3D (x-y-t) Raman imaging of tomato fruit cuticle: Microchemistry during development

Abstract The cuticle is a protective extracellular matrix that covers the above-ground epidermis of land plants. Here, we studied the cuticle of tomato (Solanum lycopersicum L.) fruits in situ using confocal Raman microscopy. Microsections from cuticles isolated at different developmental stages were scanned to visualize cuticle components with a spatial resolution of 342 nm by univariate and multivariate data analysis. Three main components, cutin, polysaccharides, and aromatics, were identified, with the latter exhibiting the strongest Raman scattering intensity. Phenolic acids and flavonoids were differentiated within the cuticle, and three schematic cuticle models were identified during development. Phenolic acids were found across the entire cuticle at the earliest stage of development, i.e. during the formation of the procuticle layer. Based on a mixture analysis with reference component spectra, the phenolic acids were identified as mainly esterified p-coumaric acid together with free p-hydroxybenzoic acid. During the cell expansion period of growth, phenolic acids accumulated in an outermost layer of the cuticle and in the middle region of the pegs. In these stages of development, cellulose and pectin were detected next to the inner cuticle region, close to the epidermal cell where flavonoid impregnation started during ripening. In the first ripening stage, chalconaringenin was observed, while methoxylated chalcones were chosen by the algorithm to fit the mature cuticle spectra. The colocation of carbohydrates, esterified p-coumaric acid, and methoxylated chalconaringenin suggests that the latter two link polysaccharide and cutin domains. Elucidating the different distribution of aromatics within the cuticle, suggests important functions: (1) overall impregnation conferring mechanical and thermal functions (2) the outermost phenolic acid layer displaying UV-B protection of the plant tissue.


Introduction
The plant cuticle is the outermost barrier, which covers the surface of aerial nonlignified organs (leaves, stems, flowers, and fruits) of land plants. This outer layer is responsible for some critical functions in the plant kingdom, such as providing mechanical support and defense against pathogens, radiation, or mechanical injuries, mediating water and gas exchange with the environment, and preventing from organ adhesion during the first stages of development (Martin and Rose, 2014). Tomato (Solanum lycopersicum L.) fruit cuticle is one of the most studied cuticles due to its economic importance, easy isolation, lack of stomata, and feasibility to perform mechanical and permeability tests (Lara et al., 2015).
The main component of the tomato fruit cuticle is the amorphous polymeric matrix named cutin, whose main monomers are 9(10),16-dihydroxyhexadecanoic acids (Heredia, 2003). As also cell wall-derived polysaccharides are present (López-Casado et al., 2007;Philippe et al., 2020), the cuticle has been defined as a cutinized cell wall (Dom ınguez et al., 2011). Additionally, waxes are present with around 3%-5% in the tomato fruit cuticle (Dom ınguez et al., 2009a) and phenolics with around 0.8%-7.5% depending on the stage of development (España et al., 2014b). Waxes can be deposited on the surface (epicuticular waxes) or embedded within the cuticle (intracuticular waxes). The phenolic fraction identified in the tomato fruit cuticle is composed of cinnamic acid derivatives such as p-coumaric acid and p-hydroxybenzoic acid, and the flavonoid chalconaringenin (Lara et al., 2019). Despite their presence as minor components, several physiologically relevant functions have been ascribed to the wax and phenolic fractions of the cuticle, such as water permeability, thermal properties, and mechanical performance (Schreiber and Schönherr, 1990;Casado and Heredia, 2001;Vogg et al., 2004;Khanal et al., 2013;España et al., 2014aEspaña et al., , 2014bJetter and Riederer, 2016;Ben ıtez et al., 2022). Additionally, the polysaccharide fraction has been shown to play an important mechanical role (López-Casado et al., 2007).
The chemical composition of the tomato fruit cuticle is known (Hunt and Baker, 1980;Vogg et al., 2004), but, there is a knowledge gap on the spatial distribution of each component due to the lack of spatial resolution of wet chemical methods, no stains and/or antibodies are available for specific components, and the fact that cutin itself masks other components and impedes penetration (Fernández et al., 2016;Segado et al., 2016). Determining the location and distribution of each component is crucial to understand their influence on the overall biophysical and macroscopic properties of the cuticle. Flavonoids have been shown to form clusters within the cuticle (Dom ınguez et al., 2009b). They have been postulated to act as fillers providing mechanical resistance (España et al., 2014a(España et al., , 2014b, similar to the role suggested for intracuticular waxes (Khanal et al., 2013). However, it remains undetermined whether these compounds are specifically located at a certain region of the cuticle or distributed throughout. Polysaccharides, on the other hand, have been traditionally assumed to be present in the inner side of the cuticle, whereas the upper section was understood as a cutin region devoid of polysaccharides (Fernández et al., 2016). In tomato, polysaccharides have been found to be an important fraction (25%-30% cuticle; Dom ınguez et al., 2008). However, immunolocation of polysaccharides within the tomato fruit cuticle has been unsuccessful (Segado et al., 2016). Only after a partial degradation of cutin, the polysaccharide domain has been detected in the cuticle of several species, showing a location across the entire cuticle width (Guzmán et al., 2014a(Guzmán et al., , 2014b. Confocal Raman microscopy (CRM) enables to reveal the distribution of plant components on the micro-scale (Gierlinger and Schwanninger, 2006;Richter et al., 2011;Mansfield et al., 2013;Mateu et al., 2016;Bock et al., 2021;Sasani et al., 2021) since it allows an in situ chemical resolution of around 300 nm (Prats-Mateu et al., 2018a). It is a rapid and nondestructive technique with simple sample preparation. The location of several components is possible simultaneously, but the multicomponent nature of biological samples results in broad and overlapping bands. To assign every Raman band to a specific component may become difficult, and previous knowledge of the chemical composition becomes helpful, as well as infrared spectroscopy studies. Infrared, as the complementary vibrational spectroscopy technique, has been widely used to study the composition and structural characteristics of the cuticle (Mazurek et al., 2013;Heredia-Guerrero et al., 2014;Farber et al., 2019aFarber et al., , 2019b and hence, numerous bands have been assigned to different cuticle components. The intensity of some of these bands can be used to image the assigned specific components by univariate approaches (Gierlinger, 2018). However, the band overlap in plant samples often needs to be overcome by multivariate approaches to retrieve a maximum of information on microchemical variation within the tissue (Gautam et al., 2015;Szyma nska-Chargot et al., 2016).
In this work, cuticle components were tracked on the microscale within the tomato fruit cuticle using CRM. We revealed the chemical microscale heterogeneity by univariate and multivariate data analysis. To find the purest component spectra, nonnegative matrix factorization (NMF) was used and to confirm specific components or reveal their mixture with other components, an orthogonal matching pursuit (OMP) algorithm in combination with a reference library was applied. With these methodologies we aim to (1) discriminate the main components, lipids, carbohydrates, and phenolics as well as (2) identify specific molecules e.g. benzoic, cinnamic acids, and flavonoids, and (3) image their quantity and location across the cuticle and (4) throughout cuticle development.

Results
Band assignment to different cuticle components and their changes during development Figure 1 shows average Raman spectra of cross sections of isolated cuticles from 8 to 55 days after anthesis (daa). Raman spectra of cuticles at the early stages of development show more noise and lower intensity, followed by a general increase in band intensity during development. Additionally, a slight shift of the Raman bands toward lower wavenumbers was observed during ripening ( Figure 1).
In the Raman spectra, the bands representative for specific components are highlighted ( Figure 1) and their band assignment is given in Table 1. The plant polyester cutin is reflected by the bands at 1,438, 1,278, 1,104, and 1,164 cm -1 (see Table 1 for assignments). Although they are representative of cutin, they can overlap with other components present in tomato fruit cuticle, mainly polysaccharides and waxes (Prinsloo et al., 2004;Heredia-Guerrero et al., 2014;Gierlinger, 2018;Farber et al., 2019a). Nevertheless, due to the higher Raman scattering intensity of polyester in comparison with signals from polysaccharides and the comparatively low amount of waxes in tomato fruit cuticle, mainly cutin contributes to these bands. Therefore, the term cutin will be used although a contribution of polysaccharides and waxes to this band cannot be excluded. Polysaccharides display strong bands at e.g. 1,095, 1,122 cm -1 (Cellulose, Wiley and Atalla, 1987) and between 853 and 856 cm -1 (pectin, Synytsya et al., 2003), but these signals have low intensity compared with those from other cuticle components such as cutin or phenolics and are consequently partially masked. Bands at 1,055, 1,060, 1,132, 1,292, and 1,438 cm -1 (Table 1) are characteristic of cuticle waxes, but their low amount ($3% of total weight; Dom ınguez et al., 2008) hampers their selective identification.
Phenolic compounds, on the other hand, exhibit strong signals in Raman spectroscopy. Tomato fruit cuticle accumulates two classes of phenolic compounds: hydroxycinnamic acid derivatives, mainly p-coumaric acid and p-hydroxybenzoic acid, present throughout development, and the flavonoid chalconaringenin that only accumulates during ripening and confers an orange-yellow color to the cuticle (España et al., 2014b). Signals centered at around 1,164 and 1,600 cm -1 (Figure 1, see Table 1 for assignment) are common for both types of phenolic compounds due to their structural similarities. The band centered around 1,630 cm -1 was selected to monitor phenolic acid derivatives, as it is strong in coumaric acid spectra and much weaker in flavonoid (chalconaringenin) spectra (Supplemental Figure S1). On the other hand, bands centered around 542 and 1,546 cm -1 appear only during ripening ( Figure 1) and can be exclusively attributed to chalconaringenin (Supplemental Figure S1). DFT calculations of free chalconaringenin revealed the nature of their bands centered at 546, 1,567, and 1,509 cm -1 in its experimental spectrum (Supplemental Figure S1). The band centered around 546 cm -1 (542 cm -1 in tomato fruit Raman spectrum) is attributed to stretching of the ring B coupled to C = C bending of the ring A and the vinyl group (see Supplemental Table S1). No band centered at 1,546 cm -1 (the emerging band in the ripe tomato fruit cuticle) was detected for free naringenin chalcone; nevertheless, in this region, two bands are exhibited by the flavonoid: one centered at 1,567 cm -1 assigned to the stretching of the C = O coupled with the in-plane O-H bending of the H bonding (González Moreno et al., 2021) and another centered at 1,509 cm -1 corresponded to the C = O stretching and O-H bending (H bond) coupled with other ring vibrations (see Supplemental Table S1 for more details).
Plotting the band area of $1,630, 1,546, and 1,438 cm -1 revealed semi-quantitatively the progression of the phenolic acids, flavonoids, and cutin, respectively ( Figure 2A). During the growing period, from 8 to 35 daa, the similar intensity of the band at 1,630 cm -1 suggested constant phenolic acid contents, while later during ripening an increase was observed. Also, the flavonoid band 1,546 cm -1 increased strongly during ripening from 40 to 55 daa, but was not present during the growing period. The 1,438 cm -1 band area increased during the 8-15 daa period and then remained more or less constant until red ripe (55 daa). To level out changes due to different focal plane or overlapping band intensities, ratios between the band areas of phenolic acids and cutin as well as between flavonoids and phenolic acids were additionally calculated ( Figure 2B). The phenolic acids/cutin ratio remained quite constant during development (including a small increase during ripening) and suggested a co-occurrence of the two from the very beginning. The flavonoid/phenolic acid ratio, on the other hand, clearly indicated that flavonoids mainly contributed to the increase in phenolic compounds during ripening and that their influence on the phenolic acid band (1,630 cm -1 ) is very small. Figure 1 Average Raman spectra of tomato fruit cuticle at different stages of development. Spectral signatures of "Cascada" fruit cuticles from 8 to 55 days after anthesis (daa) and their photographs on the right (taken from Segado et al., 2016). Band positions of characteristic component bands are highlighted (in cm -1 ) and their assignments are listed in Table 1.

Imaging the spatial distribution of tomato fruit cuticle components by band integration
To visualize the spatial distribution of the cuticle components within the cross-section of the cuticle, the bands assigned to cutin, phenolic acids, and flavonoid were integrated. The thickness of the 8 daa sample was below the limit of resolution of the technique and could not be studied with the imaging approach. Figure 3 shows the results of this univariate analysis in four selected stages of development. Raman images of the cutin depicted a homogenous distribution throughout the entire section of the cuticle in all stages of development, without any detectable differences  in intensity between the periclinal section of the cuticle and the pegs (cutinized anticlinal epidermal wall). Phenolic acids impregnate the cuticle already at 12 and 25 daa and accumulate in the outermost layer; most pronounced at 25 daa. Phenolic compounds have been reported in tomato fruit epicuticular waxes (Hunt and Baker, 1980). Thus, to determine the contribution of epicuticular waxes to the phenolic layer identified in this outermost region of the cuticle, the phenolic imaging was repeated after epicuticular wax removal. As the phenolic band neither disappeared nor was reduced after epicuticular wax extraction (Supplemental Figure S2), it is evident that phenolics are embedded within the cuticle. In later developmental stages, phenolic acids were also observed in the pegs, especially in their centers, which correspond to the outermost part of anticlinal walls and/or middle lamella. At red ripe stage, a higher band area indicates a higher accumulation in the pegs than on the outer periclinal section of the cuticle. Flavonoid accumulation showed a different picture than phenolic acid impregnation. It started later during ripening and was more pronounced at the innermost section of the cuticle at breaker stage (40 daa). At red ripe (55 daa) a dramatic increase in flavonoids was detected in the pegs. However, flavonoids showed higher intensities in the areas where, prior to ripening, phenolic acids showed low intensity, suggesting that mainly flavonoids impregnated these regions toward the epidermal layers.
Integration of three Raman bands characteristics of waxes (1,060, 1,132, and 1,292 cm -1 , Table 1) allowed to identify a very thin superficial wax layer in several stages of development (Supplemental Figure S3). Considering its outermost location and spectral wax characteristics, it corresponds to the epicuticular wax layer. The layer seemed to colocalize with the cutin, but its thickness was close or even smaller than the resolution limit and hence this colocalization should be taken with caution.

Detailed insights into microchemistry by multivariate approaches
To overcome the problem of Raman band overlaps in the band integration approach and to verify spectra and image interpretation, multivariate data analysis approaches were applied. First, the purest component spectra were identified at different stages of development by NMF (Figure 4, A-F) and the interpretation of these basis spectra verified by a mixture analysis approach (Figure 4, G-I). These purest basis spectra were used as representative component spectra in a least square fit at every pixel (basis analysis) throughout all developmental stages for comparative imaging (Supplemental Figure S4). Finally, in-depth mixture image analysis revealed specific components, including different aromatic components and carbohydrates by an OMP algorithm ( Figure 5). Optical microscopy and Raman images of tomato cuticle cross-sections at different stages of development based on univariate data analysis. "Cascada" cuticles at 12, 25, 40, and 55 daa are shown. Raman images are based on the integration of the 1,421-1,471 cm -1 band area assigned to cutin, 1,622-1,642 cm -1 band (1,618-1,638 cm -1 in 55 daa due to band shift) mainly attributed to phenolic acids and 1,536-1,566 cm -1 band corresponding to flavonoids. Color scale in Raman images is expressed in CCD cts (charge-coupled device counts); note different scaling of phenolic acids and flavonoids in the ripe stage (55 daa).
Finding the purest cuticle component spectra by NMF Due to its complex chemical nature, the average spectra of the cuticle include bands from all the components (Figure 1). Imaging with a high spatial resolution and at different developmental stages opens the chance to have at least some pixels at certain stages the one or the other component solely present. Among the different stages, the purest components have been revealed in 12 daa (phenolic acids), 25daa (cutin), and 40 daa (flavonoids; Figure 4, A-F), whereas others mainly resulted in mixed basis components (Supplemental Figure S5). In the very early stage at 12 daa, Raman spectra decomposition revealed two main basis component spectra ( Figure 4D). The first one corresponded to almost pure phenolic acids (turquoise in Figure 4D) with the characteristic bands at 1,633, 1,605, and 1,169 cm -1 . This "pure" phenolic acid spectrum was only present in the outer part of the cuticle. The second calculated component (blue in Figure 4D) mainly corresponded to cutin, but also 4 Component distribution maps and the corresponding basis spectra from unmixing analysis (NMF) of cuticles from selected stages of development. Basis spectra were assigned using a mixture analysis (OMP) based on a reference library. Component images of "Cascada" fruit cuticle cross-sections at (A) 12, (B) 25, and (C) 40 daa based on selected basis spectra (D-F) from NMF analysis. Turquoise represents spectra rich in phenolic acids (1 st line), whereby the purest phenolic acid spectrum was found at 12 daa (D). Blue represents cutin-rich spectra (2 nd in D, E, 3 rd in F), with a contribution of phenolic acids at 12 daa (D), but an almost pure cutin spectrum at 25 daa (E). The yellow spectrum (2 nd line in F) is enriched in flavonoids but with a contribution of phenolic acids and cutin (F). G-I, Mixture analysis of the three selected basis spectra to verify the purity and the composition based on reference spectra. Model fit spectra based on OMP are shown in gray. The reference spectra chosen to fit the basis spectra are shown below. A black canvas was added to the cross-sections for aesthetic reasons.
showed a band around 1,078 cm -1 that could be assigned to either polysaccharides or waxes. Hence, this basis spectrum represented cutin (with a contribution of polysaccharides and/or waxes), but still included a small contribution of phenolic acids. At 25 daa, two main components were also distinguished ( Figure 4E): one corresponding to a rich phenolic acids fraction (turquoise in Figure 4E) but with bands associated with cutin, whereas the second one (blue in Figure 4E) corresponded to an almost pure cutin spectrum with little to no contribution of phenolic acids. NMF results at 40 daa ( Figure 4F) revealed no "pure" component spectra, but comprised bands assigned to different components and thus the spectra represent mixtures of cuticle components. One is located on the top and is enriched in phenolic acids with a small contribution of cutin (turquoise in Figure 4F). The layer underneath (yellow in Figure 4F) corresponds to flavonoids with contributions of phenolic acids and cutin. The third component (blue in Figure 4F) was located on the innermost side of the cuticle and showed bands of the cutin, with little contribution of phenolic acids and flavonoids. None of them corresponded to a pure flavonoid spectrum, although this was the first stage in the flavonoids were detected. To derive a purer flavonoid spectrum for basis analysis, we, therefore, subtracted the cutin signals from the yellow spectrum ( Figure 4F; see "Materials and methods" section for details).
Verifying the purity and spectral signature of the basis spectra by OMP and reverse linear combination of spectra for component imaging The purity and composition of the basis spectra for phenolic acids and cutin from 12 and 25 daa, respectively, and the flavonoid spectrum from 40 daa were assessed by modeling them as linear combination of spectra acquired from reference compounds (Figure 4, G-I) by OMP . This analysis confirmed the three selected basis spectra as representatives of the three components: cutin, phenolic acids, and flavonoids (Figure 4, G-I). The representative cutin spectrum (blue in Figure 4E) was confirmed as cutin with small contribution of polysaccharides ( Figure 4H), while the other two (Figure 4, G and I) were elucidated as mixtures of different aromatics and still including small contributions of cutin and water. Yet, the basis, selected from 12 daa, (turquoise spectrum in Figure 4D), was dominated by phenolic acids and esters (Ethyl p-methoxycinnamate, phydroxybenzoic acid, see Figure 4G) and the one selected from 40 daa (yellow spectrum in Figure 4F) detected in a similar extent phenolics, cutin, and flavonoids (hesperidin methyl chalcone and flavokavain C, see Figure 4I). The spectrum related to the presence of flavonoids (yellow in Figure 4F) was also assessed after subtracting the cutin (Supplemental Figure S6), confirming the dominance of flavonoids in this basis spectrum.
In the next step, the verified purest component spectra were used to analyze the hyperspectral datasets of all developmental stages during the whole period of fruit growth and ripening. The three selected component spectra and a water spectrum were linearly combined at each pixel using a basis analysis to finally result in distribution maps of all four components (Supplemental Figure S4). Cutin (with a contribution of polysaccharides and/or waxes) was homogenously distributed throughout the cuticle in all developmental stages (blue in Supplemental Figure S4), whereas for phenolic acids (turquoise in Supplemental Figure S4) and flavonoids (yellow in Supplemental Figure S4) we confirmed the spatial and temporal changes as detected by band integration (Figure 3), but with more details (Supplemental Figure S4). Additionally, water was included in the fit (purple in Supplemental Figure S4) and revealed to accumulate nearby the inner side of the cuticle. The flavonoid component spectrum was not present in the datasets of earlier stages, but appeared in traces with the onset of ripening at 35 daa. At 40 daa, these compounds were clearly located underneath the phenolic acid layer and across the entire width of the pegs (see Supplemental Figure S4). With fruit ripening, flavonoids continued to accumulate across the entire cuticle. At red ripe, a contrasting pattern of phenolic acids and flavonoids was typical: phenolic acids close to the outer region and in the central region of the pegs, whereas the flavonoid signal was more intense underneath this phenolic acid layer and across the entire pegs.
In-depth hyperspectral dataset analysis by OMP to reveal specific components As the multicomponent nature of the cuticle was proven through all developmental stages and even on the microscale of 333 nm (pixel size; Figure 4, G-I), we applied the mixture analysis on the hyperspectral datasets. At every pixel, a model fit is searched for by the OMP algorithm based on our reference spectral dataset, including carbohydrates, aromatics, and lipids ( Figure 5). This analysis included the option of selectively locating the most relevant phenolics along with carbohydrates and cutin.
Among the phenolics detected throughout the development was p-coumaric acid, mainly in esterified form, and phydroxybenzoic acid ( Figure 5A). The two were colocated in a superficial layer and in the middle area of the pegs in the early stages of development, but the ester became more homogenous in the ripe stage ( Figure 5A). In the early unripe stages we could, moreover, locate carbohydrates (pectin and cellulose; Figure 5B) as well as cutin ( Figure 5C). While the latter one showed again a very homogenous distribution throughout the whole cuticle ( Figure 5C), the cellulose and pectin accumulated in contrary areas than the phenolic acids ( Figure 5B). However, the carbohydrate-rich regions toward the epidermis seem to coincide with flavonoid locations during ripening ( Figure 5D). Among the flavonoids, chalconaringenin was only detected in small amounts at 40 daa ( Figure 5D), whereas at most pixels, the cuticle spectra were better represented by methoxylated chalcones, such as flavokavain C and hesperidin methyl chalcone (see structures in Supplemental Figure S7 and their Raman spectra in Figure 4I). However, at the ripe state, the error of the fit increased and several observed band mismatches were observed. Although numerous flavonoids and aromatics were included in the library, it became more and more difficult to fit the pixel spectra by the reference spectra and, thus, pointed to more interactions in the mature stages.

Discussion
Using CRM, we successfully monitored topochemical changes of lipids, carbohydrates, and aromatic components during tomato fruit cuticle development. By combining uniand multivariate analysis we assigned and verified Raman bands and/or spectra to the main component groups and analyzed their spatial distribution as well as their changes from green to red stage (Figures 1-4). Especially, the mixture analysis combined with a comprehensive spectral reference library has provided detailed insights into the microdistribution of cutin, cellulose, pectin, and the diversity of aromatic molecules ( Figure 5). Finally, all insights were included in three schematic models to summarize our understanding of the tomato fruit cuticle and its development ( Figure 6).  Figure 6 Schematic representation of the spatial location of the main cuticle components throughout tomato fruit development. A, The earliest stages of development (8-12 daa) showed a cuticle enriched in phenolic acids (cyan). B, During most of the growing period (15-35 daa) phenolic acids were only located to the outer region of the cuticle and above a very thin superficial waxy layer was detected. C, During ripening, p-hydroxybenzoic acid preserves its initial location (cyan) while the ester of p-coumaric acid extends throughout the entire cuticle width (orange) and flavonoids showed a gradient with higher deposition toward the inner side of the cuticle (stripes). Cutin (with a contribution of cutin plus polysaccharides and/or waxes) is present across the entire cuticle width in all the stages of development (texture).

Cutin and polysaccharides: Gradual and/or uniform foundation of the cuticle?
A heterogeneous distribution of the two matrix polymers has been reported. Cutin is regarded to decrease toward the inner side of the cuticle, while at the same time polysaccharides increase (Fernández et al., 2016;Reynoud et al., 2021). Yet, Raman bands assigned to cutin (Figure 3), and also the model fits with cutin spectra (Figure 5; Supplemental Figure  S4) revealed no cutin intensity gradient in any of the stages of development. This is an indication that the amount of cutin is quite uniform within the tomato fruit cuticle. Waxes are embedded as intracuticular waxes and were additionally imaged as a very thin superficial layer (Supplemental Figure S3). The wax-specific bands are assigned to longchain alkanes or alcohols, compounds that have been reported in the epi-and intracuticular wax fractions of tomato fruit (Prinsloo et al., 2004;Vogg et al., 2004).
No strong Raman marker bands were found for carbohydrates; not for cellulose/hemicellulose and not for pectin and not at any pixel throughout all stages. As the Raman intensity is not only a matter of the amount of the molecules present, but also depends on the polarizability of the molecules, weaker Raman scatterers, like carbohydrates, can get overlaid by lipids and aromatic components. This is the reason why carbohydrates were only detected in the younger developmental stages ( Figure 5B), although they are reported to make up almost one-third of the cuticle (25%-30%; Dom ınguez et al., 2008). Due to the small contribution of carbohydrates to the overall cuticle Raman signal, no strong marker bands were detected and their distribution was only elucidated with the help of mixture analysis ( Figure 5B). With this approach we found cellulose and pectin in the younger stages especially in the pegs and toward the epidermal layer, agreeing with their previously reported location along cuticle transversal sections of some tree species (Guzmán et al., 2014a(Guzmán et al., , 2014b. In the mature stage, the higher background and signal of flavonoids mask the cutin bands partly and the carbohydrates completely. Further research to localize also polysaccharides and cutin in the mature red cuticle will be needed.

Tomato fruit procuticle is enriched in phenolic acids
Although phenolic components are reported to be around 0.8%-7.5% depending on the stage of development in the tomato fruit cuticle (España et al., 2014b), they are easily detected by Raman imaging due to their strong Raman scattering intensity resulting in numerous strong bands.
During the early stages of development, phenolics (mainly esterified p-coumaric acid at 12 daa according to Figure 5) can be located throughout the overall thickness of the procuticle. This initial and uniform procuticle is present in tomato fruit during the cell division period (Segado et al., 2016). Indeed, a small decrease in the phenolics/cutin ratio can be observed later on during the beginning of the cell expansion period ( Figure 2B) that agrees with the reduction in percentage of phenolics detected during the 15-30 daa period ($0.8%) compared to the previous stage (2.5% at 10 daa; data from España et al., 2014b). A similar percentage of phenolics was only achieved and exceeded during ripening. The increase in cuticle deposition during the expansion period can then be attributed to cutin with little participation of phenolics, thus decreasing the phenolics/cutin ratio. The notable accumulation of phenolics in the outer cuticle region remained during growth and exceeded also in the outermost layer of the cutinized anticlinal cell walls, that is the middle region of the so-called pegs. These phenolics were identified as a mixture of esterified p-coumaric acid and phydroxybenzoic acid. In several species, cinnamic acid derivatives have been identified esterified to the hydroxyl fatty acids of cutin (Cheng et al., 2013;Renault et al., 2017;Reynoud et al., 2021). In situ identification of p-coumaric acid as ethyl p-methoxycinnamate in the basis spectra and in the mixture analysis of different stages of development of tomato fruit cuticle supports the hypothesis of this molecule acting either as a nonlipid cutin monomer or as a link between the polysaccharide and cutin domains, esterified by both the acid and the hydroxyl groups.
We detected phenolic acids already in the earliest stages of development in the procuticle in high amounts. A fact, also interesting under the aspect that early land plants like different bryophytes have a high percentage of phenolics (35%) and thus a phenolic enriched cutin (Renault et al., 2017;Kong et al., 2020). This posits the idea that a high percentage of cuticle phenolics could not necessarily be associated with a specific plant group but with the initial procuticle stage and to the degree of cuticle deposition in later stages of development. Thus, in species where the cuticle would not further increase its deposition past the initial procuticle stage, or have a limited increase, a phenolic enriched cuticle could be detected, whereas in those species with notable cuticle increase after the procuticle deposition, the final result would be a cuticle with a lower percentage of phenolics. This hypothesis deserves further study as well as the link observed in tomato fruit between the procuticle and the cell division stage of growth and the cell expansion period with a massive cuticle deposition (Segado et al., 2020).
A remarkable degree of UV protection achieved by the cuticle has been reported and discussed in context with phenolic acids (Krauss et al., 1997;Ben ıtez et al., 2022;González Moreno et al., 2022). The tomato fruit cuticle was shown to block 95% of the incident UV-B light at the early stages and reaching 99% later in development, whereas in other species and organs, these results varied from an almost complete blockage to over 50%. Cinnamic acid derivatives, such as pcoumaric acid present in tomato fruit cuticle and in all the studied cuticles, considerably absorb UV-B light. Hence, these molecules give the cuticle special optical properties to protect plant tissues from harmful UV light. Thus, our identified outer layer of phenolic acids in tomato fruit cuticle throughout the development represents an optimal location to block the UV-B already at entry.

Phenolic acids and flavonoids accumulate during tomato fruit ripening
Also during ripening p-coumaric and p-hydroxybenzoic acids increase in the tomato fruit cuticle (España et al., 2014b). This is in agreement with the CRM results of the semi-quantitative mixture analysis, which showed about 4 times higher contributions of phenolic acids in the mature state than in the green state ( Figure 5A). Also in the mature stage, the OMP suggested ethyl trans-4-methoxycinnamate in esterified form and the free p-coumaric acid only at a few positions and to less extent (Figures 4, G and I and 5A). These results suggest that the huge majority of p-coumaric acid is present within the cuticle as an ester. It coexists during the whole development with the other main phenolic acid reported in the tomato fruit cuticle, the p-hydroxybenzoic acid.
However, the dramatic increase in cuticle phenolics during tomato ripening has been mainly ascribed to de novo synthesis of the flavonoid chalconaringenin (España et al., 2014a). This was proven by a 60%-70% reduction of the cuticle phenolics through transient silencing of CHALCONE SYNTHASE, which is responsible for chalconaringenin synthesis (España et al., 2014a). The expression of the next gene of the flavonoid pathway, CHALCONE ISOMERASE (CHI), is hardly detected during tomato fruit ripening (Bovy et al., 2002), hence mainly chalconaringenin is accumulated in the cuticle of ripe fruits (Hunt and Baker, 1980). Chemical isomerization between chalconaringenin and naringenin flavone (the product of CHI) has been reported (Kaltenbach et al., 2018), but recent quantum chemical analysis has shown that this chemical isomerization is very unlikely within the cuticle (González Moreno et al., 2021). Raman imaging together with univariate and multivariate analysis confirmed a higher amount of flavonoids in the cuticle during ripening (Figures 1-6). However, chalconaringenin was only selected by the OMP algorithm at 40 daa ( Figure 5), but not at ripe stages. The mismatch between the cuticle spectra and the chalconaringenin spectrum was also evident in direct comparison of the spectra. When we added also a methoxylated chalconaringenin and a methoxylated and glycosylated chalcone to the library, the OMP algorithm chose the two and the model fit explained the cuticle spectra better (see structures in Supplemental Figure S7 and their Raman spectra in Figure 4, G-I). Flavonoids located on the surface of plants have been mostly identified as aglycones, often with substituted methoxy groups (Onyilagha and Grotewold, 2004), a modification that increases molecule hydrophobicity. Hence, it is possible that the tomato fruit cuticle accumulates some methoxylated modification of chalconaringenin. But as the spectra of the ripe tomato did not fit the references perfectly, changes in the environment and even interaction with other cuticle components could be the reason behind the band differences in Raman spectra observed of pure chalconaringenin and that present in the cuticle of red ripe fruits. This means that chalconaringenin is either methoxylated, glycosylated, or covalently bond to other cuticle components over its A-ring OH-groups (at C4 and C6; the OH on C2 hydrogen bonds to the carbonyl oxygen). Impregnation with cinnamic acid derivatives and flavonoids confers mechanical resistance and structural rigidity to the cuticle during ripening (Ben ıtez et al., 2021(Ben ıtez et al., , 2022. Thus, the reported gradual changes in thermal and mechanical properties of the tomato fruit cuticle during ripening are mirrored in our results by the increasing deposition of these phenolics. At the ripe stage, the ester of p-coumaric acid was present throughout the entire width of the cuticle and pegs ( Figure 5A). Flavonoids were also distributed throughout the whole cuticle with the highest amounts toward the inner side of the cuticle ( Figure 5D). The distribution pattern across the entire cuticle supports their above-mentioned structural role as fillers that restrict the molecular motion of cutin chains. Our results revealed the carbohydrates in the earlier stages at locations ( Figure 5B) where the flavonoids start to impregnate the cuticle during ripening ( Figure 5C). Thus, flavonoids may finally fill the spaces within the more hydrophobic carbohydrate network and link them to the hydrophilic cutin polymer.
Three schematic models depict microchemistry of tomato fruit cuticles during development Based on Raman imaging of all cuticle components, we distinguish three different models during tomato cuticle development ( Figure 6). The first one corresponds to the procuticle stage in the very beginning, where phenolic acids (mainly as esterified p-coumaric acid) are present throughout the entire cuticle. The second distribution model was observed from 15 (beginning of the cell expansion period) to 35 daa (mature green stage), in the period with most fruit growth. In this stage, the cuticle showed a thin superficial waxy layer. This layer colocalized with cutin (plus a contribution of polysaccharides and/or waxes), an indication that it could be embedded within the cuticle. It was not readily identified in all the stages of development, most probably due to its thickness being close or lower to the resolution limit of CRM. Underneath the wax layer, a phenolic acid enriched cuticle layer was observed. This phenolic layer was not affected by epicuticular wax removal indicating that, at least, most of these phenolic acids are integral part of the cuticle. This phenolic acid layer was also located in the outermost region of the cutinized anticlinal cell walls. Beneath the phenolic layer, a much thicker layer corresponding to a cutin without phenolics extended to the inner side of the cuticle. Finally, during ripening, additional incorporation of phenolic acids and the deposition of flavonoids takes place. p-Hydroxybenzoic acid preserves the initial location of phenolics within the cuticle; but, p-coumaric acid (mainly as an ester) is incorporated into the inner cuticle layer, which was devoid of phenolics, thus covering the entire cuticle thickness and peg width. On the other hand, flavonoids show a gradient across the cuticle, with higher concentration toward the inner side of the cuticle, in closer contact with the epidermal cell wall, and within pegs outside the central region.
To summarize, the high Raman scattering intensity of aromatic components gives insights into the incorporation of phenolic acids and flavonoids during development. The dual location of phenolic acids identified during growth and later during ripening could be associated with different biological functions. Phenolic acids might act primarily as an outer photoprotective layer during growth and contribute to mechanical stiffening and strengthening of the cuticle during ripening, together with flavonoids. Cutin signal showed similar intensity across cuticle thickness as well as during development, whereas carbohydrates were due to their low Raman scattering only detected at early stages of development toward the inner side of the cuticle. Accumulation at regions, where later flavonoids start to impregnate the cuticle, suggests filling of hydrophilic regions with aromatics. Spatial colocations as well as spectral misfits with chalconaringenin point to interconnections of the flavonoids with other cuticle components.

Sample preparation and micro-sectioning
Tomato (Solanum lycopersicum L. "Cascada") fruit cuticles of different stages of development, from 8 daa to 55 daa (ripe red), were isolated using an enzymatic protocol (Petracek and Bukovac, 1995;España et al., 2014b). Samples were cross-sectioned into slices 10-lm thick by cryo-microtomy at -15 C (CM3050 S, Leica, Austria). The transversal sections were put onto microscope slides with a drop of D 2 O. They were covered with a cover slip 0.17-mm thick and sealed with nail polish to prevent water evaporation during measurement. Fruit cuticles from at least three different plants were studied for each developmental stage.

Confocal Raman microscopy
Raman spectra from isolated tomato cuticles were acquired by using CRM (alpha300RA, WITec GmbH, Germany). Samples were focused with a 100x oil immersion objective (numerical aperture (NA)=1.4, correction of the coverslip of 0.17 mm). A linear polarized coherent diode laser k ex = 785 nm (XTRAII, Toptica Photonics, Germany) was used as excitation line. The Raman signal was received by an optic multifiber (100 lm in diameter), transported into the spectrometer UHTS 300 (WITec, Germany; 600g.mm -1 grating, spectral resolution around 3.8 cm -1 ), and intensities measured using a CCD camera (DU401A BR DD, Andor, Belfast, Northern Ireland). The laser power was set at 150 mW and the integration time to 0.1 s. Spatial resolution of the measurement was calculated by r = 0.61k/NA, with r, k, and NA being the lateral resolution, excitation laser wavelength, and numerical aperture of the objective, respectively. Maximum spatial resolution achieved was around 342 nm. The size of evaluated scan areas was from 20 lm Â 10 lm (200 lm 2 ) to 85 lm Â 60 lm (5,100 lm 2 ). A minimum of three cross-sections from three different fruits of each stage of development have been measured to obtain robust results.

Raman band assignments
The assignment of some characteristic vibrational bands of main compounds in tomato fruit cuticle was done according to literature (Table 1) and density-functional theory (DFT) calculations in the case of chalconaringenin (predominant flavonoid extracted and characterized in tomato fruit cuticle) due to some bands not being described in literature (see Supplemental Table S1 for more details). It is worth clarifying that some wavenumbers from the literature are from infrared rather than Raman spectroscopy. Although both techniques differ in the way they excite molecular vibrations, their respective band positions are equivalent.
DFT calculations were performed using Gaussian 16 (Frisch et al., 2016) software. Raman spectra were calculated at B3LYP with 6-31 + G(d,p) basis set, applying a scale factor of 0.9613 to overcome experimental-theoretical deviations. Band assignments were carried out by GaussView software.

Spectral preprocessing and univariate approaches
Cosmic rays, which are very intense signals generated by the interaction of energetic particles with the CCD camera, were removed before processing acquired data using the software Project Four 4.1 (WITec, Germany). Pixels, which show characteristic Raman bands of cuticle, were masked in each replicate, and the average spectrum of each stage of development was calculated (throughout their three replicates) using Project Four 4.1. No normalization process was applied to these data in order to show the most realistic comparative overview. Rayleigh lines were removed, and the spectral range limited to 300-1,700 cm -1 in OPUS (Bruker, Germany). Spectra were integrated at characteristic bands using the same software. At 8 daa, chemical imaging of the sample was not possible due to the extreme thinness of the cuticle; at 10 daa, an improvement in signal/noise ratio was required to obtain a well-defined spectrum. For this reason, the ratio of pixels recovered per lm of the sample was changed from 60 px/20 lm used for all the rest of the samples to 100 px/20 lm.
Further spectral preprocessing actions before band integration, NMF and mixture analysis were smoothing and background subtraction using Project Four 4.1. The integration of specific bands (see Table 1) of each cuticular component (cutin, polysaccharides, waxes, phenolic compounds, and flavonoids), allowed to locate them along the depth of the cuticle.

Unmixing algorithm: NMF
As band integration is focusing on individual characteristic peaks only, it shows limitations due to overlapping bands in the spectrum. Spectral unmixing aims to separate measured datasets (D), such as Raman maps, into pure components (S T , also termed endmembers or basis) and their respective coefficients (C). Using this method in chemometrics, the pure components often refer to different chemical compounds while the coefficients represent concentration profiles or abundance maps of the respective compounds (van Stokkum et al., 2009). Using mathematical notation, this basic problem can be stated as follows: There are several spectral unmixing methods available, such as NMF, vertex component analysis (VCA), or multivariate curve resolution alternating least squares (MCR-ALS), which differ in their optimization approach and constraints (Prats-Mateu et al., 2018b).
NMF (Lee and Seung, 1999) might result in a more realistic representation of the spatial abundance of different compounds than univariate approaches. Runs with different numbers of basis from 2 to 5 were done and, after evaluating the pure basis spectra, the most meaningful result was kept. The maximum number of iterations was set to 10.000. From the basis spectra of all NMF runs among the different images, a reference library representing cutin, phenolics, flavonoids, and water to be used in basis analysis was compiled. To achieve a more selective flavonoid location, consecutive subtractions of cutin signature were applied to flavonoid spectrum until observing any negative band along the spectra, which was at a coefficient of 0.35 (pink spectrum in Supplemental Figure S6).
The interpretation of the obtained basis was confirmed using measured spectra from reference compounds (Figure 4, G-I). To do so, an OMP (Pati et al., 1993) was used to select predictors to be used in a linear model to approximate the respective basis spectrum.

Basis analysis
Basis analysis was used to acquire the linear combination of the Raman basis spectra (obtained by NMF) which best represents the spectrum of each pixel of the sample. These estimated coefficients are calculated by a least square fit, and are proportional to the abundance of each compound in the mixture (Schmidt et al., 2005). As each pixel can be separately analyzed, the application of basis analysis in confocal Raman measurements allows to assess the distribution of compounds in space getting color maps for each component.

Spectral reference library and mixture analysis
Isolated tomato fruit cuticle, regardless of the stage of development, accumulates phenolic compounds (España et al., 2014b). Thus, to obtain a reference spectrum of cutin, a recombinant inbred line (RIL115) derived from a cross between the domesticated tomato and a wild species (Alba et al., 2009) was employed. This line has the characteristic of having virtually zero phenolic compounds in its cuticle (Dom ınguez et al., 2009a) thanks to a specific combination of a number of genomic regions related to the accumulation of phenolic compounds in the cuticle (Barraj et al., 2021). Besides the cutin reference, the library included 148 entities, including water, aromatics, lipids, minerals, and carbohydrates.
To analyze the multicomponent nature of the cuticle spectra, we modeled them as a linear combination of measured reference spectra using the OMP (Pati et al., 1993) The OMP is an iterative and fast approach that can be applied to spectra and spectral images. At each iteration, it seeks for a component of the spectral reference library, which is best correlated with the residual of the linear combination of references selected in the preceding iterations (the experimental spectrum being the residual from an empty model, used in the first iteration) and adds it to the predictor set. A stopping criterion (e.g. a threshold for residual error or a maximal number of iterations) stops the search for additional members. The predictor set of the final model holds the selected constituents of the mixture, and the coefficients indicate the abundance of the respective compounds.

Supplemental data
The following materials are available in the online version of this article.
Supplemental Figure S1. Raman spectra of the free conformations of p-coumaric acid and chalconaringenin.
Supplemental Figure S2. Raman images corresponding to a univariate data analysis of phenolic acids in tomato cuticle cross-sections before (left) and after (right) epicuticular wax removal.
Supplemental Figure S3. Raman images based on integration of a band related to the presence of a superficial waxy layer.
Supplemental Figure S4. Basis analysis using the purest spectra obtained from the NMF algorithm.
Supplemental Figure S5. Distribution maps and Raman spectra obtained by NMF at 10 and 55 daa.
Supplemental Figure S6. Mixture analysis of the basis spectra of the flavonoid spectra after subtracting the contribution of cutin (pink spectrum).
Supplemental Figure S7. Chemical structure of phenolic compounds mentioned in this work.
Supplemental Table S1. Raman band assignment of chalconaringenin in gas phase according to DFT calculations.