenylalanine was converted to cinnamic acid and p-coumaric acid between leaflets one and two in fresh analysis. This difference suggests that the frozen method prevented early metabolic changes from stress over the course of MSI analysis.

A final concern for the new method is delocalization, wherein metabolites migrate during sample preparation, reducing the biological accuracy of the resulting ion images. Freeze–thaw cycles are a main concern: freezing causes cells to burst, and subsequent thawing allows the cell contents to migrate [14]. Leaf tissue is water-rich, so freezing and thawing can result in significant delocalization. In embedded samples, hydrophilic metabolites can also delocalize by diffusing into the embedding medium [56], although the hydrophobic barrier of the leaf cuticle may offer some protection compared to other sample types. A 2018 study by Li et al. found 10% gelatin to be the optimal embedding medium for reducing delocalization in ginkgo leaf cross-sections [14]; however, gelatin must be heated to maintain its viscosity. Since heating would be counterproductive for the purposes of this protocol, CMC was selected to mount the sample since it was the second-best option for preventing delocalization in Li et al., 2018. No thawing was visible during analysis, and the samples appeared to stay structurally intact as seen in Fig. 1, so potential sources of delocalization in this protocol should be limited to diffusion into the CMC gel used to mount the sample and migration of metabolites during ice crystal formation, particularly in the vascular tissue.

To check for potential delocalization, ion images were generated for three representative lipophilic small metabolites in Fig. 7 and three representative hydrophilic small metabolites in Fig. 8. The images were TIC-normalized to highlight differences in localization patterns rather than abundance. For each metabolite, patterns of spatial distribution were compared between methods. The fresh method provides a baseline for the localization of metabolites, since the sample was never frozen and was mounted on tape rather than CMC. The patterns of spatial localization were conserved between the fresh and frozen methods and align with current literature, indicating that the structural integrity of the sample and localization of metabolites were maintained.

TIC-normalized ion images of three lipophilic small metabolites (left column) from fresh (middle column) and frozen (right column) MSI analyses. The analytes are listed in order of increasing predicted logP from top to bottom. The name, molecular formula, adduct, m/z, class, predicted logP, and structure are provided in the left column. In each set of four ion images: the first sample is on the top row, and the second sample is on the bottom row; the leaflet analyzed first is in the left column, and the leaflet analyzed second is in the right column

TIC-normalized ion images of three hydrophilic small metabolites (left column) from fresh (middle column) and frozen (right column) MSI analyses. The analytes are listed in order of decreasing logP from top to bottom. The name, molecular formula, adduct, m/z, class, predicted logP, and structure are provided in the left column. In each set of four ion images: the first sample is on the top row, and the second sample is on the bottom row; the leaflet analyzed first is in the left column, and the leaflet analyzed second is in the right column

In Fig. 7, 2-furoic acid has a low positive predicted logP (XLOGP3 = 0.5), making it the least lipophilic of the lipophilic metabolites evaluated. It appeared to be distributed throughout the tissue and was most abundant in the veins. This pattern was conserved across both methods and aligns with literature indicating that it is localized to vascular tissue [57] and related to water tolerance [58]. Vascular tissue has a higher water content than the surrounding lamina, so consistently resolving this pattern is particularly promising, especially since 2-furoic acid is the least lipophilic metabolite—the relative hydrophilicity and localization to tissue with high water content suggest it would be the most likely to suffer from delocalization due to freeze–thaw cycles or ice crystal formation. A more lipophilic metabolite, 6-hydroxykaempferol (XLOGP3 = 1.5), was also most abundant in vascular tissue, but was much more abundant in primary veins. The smaller veins were harder to differentiate in the frozen method; however, the high abundance in primary veins suggests that this is an artifact of inconsistent detection from the overall lower abundance in frozen analysis. In both methods, the most lipophilic of the metabolites, formononetin (XLOGP3-AA = 2.8), was distributed across most of the tissue but is highly abundant in scattered areas. This pattern makes sense in the context of formononetin accumulation as a long-term stress response in areas wounded by insects and pathogens [59]. The consistency of these spatial localization patterns across methods supports that frozen analysis does not cause delocalization of lipophilic metabolites.

Patterns of spatial localization for increasingly hydrophilic amino acids were also similar between the fresh and frozen methods. In Fig. 8, the least hydrophilic amino acid, L-tryptophan (XLOGP3-AA =  − 1.1), was widely distributed across the lamina and was also highly abundant in certain areas. Like many amino acids, L-tryptophan is broadly involved in plant metabolism, but it is also important for crosstalk between stress hormones and biosynthesis of metabolites that defend against stress [60]; therefore, this pattern of distribution is expected, and the areas of high abundance likely correlate to long-term stress responses. This spatial pattern was consistent between methods. The two more hydrophilic amino acids L-leucine/L-isoleucine (XLOGP3-AA =  − 1.5) and L-threonine (XLOGP3-AA =  − 2.9), were also widely distributed across the lamina but showed even lower relative abundance in the vascular tissue. These amino acids are similarly broadly involved in metabolism and known to accumulate in response to stress [61], so this spatial distribution pattern aligns with literature and was also conserved between methods, although L-threonine was less abundant overall, resulting in less consistent detection. The similar patterns of spatial distribution of these amino acids between the fresh and frozen methods further support that even hydrophilic small molecules experience minimal delocalization.