Accepted Manuscript Title: Targeted profiling of polar intracellular metabolites using ion-pair–high performance liquid chromatography and ultra high performance liquid chromatography coupled to tandem mass spectrometry: Applications to serum, urine and tissue extracts Author: Filippos Michopoulos Nicky Whalley Georgios Theodoridis Ian D. Wilson Tom P.J. Dunkley Susan E. Critchlow PII: DOI: Reference:

S0021-9673(14)00755-9 http://dx.doi.org/doi:10.1016/j.chroma.2014.05.019 CHROMA 355411

To appear in:

Journal of Chromatography A

Received date: Revised date: Accepted date:

17-2-2014 1-5-2014 4-5-2014

Please cite this article as: F. Michopoulos, N. Whalley, G. Theodoridis, I.D. Wilson, T.P.J. Dunkley, S.E. Critchlow, Targeted profiling of polar intracellular metabolites using ion-pairndashhigh performance liquid chromatography and ultra high performance liquid chromatography coupled to tandem mass spectrometry: Applications to serum, urine and tissue extracts, Journal of Chromatography A (2014), http://dx.doi.org/10.1016/j.chroma.2014.05.019 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

6 7

Filippos Michopoulos1,2, Nicky Whalley1, Georgios Theodoridis2, Ian D Wilson3, Tom P.J.

8

Dunkley1 and Susan E. Critchlow1

cr

4

us

3

9 10

Department of Chemistry, Aristotle University of Thessaloniki, 541 24 Thessaloniki Greece. 3

M

12 13

2

Oncology iMED, Alderley Park, Macclesfield, Cheshire, SK10 4TG, UK

Department of Surgery and Cancer, Imperial College, Exhibition Rd, South Kensington, London SW7 2AZ, UK.

d

11

1

an

2

ip t

5

Targeted profiling of polar intracellular metabolites using ion-pair – high performance liquid chromatography and ultra high performance liquid chromatography coupled to tandem mass spectrometry: Applications to serum, urine and tissue extracts.

1

*Author for correspondence: Filippos Michopoulos

16

Filippos Michopoulos Email: [email protected]

17

Tel: 00441625513774

Ac ce p

18 19

te

14 15

20

Abstract

21

The effective analysis of polar ionic metabolites by LC-MS, such as those encountered in

22

central carbon metabolism, represents a major problem for metabolic profiling that is not

23

adequately addressed using strategies based on either reversed-phase or HILIC methods. Here

24

we have compared analysis of central carbon metabolites on optimized methods using HILIC,

25

porous graphitic carbon or ion pair chromatography (IPC) using tributyl ammonium as IP 1 Page 1 of 29

reagent. Of the 3 chromatographic approaches examined only IPC enabled us to obtain a

27

robust analytical methodology. This system was used to profile more than a hundred

28

endogenous metabolic intermediates in urine, serum and tissue samples. However, whilst we

29

found IPC to be the best of the approaches examined considerable care was still needed to

30

obtain robust data. Thus, in excess of 40 of representative biological samples were needed to

31

“condition” a new analytical column and further 10 matrix injections were then required at the

32

beginning of each analytical batch in order to obtain robust and reproducible chromatographic

33

separations. An additional limitation that we have found was that, for a small number of

34

phosphorylated and poly carboxylic acid metabolites, measurement was only possible if the

35

analytes were present in relatively high concentrations. We also found that, whilst this

36

methodology could be used for the analysis of both in vitro cell culture media, cell extracts,

37

tissue, and biological fluids (blood, urine), for the best results columns should only be used to

38

analyse a single matrix. However, despite the need for extensive column conditioning, and the

39

manifold disadvantages resulting from the contamination of the separation system and mass

40

spectrometer with the ion pair reagent, IPC-MS currently provides the best means of

41

analyzing these polar, ionic and problematic metabolites.

cr

us

an

M

d

te

Ac ce p

42

ip t

26

Keywords: Metabolite profiling, HILIC, porous graphitic carbon, ion pair chromatography,

43

mass spectrometry, central carbon metabolism.

44

1 Introduction

45

There is an increasing recognition of the role of energy metabolism as an important

46

discriminator of the metabolic phenotypes of tumor versus non-transformed cells and this has

47

led to a requirement for multi-analyte methods for the determination of a wide range of

48

compounds involved in central carbon metabolism [1-4]. Unfortunately the physico-chemical

49

properties of many of these compounds, such as high “polarity” and tendency for metal

50

chelation, make them unsuited to the general methods employed for LC-MS-based global

2 Page 2 of 29

metabolic profiling (metabonomics/metabolomics). Thus, despite the ability of such

52

untargeted metabolite profiling methods to measure thousands of “features” in a single

53

injection [5], a number of classes of metabolites are selectively “edited” out of the untargeted

54

profiles as a result of overall poor chromatographic performance which often means that

55

many metabolites are unretained on widely used reversed-phase (RP) liquid chromatographic

56

media. Examples of such problematic compounds include many sugars, sugar phosphates and

57

amino acids, certain carboxylic and di-carboxylic acids, phosphorylated compounds,

58

nucleotides/nucleosides and co-acyl A (CoA) derivatives. Many of these hydrophilic

59

molecules serve as intermediates within important anabolic and catabolic biological

60

processes, such as central carbon metabolism. So the development of comprehensive, multi-

61

analyte, methods that enable sensitive, selective and reliable measurements of these

62

intermediates are important for understanding basic biochemical differences between normal

63

and cancerous cells. Whilst one obvious approach is to attempt to dispense with the separation

64

entirely and analyse samples by MS alone, using direct infusion [6], the occurrence of a

65

significant number of isobaric species (e.g., the hexose phosphates glucose 1 phosphate,

66

glucose 6 phosphate, mannose 1 phosphate, galactose 1 phosphate etc.) necessitates the use of

67

a separation technique for comprehensive sample analysis.

cr

us

an

M

d

te

Ac ce p

68

ip t

51

69

In general, as indicated, these problematic compounds are hydrophilic (and often ionized) and

70

are not retained to any useful extent on RP chromatographic systems. Many analytes also

71

prove unsuited to hydrophilic interaction liquid chromatography (HILIC). So, even though a

72

number of multi-analyte methods using HILIC separations [7, 8] have been able to

73

simultaneously analyse large numbers (n>120) of primary metabolites such as organic acids,

74

amino acids, sugars and amines, these methods have met with limited success when applied to

75

phosphorylated analytes. An exception is a method that employed sequential analysis using

76

HILIC and then RPLC [9] enabling the detection of some 258 polar metabolites covering 3 Page 3 of 29

sugar phosphates, aminoacids, amines, carboxylic acids, phosphorylated compounds,

78

nucleotides/nucleosides and CoA derivatives. However, this was achieved at the expense of

79

long analysis times (ca. 2 hr/sample) employing multiple separations making it unsuited for

80

medium to high throughput screening.

ip t

77

81

Currently, for LC-MS-based methods, the best alternative for many such polar acidic

83

metabolites appears to be ion-pair chromatography (IPC) [10] which has been shown to give

84

good coverage of this area of the metabolome and provide acceptable methods of analysis.

85

Indeed a number of ion-pair chromatographic separations have been described in recent years

86

to cover this type of metabolite and all are, apparently, versatile and robust [11-17]. Recently,

87

the problems resulting from analytes with a propensity towards metal chelation have also

88

been addressed by Siegel et al.[18] who have proposed the use of acetylacetone, which

89

resulted in significant increases in peak intensities and enhanced limits of detection for the

90

affected analytes.

us

an

M

d te

91

cr

82

However, although inclusion of an IP reagent is beneficial for chromatographic separations,

93

the excess of positively charged tributylammonium ions in the mobile phase can cause severe

94

ion suppression in positive electrospray (ESI) mode rendering these methods suitable only for

95

negative ESI. Furthermore, as a rule, ion-pairing reagents contaminate the mass spectrometer

96

and effectively limit its use thereafter.

97

Ac ce p

92

98

Here we have re-examined the challenge of analyzing “problematic” polar analytes by

99

exploring the potential of a porous graphitic carbon stationary phase and re-investigating the

100

use of HILIC and IPC, thereby comparing three different chromatographic mechanisms.

101

Particular emphasis was placed on investigating the effects of “column conditioning” on the

102

chromatographic behavior of the different metabolites, and its positive role in improving 4 Page 4 of 29

system performance. We have also devoted significant effort to important aspects of method

104

validation such as the determination of limits of detection (LOD) and quantification (LOQ),

105

linear dynamic range, and retention time stability over an extended period (8 months), for a

106

large number of important analytes in samples such as cell extracts, culture media, urine and

107

serum.

ip t

103

110

2 Material and methods 2.1 Materials

us

109

cr

108

Water (18.2 M!) was obtained from a Purelab Ultra System from Elga (Bucks, UK).

112

Methanol, acetonitrile and isopropanol used for sample extraction and analysis were of HPLC

113

grade (Sigma-Aldrich, Gillingham, UK). Tributylamine (TBA), RPMI1640 cell growth

114

media, fetal calf serum, acetic acid, ammonium formate and all analytical standards, of the

115

highest purity available, were purchased from Sigma-Aldrich. The list of metabolites used in

116

the initial investigation with the HILIC and PGC chromatographic substrates are shown at

117

Table S-1. A more detailed list is given of all analytical standards (Table S-2) and their

118

abbreviations and full names (Table S-3) respectively.

120

M

d

te

Ac ce p

119

an

111

2.1.1

Surrogate matrix

121

A surrogate matrix for the determination of linearity and calculation of limits of detection and

122

quantitation was prepared from E.coli bacteria (JM109) grown in media enriched with 13C-

123

enriched glucose containing (per liter): 10 g glucose 13C6, 6 g Na2HPO4, 3 g KH2PO4, 0.5 g

124

NaCl, 1 g NH4Cl, 0.19 g MgCl2, 1.1 mg CaCl2, 1.52 mg FeSO4, 0.4 mg choline chloride, 0.5

125

mg folic acid, 0.5 mg pantothenic acid, 0.5 mg nicotinamide, 1 mg myo-inositol, 0.5 mg

126

pyridoxal-HCl, 0.5 mg thiamine, 50 µg riboflavin, 1 mg biotin, 3.7 µg (NH4)6Mo7O24, 24.7 µg

127

H3BO3, 3.9 µg CoCl2, 1.6 µg ZnSO4, 10 µg MnCl2, 1.6 µg CuSO4. To this medium 10 ml of an

128

inoculum of cultured E.coli optical density at 600nm (OD600 of 1.686 was added to 490 ml of

5 Page 5 of 29

129

13

130

culture was left to grow at 37ºC, 250 rpm, for 7 hours resulting in a final OD600 of 1.771.

131

C-enriched media in a 2 L sterile flask, to give an approximate starting OD600 of 0.034. The

2.1.1.1

Extraction of surrogate matrix

The bacterial pellet, obtained after centrifugation at 2000 g for 10 min, was extracted with 10

133

ml of cold (-20ºC) MeOH/ACN/H2O 40/40/20 (V/V/V) for 20min at -20ºC. The clear

134

bacterial extract was then obtained by centrifugation (10min centrifugation at 2000 g) and this

135

was then sub-aliquoted into 400µl portions in polypropylene tubes and stored at -20ºC until

136

required. For the linearity experiment 10 µl of this

137

temperature and re-suspended in water at a 1/20 dilution.

13

2.1.2

us

C-extract was dried under ambient

an

138 139

cr

ip t

132

In vitro cell culture

Samples from in vitro cultures of the cancer cell lines HCT116 and Calu 6 were collected

141

from cells plated in six well plates (3 x 106 per well) and cultured in RPMI1640 media

142

containing 10% fetal calf serum (FCS) and 2 mM glutamine for 24 h at 37ºC (5% CO2). The

143

culture media was changed to media containing 10% dialysed FCS (Thermo Fisher, , UK) and

144

cells were incubated further for at least 4 h. For the column conditioning investigation for the

145

ion pairing methodology the HCT116 and Calu 6 cell lines were incubated with or without an

146

AstraZeneca proprietary compound for 6, 24 and 48 hours before sample collection.

d

te

Ac ce p

147

M

140

2.1.2.1

Extraction of in vitro intracellular material

148

Extracts of tumor cell lines for the study of cellular metabolism were obtained, after cell

149

culture medium was removed and metabolism was quenched with400 ! l of the extraction

150

solvent 40/40/20 v/v/v MeOH/ACN/H2O at -20ºC added to the wells. Following cell

151

disruption with a cell scraper (BD, Oxford, UK) cells were macerated at -20ºC with the

152

extraction solvent for 20 min to extract intracellular metabolites. The contents of each well

153

were transferred to a 1.5 ml tube and were centrifuged at 16,000 g (Centrifuge 5415D,

154

Eppendorf, Hamburg, Germany) for 5 min to pellet precipitated proteins and cell debris. Clear

155

supernatants were transferred to clear tubes and stored at -20ºC until sample analysis. For LC6 Page 6 of 29

MS analysis, aliquots of 100 ! l of each extract were transferred to polypropylene HPLC vials

157

(Waters, Elstree UK) and dried at room temperature in a Savant SPD2010 SpeedVac (Thermo

158

Fisher, UK) for approximately 1 h. Metabolite extracts were then re-suspended in 50 ! l of

159

ultra pure water and centrifuged at 3270 g for 10 min at 4ºC (Allegra X12R equipped with

160

SX4750A swinging bucket rotor Beckman Coulter, USA).

161

2.1.2.2

Extraction of the in vitro culture medium

ip t

156

For the extraction of the cell culture medium a 100 ! l aliquot was treated with 400 ! l cold (-

163

20ºC) MeOH/ACN 50/50 at a 1.5 ml tube. Precipitated proteins were removed after

164

centrifugation at 16,000 g (Centrifuge 5415D, Eppendorf, Hamburg, Germany) for 5 min.

165

Clear supernatants were transferred to clear tubes and stored at -20ºC until sample analysis.

166

For LC-MS analysis, aliquots of 50 ! l of each extract were transferred to polypropylene

167

HPLC vials (Waters, Elstree UK) and dried at room temperature in a Savant SPD2010

168

SpeedVac (Thermo Fisher, UK) for approximately 1 h. Metabolite extracts were then re-

169

suspended in 100 ! l of ultra pure water and centrifuged as above at 3270 g for 10 min at 4ºC.

170 171

2.1.3

te

d

M

an

us

cr

162

Blood-urine samples

Whole blood and urine samples were collected from 12 male mice (6 weeks of age) of which

173

6 were CBAXC57BL6 control and 6 Tg197 transgenic mice. Serum was obtained from whole

174

blood samples which were left to clot for ~1 hour at room temperature. Particulate matter was

175

removed by centrifugation (3800g for 8min at 4oC). The clear supernatant was transferred to a

176

clean 1.5 mL tube and was subsequently centrifuged at 15000g for 8 min at 4oC. The resultant

177

clear supernatant was transferred into a fresh 1.5 mL tube and stored at -20oC.

178

Ac ce p

172

2.1.3.1

Extraction of serum and urine samples

179

For analysis the serum/urine samples (10 µl) were subjected to protein precipitation with the

180

addition of 40 µl cold (-20ºC) MeOH/ACN 50/50 v/v. Precipitated proteins were removed by

181

centrifugation at 21000 g, at 4ºC, for 10 min (5417R Eppendorf, Hamburg, Germany). Then

182

40 µl of clear supernatant were transferred to polypropylene HPLC vials, dried at room 7 Page 7 of 29

temperature in a Savant SPD2010 SpeedVac (Thermo Fisher, UK) for approximately 30 min.

184

Serum andurine extracts were then resuspended in 40 and 200 ! l of ultra pure water,

185

respectively and centrifuged at 3270 g for 10 min at 4ºC in an Allegra X12R centrifuge

186

equipped with SX4750A swinging bucket rotor (Beckman Coulter).

ip t

183

187 188

2.1.4

Tissue Samples

Pancreas samples were obtained from six KPC (KrasLSL.G12D/+; p53R172H/+; PdxCretg/+)

190

and six C57BL6 mice. Pancreas tissue was collected directly into 1.5 ml tubes and stored at -

191

20ºC.

us

2.1.4.1

Extraction of tissue samples

an

192

cr

189

Tissue extraction and homogenization was carried in a 2 ml CK14 lysis kit on a Precellys 24

194

device equipped with Cryolys temperature control unit (Peqlab, Southampton, UK). Pre-

195

weighed frosen tissue was extracted with 1 ml/100 mg tissue, MeOH/ACN/H2O (40/40/20

196

v/v/v) following a sequence of 3×30 second shaking cycle (5500 rpm) with 20 second pause

197

in between them. The clear supernatant obtained after centrifugation at 21000 g, 4ºC, for

198

10min (5417R Eppendorf, Hamburg, Germany) was subject to the same type of sample

199

preparation as described above for urine/serum, with the final sample diluted with 10 volumes

200

of water.

d

te

Ac ce p

201

M

193

202

For all matrices a biological quality control (QC) sample was prepared by mixing equal

203

volumes from each of the test samples and treating this pooled sample as described above. To

204

confirm metabolite retention times a test mixture containing all the metabolites to be

205

determined, at a concentration of 5 µM, in addition to a QC sample spiked with the test

206

mixture at a final concentration of 5 µM, were analysed at the beginning and the end of the

207

analytical batch. Prior to the start of the main analytical run ten injections, (5 µl) of the QC

208

samples were performed to ensure adequate system conditioning and one QC sample was

209

generally run every ten injections thereafter depending on the length of the analytical batch, 8 Page 8 of 29

210

so that a minimum number of five QC samples were assayed per run. Samples forming the

211

test set were run in a random order.

213

2.2 Methods

214

2.2.1

ip t

212

Chromatographic Separations

Chromatographic analysis was performed on a system consisting of an Ultimate 3000 RS

216

pump combined with an Ultimate 3000 autosampler operating at 4ºC through Chromeleon

217

software package (Thermo Fisher,UK).

us

cr

215

218 Porous graphitic carbon (PGC)

an

219

A porous graphitic carbon column 2.1×50 mm, 5 µm particle size, (Thermo Fisher,UK ) was

221

used with the column temperature maintained at 40±0.5°C during the analysis by using a heat

222

controller to ensure temperature stability. 10 µl of each sample was analysed using a binary

223

solvent system consisting of solvent A (0.1% Formic Acid in H2O) and solvent B (60%

224

Acetonitrile, 40% H2O with 50 mM ammonium bicarbonate). The optimum chromatographic

225

separation was achieved with a flow rate of 600 µl/min and the following gradient elution

226

profile: 0 min, 5% B; 9 min 60% B; 9.5 min 90% B; 10.5 min 90% B; 11 min 5% B; 15 min

227

5% B.

229

d

te

Ac ce p

228

M

220

HILIC

230

HILIC separations were performed on an XBridge Amide column (Waters Corp, 2.1×100

231

mm, 3.5 µm particle size). Column temperature was maintained at 40±0.5°C during the

232

analysis by using a heater controller to ensure temperature stability. 10 µl Samples were

233

analysed using a binary solvent system consisting of solvent A (95% Acetonitrile, 5% H2O

234

and 10 mM ammonium formate,) and solvent B (10 mM ammonium formate in H2O,

235

pH=6.5). The optimum chromatographic separation was achieved with a flow rate of 600

9 Page 9 of 29

236

µl/min and the following gradient elution profile: 0 min, 5% B; 9 min 60% B; 9.5 min 90% B;

237

10.5min 90% B; 11 min 5% B; 15 min 5% B.

238 239

Ion-Pair Chromatography

For IPC separations were performed on an Acquity HSS T3 UPLC column (Waters Corp,

241

2.1×100 mm, 1.8 µm particle size). Column temperature was maintained at 60±0.5°C during

242

the analysis. 5 µl Samples were injected and elution was performed using a binary solvent

243

system consisting of solvent A (H2O, 10 mM TBA, 15 mM Acetic Acid) and solvent B (80%

244

MeOH and 20% isopropanol). The optimum chromatographic separation was achieved with a

245

flow rate of 400 µl/min and the following gradient elution profile: 0 min, 0% B 0.5 min, 0%

246

B; 4 min, 5% B; 6 min, 5% B; 6.5 min, 20% B; 8.5 min, 20% B; 14 min, 55% B; 15 min,

247

100% B; 17 min, 100% B; 18 min, 0% B; 21 min 0% B.

M

an

us

cr

ip t

240

248 2.2.2

Mass spectrometry

d

249

All MS data was acquired on a QTRAP 4000 hybrid triple quadrupole linear ion trap mass

251

spectrometer operating through Analyst 1.5.1 (Applied Biosystems/MDS Sciex, Warrington,

252

U.K.) with the following settings: Turbo IonSpray voltage -3500 V, curtain gas 10 (arbitary

253

unit), temperature 550ºC, Gas1 and 2 were at 60 and 50 (arbitary unit) respectively and

254

entrance potential -10V. As factors such as Turbo IonSpray voltage, Source temperature and

255

Gas 1 and 2 parameters etc., cannot be customised for each individual metabolite a generic

256

value had to be applied. To define and optimise these generic values a number (~25) of

257

metabolites were infused at flow rate 400µl/ml 50/50 Solvent A and B at final in flow

258

concentration of 10nmol/ml. The average value of each of these parameters was used for all

259

the reported metabolites. Data was acquired in negative ionisation mode using the scheduled

260

MRM transitions given on Table S-2. To obtain the optimum MS parameters standard

261

solutions from each individual analyte were prepared in MeOH/H2O 50/50 v/v at

262

concentration of 10 µM and were infused at a flow of 10 µL/min.

Ac ce p

te

250

10 Page 10 of 29

263 264

2.2.3

Data Analysis

The raw spectrometric data was integrated with MultiQuan 2.0.2 (Applied Biosystems/ MDS

266

Sciex, Warrington, U.K.) and the results were exported to Excel for normalization and

267

univariate statistical analysis by t-test. Metabolites with a coefficient of variation (CV) lower

268

than 30% in QC samples, p-value less than 0.05 and fold increase or decrease between test to

269

control group of greater than 30% were considered to be significant. To ensure metabolite

270

identity retention time of each analyte was compared to retention time of aqueous standard

271

and spiked standard to biological matrix of interest.

273

an

272

3 Results and Discussion

M

274

us

cr

ip t

265

Given the difficulties raised by the use of IPC for the LC-MS analysis of metabolites involved

276

in central carbon metabolism, such as those associated with pathways such as glycolysis,

277

pentose phosphate (PP) and the tricarboxylic acid cycle (TCA) we initially evaluated methods

278

for their chromatographic analysis involving different separation mechanisms as described

279

below.

281 282

te

Ac ce p

280

d

275

3.1 Porous Graphitic Carbon Column (PGC)

283

PGC exhibits both hydrophilic and reversed-phase retention mechanisms, depending upon the

284

solvent system used, combined with excellent chemical stability across the entire pH range 0-

285

14. The lack of hydrogen bond interactions between analyte – stationary phase and the

286

general absence of metals from the system (PGC is packed in columns of peek tubing) should

287

act to reduce metal chelation. PGC has also been successfully used for the analysis of sugar

288

phosphates, nucleotides and nucleotide sugars [19-20]. Permanent modification of the PGC

11 Page 11 of 29

surface with cetyltrimethylammonium has also been used for the analysis of several organic

290

and inorganic acids [21]. In the present study we observed good chromatographic separation

291

for fumaric, malic, citric acid (Figure S-1) and glutamic acid and monophosphorylated

292

compounds, phosphoserine, 3-phosphoglyceric acid (Figure S-2) in the presence of 0.1%

293

formic acid in the water-acetonitrile-based solvents. However, very strong retention was

294

noted for nucleotides and other di- and triphosphorylated metabolites. Addition of ammonium

295

hydrogen carbonate [22, 22] improved the elution of NAD, NADH and their (and other)

296

monophosphorylated compounds (NADP, NADPH, AMP) as well as Acetyl CoA etc., but

297

strong retention was still observed for di- and tri-phosphorylated compounds (ADP, ATP

298

etc.,) (Figure S-2). The best overall chromatographic results were obtained after the addition

299

of 50 mM ammonium hydrogen carbonate in solvent B (as described in the experimental

300

section).. However, the methodology was less successful for metabolites such as pyruvic, and

301

lactic acids and fructose 1,6 bisphosphate which were not eluted under any of the

302

chromatographic conditions investigated and therefore no further studies with the PGC phase

303

were undertaken.

305 306

te

Ac ce p

304

d

M

an

us

cr

ip t

289

3.2 HILIC

307

In addition to PGC we investigated the potential of an amide-based HILIC phase for the

308

analysis of a range of sugar phosphates, amino acids, and nucleotides using gradient elution

309

with acetonitrile/water (ACN/H2O) modified with 10 mM ammonium formate. Typically

310

HILIC gradient elution profiles start with 95% ACN to enhance analyte retention, conditions

311

that we intially investigated but no significant difference was observed in chromatographic

312

resolution compare to the optimised gradient conditions mentioned at the experimental

313

section. Good resolution was achieved for a number of key metabolites such as the reduced

314

and oxidized forms of NAD and NADP, as well as ATP, ADP, AMP, hexose phosphates and 12 Page 12 of 29

fructose 1,6 bisphosphate. However, our results were limited by the tendency for increasing

316

peak tailing as the degree of metabolite phosphorylation increased. For example, whilst AMP

317

gave an acceptable, sharp and symmetric peak, ADP showed obvious peak tailing (Figure S-

318

2) and this was even more pronounced for ATP. Severe peak tailing was also observed for

319

some monophosphorylated metabolites such as 3-phosphoglyceric acid, with phosphoserine

320

seen to be progressively irreversibly bound to the stationary phase.

cr

ip t

315

321

Representative extracted ion chromatograms using the Amide X-bridge column are shown in

323

Figure S-2. Although this HILIC separation provided satisfactory results for a number of

324

metabolites, this methodology proved unsuited to the separation of carboxylic acids such as

325

lactate, citrate, isocitrate, malate, fumarate and succinate etc. Chromatographic separation of

326

carboxylic acids was not improved under acidic conditions when ammonium formate was

327

replaced with 0.1% formic acid. Instead rapid column deterioration was observed under these

328

conditions. Another finding was the relative short column life (up to 300 injections) under

329

neutral conditions. As with the PGC phase, no further attempts were made to optimize or

330

develop separations for these analytes using HILIC.

332

an

M

d

te

Ac ce p

331

us

322

3.3 Ion pair chromatography (IPC)

333

Thus, despite our best efforts we were unable to develop a robust analytical profiling

334

methodology for the required analytes using either HILIC or PGC. We therefore chose to

335

adapt the IPC system previously reported by Buescher et al. [13] aiming to maintain

336

chromatographic resolution whilst increasing throughput and reducing the high back pressures

337

associated with the use of gradient elution systems employing isopropanol and water. The

338

latter was achieved by i) using a shorter 10cm instead of a 15cm column, ii) replacing

339

isopropanol with a mixture of MeOH/H2O 80/20 (v:v) and iii) increasing the column

340

temperature to 60ºC. By reducing the eluotropic strength of solvent B and using pure H2O as 13 Page 13 of 29

solvent A (instead of 5% MeOH in H2O) the chromatographic performance of the system was

342

maintained. Under these conditions we were able to profile ~130 metabolites and reduce the

343

cycle time for the analysis from 36 min to 21 min per sample. Representative ion

344

chromatograms are shown in Figure 1 with clear improvement in peak shape and enhanced

345

sensitivity (~5-10 times) for most metabolites (Figure S-1, 2) compared to HILIC and PGC

346

modes.

cr

ip t

341

347

By using UHPLC we expected to enhance analytical sensitivity and chromatographic

349

resolution obtained compared to conventional HPLC and this should enable improved

350

separation of isobaric and isomeric species. For example citric and isocitric acid were almost

351

baseline separated using the UHPLC separation and a generic MRM transition (191 to 111

352

amu). The resolving power of the analytical platform was further increased when we used

353

specific MRM transitions for citric acid (191 to 87amu) and isocitric acid (191 to 73amu)

354

(Figure 2). This high resolution chromatographic separation in combination with carefully

355

selected MRM transitions were also beneficial for the separation of other isomeric species

356

such as leucine/isoleucine, six phosphorylated hexoses and three phosphorylated pentoses

357

(see Figure S-3).

an

M

d

te

Ac ce p

358

us

348

359

However, we quickly noted that certain phosphorylated compounds, such as ATP, showed a

360

non-linear signal response and high LOD’s when aqueous standards were used. For example,

361

repeated injections of 0.1, 0.4, 2, 5, and 20 µM standards ofr ATP in water resulted in an LOD

362

of 5 µM, with an average signal intensity after 3 injections of the same standard concentration

363

of 6e4 counts. The background signal was ~200 counts but surprisingly no peak was detected

364

with injections of aqueous standards with concentrations lower than 5 µM. This observation

365

led us to conclude that a threshold concentration of the analyte needs to be loaded on the

366

column before a reproducible chromatographic elution mechanism can be established. This 14 Page 14 of 29

phenomenon is well recognized with untargeted metabolite profiling where a number of

368

matrix injections are required at the beginning of the analytical run to “condition” the

369

instrument [23, 24]. We initially decided to condition the analytical system with 10 matrix

370

injections at the beginning of each analytical batch, using an aggregate “QC” sample

371

(prepared by making a pool sample by mixing equal amounts of each of the test samples).

372

However, we found that with new columns this amount of conditioning was not sufficient to

373

provide repeatable data and, even after 10 conditioning injections of matrix, the data still

374

showed significant variability in both retention time and detection signal (peak areas). We

375

further investigated this observation in an experiment where the same set of samples formed

376

two analytical batches which were analysed sequentially. The first batch was formed from

377

thirty six randomized samples and six pooled samples (QC) with the one of the latter injected

378

after every six samples, in a total of forty two injections. For the second batch the same

379

samples were analysed in the same sequence as the first, but with the insertion of 12 matrix

380

conditioning injections prior to commencement of analysis. When the data was processed via

381

PCA the scores plot (Figure 3) showed that there was very poor group clustering for the six

382

pool QC injections of the first batch (Figure 3A) indicating that the analysis was not

383

reproducible. Analytical variability was significantly reduced for the second batch (Figure

384

3B) where the QC injections formed a very tight group. Careful examination of the raw

385

UHPLC-MS data using the coefficient of variation as a measure of technical variability of

386

each metabolite peak detected in the QC samples showed much higher variability for the first

387

batch (Table S-4). Within the QC data generated from the first batch, 6 out of 10 detected

388

metabolites had peak area CV values greater than 30%. In contrast in the second batch only 6

389

out of 100 metabolites showed CV’s greater than 30%.

Ac ce p

te

d

M

an

us

cr

ip t

367

390 391

In the first batch retention time shifts of about 0.2 to 0.5 min were observed for 29 out of 91

392

metabolites, a finding which is largely in accordance with reports from the literature [7]. As 15 Page 15 of 29

above this variation was limited to the first batch, while on the second batch no metabolite

394

showed retention time drifts of greater than 0.2 min (Table S-5). From these observations it is

395

clear that for new columns more extensive column conditioning of about 40 matrix injections

396

is required prior to commencement of the analysis of real samples. To further validate that

397

this procedure had eliminated retention time drift we monitored the RT of all detected

398

metabolites on various in vitro cancer cell line extracts over a period of 8 months. For this

399

work several chromatographic columns from different batches were used by a single operator

400

and the results (Table S-6) showed only five metabolites (dCTP, glucuronic acid, GTP,

401

methylxanthine, OH-butyryl-CoA and shikimic acid) showed RT drifts of greater than 0.2

402

min. We noted that, for maximum operating life, the best results were achieved by dedicating

403

columns to particular sample matrices (e.g., in vitro cell extracts, in vitro cell culture media,

404

urine, tissue extracts or serum etc.).

M

405

an

us

cr

ip t

393

As part of the regular within batch validation a mixture of metabolite standards were first

407

analysed as a simple aqueous mixture and then reanalyzed after spiking into the matrix of

408

interest to confirm chromatographic retention time, assess potential ion suppression effects

409

and interferences from endogenous constituents. When performing this procedure we noticed

410

that the retention time for NADH in aqueous standards differed by 0.15 min compared to the

411

RT observed in the analysis of biological samples (Figure S-4). This anomalous behaviour

412

was resolved when the polypropylene HPLC vials were pre-rinsed with MeOH and sonicated

413

for 10min prior to sample aliquoting suggesting that some contaminant in the vial was

414

responsible.

Ac ce p

te

d

406

415 416

During method development it became clear that reliable determination of palmitic acid is not

417

possible by this method. We hypothesize that this acid binds strongly to the system due to

418

enhanced hydrophobicity after its interaction with the ion pair agent (TBA) resulting in large 16 Page 16 of 29

“carry over” effects from one injection to another. Carry over issues were also noticed for

420

lactic acid where physiological concentrations may exceed an order of magnitude. Rigorous

421

investigation of this problem revealed that the source of this carry over was the contamination

422

of the part of the injection valve containing the sample loop tube. To reduce this carry over

423

issue we introduced two extra washing steps with a mixture H2O/ACN/Acetone 1/1/1 v/v/v

424

and pure water, respectively. Despite utilizing strong solvents to overcome lactate carry-over,

425

this problem persisted in the long term. The problem was mitigated by, as part of system

426

maintenance, cleaning the LC module after disconnecting the TBA-equilibrated analytical

427

column and running 3-5 water injections with a binary solvent system consisting of H2O and

428

ACN both acidified with 0.1% formic acid. Prior to reconnecting the analytical column a

429

further 3-5 water injections with the ion pair solvent system were used to clear potential traces

430

of formic acid from the system (the routine addition of formic acid into washing solvents etc.,

431

should be avoided as we have observed changes in the retention of early eluting metabolites

432

in its presence). We also noted, throughout this investigation, that the lactate base line signal

433

was increased when Solvent A (H2O, 15mM acetic acid, 10mM TBA) was more than two

434

days old and recommend that Solvent A is prepared freshly for every analytical batch. Solvent

435

B (MeOH/IPA 80/20), in contrast, could be kept for a maximum of two weeks.

cr

us

an

M

d

te

Ac ce p

436

ip t

419

437

Overall, despite the need to take such extensive precautions we found the IPC-MS system to

438

be both stable and reliable, and well suited to the analysis of the polar metabolites that

439

constituted our target analytes. We have been using this system for 24 months now

440

performing more than 1200 injections per month of various matrices and for different

441

bioanalytical and life science applications. Average column life time regardless of the matrix

442

type was found to be in the range of 800 to 1000 injections. We propose the use of different

443

columns for each matrix as we have experienced enhanced data variability due to differences

444

in matrix composition when the same column was used to analyse different matrices. 17 Page 17 of 29

445 446

3.3.1

Validation of the IPC-MS Methodology

The

intended

use

of

the

proposed

bioanalytical

assay

is

to

provide

448

qualitative/semiquantitiative data with quantification relative to the control sample groups and

449

we therefore proceeded to perform a “fit for purpose”, rather than full, method validation. In

450

this respect we calculated representative limits of detection and quantification (LOD and

451

LOQ), linear dynamic range for detection and quantification in form of log2 scaled values

452

using the 13C-enriched E.coli bacterial extract. For the LOD and LOQ the signal intensity of

453

each metabolite was given a minimum signal to noise ratio of 3 and 10 respectively.

454

Moreover, all concentration points for the quantification range had coefficients of variation

455

(CV) lower than 30% between three replicate injections. As the presence of these endogenous

456

metabolic intermediates in matrices such as serum, tissue and urine etc., does not allow

457

accurate calculations of the LOD and LOQ for the proposed methodology the analytical

458

figures of were obtained after spiking the biological matrix with the isotopic (13C or 2H)

459

analogs of the metabolites of interest. For many of the metabolites listed in Table S-2 isotopic

460

species were not commercially available and so we adopted the reverse approach of spiking

461

the 12C metabolite form into a matrix that only contained the 13C species as provided by the

462

E.coli extract. This approach worked well for the calculation of these bioanalytical parameters

463

and the results are shown on Table S-2. For a few metabolites such as succinic acid, lactic

464

acid, malonic acid, Malonyl CoA and fructose 1,6 bisphosphate we were unable to reach

465

validation, due to the high concentrations of the endogenous

466

concentrations of co-eluting species. For these metabolites the in vitro cell extract was used as

467

matrix to spike in different concentrations of the 13C form to calculate sensitivity, linearity and

468

quantification parameters. Similarly for citric acid, alpha-ketoglutaric acid and NADH we

469

were unable to determine LOD, LOQ and linearity values via this approach and the values

470

presented in Table S-2 were obtained after injection of pure water standards. Phthalic acid

Ac ce p

te

d

M

an

us

cr

ip t

447

13

C analogs or high

18 Page 18 of 29

was the only metabolite that we could not obtain validation under any of the investigated

472

conditions. This is due to the high biological and environmental background concentration of

473

this intermediate.

474

As we noted above, the signal for ATP displayed non-linear behaviour in correlation to its

475

concentration. During method validation we also observed a number of additional metabolites

476

(ATP, fructose 1,6 bisphosphate, PG, dAMP, dCDP, UDP, dADP, dCTP, dATP, GTP, TTP,

477

dUTP, succinyl CoA and lactic acid, citric acid) that also showed similar anomalous behavior.

478

In particular for fructose-1,6-bisphosphate as mentioned earlier, good linearity was obtained

479

by spiking different concentrations of the

480

contained 5 µM of the

481

phosphorylated or poly carboxylic acids) not only is adequate column conditioning required

482

but a certain amount of the metabolite needs to be present in the system to facilitate

483

chromatographic elution. Under these conditions the LOD for fructose-1,6-bisphosphate was

484

improved from 2 to 0.008 µM.

13

C isotope into in vitro cell extract which also

12

Ac ce p

3.4 Applications

d

M

an

C isotope. This shows that for some metabolites (highly

te

485 486

us

cr

ip t

471

487

From the beginning of this study the primary aim was toward the analysis of serum, urine and

488

tissue samples. For all three matrixes multivariate statistical analysis (principal component

489

analysis PCA) showed clear separation between test (pancreatic cancer or arthritic) groups

490

and healthy control groups (Figure 4) with robust analytical separation forming tightly

491

clustered QC sample replicate injections. Metabolites detected for each matrix are shown in

492

Table S-2.

493

It is very important to note here the absolute requirement for protein precipitation prior to

494

urine sample analysis with the ion pair methodology unlike for untargeted metabolite

495

profiling in RPLC where simply diluting samples with water is often the standard practice

496

[23]. Whilst this works well for RPLC, and minimizes sample handling and editing, therby 19 Page 19 of 29

maintaining the key biological characteristics of the sample, we have found it quite unsuitable

498

for IPC. So it is very important to carefully precipitate proteins prior to urine analysis. Similar

499

problems can be seen with serum and tissue analysis due to high lipid content (especially

500

phospholipids) of the original sample extract. With these two matrixes the use of water as re-

501

suspension solvent following drying the sample has a beneficial effect as the majority of

502

lipids will not be dissolved, thus they will not be introduced to the analytical system.

cr

ip t

497

503

3.5 Consequences of the application of IPC for Metabolic Profiling

us

504

The results of our investigations show that IPC currently provides the most effective method

506

for determining the hydrophilic/ionic metabolites in cell extracts and other biological

507

matrices. This methodology seems well suited to the analysis of a wide range of sample types

508

and we have found it, once optimized, to be reliable and robust in general use, with a good

509

sample throughput.

510

Despite the benefits of using TBA for the analysis of polar metabolites, it should be stated

511

that ion pair reagents show high affinity towards the components of the UHPLC-MS system

512

and are thus responsible for the long term contamination of chromatographic and mass

513

spectrometry instrumentation. Based on the class of the ion pair reagent (either amine or acid

514

substituted structure), mass spectrometric detection in general is limited to negative or

515

positive ionization respectively to avoid ion suppression phenomena, high background and

516

general interference. The problem with the LC instruments is less challenging, since effective

517

clean up can, usually, be achieved with aggressive washes with either strong acids or bases. In

518

the case of persistent contamination the replacement of contaminated components can be a

519

cost effective means of eliminating the problem. Decontamination of the mass spectrometry

520

however, represents a problem of a different order. Thus after almost a year of the continuous

521

use of TBA we attempted to completely remove the ion pair reagent from the mass

522

spectrometer. However, following repeated washes with strong organic solvents and detergent

Ac ce p

te

d

M

an

505

20 Page 20 of 29

(in a sonic bath) of every single ion path component starting from the cone and reaching Q3,

524

and also replacing the whole ion source block, curtain plate and orifice with new parts the

525

final result was only a modest reduction of the background TBA signal intensity (from ca. ~e6

526

to ca. ~e5). In all probability effective decontamination of the mass spectrometer would

527

require complete replacement of the whole ion path and detector, a strategy with questionable

528

cost efficiency. Thus the adoption of IPC-MS in practice is should be considered irreversible.

cr

ip t

523

529

4

Conclusions

531

Despite much effort to find an alternative the most robust analytical methodology that we

532

have evaluated in order to reliably profile more than a hundred polar, ionic endogenous

533

metabolic intermediates in urine, serum and tissue samples required the use of IPC with

534

tributylamine. Even using this system the final methodology requires a substantial number of

535

injections of the biological (more than 40) in order to condition a new analytical column and

536

ensure robust and reproducible chromatographic separations. Following the initial column

537

conditioning phase 10 matrix injections are required at the beginning of each analytical batch

538

to achieve system equilibrium. For a small number of metabolites (highly phosphorylated and

539

poly carboxylic acids) adequate linearity could not be achieved unless these metabolic

540

intermediates were present in relatively high concentrations. The developed methodology has

541

been applied for the analysis of a range of sample types including in vitro cell culture media,

542

cell extracts, tissue extracts, and biological fluids (blood, urine). Throughout our work we

543

found that the best practice was to keep separate columns for each individual matrix.

Ac ce p

te

d

M

an

us

530

544 545

5 Acknowledgements

546

This research has been partially co-financed by the European Union (European Social Fund –

547

ESF) and Greek national funds through the Operational Program "Education and Lifelong

548

Learning" of the National Strategic Reference Framework (NSRF) - Research Funding 21 Page 21 of 29

Program: Heracleitus II. Investing in knowledge society through the European Social Fund.

550

We thank Professor Thorsten Hagemann for providing pancreas tissue specimens,

551

Biomedcode Hellas for providing the urine and serum specimens and Liz Flavell for the

552

preparation of E.Coli bacteria cultures.

ip t

549

Ac ce p

te

d

M

an

us

cr

553

22 Page 22 of 29

553 554 555

References [1] J.R. Cantor, D.M. Sabatini, Cancer Discovery 2 (2012) 881.

557

[2] M.G. Vander Heiden, Nat. Rev. Drug Discov. 10 (2011) 671.

558

[3] D.A. Tennant, R.V. Duran, E. Gottlieb, Nat Rev Cancer 10 (2010) 267.

559

[4] G. Kroemer, J. Pouyssegur, Cancer Cell 13 (2008) 472.

560

[5] G.A. Theodoridis, H.G. Gika, E.J. Want, I.D. Wilson, Anal. Chim. Acta 711 (2012) 7.

561

[6] J.C. Ewald, S. Heux, N. Zamboni, Anal. Chem. 81 (2009) 3623.

562 563

[7] H.G. Gika, G.A. Theodoridis, U. Vrhovsek, F. Mattivi, Journal of Chromatography A 1259 (2012) 121.

564 565

[8] S. Schiesel, M. Lämmerhofer, W. Lindner, Analytical and Bioanalytical Chemistry 396 (2010) 1655.

566 567

[9] B. Preinerstorfer, S. Schiesel, M. Lämmerhofer, W. Lindner, Journal of Chromatography A 1217 (2010) 312.

568 569

[10] J.M. Buescher, D. Czernik, J.C. Ewald, U. Sauer, N. Zamboni, Anal. Chem. 81 (2009) 2135.

570 571

[11] B. Luo, K. Groenke, R. Takors, C. Wandrey, M. Oldiges, Journal of Chromatography A 1147 (2007) 153.

572 573

[12] L. Coulier, R. Bas, S. Jespersen, E. Verheij, d.W. van, T. Hankemeier, Anal. Chem. 78 (2006) 6573.

574

[13] J.M. Buescher, S. Moco, U. Sauer, N. Zamboni, Anal. Chem. 82 (2010) 4403.

575 576

[14] W. Lu, M.F. Clasquin, E. Melamud, D. Amador-Noguez, A.A. Caudy, J.D. Rabinowitz, Anal. Chem. 82 (2010) 3212.

577

[15] N.B. Bennette, J.F. Eng, G.C. Dismukes, Anal. Chem. 83 (2011) 3808.

578 579

[16] G.U. Balcke, S.N. Kolle, H. Kamp, B. Bethan, R. Looser, S. Wagner, R. Landsiedel, B. van Ravenzwaay, Toxicol. Lett. 203 (2011) 200.

580

[17] P. Kiefer, N. Delmotte, J.A. Vorholt, Anal. Chem. 83 (2011) 850.

581

[18] D. Siegel, H. Permentier, R. Bischoff, Journal of Chromatography A 1294 (2013) 87.

582 583

[19] M. Pabst, J. Grass, R. Fischl, R. Leònard, C. Jin, G. Hinterkörner, N. Borth, F. Altmann, Anal. Chem. 82 (2010) 9782.

Ac ce p

te

d

M

an

us

cr

ip t

556

23 Page 23 of 29

[20] C. Antonio, T. Larson, A. Gilday, I. Graham, E. Bergström, J. Thomas-Oates, Journal of Chromatography A 1172 (2007) 170.

586

[21] K. Yoshikawa, M. Okamura, M. Inokuchi, A. Sakuragawa, Talanta 72 (2007) 305.

587 588

[22] Y. Asakawa, N. Tokida, C. Ozawa, M. Ishiba, O. Tagaya, N. Asakawa, Journal of Chromatography A 1198–1199 (2008) 80.

589

[23] H.G. Gika, G.A. Theodoridis, J.E. Wingate, I.D. Wilson, J. Proteome Res. 6 (2007) 3291.

590 591

[24] F. Michopoulos, L. Lai, H. Gika, G. Theodoridis, I. Wilson, J. Proteome Res. 8 (2009) 2114.

cr

ip t

584 585

Figure 1: Extracted ion chromatograms following separation with ion pair chromatography. In ascending retention time elution are shown 1) Glutamine, 2) Methionine, 3) Adenine, 4) Thymine, 5) Inosine, 6) Glutamic acid, 7) Phenylalanine, 8) Aspartic Acid, 9) Clucuronic acid, 10) Tryptophan, 11) Lactic Acid, 12) Galactose 1P, 13) Xylulose 5P, 14) Pyruvic acid, 15) NAD, 16) UMP, 17) GMP, 18) AMP, 19) Maleic axid, 20) Phosphocreatine, 21) Malic acid, 22) a-ketoglutarate, 23) G3P, 24) NADP, 25) FBP, 26) Isocitric acid, 27) dCTP, 28) ATP, 29) Acetyl CoA and 30) Butyryl CoA.

606 607 608

Figure 2: Separation of citrate/isocitrate using UHPLC (A) generic MRM (191 to 111amu) transition, (B) citrate specific MRM transition (191 to 87amu) and (C) isocitrate specific MRM transition (191 to 73amu).

609 610 611 612

Figure 3: Effect of column conditioning on data quality. PCA scores plots of the same set of samples analysed twice (A, B) consecutively on a brand new column. Triangles (▲), dots (●) and boxes (□) represent quality control, HCT116, and Calu6 cell extracts samples respectively.

613 614

Figure 4: PCA scores plots: A) Pancreas Tissue, B) Serum, C) Urine. (□) Quality control samples, (▲) Untreated (test) samples, (●) (Healthy) Control Samples

an

M

d

te

Ac ce p

615

Captions

us

592 593 594 595 596 597 598 599 600 601 602 603 604 605

24 Page 24 of 29

615 616 617 618 619 620 621

Highlights  Different chromatographic approaches investigated for endogenous metabolite analysis  Ion pair chromatography showed the best metabolite coverage

623

 40 matrix injections required to “condition” a new chromatographic column

624

 Analytical platform applied for urine, plasma and tissue extracts analysis

cr

ip t

622

Ac ce p

te

d

M

an

us

625

25 Page 25 of 29

Ac

ce

pt

ed

M

an

us

cr

i

Figure 1

Page 26 of 29

Ac ce p

te

d

M

an

us

cr

ip t

Figure 2

Page 27 of 29

Ac ce p

te

d

M

an

us

cr

ip t

Figure 3

Page 28 of 29

Ac ce p

te

d

M

an

us

cr

ip t

Figure 4

Page 29 of 29