Differential localization of flavonoid glucosides in an aquatic plant implicates different functions under abiotic stress

Flavonoids may mediate UV protection in plants either by screening of harmful radiation or by minimizing the resulting oxidative stress. To help distinguish between these alternatives, more precise knowledge of flavonoid distribution is needed. We used confocal laser scanning microscopy (cLSM) with the “ emission fingerprinting ” feature to study the cellular and subcellular distribution of flavonoid glucosides in the giant duckweed ( Spirodela polyrhiza ), and investigated the fitness effects of these compounds under natural UV radiation and copper sulphate addition (oxidative stress) using common garden experiments indoors and outdoors. cLSM “ emission fingerprinting ” allowed us to individually visualize the major dihydroxylated B-ring-substituted flavonoids, luteolin 7-O-glucoside and luteolin 8-C-glucoside, in cross-sections of the photosynthetic organs. While luteolin 8-C-glucoside accumulated mostly in the vacuoles and chloroplasts of mesophyll cells, luteolin 7-O-glucoside was predominantly found in the vacuoles of epidermal cells. In congruence with its cellular distribution, the mesophyll-associated luteolin 8-C-glucoside increased plant fitness under copper sulphate addition but not under natural UV light treatment, whereas the epidermis-associated luteolin 7-O-glucoside tended to increase fitness under both stresses across chemically diverse genotypes. Taken together, we demonstrate that individual flavonoid glucosides have distinct cellular and subcellular locations and promote duckweed fitness under different abiotic stresses.

Flavonoids are known to serve diverse functions in plants including protection against insect predators, attraction of pollinators, defence against microbes, pollen germination and modification of development by involvement in auxin transport (Mouradov & Spangenberg, 2014;Panche, Diwan, & Chandra, 2016). It is generally agreed that flavonoids also function to protect plants against UV and other radiation, but whether these compounds function as screening pigments or antioxidants has been controversial (Agati et al., 2009(Agati et al., , 2013Burchard, Bilger, & Weissenböck, 2000;Di Ferdinando, Brunetti, Agati, & Tattini, 2014;Harborne & Williams, 2000;Hernández, Alegre, Van Breusegem, & Munné-Bosch, 2009;Mouradov & Spangenberg, 2014;Robson, Klem, Urban, & Jansen, 2015). At one time, the role of flavonoids was considered to be primarily as filters of UV radiation, consistent with their frequent location in cell walls and vacuoles of epidermis cells or trichomes (Barnes et al., 2016;Hutzler et al., 1998;Schnitzler et al., 1996;Strack et al., 1988). However, more recent studies challenged this view, as flavonoids accumulate also in vacuoles of mesophyll cells (Agati et al., 2002;Liu, Gitz, & Mcclure, 1995), chloroplasts (Agati, Matteini, Goti, & Tattini, 2007;Saunders & McClure, 1976) and in the nucleus (Feucht, Schmid, & Treutter, 2014). Furthermore, flavonoids are relatively poor UV-B absorbers compared to other phenylpropanoid classes such as hydroxycinnamic acids (Agati et al., 2013). In addition, the ratio of flavonoids to hydroxycinnamates strongly increases upon exposure to UV-B or strong sunlight: for instance, the flavonol quercetin replaced hydroxycinnamic acid derivatives during shade-to-sun transition (Agati et al., 2009(Agati et al., , 2013. These observations suggest that UV-B screening is not the sole function for flavonoids. In particular, flavonoids may attenuate oxidative stress that is caused by UV radiation (Ravanat, Douki, & Cadet, 2001). The role of flavonoids in alleviating oxidative stress is further supported by the observation that flavonoid biosynthesis is up-regulated by a plethora of abiotic and biotic stresses that all lead to the generation of reactive oxygen species (ROS) Babu et al., 2003;Vogt, Gülz, & Reznik, 1991). For example, copper, a metal that has dramatically increased in the environment due to the present industrial and agricultural practices (Fernandes & Henriques, 1991), generates ROS by auto-oxidation and metaldependent Fenton reactions (Michalak, 2006;Sytar et al., 2013). Flavonoids, particularly the ortho-dihydroxylated B-ring-substituted flavonoids, are assumed to benefit plant performance under copper addition by scavenging ROS and suppressing ROS-formation as chelating agents (Babu et al., 2003;Brown, Khodr, Hider, & Rice-evans, 1998;Rice-Evans, Miller, & Paganga, 1996). The structural diversity and the cellular and subcellular distribution of individual flavonoids may strongly affect their ability to function as radiation filters or antioxidants.
Here, we aimed to explore the individual localization of the four major flavonoid glucosides in S. polyrhiza under copper sulphate and natural sunlight treatment to learn about their tissue-specific and subcellular distribution. We hypothesized that variation in the cellular or subcellular localization of specific flavonoids with different anti-oxidative capacity but similar UV absorption would indicate that the flavonoids act at least partially as antioxidants. We first developed a cLSM method to determine the cellular and subcellular distribution of individual flavonoids without interference from one another. We then determined the  (Appenroth, Teller, & Horn, 1996). The Spirodela polyrhiza genotype 7,498 originating from North Carolina (United States) was used for cLSM analysis. A full list of the registered accessions involved in the fitness assays is provided in supplemental Table S1 (Zhao, Appenroth, Landesman, Salmeán, & Lam, 2012).

| Statistical analyses
All statistical analyses were performed in R version 3.3.1 (R Core Team, 2016). Pairwise comparisons were performed with the multcomp package (Hothorn, Bretz, & Westfall, 2008). Linear mixed model analyses were performed using the lme4 package (Bates, Mächler, Bolker, & Walker, 2015). More details are given in the experimental section below with further references to detailed information in the supplemental materials. Each synthetic standard (5 mM in DMSO) was dissolved 1:1 with 0.25% 2-aminoethyl diphenylborinate (2-APB) resulting in a final concentration of 2.5 mM and was excited at 405 nm. The chlorophyll spectrum was extracted from an unstained cross-section and excited at 405 nm. Images were acquired in the λ-mode and emission spectra were extracted via manual landmarks in the cLSM software and stored in the software database to subsequently enable "emission fingerprinting" in 2-APB-stained cross-sections of S. polyrhiza [Colour figure can be viewed at wileyonlinelibrary.com] concentration of 2.5 mM each. Using a confocal laser scanning microscope (ZEISS 880, Carl Zeiss, Oberkochen, Germany) 40 μl of 2-APB mixed standards were placed on a microscopy slide, covered with a 22×22 mm coverslip, excited at 405 nm and emission was detected between 410 and 695 nm in 9 mm bins (λ-mode). Firstly, an image in the λ-mode was acquired to record an emission spectrum of each compound.
The acquired spectra were stored in the spectra database (which included the autofluorescence spectrum of chlorophyll from a crosssection excited with 405 nm) and subsequently used for the online emission fingerprinting mode. In this mode, the fluorescence emission from the cross-sections, containing all four flavonoids, excited by 405 nm is compared to the previously acquired and stored spectra of the standards and sorted/unmixed into the respective channels based on the similarity of the spectra pixel-by-pixel. The emission fingerprinting tool is based on spectral differences among compounds such as curve shape/skewness and the positions of peaks, but not intensity differences. The microscope was equipped with an EC-Plan-Neofluar 10×/0.3 dry type objective and a C-Apochromat 40×/1.20 W Korr M27 water immersion type objective and ZEN 2.1 [black, 64-bit] software. Identical settings were used for all images and channels used for fluorescence quantification: excitation wavelength was 405 nm using a laser diode, 10× objective, 21% transmission, 625 gain, pinhole set to 1 Airy Unit (39 μm). The focus for the image recordings was chosen (a) by finding the maximum when using a solution drop or (b) by focusing on the chloroplasts for in vivo images of S. polyrhiza cross-sections (details in Text S2). In order to test the robustness of our recorded emission spectra, we performed additional tests extracting spectra under varying pH conditions, 2-APB concentrations and DMSO levels. Furthermore, we tested for the association of compound concentration and fluorescence intensities (details in Text S2).

| Visualization and quantification of secondary fluorescence across tissue layers of the S. polyrhiza frond
To qualitatively and quantitatively investigate the distribution of the flavonoid glucosides across the S. polyrhiza frond, we visualized the compounds in cross-sections using cLSM with the established fluorescence database. Mature fronds were freehand sectioned along a distinct peripheral axis ( Figure S1) with a section thickness of about 0.5 mm. Cross-sections were stained with 0.25% 2-APB for 1 min and immediately analysed without washing the sections under the same cLSM settings as described above. Chloroplasts seen by autofluorescence of chlorophyll were chosen to define the focal plane for all recorded in vivo images. The "online emission fingerprinting" feature of cLSM was used to separate the mixed signals of stained cross-sections pixel-by-pixel using the entire emission spectrum of each of the previously stored reference spectra.
Fluorescence intensities in the adaxial epidermis, mesophyll and abaxial epidermis were estimated by randomly arranging five distinct circular regions of interest (ROI) within each of these tissue layers using ImageJ. The ROI size (adjusted to the size of each layer) was 365, 5,795 and 365 square microns for the adaxial epidermis, mesophyll and abaxial epidermis, respectively. In this work, mesophyll refers to the area between the upper epidermis and beginning of aerenchyma in cross-sections. ROIs were placed randomly across each tissue layer except that aerenchyma was avoided. The ROI calculation output is the mean grey value per pixel within a ROI ranging from 0 to 255. Averaging the means of five ROIs within one layer resulted in the final fluorescence intensity per replicate. For statistical analysis of this experiment, see Text S3.

| Verification of cLSM-based flavonoid visualization in vivo
To verify our visualization method in planta, we first investigated the effect of the phenylalanine ammonia lyse (PAL)-inhibitor, 2-aminoidane-2-phosphonic acid (AIP), an inhibitor of the first step on the phenylpropanoid pathway (Gitz et al., 2004), in S. polyrhiza on both fluorescence signals and HPLC-based flavonoid content analysis. For details on the plant handling, HPLC and statistical analysis, see Text S4.

| Effect of frond age on flavonoid glucoside accumulation
To investigate the effect of frond age on flavonoid glucoside accumulation, 12 mother fronds with a small attached daughter frond each were divided up among three bowls and grown for 14 days under control conditions. On days 0, 1, 2, 3, 5, 7 and 14, one daughter frond from each bowl was harvested for HPLC analysis (remaining fronds were grown as backups). The medium was replaced once a week and additional generations of fronds were discarded to ensure space and nutrients for fronds to be analysed. To test for differences in flavonoid content over the lifespan of fronds, we compared mean flavonoid levels between different days using ANOVA and Tukey's post hoc test.

| Investigation of subcellular localization of flavonoids
To investigate the subcellular localization of the flavonoids, we additionally stained cross-sections with the vacuole specific staining agent, neutral red. Cross-sections of mature plants were incubated for 3 min in 0.01% neutral red (3-amino-7-dimethylamino-2-methylphenazine hydrochloride, Sigma Aldrich, Steinheim, Germany) in purified water (w/v). Sections were then washed three times in tap water for 2 min prior to cLSM microscopy. Samples were excited with a 543 nm HeNe laser (40× objective, 0.6 zoom).

| Tissue-specific induction of flavonoids under copper sulphate treatment
To investigate the magnitude and tissue layer specificity of flavonoid glucoside induction under copper sulphate addition, we performed cLSM and HPLC analyses on plants growing in the presence and absence of CuSO 4 . Three single fronds were placed into 250 ml plastic bowls filled with 180 ml full nutrient medium with and without 10 μM CuSO 4 (n = 12). After 5 days, one mature granddaughter fronds ( Figure S1) from half of the bowls of each treatment was used for cLSM analysis as described above. At the same time, one mature granddaughter of each of the remaining bowls was harvested and analysed by HPLC as described above. In addition, the fresh mass of all emerged fronds was recorded. For statistical analysis procedure, see Text S5.

| Induction of flavonoids under natural UV light
To investigate the effect of natural midsummer UV radiation on flavonoid content and accumulation pattern, we performed outdoor experiments under natural UV light exposure with UV-shielded conditions as control. Plants were transferred from the growth chamber to a sunexposed field site in Jena, Germany, in July 2017. Plants were grown in full nutrient medium-filled 200 ml plastic cups (diameter 5/6.5 cm (bottom/top, height 8.5 cm, Pöppelmann, Lohne, Germany) that were fitted into the cut holes of white polyvinyl chloride inserts (3 mm thickness) floating inside water-filled 10 L buckets (weather data see Table S2). To prevent excessive heating of the medium by the sun, buckets were buried to the rim in soil. UV radiation was manipulated by covering the buckets with either UV-blocking (UV Gallery100, Sandrock Kunststoffe, Germany) or UV-transmitting poly(methyl methacrylate) sheets (GS 2458, Sandrock Kunststoffe, Germany) leaving a 2-3 cm gap to allow for airflow. Sheets differed mostly in UV-A but also in UV-B transmission ( Figure S1 in [Xu et al., 2019]). To allow for acclimation of the plants to outdoor conditions prior to the start of the experiments, plants were cultivated under UV-blocking sheets that were covered with two layers of green plastic foil, which were successively removed after 1 and 2 days, respectively. Fronds were pre-cultured outdoors for another 5 days under UV-blocking sheets.
Due to enhanced evaporation outdoors, medium was replaced twice a week during pre-cultivation and experiment. Subsequently, three single fronds were placed into full nutrient medium-filled plastic cups floating inside UV-transmitting and UV-blocking buckets (n = 10).
After 7 days, one mature granddaughter frond from half of the replicates was used for cLSM as described above and one granddaughter of the remaining replicates were used for HPLC analysis as described above, except that frozen fronds were ground for 1.5 min in 5 ml tubes since fronds grown outdoors were firmer. Biomass production of all emerged fronds was recorded. For statistical analysis procedure, see Text S6.
2.10 | Fitness assay of genotypes exposed to copper sulphate To test whether individual flavonoid content correlates to plant fitness under copper sulphate treatment, 53 worldwide distributed S. polyrhiza genotypes were grown in the presence and absence of copper sulphate. Genotypes, which were selected based on their geographically dispersed origin (Table S1), were pre-cultivated for 7 days in 250 ml Erlenmeyer flasks containing 100 ml full nutrient medium in the above-mentioned climate chamber. For each genotype, five mature plants with a small daughter frond each were placed into 250 ml plastic bowls that were filled with 150 ml full nutrient medium with and without 10 μM CuSO 4 (n = 3) in a climate chamber operating under the following conditions: 26 C constant, 16:8 hr light:dark, illumination at 160 μmol m −2 s −1 supplied by metal halide lamps (mt400DL, EYE Iwasaki). The bowls were covered with transparent and perforated plastic lids and a white fleece (Agrarvlies, 17 g/m 2 ) to avoid water condensation on the lid. After 7 days, plants were harvested, weighed and approximately 50 mg fresh mass frozen in liquid nitrogen. Samples were stored at −20 C until further analysis.
Methanol extractions and HPLC analysis to measure flavonoid concentrations were performed as described above. Detailed information on statistical analysis can be found in Text S7.

| Fitness assay of genotypes exposed to ambient UV light
To test whether individual flavonoids benefit plant fitness under natural UV light, 38 worldwide distributed S. polyrhiza genotypes, including 33 genotypes used in the copper sulphate experiment (Table S1) Table S2). Genotype position was randomized in each box. The medium was exchanged after 1 week. After 2 weeks, fronds were col-   Figure S9). In addition, the fluorescence of lut 8-C-glc overlapped with that of the chloroplasts (Figure 3, Figure S9). Apigenin glucosides on the contrary were associated with chloroplasts only (Figure 3, Figure S6). Views of the adaxial epidermis of neutral red and 2-APB-stained sections revealed a lack of fluorescence signals in cell wall regions ( Figure S10).

| Copper sulphate addition but not UV radiation induces flavonoid glucoside concentrations in a tissue-specific manner
To investigate whether the copper sulphate addition affects flavonoid glucoside accumulation in S. polyrhiza, we analysed flavonoid glucoside concentrations using HPLC and determined the individual distribution patterns of compounds using cLSM in fronds grown after copper sulphate treatment indoors. Under the experimental conditions used, copper sulphate addition reduced plant biomass production by 54% ( Figure S11

| Certain flavonoid glucosides correlate with resistance to copper sulphate addition and UV stress
The distinct cellular and subcellular distribution of the four major flavonoids of S. polyrhiza, as well as their differential and partially tissuespecific induction under copper sulphate addition suggest that the compounds may have differential fitness effects under natural UV light exposure and copper sulphate addition. We tested this hypothesis by growing chemically diverse S. polyrhiza genotypes under these conditions. We standardized growth across genotypes by comparing the mean biomass production of exposed plants (with copper sulphate or with UV light) of each genotype to the mean biomass production of non-exposed plants (without copper sulphate addition; without UV light) of each genotype ("relative plant fitness"; values below one indicate growth suppression, whereas values above one imply growth promoting effects under copper and UV light, respectively).
After 7 days of growth indoors, copper sulphate addition reduced biomass production by average of 30% across 53 S. polyrhiza genotypes (p = 2.2e −16 , paired t test, Figure S13). The constitutive concentrations of both luteolin glucosides positively correlated to relative plant fitness under copper sulphate addition across the genotypes (lut 7-O-glc, p = .046; lut 8-C-glc p = .04, linear models, Figure 6a), F I G U R E 3 Subcellular distribution of individual flavonoid glucosides based on cLSM false-colour-images of a (a) 2-APB and a (b) neutral red stained Spirodela polyrhiza cross-section. Both luteolin glucosides accumulated mostly in vacuoles, and luteolin 8-C-glucoside was associated with chloroplasts. Apigenin glucosides were almost exclusively associated with chloroplasts. cLSM was performed with the "online emission fingerprinting" feature again based on the stored emission spectra of the respective standards shown in Figure 1b. 2-APB-stained cross-sections were excited with 405 nm; in another cross-section, vacuoles were stained with neutral red and excited with 543 nm. Brightness and contrast of all images were adapted in ImageJ for illustrative purpose. 2-APB = 2-aminoethyl diphenylborinate. Additional images of merged channels are shown in Figure S9 [Colour figure can be viewed at wileyonlinelibrary.com] whereas no correlations were observed for the apigenin glucosides (p > .4, linear models, Figure 6a), despite the constitutive concentrations of all flavonoid glucosides being highly correlated to each other ( Figure S14, p < .001, Pearson's moment correlations). Total constitutive flavonoid glucoside concentration was weakly correlated to relative fitness (p = .06, linear model, Figure S15). Similar patterns for relative plant fitness and individual and total flavonoid glucoside concentrations were found when the induced metabolite concentrations were analysed ( Figure S16a). In absolute terms, both luteolin glucosides were negatively correlated to biomass production under control conditions (lut 7-O-glc: p = 5.72e −07 , lut 8-C-glc: p = 1.89e −05 , linear models) and only weakly negatively (lut-7-O-glc, p = .099) or not correlated to biomass production (lut 8-C-glc, p = .39) under copper sulphate addition (interaction metabolite concentration * copper p < .0014 for both luteolin glucosides, linear models, Figure S17). Ap 7-O-glc was negatively correlated to plant biomass production regardless of treatment (linear models, p = .017 with copper sulphate addition, p = .002 without copper sulphate addition, interaction ap 7-Oglc * treatment = 0.13, Figure S17). Ap 8-C-glc was not correlated to biomass production in the presence or absence of copper sulphate addition (p > .76 with and without copper sulphate addition, interaction ap 8-C-glc * treatment p = 0.77, Figure S17). F I G U R E 5 Induction of flavonoid glucosides in different tissues of Spirodela polyrhiza after treatment with copper sulphate (a) natural UV light (b). Copper but not UV treatment had tissue-specific effects on flavonoid accumulation, particularly on luteolin 8-C-glucoside levels. Plants were grown in the absence and presence of copper sulphate or natural UV light until the granddaughter generation had matured, which was subsequently free hand-sectioned and 2-APB-stained for cLSM analysis. p values of likelihood ratio tests comparing two linear mixed effect models that differed in the interaction term of treatment and tissue layer are displayed above each figure. Wilcoxon-Mann-Whitney tests were then used to test for differences of individual flavonoid fluorescence intensities between treatments within a tissue layer. n = 6 (a), n = 5 (b), ctr = control [Colour figure can be viewed at wileyonlinelibrary.com] We next assessed the correlation of the individual flavonoids to plant fitness in plants growing under UV-exposed and -shielded conditions outdoors. After 14 days of growth outdoors, natural UV light reduced biomass production an average of 8% across 38 S. polyrhiza genotypes ( Figure S18, p = .013, paired t test). Ultraviolet light reduced the biomass production of most genotypes, but positive effects of UV light on plant growth were also observed in about 30% of the genotypes ( Figure S19). The concentration of the adaxial epidermis-enriched lut 7-O-glc (measured under UV-shielded conditions) tended to positively correlate to relative plant fitness under UV light (p = .08, linear model, Figure 6b), whereas no correlations were observed for the other flavonoid glucosides (p > .2, linear models, Figure 6b). Total flavonoid concentration was not correlated to relative plant fitness (p = .16, linear model, Figure S20). Similar but nonsignificant associations between relative plant fitness and individual or total flavonoid concentrations were found when metabolite concentrations under UV-exposed were analysed ( Figure S16b). In absolute terms, both luteolin glucosides were negatively correlated to plant biomass production in the presence and absence of UV light (lut 7-O-gluc: p < .0005, lut 8-C-glc: p < .06, linear models; p > .82 for interaction of metabolite concentration * treatment, linear models; Figure S21). Ap 7-O-glc concentration was not correlated to biomass production regardless of UV treatment (p > .18 linear models, p = .64 for interaction metabolite concentration * treatment, linear model, Figure S21). Ap 8-C-glc was positively associated with plant fitness in the presence and absence of UV light (p < .002, linear models; p = .57 for interaction metabolite concentration * treatment, linear model, Figure S21).
Taken together, these data provide evidence that lut 8-C-glc and possibly lut 7-O-glc benefit plant fitness under copper sulphate addition, while lut 7-O-glc may alleviate the negative effects of UV radiation.

| DISCUSSION
The ways in which flavonoids protect plants under UV radiation have long been controversial with opinion divided between UV screening and anti-oxidative mechanisms. Here, we show that both possibilities F I G U R E 6 Correlation of individual flavonoid glucosides to relative plant fitness across Spirodela polyrhiza genotypes in the presence and absence of copper sulphate (a) and natural UV light (b). Relative fitness is the mean biomass of exposed plants (A: with copper sulphate; B: with natural UV light) of each genotype compared to the mean biomass production of control plants (A: without copper sulphate; B: without natural UV light) of each genotype. Data points below the horizontal dashed line indicate reduction in biomass under copper excess and UV light, respectively. Fifty-three genotypes were grown indoors for 7 days in the presence and absence of 10 μM copper sulphate, whereas 38 genotypes were grown outdoors for 14 days in the presence and absence of natural UV light. Each data point represents the mean of one genotype.  (Saunders & McClure, 1976). In contrast to our work, lut 7-O-glc was previously identified in isolated chloroplast preparations of S. polyrhiza (Saunders & McClure, 1976). However, a microscopic survey of fresh, unwashed sections, such as performed here, may be more accurate for localization of low molecular weight metabolites than subcellular fractionation due to the tendency of many metabolites to adhere to membrane fractions. Flavonoid glucosides in S. polyrhiza may be additionally associated with other cell compartments in concentration levels not detectable with our methods, especially as our protocols are less sensitive to the apigenin glucosides and other non-orthodihydroxylated B-ring flavonoids that do not form adducts with 2-APB at typical cellular concentrations (Agati et al., 2009).
Uneven distribution of flavonoid glucosides across plant tissue layers was reported before, for example, for Phillyrea latifolia (Agati et al., 2002(Agati et al., , 2012 and Ligustrum vulgare (Tattini et al., 2004), and several studies suggest that individual flavonoids occur in different subcellular compartments using HPLC, cLSM and organelle isolation (Agati et al., 2012;Hernández et al., 2009). Here, we showed that closely related flavone glucosides had partially contrasting tissue and subcellular locations, which suggest different functions in the plant.

| The cLSM software enables visualization of individual flavonoids in planta
In demonstrating the differential localization of individual flavonoid glucosides, we showed for the first time, to our knowledge, that individual flavonoid glucosides can be selectively visualized in planta using the flavonoid complexing agent 2-APB, which forms a fluorescent adduct. We employed the linear unmixing feature of the cLSM software to separate a spectral signal recorded from a single or a group of pixels containing various fluorophores into a separate intensity signal for each compound (Talamond, Verdeil, & Conéjéro, 2015). In Coffea species, this unmixing feature was used to visualize distinct fluorescence signals of the phenolic compounds chlorogenic acid and mangiferin (Conéjéro, Noirot, Talamond, & Verdeil, 2014). In our study, lut 8-C-glc and lut 7-O-glc were successfully distinguished in crosssections of S. polyrhiza fronds despite largely overlapping emission curves that would have hindered selective visualization using traditional cLSM techniques. Furthermore, we were able to locate monohydroxylated B-ring flavonoid glucosides, which are also known to fluoresce when stained with 2-APB albeit upon excitation with 365 nm instead of 405 nm as used in our experiments (Agati et al., 2002). Experiments to determine whether the linear unmixing algorithms are sufficient to individually trace other flavonoids will help to assess the suitability of the described methods for other organisms.

| Individual duckweed flavonoid glucosides may have separate roles in resistance to copper and UV stress
Luteolin 7-O-and 8-C-glucoside share many chemical features that may be important for copper and UV stress resistance. Firstly, these compounds have the same aglycone moieties, and thus identical conjugated double bond systems and arrangements of functional groups except glucose, resulting in nearly identical UV-absorption spectra ( Figure S22). Secondly, both compounds share the same orthodihydroxylated B-ring patterns, which are important for their radical scavenging capacity (Burda & Oleszek, 2001;Sekher, Chan, O'Brien, & Rice-Evans, 2001), Cu + -chelating activity (Brown et al., 1998) and the suppression of the Fenton reaction (Cheng & Breen, 2000). Although the ortho-dihydroxylated B-ring (catechol group) is very important for the antioxidant properties of these compounds, the glycosylation site still may affect antioxidant capacity. Indeed, based on 2,2-diphenyl-1-picrylhydrazyl (DPPH), 2,2-azinobis (3-ethyl-benzothiazoline-6-sulfonic acid (ABTS) and FRAP assays, luteolin-8-C glucoside has higher radical scavenging and antioxidant potential than its 7-Ocounterpart (Zhang et al., 2014).
Indeed, the two luteolin glucosides appear to have partially different functions in S. polyrhiza fronds (Table 1). In a survey of chemically diverse S. polyrhiza genotypes, both compounds were correlated with high fitness under copper sulphate addition and both were induced by this treatment. This is in line with previous studies showing that ortho-dihydroxylated B-ring-substituted flavonoids have high chelating and antioxidant potential that may reduce heavy metal stress (Brown et al., 1998;Mira et al. 2002), and that flavonoid deficient A. thaliana plants are more susceptible to cadmium and zinc stress than controls (Keilig and Ludwig-Müller, 2009). However, lut 8-C-glc accumulated mainly in the mesophyll, while lut 7-O-glc accumulated mostly in the adaxial epidermis. Although both compounds increased in the mesophyll under copper sulphate treatment, the fact that lut 8-C-glc is the dominant compound points to its importance for copper stress resistance. At the subcellular level, lut 8-C-gluc was stored mainly in vacuoles and was also associated with chloroplasts, in which flavonoids have been suggested to reduce hydrogen peroxide (H 2 O 2 ) and to maintain ROS concentrations at sub-lethal levels in combination with vacuolar class III peroxidases (Ferreres et al., 2011). The sites of lut 8-C-glc in vacuoles and chloroplasts are also locations where copper from the environment can be accumulated (Printz, Lutts, Hausman, & Sergeant, 2016). However, in both compartments, the pH is compatible with the formation of flavonoid-copper complexes (Berkowitz & Wu, 1993;Malešev & Kunti c, 2007), allowing the chelation of copper leading to copper stress resistance. The localization and induction patterns of lut 8-C-glc, its high radical scavenging potential relative to lut 7-O glc (Zhang et al., 2014) and its correlation with fitness under copper sulphate addition suggest that this compound alleviates oxidative stress that may be induced by copper treatment.
In contrast to lut 8-C-gluc, lut 7-O-glc not only correlated to plant fitness under copper sulphate but also showed a tendency to be correlated to fitness under natural UV light exposure in a survey of chemically diverse S. polyrhiza genotypes. In addition, lut 7-O-glc accumulated not in the mesophyll, but mostly in the adaxial epidermis, the cell layer that receives most UV radiation through its exposure to sunlight. This O-glucoside was mainly stored in vacuoles, and so may shield the underlying cell layers from UV radiation as vacuoles occupy almost the entire surface of the epidermis (Figure 3, Figure S10). On the other hand, lut-7-O-glc may also act by scavenging light-induced ROS. Thus, this compound may protect the adaxial epidermal cells from oxidative damage both through reducing the formation of ROS (UV screening) and through quenching the radicals (antioxidant capacity). Protection of the epidermal cells may be particularly important in S. polyrhiza due to the proximity of the underlying reproductive cells.
In recent years, some researchers suggested that the UVprotecting role of flavonoids is due more to their antioxidant than UVscreening properties since though the maximum absorption of flavonoids lies in the UV-A region, they are induced by light in UV-B region and visible ranges also (Agati et al., 2012;Brunetti, Fini, Sebastiani, Gori, & Tattini, 2018). However, the fact that lut 7-O-glc, primarily localized in the adaxial epidermis, was associated with greater fitness under both UV radiation and copper sulphate treatment, but that lut 8-C-glc, localized in the mesophyll, was associated with greater fitness only under copper sulphate addition, indicates that individual flavonoids may have antioxidant and/or UV-screening functions, depending on their location. The localization of apigenin glucosides, which are considered to have low anti-oxidative potential (Rice-Evans et al., 1996), in the chloroplast, a cellular compartment that generates oxidative stress appears contradictive and raises questions about the role of apigenin derivatives. Targeted manipulation of apigenin derivatives through genetic engineering may provide insights into the compounds' function in vivo.
Whatever the mechanism of protection, flavonoids are known to alleviate the effects of UV radiation indoors and under natural settings outdoors Jansen & Bornman, 2012;Kliebenstein, 2004;Robson et al., 2015). Experiments that specifically manipulate the major S. polyrhiza flavonoids through genetic engineering or chemical modification may provide further insights into the ecological relevance of these compounds under different stresses and their functional specificity.
As one of the most common and phylogenetically widespread groups of secondary metabolites, flavonoids have received much attention in recent years due to their potentially positive effects on T A B L E 1 Summary of tissue-specific distribution, subcellular occurrence, induction and fitness benefits of the major flavonoid glucosides in Spirodela polyrhiza Note: Classification of the tissue-specific distribution flavonoids in fronds was based on Tukey's HSD tests (genotype 7498, from Figure 2). Flavonoid induction under stress is based on Student's t test comparing induced and non-induced flavonoid levels determined by HPLC analysis (genotype 7,498, from Figure 4). p values of fitness benefits under stress relate to linear models correlating relative fitness to individual metabolite concentration across worldwide distributed S. polyrhiza genotypes (from Figure 6). ++, high concentrations; +, low concentrations; ns, not significant.
human health (Panche et al., 2016). Yet, the ecological roles of flavonoids remain controversial, partially due to their structural diversity, different accumulation patterns and varying abundance, all of which affect their biological activity (Hernández et al., 2009). By using a combination of cLSM and common garden experiments with chemically diverse S. polyrhiza genotypes, we showed that chemically similar flavonoids have distinct tissue-specific and intra-cellular distributions that may be associated with distinct functions in abiotic stress resistance. Knowledge of the localization of plant metabolites can be very valuable in understanding their ecological relevance.

ACKNOWLEDGMENTS
We would like to thank Daniel Veit for crafting equipment for the growth chamber and the set-up of the outdoor equipment, Grit