Spatiotemporal Study of Galactolipid Biosynthesis in Duckweed with Mass Spectrometry Imaging and in vivo Isotope Labeling

: Isotope labeling coupled with mass spectrometry imaging (MSI) presents a potent strategy for elucidating the dynamics of metabolism in cellular resolution, yet its application to plant systems is scarce. It has the potential to


Introduction
Over the last two decades, metabolomics has elucidated the metabolic responses of plants to various perturbations and provided a deeper understanding of metabolic networks (Alseekh and Fernie 2018).Further, many gene functions were able to be annotated through the metabolomics of knock-out mutants, quantitative trait loci mapping, and genome-wide association studies.Most of these studies are based on chromatographic separation followed by mass spectrometry (MS) analysis of whole tissue extracts and are unable to distinguish metabolites from different cell types.Additionally, metabolite concentrations alone do not directly probe metabolic activities (Jang et al. 2018).Two new strategies have recently emerged to enlighten the spatial or temporal dynamics of plant metabolism, which was not possible with traditional metabolomics tools.One is isotope labeling and tracing labeled metabolites, which was used to annotate metabolites (Huang et al. 2014), elucidate pathway structures, and analyze metabolite flux (Allen et al. 2015;Wang et al. 2018).The other is mass spectrometry imaging (MSI), which visualizes each metabolite directly on tissues at single-cell resolution (Lee et al. 2012).Both techniques rely on MS when a stable isotope is used, but the combination of the two, namely MSI with in vivo isotope labeling which we call MSIi, has been rarely applied to plant systems.
There are two notable studies of MSIi in plants.We have previously visualized free amino acids in maize root sections grown in 15 NH 4 Cl medium to differentiate external nitrogen from those transported from seeds.This study demonstrated the genotypic difference in amino acid localization and their indifference to the nitrogen source (O'Neill and Lee 2020).Visualization of 13 C-labeled phosphatidylcholine (PC) species in developing Brassica seeds (camelina and pennycress) was achieved by Romsdahl and coworkers by feeding siliques with 13 C-glucose (Romsdahl et al. 2021).They revealed a greater 13 C-labeling in cotyledons than in embryonic axis, and in PC species with saturated and longer acyl chains.Other MSIi studies in plants include 15 N-labeling of Catharanthus roseus to assist nitrogen-containing metabolite identifications (Nakabayashi et al. 2020) and the labeling of lemna and tomato with D 4 -and 13 C 9 -tyrosine to study tyrosine-derived metabolic network (Feldberg et al. 2018).
To further explore the potential of MSIi, here we adopt D 2 O labeling to study the galactolipid biosynthesis in Lemna minor.Unlike other isotope labeling, D 2 O is a global labeling agent in plants as all hydrogen atoms are originated from water, and "fixed" during photosynthesis (Nett et al. 2018).Heavy water (D 2 O) labeling is previously used in plants to produce deuterated biomass in annual ryegrass (Evans et al. 2014) and Lemna minor (Evans et al. 2019), to investigate de novo synthesis of volatile terpenes in Achryanthes bidentata (Tamogami et al. 2013), to measure turnover rates of Arabidopsis proteins (Yang et al. 2010), and to measure biosynthetic rate of the Downloaded from https://academic.oup.com/pcp/advance-article/doi/10.1093/pcp/pcae032/7642521 by guest on 18 April 2024 cytokinin class of plant hormones in Arabidopsis thaliana (Astot et al. 2000) and various natural products in medicinal plants (Nett et al. 2018).However, it has not been used in MSIi of plants, although used in mouse to visualize rapidly growing tumor region (Louie et al. 2013).
As an aqua plant, duckweed, Lemna, is very attractive for MSIi using D 2 O. High D 2 O concentration is toxic to all organisms, but Lemna species can grow in up to 65% D 2 O with supplementation (Cope et al. 1965).It can avoid problems associated with D 2 O, such as germination rupture or root elongation since it grows by budding from fronds (Evans and Shah 2015).Evans and coworkers have successfully cultivated L. minor in 50-60% D 2 O and studied morphological and biochemical properties of D-labeled plants (Evans et al. 2019).Fatty acid and protein syntheses are commonly studied with D 2 O as an isotopic tracer to measure pathway activities (Jang et al. 2018), but they often require saponification or hydrolysis, not adequate for MSI with direct sampling on tissue surfaces.Instead, galactolipids, the most abundant chloroplast membrane lipids, are chosen in this study because of their high abundance in matrix-assisted laser desorption/ionization (MALDI)-MS, a technical platform used in this study for MSI.A partial D 2 O labeling (e.g., 50%) leads to low signals for each isotopologue due to the binomial distribution of H-vs D-labeling; as a result, it was essential to study highly abundant lipids in this first application of D 2 O labeling for MSIi.
Chloroplasts are home of thylakoids where photosynthesis and many essential biosynthesis occur in plants.Two galactolipids are the major components of thylakoid membrane lipids, monogalactosyldiacylglycerol (MGDG) and digalactosyldiacylglycerol (DGDG).Most steps in galactolipid biosynthesis are localized to chloroplasts, including photosynthesis (glucose synthesis), fatty acid synthesis, and the final step of galactosylation.However, the utilization of fatty acids is different between prokaryotic and eukaryotic pathways.In the prokaryotic pathway, de novo synthesized 16:0 and 18:1 fatty acids (FAs) are directly used to synthesize galactolipids within the plastid envelope.In the eukaryotic pathway, they are exported to endoplasmic reticulum (ER) for eukaryotic phospholipid synthesis, such as phosphatidylcholine (PC) and phosphatidylethanolamine (PE).Some portion of diacylglycerol (DAG) precursors, e.g., PC or phosphatidic acid (PA), are returned to the chloroplast to produce galactolipids (Hölzl and Dörmann 2019).While the biosynthetic pathways are well-established, there has been no study on the spatiotemporal dynamics during the leaf development.Lemna proves to be an excellent system to investigate this aspect with MSIi, given that its budding process enables the observation of the spatiotemporal advancements in new lipid biosynthesis.

MALDI-MS of Lemna minor grown in 50% D 2 O medium
In the current simple experimental setup, we opted to grow duckweeds in a Petri dish with 0.5x Shenk and Hildebrandt (SH) medium on a lab bench.As expected, room humidity affected D 2 O concentration overtime due to water vapor exchange, but it was minimal when the humidity was low, which will be further discussed in the later section.Unlike the previous work of Evans and coworkers (Evans et al. 2019), supplementation with 0.5% glucose was excluded to avoid fungal growth and its effect in lipid biosynthesis.Initially, we added glucose into both H 2 O and 50% D 2 O media and sterilized plants with multiple methods, but fungal growth was not avoidable in this setup.As shown in Figure 1A and Supporting Video S1, the growth of L. minor was about twice as slow in 50% D 2 O medium than in H 2 O. Starting with mature fronds of the same size, the second daughter fronds appeared after 3 days in 50% D 2 O media, which were 2 days later than those grown in H 2 O.At the end of the week, the second daughter fronds in 50% D 2 O reached about a quarter size of mature frond, whereas those in the H 2 O media were slightly smaller than the mature one.Additionally, the fronds grown in 50% D 2 O were thicker and smaller (Figure 1B).Lastly, roots in 50% D 2 O medium were hardly seen, while they were about 1 mm in the control condition.Nevertheless, L. minor looked healthy otherwise and could grow in 50% D 2 O indefinitely with regular medium change.Despite some differences, growing L. minor in D 2 O medium seems to have relatively minor adverse effects by D 2 O stress and could be used to monitor their D-labeling in lipid biosynthesis.
To obtain the MALDI-MSI data of the thylakoid membrane lipids across the lateral dimension of duckweed fronds, a fracturing method is employed to horizontally split L. minor fronds into two halves (See Methods for the Downloaded from https://academic.oup.com/pcp/advance-article/doi/10.1093/pcp/pcae032/7642521 by guest on 18 April 2024 details).As we have previously demonstrated (Klein et al. 2015), the exposed internal mesophyll layers produce mostly hydrophobic lipid profiles when interrogated by MALDI-MS.MALDI-MS spectra were acquired throughout the fractured plant, pixel by pixel, with each x, y information for the later imaging analysis.Figures 1C and 1D compare the averaged MALDI-MS spectra of the L. minor grown in H 2 O and D 2 O media for four weeks, typical life span of L. minor under laboratory conditions (Ashby et al. 1949), after prepared by the fracturing method.The MALDI-MS spectrum in H 2 O condition (Figure 1C) was dominated by galactolipids, MGDG 36:6, DGDG 36:6, DGDG 34:3, and pheophytin a.The abundance of MGDG 36:6 and DGDG 36:6 was significantly greater than DGDG 34:3, consistent with the fact that L. minor was reported as 18:3 plant with C 18 FA at sn-2 position through the eukaryotic pathway (Mongrand et al. 1998).In MALDI-MS condition, chlorophyll a readily loses noncovalently bound Mg 2+ and is detected as pheophytin a.After growing in the 50% D 2 O medium for four weeks, the galactolipids and pheophytin a were significantly labeled by deuterium as shown in Figure 1D.Notably, no unlabeled, monoisotopic peaks were left in the mass spectrum, because parent fronds died and were separated out.D-labeling efficiency of galactolipids seems to be close to 100% considering the mass difference between the monoisotope in H 2 O medium and the most abundant isotope in 50% D 2 O medium.For example, MGDG 36:6 has the mass difference of ~34.2 Da corresponding to ~97% D-labeling efficiency out of 50% (D 2 O concentration) of its 70 carbon-bound hydrogens (C-H): 34.2 Da/(50% x 70 x 1.006277 Da) where 1.006277 Da corresponds to the mass difference between H and D atoms.

Temporal change of galactolipid isotopologues with D 2 O labeling
MALDI-MS data of L. minor were obtained after various time points grown in 50% D 2 O medium to monitor the temporal change of D-labeling in these lipids.Figure 2A and 2B show the isotopologue distributions of MGDG 36:6 and DGDG 36:6 on Day 5.Because of insufficient mass resolution to separate 13 C-or other natural isotopes from D-labeled peaks, natural isotope abundances were subtracted as described in Supporting Methods and Figure S1 to obtain D-labeling isotopologue distributions.There might be some subtraction errors in this process due to the isotope abundance measurement error in mass spectrometers.MGDG 36:6 exhibited three binomial isotopologue distributions centered around the average D-labeling of 3.5, 18, and 35, which were referred to Group 1, 2, and 3, respectively (Figure 2A).The three distinct D-labeling isotopologue groups were attributed to the D-labeling of different structural moieties in MGDG (Figure 2A, 2E).Assuming 100% D-labeling efficiency, Group 1 corresponds to the D-labeling of the galactosyl group; when half of the seven carbon-bound hydrogen atoms (7 C-H) in the galactose moiety are deuterium-labeled, the count amounts to 3.5.Group 2 corresponds to the D-labeling of the galactosyl group and a fatty acyl chain (FA); half of 36 C-H in FA 18:3 and galactose is 18.Group 3 corresponds to the D-labeling of the whole molecule; half of 70 C-H in MGDG 36:6 is 35. Figure S2 shows the binomial distributions simulated for Group 1, 2, and 3, assuming p (D 2 O concentration) of 0.5 and n (number of C-H) of 7, 36, and 70, respectively.They are very closely matching with the experimental isotopologue distributions, especially at the peak positions, although experimental data are slightly broader for Group 2 and 3.
DGDG had a similar labeling trend to MGDG but mostly with two groups, Group 1 and 3 (Figure 2B).Group 2 was barely present in a very low abundance, and it seemed real according to MALDI imaging (see next section).It could be clearly seen in electrospray ionization (ESI)-MS analysis in Figure S3 when multiple plants are combined for the lipid extraction.In a closer look, the D-labeling of Group 1 was extended to at least ten deuterations for DGDG 36:6, which was attributed to the 2nd galactosyl group labeling with the average D-labeling of 7, half of 14 C-H in di-galactose.The contribution from the second galactose labeling increased over time and became clearer in later days as shown in Figure S4, but the first galactose labeling was still dominant even on Day 15.
The temporal change of three D-labeling groups is shown in Figure 2C for MGDG 36:6, along with the unlabeled monoisotope.Group 1 emerged rapidly on the first day, increased slightly until Day 3, then stayed at the similar level until Day 10.Group 2 showed up on Day 2, increased slowly until Day 5, then decreased.Group 3 did not clearly appear until Day 3 but rapidly increased and became dominant by Day 15.It should be noted that unlabeled galactolipids were almost gone on Day 15 because the parent frond was separated out from the daughter fronds by then.DGDG 36:6 showed similar temporal changes to that of MGDG for Group 1 and 3 (Figure 2D) but the relative abundance of Group 1 is more than twice that of MGDG Group 1. Unlike galactolipids, pheophytin a had only a single isotopologue distribution corresponding to the D-labeling of the entire molecule over the fifteenday period (Figure S5).

Spatial distribution of galactolipid isotopologues with D 2 O labeling
From the MALDI-MS data obtained across the entire tissues, MS images can be constructed for each m/z species using x, y information of each mass spectrum.Figure 3 shows the MS images of three major galactolipids and pheophytin a for a duckweed grown in 50% D 2 O for five days, the same data used in Figure 2A and 2B.The group images were obtained by combining all the deuterated peaks for the same group, and the MS images were essentially identical within the same group (Supporting Video S2).Some contamination of natural isotopes was unavoidable in the imaging construction, especially 13 C 1 -and 13 C 2 -contribution to D 1 and D 2 -labeling in Group 1; however, its contribution is mostly ignorable in the overall images.
As can be seen in the optical image of a fractured half (Figure 3A), a duckweed plant typically has three fronds, one parent and two daughter fronds.The unlabeled monoisotope peaks were present almost exclusively in the parent frond, which was grown in H 2 O medium before transferring into the D 2 O medium (Figure 3B, 3C, 3D).Group 1 of all galactolipids were co-localized to the unlabeled monoisotope, whereas Group 3, the labeling of the whole molecule, was present only in the daughter fronds, newly grown tissues in D 2 O medium.Lastly, the Group 2 of MGDG showed an intermediate behavior between Group 1 and 3 images, present in the newer part of parent frond (i.e., near the base) or the older part of daughter fronds (i.e., near the margin).Interestingly, pheophytin a has only Group 3 appeared not only in the new daughter fronds but also in intermediate tissue regions including the base of parent frond (Figure 3E).This suggests chlorophyll a was rapidly synthesized with full D-labeling even in the intermediate tissues, which agrees with the fact that pheophytin a is fully D-labeled even on Day 2 (Figure S5).
This observation in MS imaging was further supported by the electrospray ionization (ESI)-MS of lipid extracts from the L. minor grown in 50% D 2 O medium for five days (Figure S3).For the lipids extracted from the parent fronds, the isotopologue distribution was dominated by unlabeled monoisotope and Group 1 but also with a little amount of Group 2, whereas the lipids extracted from daughter fronds were mostly Group 3 and a small amount of Group 1 and 2. This trend is expected from MALDI-MS images in Figure 3.

Preliminary 13 CO 2 -labeling experiment
We have shown D 2 O labeling could successfully capture some intermediates of galactolipid biosynthesis and their spatiotemporal behaviors.There is a major concern that the observation of three isotopologue groups is a result of D 2 O-induced artifact.To test whether the observed spatiotemporal behavior is consistent without D 2 O stress, we performed a simple experiment of 13 CO 2 -labeling by growing duckweeds in a small chamber with 13 CO 2 .It is described in detail in the experimental section but in short duckweeds were grown in a small beaker with 0.5x SH medium inside a sealed large flask (Figure S6).The ambient air inside the flask was flushed out with CO 2 -free air and 13 CO 2 was produced using Ba 13 CO 3 .As there was a minor leak in the system, air flush and new 13 CO 2 production were repeated every three hours during the daytime, with the first cycle of a day coinciding with the activation of LED lights, the onset of photosynthesis.The 13 CO 2 concentration was estimated to be around 1,500 ppm when freshly produced.It was a much higher CO 2 concentration than in ambient air but on purpose to minimize 12 CO 2 -labeling due to the leak.
Unlike D 2 O labeling, no visual difference was apparent between control and 13 C-labeled fronds.Three replicate plants each were subjected to MALDI-MSI analysis after growing in 13 CO 2 for 1-3 days.The 13 Cisotopologue distributions of MGDG 36:6 and DGDG 36:6 on Day 3 (Figure 4A, 4B) have three isotopologue groups similar to D 2 O labeling.Unlike symmetric distributions in D-labeling, isotopologue distributions in 13 Clabeling were skewed toward higher isotope due to a much higher 13 CO 2 concentration than 12 CO 2 .In addition to galactose labeling of Group 1, glycerol backbone labeling was observed marked as Group 1', which is further explained in the discussion.Temporal change of the three isotopologue groups (Figure 4C, 4D) also shows similar trend as in D-labeling with unlabeled decreasing and Group 3 increasing over time, while Group 1 and 2 increased fast initially then stabilized.A similar trend is expected to continue in a longer experiment although we did not perform longer than three days due to the experimental difficulty in the current setup.The decrease of unlabeled and the increase of fully labeled were much faster than in D-labeling.For example, Group 2 vs Group 3 ratio of MGDG 36:6 in Day 2 13 C-labeling (Figure 4C) is already comparable to that on Day 5 of D-labeling (Figure 2C).It was mostly attributed to a faster growth of L. minor in H 2 O medium (Figure 1A), but also partially to a much higher 13 CO 2 concentration.
MALDI-MS images are shown in Figure 5 for each 13 C-isotoplogue group of MGDG 36:6 and DGDG 36:6.Overall, they show similar images with D-labeling in Figure 3. Unlabeled galactolipids were present mostly in parent frond as well as Group 1 and 1', while Group 3 was present almost exclusively in daughter fronds.Group 2 showed an intermediate behavior between Group 1 and 3 for both galactolipids but a little more localized to the parent frond for MGDG and to the daughter fronds for DGDG.It is consistent between 13 C-and D-labeling that one and two fatty acid(s) labeling represented by Group 2 and 3, respectively, have appearance in newer tissues, distinct from old tissues at the time of starting the labeling experiments.

Backward labeling
To further support the observed isotopologue distributions, we have performed another supporting experiment by performing backward labeling.In this experiment, we have established L. minor adapted to D 2 O stress via multiple generations of culture in 50% D 2 O medium for three months.They were then moved back to H 2 O media and monitored the change of their isotopologue patterns over a fifteen-day period.In this case, we expect the labeling would occur reversely, in which more unlabeled moieties and molecules would be produced over time.Isotopologue distributions on Day 3 are shown in Figure 6A and 6B for MGDG 36:6 and DGDG 36:6, respectively.Fully labeled L. minor grown in 50% D 2 O for three months had only Group 3 isotopologue distributions with the average number of deuterations of ~34.5 and ~38 for MGDG 36:6 and DGDG 36:6, respectively (not shown).To make it easier to understand the backward labeling data, a secondary x-scale is shown in the reverse order as the number of D-removed.The number of D-removed is defined here as the average full deuteration (34.5 for MGDG 36:6 and 38 for DGDG 36:6) subtracted by the number of deuterations.As expected, three groups of isotopologue distributions are observed similar to forward labeling (i.e., D 2 O labeling) corresponding to D-removal of galactose (Group 1), intermediate D-removal (Group 2), and full D-removal (Group 3).It should be noted that although we call it 'D-removal' for the convenience of explanation, it is actually H-labeling in new biosynthesis that dominates Dlabeling in the total population.The major differences between the forward (Figure 2) and backward labeling (Figure 6) are 1) the isotopologue distribution was very broad in Group 1 and became narrower as D-removed (or H-labeled) in Group 3, and 2) D-removal (or H-labeling) was much faster than D-labeling that Group 3 was already ~40% of total on Day 3 for both MGDG 36:6 and DGDG 36:6, similar level on Day 5 of D-labeling.The labeling time difference could be attributed to the D 2 O-induced stress and corresponding slower growth in D 2 O (Figure 1A).As another and more quantitative measure of D 2 O-stress induced growth delay, we calculated the average mass from the entire isotopologue and traced their changes over 15 day-period as shown in Figure 6C and 6D for MGDG 36:6 and DGDG 36:6, respectively.As expected, the average mass increased over time for forward labeling and decreased for backward labeling.Assuming half-way change of the average mass represents the overall growth rate or total D-(or H-) labeling rate, there is about 1.5~2 days of delay in forward labeling compared to backward labeling, presumably due to D 2 O-induced stress.Interestingly, the trend of backward labeling has a simple gradual decrease that can be fitted with an exponential decay function at R 2 of >0.99; however, the forward labeling shows a S-curve trend suggesting the two-day delay might be mostly due to the initial adjustment time to the new D 2 O stress environment.

Spatiotemporal pattern of D-labeled isotopologue groups follow linear pathway of galactolipid biosynthesis
In this work, we have successfully demonstrated that in vivo D 2 O-labeling can be used to capture metabolite intermediates in galactolipid biosynthesis and monitor their temporal changes and spatial distributions.The spatiotemporal pattern of three D-labeled isotopologue groups (Group 1, 2, and 3 in Figure 2 and Figure 3) can Downloaded from https://academic.oup.com/pcp/advance-article/doi/10.1093/pcp/pcae032/7642521 by guest on 18 April 2024 be explained by galactolipid biosynthesis (Ohlrogge and Browse 1995) illustrated in Figure 7 with the isotope labeled building block indicated in red.The major difference among the isotopologue groups is the number of fatty acyl chains labeled and their localizations: Group 1, 2, and 3 has no, one, and two fatty acyl chains labeled, respectively, with the localization in old, intermediate, and new tissues.The interpretation of Group 1, 2, and 3 are straightforward as they are in the reverse order in the biosynthesis.Namely, Group 1 labeling occurs in the old tissues via the last step of labeled galactose attachment to the unlabeled precursors.Group 2 labeling starts with unlabeled LPA as a precursor that was available in the intermediate tissues by adding a deuterated fatty acid and galactose(s).Lastly, Group 3 is observed only in new tissues that were grown after being moved to D 2 O medium as the entire molecule is D-labeled.Their temporal appearances are also in the order of Group 1, 2, and 3, as expected.In a closer look, there is some partial overlap in the localization between Group 1 and 2 as well as between Group 2 and 3 (Figure 3, Figure 5).This suggests that the boundary of some intermediate tissues is not clear.
The observation of three distinct major isotopologue groups, referred to as Group 1, 2, and 3, in in vivo isotope labeling of galactolipids was possible as a result of 1) the time scale difference in biosynthesis of their building blocks, 2) the high enough abundance of precursors, and 3) sufficient separation in mass dimension.First, sugars, such as galactose, are rapidly synthesized and its attachment to DAG or MGDG to produce MGDG Group 1 or DGDG Group 1 appears just one day after moving to D 2 O medium.Meanwhile, it took one and two more days for the obvious presence of Group 2 and 3, respectively.If the overall biosynthesis occurred rapidly without distinct time scale difference between each step, only Group 3 would have been observed as in pheophytin a (Figure 3, Figure S5).Second, each isotopologue group would have not been observed above the detection limit if there were insufficient amounts of precursors.For example, Group 1 of MGDG is composed of unlabeled DAG and deuterated galactose (dGal), in the form of 'DAG-dGal' (Figure 7).Hence, the presence of a large pool of unlabeled DAG precursors (e.g., unlabeled PC or PA) in old tissues would be necessary for the observation of MGDG Group 1. Similarly, the observation of Group 2 suggests the presences of a relatively high abundance of unlabeled LPA in intermediate tissues so that D-labeled fatty acid and D-labeled galactose can be attached to produce MGDG or DGDG Group 2. Finally, the observation of three isotopologue groups was not possible without clear mass difference between them.We used almost the maximum D 2 O concentration possible for L. minor; yet a broad binomial distribution is unavoidable in this study.Fortunately, many C-H in fatty acyl chains (29 for FA 18:3) was sufficient to separate at least three isotopologues groups with no (Group 1), one (Group 2), and both (Group 3) fatty acyl chains labeled.

Partial glycerol labeling is present in all isotopologue groups.
In addition to Group 1, 2, and 3 observed in D 2 O labeling, 13 C-labeling and backward labeling uncovered other isotopologue groups named as Group 1', 2', and 3' that are hidden within the wide isotopologue distributions of D-labeling.They contain additional building blocks that are not in the order of galactolipid biosynthesis, suggesting the mixed presence of labeled and unlabeled building blocks.First, 13 C-labeling revealed the presence of Group 1' corresponding to the labeling of glycerol in addition to galactose labeling.The three 13 C-isotopologue distributions for MGDG 36:6 in Figure 4A could be explained by the simulated binomial distributions in Figure S7A using p ( 13 CO 2 concentration) of 0.9 and n (carbon number) of 6 (galactose), 24 (galactose + FA 18:3), and 45 (entire molecule) for the Group 1, 2, and 3, respectively.Experimental isotopologue distributions were broader for Group 2 and 3 than in the simulation, most likely because 13 CO 2 vs 12 CO 2 concentration was continuously changing over time due to the leak.An interesting observation was a distinct distribution of 13 C 7 -13 C 9 isotopologues (or M7-M9) in Figure 4A, named as Group 1' because of its co-localization with Group 1 (Figure 5).The simulation in Figure S7A with n = 9 (galactose + glycerol) can explain for the presence of Group 1' (M7-M9 isotopologue).Galactose and glycerol backbone labeling was not observed in D-labeling partially because of the broad nature of D-labeling isotopologue distributions.Group 1' was also observed for DGDG 36:6 although it overlapped with 13 C-labeling of the second galactose (Figure 4B).Unlike the simulation, 13 C 7 -labeling has a higher abundance than 13 C 8 or 13 C 9labeling, suggesting there might be some contribution from the recycling of unlabeled moieties.The presence of Group 1' in old tissue does not follow the linear pathway of galactolipid biosynthesis as illustrated in Figure 7; Downloaded from https://academic.oup.com/pcp/advance-article/doi/10.1093/pcp/pcae032/7642521 by guest on 18 April 2024 namely, unlabeled fatty acids are attached to newly synthesized 13 C-glycerol backbone.This may suggest membrane restructuring is continuously occurring in old tissues but recycling unlabeled fatty acids (Yu et al. 2021).
Second, 13 C-labeling and the backward labeling revealed the presence of Group 2', another intermediate labeling with not only one fatty acid and galactose but also glycerol backbone labeled.In a closer look of 13 Cisotopologue profile (Figure 4), Group 2 peak positions of 24 and 30 for MGDG 36:6 and DGDG 36:6, respectively, were slightly higher than 22 and 28 in the simulated peak positions of Group 2 (Figure S7).They were rather closer to 25 and 30 in the simulated peak positions of Group 2' calculated with n = 27 (galactose + FA 18:3 +glycerol) and 33 (two galactose + FA 18:3 + glycerol), respectively, corresponding to glycerol labeling in addition to Group 2. It may suggest Group 2 also had some contribution of 13 C-labeled glycerol backbone, although they cannot be separated from Group 2 due to the wide isotopologue distributions.Another evidence of Group 2' can be found in the backward labeling data (Figure 6).Group 2 was shifted by ~19.7 D-removed (or H-labeling) for MGDG 36:6 and ~22.6 D-removed (or H-labeling) for DGDG 36:6 from the fully D-labeled initial positions.These shifts were larger than the labeling of galactose and a fatty acyl chain in forward labeling; 18 and 21.5 for MGDG and DGDG, respectively (half of 36 C-H and 43 C-H).Similar to 13 C-labeling, this is attributed to another intermediate labeling, Group 2', with the labeling of glycerol backbone (half of 5 C-H is 2.5) in addition to the labeling of galactose and a fatty acyl chain in Group 2. Group 2' would have 20.5 and 23 D-labeling for MGDG 36:6 and DGDG 36:6, respectively.The observed Group 2 in backward labeling might be a mixture of Group 2 (galactose and a FA labeling) and 2' (glycerol, galactose, and a FA labeling).The presence of Group 2' is attributed to a large pool of both labeled and unlabeled fatty acids in intermediate tissues, which can be randomly added to sn-1 and sn-2 position.
Third, the backward labeling also revealed Group 3' that contains the recycling of unlabeled glycerol in the biosynthesis of new lipids.The non-binomial distribution of Group 3 in the backward labeling (Figure 6) is not surprising as the maximum number of removable D is limited, but it has a broader distribution than natural isotope distribution in fully H-labeled galactolipids.For example, MGDG 36:6 has the natural isotope abundance of 100:51:9 for M0:M1:M2 as seen in Figure 1C or Figure S1, but Group 3 in Figure 6A has much higher M1 and M2 as well as significant signals for M3-M6.This suggests that the newly synthesized Group 3 is not purely Hlabeled.Here they are referred as Group 3', as illustrated in Figure 7 with only glycerol unlabeled.Similar labeling is expected in forward labeling but not observed due to the very broad nature of Group 3. It is odd to detect unlabeled (or pre-existing D-labeled in the backward labeling) glycerol in new tissues, but we speculate they might have come from parent fronds where they were reserved as starch or other forms of oligosaccharides (Dörmann and Benning 1998).

D 2 O stress and 13 C-labeling
Despite its usefulness as a global isotope labeling strategy, in vivo D 2 O labeling has a critical downside due to its toxicity (Evans and Shah 2015).While the use of low D 2 O concentration can minimize adverse effects, we have chosen to use the maximum D 2 O concentration, 50%, that is possible without supplementation.It was an inevitable choice to maximize the separation of intermediate isotopologue groups.For example, a significant overlap is expected among the isotopologue groups with 25% D 2 O concentration as simulated in Figure S8.Another benefit of using 50% D 2 O concentration is the symmetric isotopologue distributions, making the data interpretation easier.Although L. minor could propagate indefinitely in this high D 2 O concentration, D 2 O induced stress is apparent as observed in the slow growth rate and smaller frond sizes (Figure 1A, 1B), similar to reported by others (Evans et al. 2019;Cooke et al. 1979;Cope et al. 1965).Evans et al. suggested the growth inhibition by high D 2 O concentration might be due to the impacts on multiple metabolic pathways (Evans et al. 2019).We did not perform detailed physicochemical analysis of L. minor grown in 50% D 2 O medium because it was thoroughly performed previously by Evans and coworkers (Evans et al. 2019).When compared to L. minor grown in normal media, they found no obvious abnormalities in cellular morphology, a slight decrease in the degree of polymerization in cellulose (15%), and similar average weight (0.031 g vs 0.027 g for 35 fronds).Notable differences they reported include slower growth rate (40%), lower chlorophyll content (0.065% vs 0.113% of total weight) and lower lignin content (8% vs 18% of total weight).A significant deuterium substitution, 40-50%, is found in the solid-state NMR analysis, corresponding to 80-100% of D-labeling efficiency.This D 2 O stress did not affect the presence of three distinct isotopologue groups as supported by the backward labeling and 13 C-labeling; however, their temporal changes are affected by about two days of delay according to the backward labeling experiment (Figure 6).Cooke et al. observed a rapid loss of soluble protein and an increase in free amino acids when L. minor were transferred to 50% D 2 O.It was recovered in two days, but the protein synthesis was still inhibited by 20% which would consequently lead to the slow growth (Cooke et al. 1979).This is consistent with the result of our forward labeling experiment, in which the growth delay might be occurring only initially to adjust to the new stressed environment.Another downside of D 2 O labeling is the wide binomial distribution.This is because the use of 50% D 2 O concentration produces half and half chance of hydrogen vs deuterium-labeling for every carbon bound hydrogens that are being synthesized if we ignore kinetic isotope effect, resulting in symmetric Gaussian-like distribution.If it were possible to use 99% D 2 O, an almost complete Dlabeling would have been achieved with a much narrower isotopologue distribution that can reveal more details of in vivo isotope labeling of each building block.Similarly, ~90% 13 CO 2 -labeling resulted in a much narrower isotopologue distribution than in 50% D 2 O labeling (Figure 4).
Considering the wide isotopologue distributions and D 2 O induced stress, 13 CO 2 -labeling might be more effective than D 2 O labeling.As our preliminary experiment demonstrated, it can not only separate the three isotopologue groups but also glycerol backbone labeling of Group 1' and 2' that was not observed in D 2 O labeling.It was not possible, however, to perform almost complete (e.g.99%) 13 CO 2 -labeling in this preliminary experiment because of the need for a completely air-tight system.Such systems have been previously demonstrated, but it often requires many years of efforts to build an air-tight growth chamber with the precise environmental control including 13 CO 2 concentration (Ćeranić et al. 2020;Peters et al. 2018).We were able to achieve an average of ~90% 13 Clabeling when compared to a simulation but a broad distribution was unavoidable for Group 2 and 3 as 13 CO 2 vs 12 CO 2 concentration was continuously changing.Nevertheless, carbon metabolism is not the same as hydrogens.Carbons in 13 CO 2 -labeling enter the metabolic pathway of a plant system only through Calvin cycle (Leegood 2013), but hydrogens in D 2 O-labeling can enter the metabolic pathway through multiple different pathways.As a result, the carbon flux and hydrogen flux would not be the same even when monitoring the same metabolic pathway.A careful comparison of mixed in vivo labeling, e.g., 13 CO 2 + H 2 O, 12 CO 2 + D 2 O, and 13 CO 2 + D 2 O, may reveal some differences between the two fluxes.

Metabolite flux information
The ultimate goal of MSI with in vivo isotope labeling would be to obtain the imaging of metabolite flux or tissue-specific/cell-specific flux information.Recently, Wang and coworkers illustrated an example of localizationspecific flux information using MSI of rat kidney with in vivo isotope tracing (Wang et al. 2022).They have monitored glycolysis metabolism by measuring M6 UDP-glucose while infusing [U-13 C]-glucose and gluconeogenesis metabolism by measuring M3 UDP-glucose while infusing [U-13 C]-glycerol.MSI data revealed high glycolysis activity in the medulla of rat kidney but high gluconeogenesis activity in the cortex.Similarly, in the current study, MSI with D 2 O labeling of L. minor revealed there is no Group 2 or 3 in old parent tissues, meaning new fatty acid biosynthesis does not occur in mature tissues or at least not involved in galactolipid biosynthesis.This is despite the fact that lipid restructuring is still being made in old tissues by recycling unlabeled "old" fatty acids evidenced by the presence of Group 1'.
Assuming that the mature plant tissues moved to 50% D 2 O medium were in pseudo-metabolic steady state because they almost stopped growing (see Supporting Video S1 for example), unlabeled and Group 1 of MGDG 36:6 are plotted between the two over time to provide flux information.These two groups are present only in old parent tissues unlike other isotopologue groups.As shown in Figure S9A and S9C for MALDI-MS and ESI-MS data, respectively, the relative abundance of the labeled MGDG (Group 1) increases exponentially with the asymptotic value of 28-35%, while a majority, 65-72%, of pre-existing (unlabeled) MGDG in old tissues does not further metabolize.Similar exponential changes are observed for DGDG 36:6 (Figure S9B, S9D) but only 36-42% of the pre-existing (unlabeled) DGDG are present in old tissues at the asymptote.From ESI-MS data, the influx is estimated to be ~22% per day and ~30% per day for the synthesis of MGDG 36:6 and DGDG 36:6, respectively, assuming t 1/2 = [Pool size]/F in x ln(2) (Jang et al. 2018).This is only a rough estimation of the daily rate of synthesis and incorporation of labeled galactose into the lipid pools under D 2 O stress environment.

Limitations of this study
Despite its success, the current study has many limitations.Most importantly, D 2 O stress might have induced some differences from unstressed normal physiological conditions.While most qualitative observations are expected to be similar, some quantitative information may not be the same.One confirmed example is two days of delay in forward labeling compared to backward labeling (Figure 6).The estimated influx of 22% and 30% per day for MGDG 36:6 and DGDG 36:6 synthesis in the final galactosylation step (Group 1) might be also slightly higher in unstressed condition (H 2 O media).There is a technical difficulty or consideration in this experiment coming from the effect of environmental humidity in D 2 O concentration.A simple in vivo D 2 O labeling system adopted in this study using a Petri dish was exposed to ambient air for the continuous gas exchange.Although the lid was almost closed to minimize water vapor exchange, some exchange was inevitable during the multiple days of plant growth.As a result, D 2 O concentration was decreased over time and subject to the lab humidity as shown in Figure S10.At 70% lab humidity, 50 mol% of initial D 2 O concentration is decreased to ~30 mol% in about a week, whereas 50 mol% of initial D 2 O concentration is decreased to only about ~46 mol% at 25% humidity.The peak positions of each isotopologue group were greatly affected by the change in D 2 O concentration, but the peak area seemed to be rather insensitive.Most experiments used in this study were obtained when the lab humidity is relatively low.When necessary, experiments were performed inside an environmental chamber to control the humidity.

Future outlook
The current study exemplified some potentials of MSIi to elucidate the spatiotemporal details of galactolipid biosynthesis in duckweeds.We expect that there will be a lot more this technology can offer to better understand plant metabolic biology.Some limitations, technical and biological, have been identified but they could be overcome with improvements or with the help of other alternative approaches in parallel.For example, phospholipids, key intermediates of galactolipid biosynthesis, were not detected in the current experimental setup.According to the preliminary thin layer chromatography-ESI-MS analysis, their isotopologue analysis turned out to be very complicate and almost impossible with the current instrumentation because of overlapping isotopologues with multiple unsaturation.Future study will include additional method development and ultrahigh-resolution MS analysis for MSIi of phospholipids.Despite its advantage as a global labeling agent, D 2 O labeling has several limitations such as D 2 O-induced stress and broad isotopologue distributions.Other isotope labeling has their own pros and cons; 13 CO 2 -labeling is also attractive in the study of plant biology, but technically challenging due to the need of gas tight growth chamber.In the future, we will build a relatively simple system for the 13 CO 2 -labeling of duckweeds that can achieve ~99% 13 CO 2 concentration while controlling its concentration with CO 2 sensor.It may need the scrubbing of volatile organic compounds for a long-term plant growth (Peters et al. 2018) but may not be necessary for a short term experiment.However, 13 CO 2 -labeling may not be completely compatible with D-labeling as it is carbon-centric.The use of organic precursors, e.g., [U-13 C]-glucose or [1,2]-13 C 2 -glucose, is another option for 13 C-labeling that can also specifically define the reaction pathways that are being studied while limiting the scope of the study (Dellero et al. 2024;Lee et al. 1998).
Some technological limitations in MSIi are coming from the limitations in mass spectrometry.Ultrahigh mass resolution is necessary to separate a complex mixture of various isotope labeling.For example, to distinguish the same two lipids with 13 C 1 vs D 1 labeling (2.9 mDa difference; e.g., 13 C 1 -MGDG 36:6 vs D 1 -MGDG 36:6) requires the mass resolution higher than 280,000, which is not obtainable with the Orbitrap used in this study.It will be possible but challenging even with the most advanced mass spectrometry technology such as Fourier Transform Ion Cyclotron Resolution (FTICR) and would not be routinely available to many researchers.Further, a trapping type of mass spectrometers such as Orbitrap and FTICR has isotope abundance errors due to space charge effect when storing ions for a long period to obtain higher mass resolution (Hohenester et al. 2020).As a result, one needs to be very cautious in data analysis, not to overinterpret the result and clarify the quantitative limitations.Some fundamental study would be necessary to define the limitations and find the practical alternatives.Regardless, MSIi is a newly emerging technology and expected to provide insights in metabolic biology that have not been previously available in currently available tools.Potential applications include MSIi of crop seeds under stress conditions or with genetic engineering, which may provide further insights in spatiotemporal change of lipid biosynthesis and provide solutions in overcoming the current bottleneck in genetic engineering.

Chemicals
LC/MS-grade isopropanol, chloroform, methanol and water, sodium acetate and potassium acetate, and lactic acid were obtained from Fisher Scientific (Hampton, NH, USA).Isotope products, 70 or 99.9% D 2 O and 99.8% Ba 13 CO 3 , were purchased from Cambridge Isotope Laboratories, Inc. (Tewksbury, MA, USA).MS-grade thin-layer chromatography from Sigma-Aldrich (St. Louis, MS, USA) was used for lipid separation.

Plant materials
Lemna minor plant was originally obtained from VWR (catalog number 470194-692) and propagated in 0.5x Schenk and Hildebrandt (SH) (Phyto Technology Laboratories) for many months on a lab bench.Plant media and growing beakers were sterilized by autoclaving under a liquid cycle.A cool white LED shop light was used as a light source with the photon density of 60 µmol m -2 s -1 and the light cycle of 16-h/8-h.The ambient conditions were largely relying on the building air circulation system.The laboratory temperature was typically at 20 -25 o C and the ambient humidity varied between 20-35% in winter and 55-70% in summer.The medium is regularly replaced every ~7 days.
For the forward D 2 O-labeling experiment, 15-20 healthy fronds grown in control media were transferred into each of six small Petri dishes with 14 mL of 50% D 2 O, H 2 O:D 2 O (50:50, mol/mol), 0.5x SH media and grown for different periods, 1, 2, 3, 5, 10, and 15 days.For the backward labeling, three replicates of 15-20 plants grown in 50% D 2 O for 3 months were transferred to each petri dish of H 2 O media.
It should be noted that high room humidity, especially on rainy summer days could greatly affect Dlabeling (Figure S10).Both forward and backward labeling experiments were performed at low humidity (~25 %) and its effect on D-labeling is expected to be minimal, an average of ~2% for the 5-day labeling data.The D 2 O concentration of media was measured with infrared spectroscopy, Bruker Tensor 37 FTIR.An environmental chamber (ICH110L; Memmert, Schwabach, Germany) was used to control humidity for the forward labeling experiment in Figure 6, in which temperature and humidity was controlled at 20°C and 25%, respectively.
An Erlenmeyer flask (250 mL) with a three-hole screw cap was set up and connected with CO 2 -free air and lactic acid-containing syringe as shown in Figure S6.The third hole is blocked with an end cap which was used to release or seal the air inside.The flask contained two 10-mL beakers, one for 5-7 plants in 5 mL 0.5x SH medium and one for ~1 mg Ba 13 CO 3 powder.At first, CO 2 -free air was introduced into the flask to flush out ambient 12 CO 2 gas inside the flask.After two minutes, the cylinder valve and the end cap were closed almost simultaneously, then ~60 µL of lactic acid (3.3%, w/v) was introduced into the Ba 13 CO 3 beaker to produce 13 CO 2 (Ida and Kudo 2008).As the system was not perfectly sealed and there might be a build-up of organic gases over time, this process was repeated every 3 hours to flush out the air and freshly produce 13 CO 2 by adding lactic acid during 16 hours of daylight for 0-3 days.For MALDI-MSI samples, three healthy individual plants were selected at each time point.The fracturing method was employed to expose the middle layer of the frond as described elsewhere (Klein et al. 2015).Briefly, the plants were attached onto a packing tape, vacuum dried, enclosed the tape to also attach the other side of fronds with the tape, passed through a rolling mill to make a mechanical damage to the internal tissues, then pulled over two ends of the tape piece to produce two split half-fronds exposing the internal mesophyll layers.The top half layer with adaxial side attached to the tape was used for the MALDI-MSI.The tissue samples were sputtered with gold for 20 seconds at 40 mA (Cressington 108; Ted Pella, Redding, CA, USA) to provide electrically conductive surface and then sprayed with 2,5-dihydroxybenzoic acid using a TM sprayer (HTX Technologies, Chapel Hill, NC, USA).

MALDI-MS imaging
An orbitrap mass spectrometer QExactive HF (Thermo Scientific, San Jose, CA, USA) equipped with a medium pressure (~8 torr) MALDI source (Spectroglyph, Kennewick, WA, USA) was used in this study.A 349 nm Nd:YLF laser (Explorer One, Spectra Physics, Milpitas, CA) was used with the laser energy of ~1 µJ per pulse and the spot size of 15 µm.MALDI-MSI data set acquired for m/z 100-1200 in positive mode with a raster step of 30 µm and the mass resolution of 120,000 at m/z 200.Data was processed with MSiReader (Robichaud et al. 2013) to generate MS images, Xcalibur (Thermo) to extract raw data, in-house Python code for filtering isotopic peaks, and MATLAB for deconvolution of Group 2 vs Group 3 isotopologue distributions.

ESI-MS analysis
Lipid extraction followed by ESI-MS analysis was used for the isotopologue group comparison between parent vs daughter fronds (Figure S3) and the backward labeling experiment (Figure 6).The lipid extract method was derived from the single-extraction method by Shiva and coworkers (Shiva et al. 2018).Harvested L. minor plants were gently dried with Kimwipes, incubated in preheated 75°C solution of 0.01% BHT isopropanol (w/v) for 15 min, and cooled to room temperature.Chloroform, methanol, and water were added to make the final solvent mixture with isopropanol: chloroform: methanol: water in the ratio of 31/28/38/3 (v/v/v/v).The extract was shaken at ~500 rpm for 12 hours.The parent vs daughter frond experiment was performed by directly infusing the lipid extract to the orbitrap MS with ESI using a sample loop after adding potassium acetate solution to a final concentration of 2 mM potassium.
For the backward labeling experiment, thin-layer chromatography (TLC) was employed to separate lipid species.The lipid extracts from both forward and backward labeling sample sets were dried under nitrogen and resuspended with 0.16 volume of chloroform: methanol (2/1, v/v) mixture.Approximately 10 µL of condensed lipid extract was deposited onto a TLC plate predeveloped with chloroform: methanol (1/1, v/v) to remove impurities.A mixture of chloroform: methanol (72/28, v/v) was used as the developing solvent (Hölzl and Dörmann 2021).After separation, lipid bands of MGDG and DGDG were scraped off and dissolved in mixture of chloroform: methanol (2/1, v/v), and gently shaken for 15 minutes.When silica gel was settled at the bottom of the test tubes, the transparent solution was taken and added sodium acetate solution to 2 mM final concentration, then injected to the ESI loop.All lipid data was acquired in the m/z range of 600-1100 with positive mode and resolution of 240,000 at m/z 200.

Figure 1 .
Figure 1.Growing L. minor in D 2 O medium.(A) The growth rate of L. minor in the H 2 O vs 50% D 2 O 0.5x Schenk and Hildebrandt (SH) medium.(B) The photos of L. minor after growing in H 2 O (left) vs 50% D 2 O medium (right) for 4 weeks.(C, D) The averaged MALDI mass spectra of L. minor grown in (C) H 2 O medium vs (D) 50% D 2 O medium.All galactolipids and pheophytin a are detected as potassium adducts.There are some background peaks (*) coming from non-tissue area.

Figure 2 .
Figure 2. D 2 O-labeled galactolipids and their temporal changes.(A, B) D-labeling isotopologue distribution of (A) MGDG 36:6 and (B) DGDG 36:6 after growing L. minor for five days in 50% D 2 O medium. 13C-or other natural isotope abundance is subtracted as described in Supporting Methods.(C, D) The temporal change of each D-labeling group over time for (C) MGDG 36:6 and (D) DGDG 36:6.Three replicates of L. minor were used for each time point.(E, F) The structures of (E) MGDG 36:6 and (F) DGDG 36:6 indicate each D-labeled structural moiety.

Figure 3 .
Figure 3. MALDI-MS images of L. minor after grown in 50% D 2 O for five days.(A) Optical image of the top half fractured and (B-E) MS images of each D-labeling group for (B) MGDG 36:6, (C) DGDG 36:6, (D) DGDG

Figure 4 .
Figure 4. 13 CO 2 -labeled galactolipids and their temporal changes.(A, B) Isotopologue distribution of (A) MGDG 36:6 and (B) DGDG 36:6 after growing L. minor for three days in 13 CO 2 .Natural isotope abundance was not subtracted to avoid confusion.(C, D)The temporal change of each 13 C-labeling groups over time for (C) MGDG 36:6 and (D) DGDG 36:6.Group 1' is combined with Group 1 in the temporal change as it cannot be separated in DGDG, and they both are present in the parent frond.Three replicates of L. minor were used for each time point.

Figure 5 .
Figure 5. MALDI-MS images of L. minor after growing in 13 C-chamber for three days.The max scale is arbitrarily adjusted for each image.

Figure 6 .
Figure 6.Backward labeling of L. minor.(A, B) Isotopologue distribution of (A) MGDG 36:6 and (B) DGDG 36:6 three days after moving L. minor grown in 50% D 2 O for three months to H 2 O medium.A secondary x-scale is shown in the reverse order as the number of D-removed, defined as the average number of full deuteration (34.5 for MGDG 36:6 and 38 for DGDG 36:6) subtracted by the number of deuterations.The natural isotope abundance was not subtracted to avoid confusion and calculation error.(C, D) Temporal change of the average mass of (C) MGDG 36:6 and (D) DGDG 36:6.This data was obtained by TLC-ESI-MS of lipid extract from three replicates of 15-20 plants at each time point.In forward labeling, the average mass of MGDG 36:6 and DGDG 36:6 was reached 1.5-2 days later than in backward labeling.For the backward labeling, average mass change is fitted with an exponential decay function at R 2 > 0.99.