Pathway-specific metabolome analysis with 18O2-labeled Medicago truncatula via a mass spectrometry-based approach

Introduction Oxygen from carbon dioxide, water or molecular oxygen, depending on the responsible enzyme, can lead to a large variety of metabolites through chemical modification. Objectives Pathway-specific labeling using isotopic molecular oxygen (18O2) makes it possible to determine the origin of oxygen atoms in metabolites and the presence of biosynthetic enzymes (e.g., oxygenases). In this study, we established the basis of 18O2-metabolome analysis. Methods 18O2 labeled whole Medicago truncatula seedlings were prepared using 18O2-air and an economical sealed-glass bottle system. Metabolites were analyzed using high-accuracy and high-resolution mass spectrometry. Identification of the metabolite was confirmed by NMR following UHPLC–solid-phase extraction (SPE). Results A total of 511 peaks labeled by 18O2 from shoot and 343 peaks from root were annotated by untargeted metabolome analysis. Additionally, we identified a new flavonoid, apigenin 4′-O-[2′-O-coumaroyl-glucuronopyranosyl-(1–2)-O-glucuronopyranoside], that was labeled by 18O2. To the best of our knowledge, this is the first report of apigenin 4′-glucuronide in M. truncatula. Using MSn analysis, we estimated that 18O atoms were specifically incorporated in apigenin, the coumaroyl group, and glucuronic acid. For apigenin, an 18O atom was incorporated in the 4′-hydroxy group. Thus, non-specific incorporation of an 18O atom by recycling during one month of labeling is unlikely compared with the more specific oxygenase-catalyzing reaction. Conclusion Our finding indicated that 18O2 labeling was effective not only for the mining of unknown metabolites which were biosynthesized by oxygenase-related pathway but also for the identification of metabolites whose oxygen atoms were derived from oxygenase activity. Electronic supplementary material The online version of this article (10.1007/s11306-018-1364-6) contains supplementary material, which is available to authorized users.


Introduction
It is well known that plants biosynthesize a wide variety of metabolites. A statistical approach has estimated that there are over 1,060,000 metabolites in plant species (Afendi et al. 2012). Although these are often beneficial in both plant physiology and for drug development (Newman and Cragg 2012). only a small proportion of these metabolites (50,897 metabolites were recorded in KNApSAcK) have been identified (Nakamura et al. 2014). Therefore, it is highly probable that metabolome analyses will lead to the identification of the unknown metabolites, uncover unknown biological phenomena and lead to advancements in drug development.
For large-scale and high-throughput metabolite analyses, chromatography-coupled to mass spectrometry (MS) is a very powerful tool due to its ability to detect a wide range of metabolites . However, MS merely provides information on the mass and intensity of a large number of metabolite-derived peaks for metabolite, and 'metabolite annotation' is a crucial aspect of MS-based metabolomics. Annotation is typically performed by comparing tandem mass spectrometry (MS/ MS) spectra of unknowns with spectra of known metabolites (Sumner et al. 2007). However, this approach cannot be applied to many rare plant-specific metabolites, because corresponding authentic standards are not available in many cases. Moreover, unknown metabolites are often annotated as 'unknown', and the information is not effectively used for further estimation (Iijima et al. 2008). Currently, advanced MS instrumentation with high accuracy and resolution enables a standard compound-independent annotation strategy in which the elemental composition is estimated based on accurate mass values (Makarov and Scigelova 2010;Ohta et al. 2010). To avoid incorrect assignment of elemental compositions, the natural average abundance of stable isotopes has been used (Kind andFiehn 2006, 2007).Using a different approach, Giavalisco et al. prepared Arabidopsis thaliana labeled with 13 C, 15 N, and 34 S, and estimated the elemental composition using the number of stable isotopes incorporated in the compound as an index (Giavalisco et al. 2011). Therefore, use of stable isotopes is effective for the estimation of elemental composition. 13 C, 15 N, and 34 S have also been used as a tag or tracer, and are sometimes enriched by feeding plants with 13 CO 2 ,or 34 S-inorganic salts during cultivation (Glaser et al. 2014;Harada et al. 2006;Huege et al. 2007;). In addition, 18 O also has potential roles in metabolite flux analysis and gene mining. In higher plants, oxygen atoms are incorporated into compounds from carbon dioxide, water, or molecular oxygen by each responsible enzyme. Among them, oxygenases (EC1.13 and EC1.14) are well known for transferring an oxygen atom from molecular oxygen, and not from carbon dioxide or water, into metabolites (Hayaishi et al. 1955;Mason et al. 1955 with 18 O 2 (molecular oxygen) in vivo and analyzed the origin of the oxygen atoms of humulone and cohumulone by nuclear magnetic resonance (NMR) spectrometry (Hecht et al. 2004). The result indicated that oxygen atoms from 18 O 2 were distinguishable from unlabeled O 2 . Further, in vivo labeling with 18 O 2 can be effective for suggesting the existence of an oxygenase in a biosynthetic pathway. However, there are no reports that have applied 18 O 2 (molecular oxygen) for MS-based comprehensive metabolome analysis.

Analysis of the labeling efficiencies for 13 C and 15 N by sealed combustion method
The whole plants were lyophilized and ground by mortal. Samples were placed in quartz tubes that contained copper oxide. In the presence of Cu and Ag foil, the tubes were evacuated, sealed, and heated at 500 °C for 30 min and at 850 °C for 2 h, followed by cooling to room temperature. After combustion, the sample gas purified by cold trapping technique was analyzed by mass spectrometry RMI-2 (Hitachi, Japan).

Liquid chromatography-mass spectrometry (LC-MS) measurement
The analysis was performed as described previously (Kera et al. 2014

Data analysis
Data analysis was performed as described previously (Kera et al. 2014). All raw data files were converted into text files using MSGet (http://www.kazus a.or.jp/komic s/softw are/ MSGet ), and peaks were extracted by PowerFT (http:// www.kazus a.or.jp/komic s/softw are/Power Get/) (Sakurai et al. 2014). Unlabeled and blank data were aligned using PowerMatch (http://www.kazus a.or.jp/komic s/softw are/ Power Get/) with manual curation, and valid peaks were selected after background correction. In the annotation process, the primary database search and the primary estimation of elemental composition were performed using PowerMatch and MFSearcher within a mass accuracy of 3 ppm (http://webs2 .kazus a.or.jp/mfsea rcher /) (Sakurai et al. 2013). Labeling data were further analysis by Shift-edIonsFinder (Kera et al. 2014).
To find the isotopic peaks in the labeled data and associate them with monoisotopic peaks from the unlabeled data, the labeled peak search was performed with the parameters as follows: Max fold, C = 100, N = 50, O = 50, S = 50; Mass difference = 3 ppm; and RT difference = 1. In flavonoid-like peak analysis, candidate peaks with chemical modification, such as glycosylation and acylation, were searched by comparing the peak list containing mass information of aglycone of typical flavonoids against the peak list from an unlabeled sample using ShiftedIonsFinder. Xylosylation (Xyl) ( , m/z 100.016044) were selected as modification groups. The settings were as follows: Max fold, Glc = 3, Rha = 3, GlcUA = 3, Cinnamoyl = 3, Coumaroyl = 3, Caffeoyl = 3, Feruloyl = 3, Malonyl = 3, Succinyl = 3; Mass difference = 3 ppm; RT difference = 60. The exported file from ShiftedIonsFinder was arranged by Excel.

UHPLC-ESI-QTOF-MS/MS measurement
The metabolites, extracted with 80% methanol were injected in a Waters ACQUITY UPLC (Waters, USA) coupled to a Bruker maXis Impact ESI-QTOF-MS system having a mass resolution of ∼ 40,000 (Bruker Daltonics, USA). The identification of the metabolite was confirmed by NMR following UHPLC-solid-phase extraction (SPE) using a Bruker/Spark Holland Prospekt II SPE system as described previously (Qiu et al. 2016). Details were provided in supplemental data 4.

Establishment of an economical sealed-glass bottle system
Although many stable isotopes, such as, 13 (Hecht et al. 2004). In their system, several cones were enclosed in glass vessels filled with 20% 18 O 2 and 80% argon, and were incubated for 14 days without gas exchange. Therefore, we used a sealed-glass bottle with air filters (ADVANTEC, Japan) ( Fig. 1) for gas exchange. Additionally, unlabeled and 13 C-, 15 N-, and 34 S-labeled samples were prepared using the same setup, respectively. The labeling efficiencies for 13 C-labeled and 15 N-labeled samples were 59.6 and 78.8%, respectively, while those of unlabeled sample were 1.09 and 0.365%, respectively. These were not as high compared with those in previous reports, in which greater than 90% efficiency was achieved using a growth chamber (Giavalisco et al. 2011). In order to grow each labeling plants in equal condition, 13 C-labeled air was exchanged daily in this study. However, continuous gas flow might improve it. For 15 N-labeling, exchanging medium during long cultivation might effective. To enable exchange of medium, we plan to use liquid medium and modify Leonard jar (Leonard 1943) to keep sealed and exchange the gas in the future. Moreover, 78.8% of 15 N labeling efficiency was not perfect but it seemed to be enough for further curation of the elemental composition. Because incorporation of 18 O or 34 S cannot be measured directly, the 80% methanol extraction of unlabeled, 18 O-labeled, and 34 S-labeled shoots were analyzed by LC-Orbitrap-MS using a mass accuracy of 3 ppm. The incorporation was manually confirmed in the mass spectra using Xcalibur (Thermo Fisher Scientific, USA) (data not shown). These results suggest that this labeling system works, although that there is still room for improvement in labeling efficiency.

Untargeted metabolite annotation of peaks able to be labeled by 18 O 2
For target analysis, whether a target metabolite is labeled by 18 O 2 or not, was investigated by searching for the calculated m/z values of 18 O-labeled peaks in the 18 O 2 labeled sample.
Here, this is calculated by adding the m/z difference between 16 O and 18 O of 2.0042462 to that of a target monoisotopic peak. However, this approach is not applicable to non-target analyses, because the monoisotopic m/z values of peaks that can be labeled by 18 O 2 are not assigned. One effective approach is to compare mass spectra from an unlabeled sample with those from a 18 O 2 labeled sample, and associate the peaks in the unlabeled sample with the peaks in the 18 O 2 labeled sample based on the specified m/z difference responsible for 18 O-labeling. Therefore, the comprehensive search for peaks labeled by 18 O 2 can be accomplished in four steps: (1) preparation of an unlabeled peak list from unlabeled samples with tight intensity filter to reduce noise: (2) preparation of an 18 O-labeled peak list from 18 O 2 labeled samples with a loose intensity filter to pick tiny peaks; (3) search for peaks labeled by 18 O 2 by association of the unlabeled peak list with the 18 O-labeled peak list using ShiftedIonsFinder (Kera et al. 2014), and (4) curation of the search results by manually checking the mass spectra. Initially, a total of 3,417 peaks and 3,518 peaks were extracted as unlabeled peaks from M. truncatula shoots and roots, respectively. Because analysis using ShiftedIonsFinder requires manual curation of the mass spectra, we selected peaks labeled by 18 O 2 according to the search result of ShiftedIonsFinder and the existence of MS 2 information for further annotation steps. Finally, a total of 511 peaks and 353 peaks were extracted as peaks labeled by 18 O 2 from M. truncatula shoots and roots, respectively (Supplement data 1, 2). The elemental compositions of peaks labeled by 18 O 2 were first predicted using MFSearcher (Sakurai et al. 2013). Because multiple elemental compositions are generally predicted for peaks with high m/z values, even if the deviation of m/z value is within 1 ppm, the predicted elemental compositions were manually curated according to a strategy in which the number of stable isotopes incorporated in the compound is used as an index (Giavalisco et al. 2011;Kera et al. 2014). As well as detecting 18 O-labeled peaks, the corresponding 13 C-, 15 N-, and 34 S-labeled peaks labeled by 18 O 2 were found from each labeled sample by comparing with unlabeled peak list using ShiftedIonsFinder. Considering that the labeling efficiency was not enough to determine the max number of each element, we excluded the elemental compositions in which the number of each element was less than that of detected.  The sealed-glass bottle system for economical labeling. The normal or labeled air was blown into the bottle through an air filter. After gas exchange, the flow channels of the filter were closed. The other labeled nutrients were supplied in the medium beforehand and methods"), suggesting that many secondary metabolites can be labeled with 18 O 2 by oxygenases (Table 2).

Identification of 18 O-labeled unknown flavonoids
As described in a previous report, ShiftedIonsFinder is useful for finding candidates of known and unknown flavonoids by calculating theoretical m/z values by adding modification groups to aglycone like a building block (Kera et al. 2014). Potential flavonoids were searched for within the 511 peaks of M. truncatula shoot with regard to some modification groups. As a result, 26 peaks were picked up as flavonoid candidates (Supplement data 3). As an example, we focused on peak No. 312 (ion 1: m/z 769.1623 in Fig. 2a), one of the flavonoid candidates whose mass was not matched to referenced databases. We first attempted to estimate its structure by a MS-based approach. This peak yielded three candidate elemental compositions, C 47 H 29 O 9 P 1 , C 36 H 32 O 19 , and C 54 H 24 O 6 within 3 ppm, and 38 candidate flavonoids. In order to estimate its detailed structure, additional MS n analyses with high resolution and high mass accuracy using Orbitrap Fusion (Thermo Fisher Scientific, USA) were performed. At first, three major fragments (ion 2: m/z 271.0596; ion 3: m/z 323.0756; and ion 4: m/z 447.0916) were detected by MS 2 analysis from the parent monoisotopic ion (Fig. 2b).
In positive electrospray ionization-mode fragmentation of flavonoid glycosides, it is well known that after cleavage of glycosidic bonds between the glycoside group and aglycone (Cuyckens and Claeys 2004), the product ion for a flavonoid aglycone results. It is strongly suggested that ion 2 at m/z 271.0596 is the aglycone. Among the 38 candidate flavonoids suggested by ShiftedIonsFinder, the structure having such an aglycone was [Aglycone + 2 GlcUA + Coumaroyl + H] + (  (Sumner et al. 2014(Sumner et al. , 2015. To date, many types of apigenin glucuronides have been reported from M. truncatula (Kowalska et al. 2007;Marczak et al. 2010) and Medicago sativa (Stochmal et al. 2001a, b (Fig. 2c). In some, but not all, cases, MS n analysis can indicate a more confined 18 O-labeled position. In MS 3 analysis of apigenin (ion 2), two characteristic fragments (ion 13: m/z 119.0631 and ion 14: m/z 153.0669) are known retro-Diels-Alder fragment ions (Fig. 2d). Focusing on the responsible ions (ion 15: m/z 121.2355 and ion 16: m/z 153.0940) in MS 3 analysis of ion 10, the 18 O responsible mass shift was detected between ion 13 and ion 15, but not between ion 14 and ion 16 (Fig. 2d).
In addition to ion 13, other fragment ions containing a 4′-hydroxy group (ion 17: m/z 145.0565 and ion 18: m/z 229.0727) also showed the 18 O mass shift, whereas ion 16 contains all four oxygen atoms except the 4′-hydroxy group. Hence, the 18 O atom was specifically incorporated into the 4′-hydroxy group of apigenin. According to the apigenin biosynthetic pathway, the hydroxy groups at positions 1, 4, 5, and 7 are derived from malonyl CoA, which is synthesized by an oxygenase-independent pathway, and the 4′-hydroxy group is derived from p-coumaroyl CoA by cinnamic acid 4-hydroxylase (C4H), which is a cytochrome P450 monooxygenase ). Thus, specific incorporation into the 4′-hydroxy group is reasonable and indicates that nonspecific incorporation of an 18 O atom by recycling during one month of labeling is unlikely compared with the oxygenase-catalyzing reaction. On the other hand, by MS 3 analysis of ion 3: m/z 323.0756 and ion 11: m/z 325.0797, one major fragment (m/z 147.0439 [Coumaroyl]) was detected from ion 3, and one major fragment (m/z 149.0479) was detected from ion 11, indicating that one 18 O atom in ion 11 was derived from [Coumaroyl] (Fig. 2e). We could not determine the detailed labeling position by MS n analysis, but an 18 O atom was also thought to be incorporated at the C-4 position of [Coumaroyl] by C4H. Finally, tiny ion 8: m/z 775.1750 and ion 9: m/z 777.1797 were responsible for the incorporation in [GlcUA] (data not shown). Since UDP-glucuronic acid is biosynthesized in plants through both a UDP-glucose 6-dehydrogenase-catalyzing reaction, in which H 2 O is used as an oxygen source, and a myo-inositol oxygenase-catalyzing reaction, in which O 2 is used (Roberts 1971), these weak ions might reflect the biosynthetic pathway in M. truncatula. Consequently, these data well supported that ion 7 containing two 18 O atoms from [Coumaroyl] was more abundant.

Concluding remarks
We have studied pathway-specific metabolome analysis     (Kera et al. 2014) pathway-specific metabolome analysis can be a key for linking metabolite information to enzymatic information.