Discovering Regulated Metabolite Families in Untargeted

Subscriber access provided by STEPHEN F AUSTIN STATE UNIV

Article

Discovering regulated metabolite families in untargeted metabolomics studies Hendrik Treutler, Hiroshi Tsugawa, Andrea Porzel, Karin Gorzolka, Alain Tissier, Steffen Neumann, and Gerd Ulrich Balcke Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.6b01569 • Publication Date (Web): 24 Jul 2016 Downloaded from http://pubs.acs.org on July 25, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Analytical Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 18

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

1

Discovering regulated metabolite families in

2

untargeted metabolomics studies

3

Hendrik Treutler2, Hiroshi Tsugawa4, Andrea Porzel3, Karin Gorzolka2, Alain Tissier¹, Steffen

4

Neumann2, and Gerd Ulrich Balcke¹*

5 6 7 8 9 10 11

¹Leibniz Institute of Plant Biochemistry, Dept. of Cell and Metabolic Biology, Weinberg 3, D-06120 Halle/Saale, Germany 2

Leibniz Institute of Plant Biochemistry, Dept. of Stress and Developmental Biology, Weinberg 3, D-06120

Halle/Saale, Germany 3

Leibniz Institute of Plant Biochemistry, Dept. of Bioorganic Chemistry, Weinberg 3, D-06120 Halle/Saale,

Germany 4

RIKEN Center for Sustainable Resource Science, Yokohama, Kanagawa, 230-0045, Japan

12

ABSTRACT: The identification of metabolites by mass spectrometry constitutes a major bottleneck which considerably limits the throughput of metabolomics studies in biomedical or plant research. Here, we present a novel approach to analyze metabolomics data from untargeted, data-independent LC-MS/MS measurements. By integrated analysis of MS1 abundances and MS/MS spectra, the identification of regulated metabolite families is achieved. This approach offers a global view on metabolic regulation in comparative metabolomics. We implemented our approach in the web application ‘MetFamily’, which is freely available at http://msbi.ipb-halle.de/MetFamily/. MetFamily provides a dynamic link between the patterns based on MS1-signal intensity and the corresponding structural similarity at the MS/MS level. Structurally related metabolites are annotated as metabolite families based on a hierarchical cluster analysis of measured MS/MS spectra. Joint examination with principal component analysis of MS1 patterns, where this annotation is preserved in the loadings, facilitates the interpretation of comparative metabolomics data at the level of metabolite families. As a proof of concept, we identified two trichomespecific metabolite families from wild-type tomato Solanum habrochaites LA1777 in a fully unsupervised manner and validated our findings based on earlier publications and with NMR. 13 14

ACS Paragon Plus Environment


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

15

INTRODUCTION

16

Metabolomics experiments provide small molecule measurements from biological samples in

17

a broad range of applications including cancer research, drug development, and plant

18

science1-5. Mass spectrometry (MS) coupled to liquid chromatography (LC) is an essential

19

analytical technology to acquire a snapshot of the metabolic state of a sample. Based on

20

untargeted MS measurements, it is possible to measure thousands of detectable signals as

21

MS1 features per chromatographic run and to acquire signal profiles of small molecules

22

based on retention time (RT), accurate mass-to-charge ratio (m/z), and abundance6.

23

Univariate or multivariate statistical analysis is then applied to signal profiles of different

24

sample groups to detect MS1 features that are group-discriminating or of interest based on

25

the experimental design.

26

Hints for the structural characterization or even identification of MS1 features are obtained

27

from tandem MS measurements (MS/MS), where the metabolites undergo fragmentation

28

resulting in MS/MS spectra. MS/MS spectra can be collected by data-dependent acquisition

29

(DDA) or in data-independent acquisition (DIA) mode, requiring a trade-off between dwell

30

time and spectral purity7,8. Using DIA, it is possible to collect thousands of MS1 features from

31

a single LC run as well as the associated MS/MS spectra9. However, in most studies, the

32

identity of the vast majority of MS1 features is unknown. Structure elucidation of each

33

individual MS1 feature from complex biological samples, e.g. by NMR and interpretation of

34

MS/MS spectra, is currently out of reach. Thus, the biochemical relation between MS1

35

features remains largely unexplained.

36

Group-discriminating MS1 features are often structurally related, e.g. if particular metabolic

37

pathways are differentially regulated as a consequence of disease10, stress11, genetic

38

manipulation12 or in the case of organ-specific accumulation of structurally related

39

metabolites13. Structurally related metabolites often exhibit latent similarity in their MS/MS

40

spectra in which characteristic fragmentation patterns arise from common functional groups

41

or structural features. For instance, upon negative mode ionization and collision-induced

42

dissociation (CID), adenylated metabolites such as adenyl nucleotides, CoA esters, and

43

NAD cofactors form a fragment ion of m/z 134.0472 Da (C5N5H4-), which corresponds to the

44

mass of the purine core element. Under the same conditions, glucosides often form a

45

fragment ion of m/z 161.0455 Da (C6H9O5-), characteristic of the hexose side-chain. Thus,

46

based on existing information, precursor ions showing these characteristic fragments could

47

be grouped together as metabolites sharing common structural features, or metabolite

48

families. However, even pre-existing MS/MS information characteristic of certain metabolite

49

families is sparse. Hence, novel approaches that turn MS1- and MS/MS-features into

50

interpretable information within a reasonable amount of time are urgently needed. These

51

approaches should be able to relate MS1 abundances to latent similarity at the MS/MS

52

spectral level.


Page 2 of 18

Page 3 of 18

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


53

Recently, several studies reported on the organization of hundreds of MS1 features by

54

molecular networking depicting relationships between structurally related molecules based

55

on their spectral similarity14-17. However, an explicit assignment of MS1 features to similarity

56

clusters and the source of structural similarity between up- or downregulated MS1 features

57

was not apparent. Previously, Wagner et al. used GC-MS data for hierarchical cluster

58

analysis (HCA) to arrange known and structurally related metabolites18. Using HCA, it was

59

possible to identify structural classes amongst 59 metabolites. Rasche et al. described FT-

60

BLAST19 to compare spectra and computationally derived fragmentation trees, revealing

61

clusters of structurally closely related compounds. However, neither Wagner et al. nor

62

Rasche et al. considered the abundance of MS1 features in different samples.

63

Inspired by the idea to comprehensively analyze molecular networks and to explicitly group

64

MS1 features, we performed HCA across hundreds of MS/MS spectra obtained from

65

glandular trichomes of wild-type tomato Solanum habrochaites LA1777. Glandular trichomes

66

of vascular plants such as tomato are metabolic factories producing a plethora of secondary

67

metabolites involved in plant defense and the communication with its environment13,20. We

68

considered characteristic fragments prevalent in MS/MS similarity clusters to assign MS1

69

features to certain trichome-specific metabolite families. In addition, we applied principal

70

component analysis (PCA) to metabolite profiles for the discovery of group-discriminating

71

MS1 features and combined the information on metabolite families obtained from HCA

72

(MS/MS feature similarity) with the PCA loadings (sample-specific MS1 abundance). This

73

combination of statistical analyses of MS1 feature abundances and MS/MS structural

74

annotations can not only speed-up the individual analysis steps, but allows to address new

75

questions, such as the discovery of group-discriminating metabolite families with biochemical

76

relevance. Here, we exemplarily selected two metabolite families being produced by tomato

77

glandular trichomes which play important roles in the plant defense against herbivores,

78

namely the branched chain acyl sugars21-24 and the sesquiterpene glucosides which are

79

potentially poisonous to plant herbivores25,26. We implemented the proposed methodology in

80

the Open Source web application ‘MetFamily’ and made our approach freely available

81

(accessible via http://msbi.ipb-halle.de/MetFamily/).

82

MATERIALS AND METHODS

83

Fragment matrix assembly. MetFamily processes a metabolite profile of a set of MS1

84

features together with an MS/MS library comprising MS/MS spectra for these MS1 features.

85

We obtain both data sets as output of MS-DIAL9, where the metabolite profile contains

86

extracted m/z / retention time features from MS1 scans with the corresponding feature

87

abundances (Data S-1) and the MS/MS library contains deconvoluted MS/MS spectra of the

88

MS1 features with relative intensities of the fragment ions (Fig. 1, Data S-2). Instead of MS-

89

DIAL, other tools can produce similar input data as described in Note S-3. Upon data import,

90

MetFamily aligns all MS/MS spectra with a user-defined m/z error to create the fragment



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 4 of 18

91

matrix as shown in Fig. 2, where the relative intensity of unique MS/MS fragments is

92

associated with the corresponding MS1 feature (i.e. precursor ion) and its MS1 abundance in

93

individual samples (Data S-3). For our showcase this preprocessing step takes one or two

94

minutes. The fragment matrix is assembled as follows.

95

First, we process the set of all fragments. Here, we remove fragments with an intensity

96

below a user-defined noise threshold. We normalize fragment intensities within each MS/MS

97

spectrum to a maximum of 1 (base peak). In addition, we add one neutral loss (NL) for each

98

fragment by calculating the mass of the neutral loss as the difference of fragment m/z and

99

precursor m/z in MS1 (intentionally a negative m/z value). The intensity of the NLs is chosen

100

equal to the intensity of the corresponding fragment. In this manuscript, we treat fragments

101

and NLs equally by denoting both as fragments.

102

Second, we align the individual MS/MS spectra (Figures 1 and 2). Here, we match fragments

103

from different MS/MS spectra with similar m/z and merge these to fragment groups of unique

104

m/z. We call the mean of all fragment m/z’s of one fragment group the fragment group mean.

105

For the alignment of the individual MS/MS spectra, we use an efficient algorithm

106

implemented in the R package xcms31 (version 1.44.0). This algorithm avoids the usage of

107

fixed m/z bins with a heuristic approach that groups fragments with similar m/z and

108

decomposes contiguous fragment groups using hierarchical clustering. Here, a fragment m/z

109

matches a fragment group, if

110

|

−

|≤

/

+

∗

/

/1 6,

111

where

112

parameter representing the absolute m/z error, and

113

m/z error in ppm (parts per million). See Table S-3 for a summary of user-customizable

114

parameters. After fragment group assembly, we remove fragment groups which correspond

115

to isotopic ions. Specifically, we detect fragment groups with a m/z difference of 1.0033 Da

116

(regarding the fragment group means +/- m/z error) which correspond to 13C isotopes. Third,

117

we create the fragment matrix with one row for each unique MS1 precursor and columns of

118

fragment groups (Fig. 2). We register the intensity of each fragment in the row and column of

119

the corresponding MS1 feature and fragment group, respectively. For each MS1 feature, we

120

generate an ID given by “m/z / retention time” in MS1.

121

Finally, we add the set of MS1 abundances in all samples and other annotations to each row

122

resulting in a combined data matrix. The combined data matrix represents the data basis for

123

subsequent analyses and can be examined in a spreadsheet program for complementing

124

analyses (Fig. 2 and Data S-3).

125

MS1 / MS/MS combined data analysis. A principal component analysis (PCA) for the set of

126

is the fragment m/z,

MS1 features in

is the fragment group mean, /

samples is performed as follows. Given the


/

is a

is representing the relative

by

matrix of scaled MS1

Page 5 of 18

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


127

abundances, we calculate the scores and the loadings. Here, MetFamily supports the

128

scaling functions log2 transformation, Pareto scaling, Centering, and Autoscaling27. The

129

scores comprise one data point per sample and reflect differences between samples. The

130

loadings comprise one data point per MS1 feature and emphasize MS1 features with

131

differential abundance between samples.

132

We perform a hierarchical cluster analysis (HCA) on MS/MS spectra of a set of MS1

133

precursor features as follows. We calculate the distance matrix of pairwise dissimilarities

134

between the MS/MS spectra of all MS1 features. Here, we provide different distance

135

functions to score common and distinct fragments. Specifically, we recommend the distance

136

function ‘Jaccard (intensity-weighted)’, which sums the intensities of common and disjoint

137

fragments:

138

!( !# ( $ ∩ !( !# ( $ ∪

( , )=1−

& ))

,

& ))

are the fragments in the MS/MS spectrum of MS1 feature ( and ) . To

139

where

140

suppress noise and emphasize the importance of intense fragments,

141

intensities of the fragments as follows. Intensities smaller than 0.2 are mapped to 0.01,

142

intensities greater or equal than 0.2 and smaller than 0.4 are mapped to 0.2, and intensities

143

greater or equal than 0.4 are mapped to 1. Given the distance matrix, we calculate a

144

hierarchical cluster dendrogram where each cluster of MS1 features represents a putative

145

metabolite family.

146

For each cluster of MS/MS spectra, we calculate the cluster-discriminating power for

147

prevalent fragments as follows. For each fragment present in more than 50% of the MS/MS

148

spectra in a cluster, we measure the ability of this fragment to discriminate spectra in the

149

cluster from spectra outside the cluster as

and

,-+(

.,/ )

=

+0−+

*+ discretizes the

1

150

where

151

3-th cluster containing the fragment

152

th cluster containing the fragment

153

cluster. If +

154

fragment is in the range from zero to one and a fragment with a cluster-discriminating power

155

close to one indicates a very specific fragment.

156

Clusters containing fragments with a cluster-discriminating power close to one indicate

157

metabolite families. Currently, the annotation of metabolite families based on characteristic

158

MS/MS fragments is performed by a mass spectrometry expert who manually evaluates the

159

hierarchy of putative metabolite families and labels a set of clusters with functional and/or

.,/

is the 2-th fragment of the 3-th cluster, + 0 is the number of MS/MS spectra in the

1

.,/ ,

.,/ ,

+

and

> + 0 , then we define ,-+(

1

is the number of MS/MS spectra outside the 3is the total number of MS/MS spectra in the 3-th

.,/ )

= 0. The cluster-discriminating power of a



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

160

structural annotations based on characteristic fragment patterns. Each MS1 feature can be

161

labeled with one annotation, i.e. membership in a metabolite family.

162

Plant growth and harvest. Solanum habrochaites LA1777 was grown on soil in a

163

greenhouse (65% humidity, light intensity: 165 µmol s-1 mm2, 21-24 °C, 16 h light period) and

164

watered with tap water every two days. The plant material was harvested 12 weeks after

165

germination during the light phase in the early afternoon. For trichome harvest, tomato

166

leaves were put on the hand palm (using gloves) and trichomes were quickly brushed off the

167

leaves by a 2 cm broad paint brush which was dipped in liquid nitrogen. The frozen

168

trichomes were collected in a mortar filled with liquid nitrogen. Trichomes from 15 plant

169

leaves were pooled under cryogenic conditions and further purified by sieving through steel

170

sieves of 150 µm mesh width (Retsch, Hahn, Germany). After removal of trichomes, the

171

plant leaves were immediately quenched in liquid nitrogen. Pooled leaves were ground in a

172

mortar under liquid nitrogen conditions. After evaporation of all liquid nitrogen during storage

173

at -80 °C leaves and trichomes were lyophilized overnight and stored in a deep freezer until

174

extraction.

175

Metabolite extraction. Using wall-reinforced cryo-tubes of 1.6 mL volume (Precellys Steel

176

Kit 2.8mm, Peqlab Biotechnologie GmbH, Erlangen, Germany) filled with 5 steel beads (3

177

mm), 25 mg aliquots of dry leaf or trichome powder was suspended in 900 µL

178

dichloromethane/ ethanol (-80 °C). Then, 200 µL of 50 mM aqueous ammonium formate/

179

formic acid buffer (0 °C, pH 3) was added to each vial and two rounds of cell rupture/

180

metabolite extraction were conducted by FastPrep bead beating (60 s, speed 5.5 m/s, 1st

181

round -80 °C, 2nd round room temperature, FastPrep24 instrument with cryo adapter, MP

182

Biomedicals LLC, Santa Ana, CA, USA). After phase separation by centrifugation at 20,000

183

g (2 min, 0°C) the aqueous phase was removed and 600 µL of the organic phase was

184

collected. Following, 500 uL tetrahydrofuran (THF) was added to exhaustively extract

185

hydrophobic metabolites and the Fastprep and centrifugation were repeated accordingly.

186

The THF supernatant was combined with the first organic phase extract and dried in a

187

stream of nitrogen gas. The dried extract was re-suspended in 150 µL 75% methanol

188

(aqueous) and filtered over 0.2 µm PVDF.

189

Analytical conditions for liquid chromatography and mass spectrometry. 0.5 µL

190

methanolic extract was injected into an Acquity-UPLC (Waters Inc.) and separated on a

191

Nucleoshell RP18 (150 mm x 2 mm x 2.7 µm; Macherey & Nagel, Düren, Germany) at 40°C.

192

The mobile phase A was 0.33 mM ammonium formate with 0.66 mM formic acid in water;

193

mobile phase B was acetonitrile. The gradient was 0 min, 5% B; 2 min, 5 % B; 19 min, 95%

194

B; 21 min, 95% B; 21.1 min, 5% B; 24 min, 5% B. The column flow rate was 0.4 mL/min, the

195

autosampler temperature was 4 °C.


Page 6 of 18

Page 7 of 18

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


196

ESI-(-)-Mass spectrometry was performed on an AB Sciex TripleTOF 5600 system (Q-TOF)

197

equipped with a DuoSpray ion source. All analyses were performed at the high sensitivity

198

mode for both TOF MS1 and product ion scan. The mass calibration was automatically

199

performed every 20 injections using an APCI calibrant solution via a calibration delivery

200

system (CDS). The instrument (TripleTOF 5600, Sciex, Toronto, Canada) was configured to

201

simultaneously acquire high resolution MS/MS spectra for all MS1 features (sequential

202

window acquisition of all theoretical fragment-ion spectra, SWATH)28 (Fig. S-1). The SWATH

203

parameters were MS1 accumulation time, 150 ms; MS2 accumulation time, 20 ms; collision

204

energy, -45 V; collision energy spread, 35 V; cycle time, 1160 ms; Q1 window, 25 Da; mass

205

range, m/z 65–1250. The other parameters were curtain gas, 35; ion source gas 1, 60; ion

206

source gas 2, 70; temperature, 600 °C; ion spray voltage floating, -4.5 kV; declustering

207

potential, 35 V.

208

Raw data processing. After measurement, raw data of triplicate trichome and trichome-free

209

leaf material was converted from the vendor file format (in our case *.wiff) into the common

210

file format of Reifycs Inc. (Analysis Base File format *.abf) using the freely available Reifycs

211

ABF converter (http://www.reifycs.com/AbfConverter/index.html).This process took about one

212

minute per sample. After conversion, the freely available MS-Dial software was used for

213

feature detection, ion species annotation, compound spectra extraction, and peak alignment

214

between samples9. Data processing by MS-Dial using the parameters in Table S-1 took

215

about 30 min. Data processing by MetFamily using the parameters in Table S-2 took 1 min.

216

Notably, neither the use of SWATH-triggered CID fragmentation nor the use of MS-Dial are

217

prerequisite to run MetFamily. Any data independent or data dependent acquisition to collect

218

MS/MS spectra and other peak picking and deconvolution software can alternatively be used

219

29-32

220

and a msp-type spectral library which are formatted as exemplified in Data S-1 and Fig. 1,

221

and described in Note S-3. However, as unique feature, MS-Dial jointly deconvolutes MS1

222

and MS/MS features and automatically predicts the precursor ion when DIA was applied. Via

223

the Reifycs ABF converter, MS-DIAL accepts all of major MS vendor-formats as well as the

224

common mzML data and is applicable to either DIA or DDA MS/MS fragmentation methods.

225

Substance Purification. Since NMR requires purified analytes in the upper µm range, 1 kg

226

of LA1777 leaf material was surface-extracted with methanol for 2 h. After evaporation, a

227

methanolic concentrate of this extract was produced and injected into a LC system in 100 µL

228

increments. For peak separation using semi-preparative HPLC and an analysis by mass

229

spectrometry (1260 Infinity system, Agilent), a full scan between 200-800 m/z was performed

230

after negative electrospray ionisation (ion source: API-ES, gas temperature: 350 °C, drying

231

gas 10 mL/min, nebulizer pressure 35 psig, capillary voltage 4500 V). For HPLC, a XTerra

232

prep MS C18 column (5 µm x 7.8 mm x 150 mm; Waters) was used and run at a flow rate of

233

6 mL/min at 25 °C. Solvent A was 0.3 mM ammonium formate acidified with formic acid to

. In that case, their output has to be provided as a text file containing the peak intensities



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

234

pH 6.2. Solvent B was acetonitrile. Gradient conditions were: 0-5 min 5% B; 5-87 min linear

235

gradient to 95% B; 87-88 min 95% B; 88-90 min 5% B. For fractionation, m/z 605.5, 737.5,

236

and 751.5 triggered the selective collection. A make-up pump that transferred an aliquot of

237

the eluate to the mass analyser was set to 0.5 mL/min 50% A - 50% B. Subsequently, all

238

collected fractions were dried by lyophilisation prior to NMR analysis.

239

Analytical conditions for NMR. NMR spectra were recorded on an Agilent/Varian VNMRS

240

600 NMR spectrometer operating at a proton NMR frequency of 599.83 MHz using a 5 mm

241

inverse detection cryoprobe. 2D NMR spectra were recorded using standard pulse

242

sequences (gDQCOSY, zTOCSY, gHSQCAD, gHMBCAD) implemented in Agilent (Varian)

243

VNMRJ 4.2A (CHEMPACK 7.1) spectrometer software. A TOCSY mixing time of 80 msec

244

was used. HSQC experiments were run with multiplicity editing and optimized for 1JCH = 146

245

Hz. HMBC experiments were optimized for a long-range coupling constant of 8 Hz; a 2-step

246

1

247

internal TMS (0 ppm).

248

RESULTS and DISCUSSION

249

As a proof of concept, we applied MS signal profiles to compare the metabolism of a special

250

plant organ in tomato, the glandular trichomes, to tomato leaves. Plant glandular trichomes

251

are secretory cells that protrude from the epidermis of many vascular plants. As “metabolic

252

factories”, they produce important drugs such as the antimalaria artemisinin or compounds

253

known to be involved in plant defense20,33. Here, we used Solanum habrochaites LA1777, a

254

wild type tomato accession with a rich profile of secondary metabolites produced in the

255

glandular trichomes34. We used six UPLC-(-)ESI-SWATH-MS/MS runs of triplicate trichome

256

and trichome-free leaf extracts (cf. Materials and Methods). However, MetFamily is

257

applicable to a larger number of samples and sample groups. We used MS-DIAL9 for data

258

preprocessing and exported (i) a signal profile with MS1 features and (ii) a spectral library

259

with deconvoluted MS/MS spectra extracted from the raw data (Data S-1 and Data S-2).

260

Using the software MetFamily, we aligned the MS/MS spectra of the spectral library resulting

261

in a novel fragment matrix structure and we fused this fragment matrix with the matching set

262

of MS1 features from the six individual samples to a single matrix (cf. Materials and Methods,

263

Figures 1 and 2, Data S-3, Table S-3).

JCH filter was used (130–165 Hz). Proton and carbon chemical shifts are referenced to


Page 8 of 18

Page 9 of 18

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


264 265

Figure 1. MS/MS library format before upload into MetFamily.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 10 of 18

266 267

Figure 2. Combined data matrix after data pre-processing by MetFamily. The quantification

268

part (red, left) contains the MS1 features (rows; precursor ions) and the MS1 abundances in

269

individual samples. In the fragment part (green, right), the column headers are the mean of

270

binned MS/MS features (m/z or neutral loss) from the MS/MS library. Upper zoom: m/z;

271

retention time of feature (628.2452; 9.16) and its respective peak heights in two trichome

272

samples. Lower zoom: relative MS/MS intensities of fragment ion m/z 323.09570 Da. Arrows

273

to the left and to the right: MS1 abundances are analyzed using PCA and MS/MS spectra are

274

analyzed using HCA.

275


Page 11 of 18

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


276

MetFamily provides options to perform principal component analyses (PCA). Here, we

277

performed a PCA on 2585 MS1 features detected in glandular trichomes or leaves of LA1777

278

using Pareto-scaled data. In our example, PC1 shows a clear separation between trichomes

279

and leaves with R2=0.90, Q2=0.82 and a large number of MS1 features more abundant in

280

glandular trichomes (Fig. 3A and B). A Scree plot on additional principal components is

281

provided in Fig. S-2. Up to this point, all data has been acquired in a fully untargeted manner

282

and traditionally this is where group-discriminating MS1 features would be subjected to

283

tedious manual structure elucidation. In our approach, we amended the loadings plot of the

284

PCA (Fig. 3B) with a set of structural annotations based on characteristic MS/MS fragments

285

which we identified in different signal-clusters using HCA (Fig. 4). Using MetFamily, we

286

performed a hierarchical cluster analysis (HCA) on MS/MS spectra of the fragment matrix

287

(Data S-3). For PCA as well as for HCA, MetFamily allows the usage of thresholds for the

288

MS1 abundance of individual MS1 features (average of all samples) and for the log2-fold

289

change between the average MS1 abundance of two sample groups. Since we were

290

interested in abundant trichome-specific metabolites, we retained 135 MS1 features in the

291

HCA with MS1 abundances ≥20,000 counts and a log2-fold change ≥2 comparing trichomes

292

versus leaf. After hierarchical cluster analysis, the resulting dendrogram indicated a clear

293

segregation into two main clades with internal spectral similarity (Fig. 4).

294

The first signal-cluster contained 73 MS1 features which correspond to short branched chain

295

acyl sugars21 (AS, blue in Fig. 4). The structural similarities among members of this clade

296

was supported by prevalent fragment ions 87.0451 Da (theoretical mass for C4H7O2- is

297

87.0452) and 101.0603 Da (theoretical mass for C5H9O2- is 101.0608), which are indicative

298

for short branched acyl groups. These acyl moieties were esterified to sucrose as reflected

299

by the fragments 323.0957 Da (theoretical mass for C12H19O10- is 323.0984; sucrose-H2O-H-)

300

and 305.0864 Da (theoretical mass for C12H17O9- is 305.0878; sucrose-2H2O-H]-). MS/MS

301

fragmentation patterns and NMR analysis of two selected MS1 features of this clade ([m/z;

302

RT]: [737.3578; 14.65] and [751.3749; 15.64]) confirmed the membership to the metabolite

303

family of short branched chain acyl sugars (Figures S-3, S-4, S-7, S-8, S-11 - S-14, S-15 - S-

304

19 and Tables S-5, S-6). Our NMR analysis revealed that the feature [737.3578; 14.65]

305

comprised an isomeric mixture of isobutyl, isopentyl, and anteisobutyl acyl moieties, which

306

were not resolvable using our chromatography. MS/MS fragmentation and NMR of various

307

AS have been thoroughly studied earlier by Ghosh et al., where compounds selected here

308

for analysis were annotated as acylsucrose S4:21[2] (theoretical m/z:737.36012 Da (formate

309

adduct-H)) and acylsucrose S4:22[6] (theoretical m/z: 751.37577 Da (formate adduct-H)),

310

respectively21.

311

The second signal-cluster contained a group of four MS1 features which correspond to

312

sesquiterpene glycosides (SQT-glucosides, red in Fig. 4). The structural similarities among

313

members of this clade was supported by three prevalent fragment ions: m/z 401.2548 Da

314

(theoretical mass for C21H37O7- is 401.2545), 563.3051 Da (theoretical mass for C27H47O12- is



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 12 of 18

315

563.3073), and 605.3176 Da (theoretical mass for C29H49O13- is 605.3179) (Fig. S-5).

316

Recently, Ekanayaka et al. identified a novel class of trichome-specific sesquiterpene

317

glucosides from S. habrochaites using these fragment ions and elucidated the structures of

318

purified representatives by NMR26. In our study, CID fragmentation and preparative isolation

319

of MS1 feature [605.3160; 7.07, an abundant in-source fragment] with subsequent NMR

320

confirmed

321

glucopyranosyl)-campherenane-2-endo,12-diol, a member of the novel sesquiterpene

322

glucoside metabolite family (Figures S-3 - S-6, S-9, S-10 and Tables S-3, S-4).

323

After annotation of both metabolite families, the corresponding MS1 features are highlighted

324

by their color-code in the PCA loadings (Fig. 3B). In our case, it was evident that the

325

representatives of both metabolite families were enriched in glandular trichomes, indicating a

326

trichome-specific upregulation of short branched chain acyl sugars and sesquiterpene

327

glucosides. Please note that the hierarchical cluster dendrogram comprised more clades

328

with internal spectral similarity, but we concentrated on the short branched chain acyl sugars

329

and sesquiterpene glucosides whose structures were confirmed by NMR. A detailed

330

workflow exemplified here is given in Fig. S-1, and the full showcase protocol is given in

331

Note S-1. A general user guide for MetFamily is given in Note S-2.

the

structure

of

12-O-(6’’-O-malonyl-β-D-glucopyranosyl-(1→2)-β-D-

Figure 3. Principal component analysis of metabolite extracts of glandular trichomes and leaves of Solanum habrochaites LA1777. Comparison of 2585 MS1 features from TOF-MS measurements (n=6). a: scores, b: loadings with annotations. The PCA loadings with annotations indicate a predominant enrichment of acyl sugars in glandular trichomes. AS: acyl sugars, SQT-glucosides: sesquiterpene glucosides, Unknown: Not characterized


Page 13 of 18

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


here. 332 333

Figure 4. Hierarchical cluster analysis of 135 trichome-specific MS1 features using the corresponding MS/MS spectra obtained from organic extracts of S. habrochaites LA1777. For comparison of the groups trichomes versus leaf focusing on trichome-specific features, the set of 2585 MS1 features was filtered using an MS1 abundance threshold of 20,000 counts and a log2-fold change (LFC) of two. The heatmap below depicts the LFC and the absolute MS1 abundance in glandular trichomes (TRI) and trichome-free leaves (LVS), respectively. The 135 filtered MS1 features clearly segregated into two main signal-clusters which in turn further segregated into signal-clusters with different levels of similarity between MS/MS spectra. Specifically, we identified a cluster of 73 short branched chain acyl sugars (AS, in blue) and a cluster of four sesquiterpene glucosides (SQT-glucosides, in red) on the basis of a set of characteristic fragments which were prevalent in both clusters (see legend ‘Annotations’ on the right). Both signal-clusters show characteristic fragments with a cluster-discriminating power of 80% and more (size of the branch nodes, see legend ‘Cluster-discriminating power’ on the right). 58 trichome-specific MS1 features partially showed further clusters, but remained uncharacterized in this study (Unknown in black). 334



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 14 of 18

335

Additional features of MetFamily. MetFamily also supports semi-targeted analyses. In this

336

case, sets of MS1 features can be selected by certain fragment masses, neutral losses or

337

combinations thereof within a user-defined mass error in ppm as filter criteria. Using this

338

option, only selected MS1 features are considered in subsequent PCA or HCA calculations

339

and the data analysis is consequently constrained to selected metabolite families. For

340

example, to isolate only glycosylated MS1 features from all data the user can specify a

341

fragment ion of m/z 161.0455 Da (C6H9O5-) from MS/MS spectra in negative mode and can

342

then focus on the regulation of enzymatic glycosylations in a biological context (for details

343

see the MetFamily user guide in Note S-2). When we applied this filter with a mass error of

344

25 ppm, we obtained 568 MS1 features from our example data, presumably containing a

345

hexose as a structural moiety. In addition, it is possible to search MS1 features with certain

346

fragments or neutral losses post-analysis. The corresponding MS1 features can then be

347

jointly visualized in the PCA loadings and the hierarchical cluster dendrogram.

348

It is possible to export different kinds of results from MetFamily. Selected sets of precursor

349

ions can be exported and, e.g., reloaded into the original MS data acquisition software.

350

Further, it is possible to export both the hierarchical cluster dendrogram and the PCA plots

351

as publication-ready high quality images. The set of parameters used for the initial data

352

import can be exported and imported. Finally, it is possible to export the whole project

353

(including all annotations and color codes) to enable the user to share the project or to

354

continue the data analysis at a later time (Data S-4).

355

CONCLUSIONS

356

The web application ‘MetFamily’ presented here constitutes a novel approach to analyze

357

metabolomics data from untargeted, data-independent LC-MS/MS measurements. Rather

358

than relying on the time-consuming structure identification of individual metabolites,

359

MetFamily assists in the interpretation of complex metabolomics data by identifying

360

metabolite families through patterns in MS/MS. These are generated by similarity clustering

361

of associated MS/MS spectra and can be annotated with names and colors. After pre-

362

processing of LC-MS/MS raw data, MetFamily performs a joint data analysis of MS1

363

abundances and MS/MS spectra in which the annotation of metabolite families facilitates the

364

interpretation of comparative data sets. Structure elucidation at the metabolite level can be

365

performed afterwards in a much more focused way. As a proof of concept, we identified two

366

trichome-specific metabolite families from wild type Solanum habrochaites LA1777 in a fully

367

unsupervised manner and validated our findings based on earlier publications and with

368

NMR. The plethora of identified trichome-specific acyl sucroses correlates with upregulation

369

of acyltransferases of the BAHD family in tomato glandular trichomes (Schilmiller 2012). In

370

addition, the size of the clade “acyl sugar” is related to a low substrate specificity of BAHD


Page 15 of 18

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


371

acyltransferases, illustrating that MetFamily can uncover links between enzymatic

372

promiscuity and organ-specific regulation of enzymes.

373

Using the proposed approach, it is now possible to obtain a comprehensive overview of data

374

sets containing thousands of mass features within a reasonable amount of time. Thus, by

375

providing a dynamic link between structural similarity at the MS/MS level (HCA) and the

376

corresponding MS1-signal intensity-based patterns (PCA) we bridge the gap between raw

377

data and structural information. Moreover, using MetFamily, precursor ions can now be

378

filtered via combinations of fragment ions and neutral losses, permitting the selection of

379

metabolite families based on characteristic fragmentation patterns.

380

While traditional compound identification is based on the comparison of MS/MS spectra (or

381

electron impact MS spectra) with reference spectra from known compounds, future

382

developments should exploit spectral patterns of MS/MS features being characteristic of

383

certain metabolite families. Public knowledge on such characteristic fragment ions or neutral

384

losses, e.g. based on metabolite families, can assist mass spectrometry specialists in the

385

elucidation of unknown features and will open new perspectives in life science.

386 387

AVAILABILITY

388

Project name: MetFamily

389

Source code: https://github.com/Treutler/MetFamily

390

Availability: http://msbi.ipb-halle.de/MetFamily/

391

Operating system(s): Platform independent

392

Programming language: R

393

Other requirements: Installation of R 3.2.2 or higher; License: GPL 3

394

Any restrictions to use by non-academics: None

395

ASSOCIATED CONTENT

396

Supporting Information

397

General work flow as flow chart (Fig. S-1) Scree plot of the first five principle components of

398

Figure 4. Exported parameter file of MS-DIAL (Table S-1) and exported parameter file from

399

MetFamily and parameter explanation (Tables S-2, S-3). Structure elucidation of three

400

selected MS1 features (Figures S-3 - S-19, Tables S-4 - S-6). Protocol for the presented

401

showcase (Note S-1) and a user guide for MetFamily (Note S-2). Detailed specification of

402

MetFamily input files (Note S-3). Metabolite profile of the showcase (Data S-1), MS/MS

403

library of the showcase (Data S-2), matrix of the showcase (Data S-3), and annotated

404

MetFamily project file (Data S-4).



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 16 of 18

405

AUTHOR INFORMATION

406

Corresponding Author

407

*E-mail: [email protected]

408

Notes

409

The authors declare that they have no competing financial interests.

410

ACKNOWLEDGMENTS

411

We would like to thank Anja Ehrlich for the preparative isolation of metabolites, Anja Henning

412

and Nick Bergau for trichome harvest and data acquisition. We thank Prof. Masanori Arita

413

and Dr. Karin Gorzolka for fruitful discussions and review of the manuscript.

414

REFERENCES

415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447

(1) Jorge, T. F.; Rodrigues, J. A.; Caldana, C.; Schmidt, R.; van Dongen, J. T.; ThomasOates, J.; Antonio, C. Mass spectrometry reviews 2015. (2) Tonoli, D.; Varesio, E.; Hopfgartner, G. Chimia (Aarau) 2012, 66, 218-222. (3) Wishart, D. S.; Mandal, R.; Stanislaus, A.; Ramirez-Gaona, M. Metabolites 2016, 6. (4) Suhre, K.; Shin, S. Y.; Petersen, A. K.; Mohney, R. P.; Meredith, D.; Wagele, B.; Altmaier, E.; CardioGram; Deloukas, P.; Erdmann, J.; Grundberg, E.; Hammond, C. J.; de Angelis, M. H.; Kastenmuller, G.; Kottgen, A.; Kronenberg, F.; Mangino, M.; Meisinger, C.; Meitinger, T.; Mewes, H. W.; Milburn, M. V.; Prehn, C.; Raffler, J.; Ried, J. S.; RomischMargl, W.; Samani, N. J.; Small, K. S.; Wichmann, H. E.; Zhai, G.; Illig, T.; Spector, T. D.; Adamski, J.; Soranzo, N.; Gieger, C. Nature 2011, 477, 54-60. (5) Fiehn, O. Plant Mol Biol 2002, 48, 155-171. (6) Fernie, A. R.; Trethewey, R. N.; Krotzky, A. J.; Willmitzer, L. Nature reviews. Molecular cell biology 2004, 5, 763-769. (7) Roemmelt, A. T.; Steuer, A. E.; Poetzsch, M.; Kraemer, T. Anal Chem 2014, 86, 1174211749. (8) Zhu, X.; Chen, Y.; Subramanian, R. Anal Chem 2014, 86, 1202-1209. (9) Tsugawa, H.; Cajka, T.; Kind, T.; Ma, Y.; Higgins, B.; Ikeda, K.; Kanazawa, M.; VanderGheynst, J.; Fiehn, O.; Arita, M. Nature methods 2015, 12, 523-526. (10) Ferslew, B. C.; Xie, G.; Johnston, C. K.; Su, M.; Stewart, P. W.; Jia, W.; Brouwer, K. L.; Sidney Barritt, A. t. Digestive diseases and sciences 2015, 60, 3318-3328. (11) Kaling, M.; Kanawati, B.; Ghirardo, A.; Albert, A.; Winkler, J. B.; Heller, W.; Barta, C.; Loreto, F.; Schmitt-Kopplin, P.; Schnitzler, J. P. Plant Cell Environ 2015, 38, 892-904. (12) Qu, G.; Quan, S.; Mondol, P.; Xu, J.; Zhang, D.; Shi, J. J Integr Plant Biol 2014, 56, 849-863. (13) Glas, J. J.; Schimmel, B. C. J.; Alba, J. M.; Escobar-Bravo, R.; Schuurink, R. C.; Kant, M. R. Int J Mol Sci 2012, 13, 17077-17103. (14) Watrous, J.; Roach, P.; Alexandrov, T.; Heath, B. S.; Yang, J. Y.; Kersten, R. D.; van der Voort, M.; Pogliano, K.; Gross, H.; Raaijmakers, J. M.; Moore, B. S.; Laskin, J.; Bandeira, N.; Dorrestein, P. C. Proc Natl Acad Sci U S A 2012, 109, E1743-1752. (15) Garg, N.; Kapono, C. A.; Lim, Y. W.; Koyama, N.; Vermeij, M. J. A.; Conrad, D.; Rohwer, F.; Dorrestein, P. C. Int J Mass Spectrom 2015, 377, 719-727. (16) Nguyen, D. D.; Wu, C. H.; Moree, W. J.; Lamsa, A.; Medema, M. H.; Zhao, X. L.; Gavilan, R. G.; Aparicio, M.; Atencio, L.; Jackson, C.; Ballesteros, J.; Sanchez, J.; Watrous,


Page 17 of 18

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486


J. D.; Phelan, V. V.; van de Wiel, C.; Kersten, R. D.; Mehnaz, S.; De Mot, R.; Shank, E. A.; Charusanti, P.; Nagarajan, H.; Duggan, B. M.; Moore, B. S.; Bandeira, N.; Palsson, B. O.; Pogliano, K.; Gutierrez, M.; Dorrestein, P. C. P Natl Acad Sci USA 2013, 110, E2611-E2620. (17) Li, D.; Baldwin, I. T.; Gaquerel, E. Proc Natl Acad Sci U S A 2015, 112, E4147-4155. (18) Wagner, C.; Sefkow, M.; Kopka, J. Phytochemistry 2003, 62, 887-900. (19) Rasche, F.; Scheubert, K.; Hufsky, F.; Zichner, T.; Kai, M.; Svatos, A.; Bocker, S. Anal Chem 2012, 84, 3417-3426. (20) Tissier, A. Plant J 2012, 70, 51-68. (21) Ghosh, B.; Westbrook, T. C.; Jones, A. D. Metabolomics 2014, 10, 496-507. (22) Kim, J.; Kang, K.; Gonzales-Vigil, E.; Shi, F.; Jones, A. D.; Barry, C. S.; Last, R. L. Plant Physiol 2012, 160, 1854-1870. (23) Schilmiller, A.; Shi, F.; Kim, J.; Charbonneau, A. L.; Holmes, D.; Jones, A. D.; Last, R. L. Plant J 2010, 62, 391-403. (24) Schilmiller, A. L.; Charbonneau, A. L.; Last, R. L. P Natl Acad Sci USA 2012, 109, 16377-16382. (25) Ekanayaka, E. A.; Celiz, M. D.; Jones, A. D. Plant Physiol 2015. (26) Ekanayaka, E. A. P.; Li, C.; Jones, A. D. Phytochemistry 2014, 98, 223-231. (27) van den Berg, R. A.; Hoefsloot, H. C.; Westerhuis, J. A.; Smilde, A. K.; van der Werf, M. J. BMC genomics 2006, 7, 142. (28) Gillet, L. C.; Navarro, P.; Tate, S.; Rost, H.; Selevsek, N.; Reiter, L.; Bonner, R.; Aebersold, R. Mol Cell Proteomics 2012, 11, O111 016717. (29) Lommen, A. Anal Chem 2009, 81, 3079-3086. (30) Lommen, A.; Kools, H. J. Metabolomcis 2012, 8, 719-726. (31) Smith, C. A.; Want, E. J.; O'Maille, G.; Abagyan, R.; Siuzdak, G. Anal Chem 2006, 78, 779-787. (32) Kuhl, C.; Tautenhahn, R.; Bottcher, C.; Larson, T. R.; Neumann, S. Anal Chem 2012, 84, 283-289. (33) Paddon, C. J.; Westfall, P. J.; Pitera, D. J.; Benjamin, K.; Fisher, K.; McPhee, D.; Leavell, M. D.; Tai, A.; Main, A.; Eng, D.; Polichuk, D. R.; Teoh, K. H.; Reed, D. W.; Treynor, T.; Lenihan, J.; Fleck, M.; Bajad, S.; Dang, G.; Dengrove, D.; Diola, D.; Dorin, G.; Ellens, K. W.; Fickes, S.; Galazzo, J.; Gaucher, S. P.; Geistlinger, T.; Henry, R.; Hepp, M.; Horning, T.; Iqbal, T.; Jiang, H.; Kizer, L.; Lieu, B.; Melis, D.; Moss, N.; Regentin, R.; Secrest, S.; Tsuruta, H.; Vazquez, R.; Westblade, L. F.; Xu, L.; Yu, M.; Zhang, Y.; Zhao, L.; Lievense, J.; Covello, P. S.; Keasling, J. D.; Reiling, K. K.; Renninger, N. S.; Newman, J. D. Nature 2013, 496, 528-532. (34) McDowell, E. T.; Kapteyn, J.; Schmidt, A.; Li, C.; Kang, J. H.; Descour, A.; Shi, F.; Larson, M.; Schilmiller, A.; An, L. L.; Jones, A. D.; Pichersky, E.; Soderlund, C. A.; Gang, D. R. Plant Physiol 2011, 155, 524-539.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

487

for TOC only

488


Page 18 of 18

Discovering Regulated Metabolite Families in Untargeted

Recommend Documents