ACTAUNIVERSITATIS

UPSALIENSISUPPSALA

2017

Digital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Pharmacy 232

Selectivity in NMR and LC-MSMetabolomics

The Importance of Sample Preparation andSeparation, and how to Measure Selectivity in LC-MS Metabolomics

ALBERT ELMSJÖ

ISSN 1651-6192ISBN 978-91-554-9879-5urn:nbn:se:uu:diva-318296

Dissertation presented at Uppsala University to be publicly examined in B41, BMC,Husargatan 3, Uppsala, Friday, 19 May 2017 at 10:15 for the degree of Doctor of Philosophy(Faculty of Pharmacy). The examination will be conducted in Swedish. Faculty examiner:Professor Thomas Moritz (Swedish Metabolomics Centre, Swedish University of AgriculturalSciences, Umeå, Sweden).

AbstractElmsjö, A. 2017. Selectivity in NMR and LC-MS Metabolomics. The Importance of SamplePreparation and Separation, and how to Measure Selectivity in LC-MS Metabolomics.Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Pharmacy232. 40 pp. Uppsala: Acta Universitatis Upsaliensis. ISBN 978-91-554-9879-5.

Until now, most metabolomics protocols have been optimized towards high sample throughputand high metabolite coverage, parameters considered to be highly important for identifyinginfluenced biological pathways and to generate as many potential biomarkers as possible. Froman analytical point of view this can be troubling, as neither sample throughput nor the numberof signals relates to actual quality of the detected signals/metabolites. However, a method’sselectivity for a specific signal/metabolite is often closely associated to the quality of that signal,yet this is a parameter often neglected in metabolomics.

This thesis demonstrates the importance of considering selectivity when developingNMR and LC-MS metabolomics methods, and introduces a novel approach for measuringchromatographic and signal selectivity in LC-MS metabolomics.

Selectivity for various sample preparations and HILIC stationary phases was compared.The choice of sample preparation affected the selectivity in both NMR and LC-MS. For thestationary phases, selectivity differences related primarily to retention differences of unwantedmatrix components, e.g. inorganic salts or glycerophospholipids. Metabolites co-eluting withthese matrix components often showed an incorrect quantitative signal, due to an influencedionization efficiency and/or adduct formation.

A novel approach for measuring selectivity in LC-MS metabolomics has been introduced. Bydividing the intensity of each feature (a unique mass at a specific retention time) with the totalintensity of the co-eluting features, a ratio representing the combined chromatographic (amountof co-elution) and signal (e.g. in-source fragmentation) selectivity is acquired. The calculatedco-feature ratios have successfully been used to compare the selectivity of sample preparationsand HILIC stationary phases.

In conclusion, standard approaches in metabolomics research might be unwise, as eachmetabolomics investigation is often unique. The methods used should be adapted for theresearch question at hand, primarily based on any key metabolites, as well as the type of sampleto be analyzed. Increased selectivity, through proper choice of analytical methods, may reducethe risks of matrix-associated effects and thereby reduce the false positive and false negativediscovery rate of any metabolomics investigation.

Keywords: Metabolomics, NMR, LC-MS, HILIC, UHPLC, Q-ToF, selectivity, co-featureratio, method evaluation, data evaluation

Albert Elmsjö, Department of Medicinal Chemistry, Analytical Pharmaceutical Chemistry,Box 574, Uppsala University, SE-75123 Uppsala, Sweden.

© Albert Elmsjö 2017

ISSN 1651-6192ISBN 978-91-554-9879-5urn:nbn:se:uu:diva-318296 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-318296)

A good hockey player plays wherethe puck is.

A great hockey player plays wherethe puck is going to be.

Wayne Gretzky

List of Papers

This thesis is based on the following papers, which are referred to in the text by their Roman numerals.

I Engskog, M.K.R., Karlsson, O., Haglöf, J., Elmsjö, A., Brittebo, E., Arvidsson, T., Pettersson, C. (2013) The cyanobacterial amino acid β-N-methylamino-l-alanine perturbs the intermediary metabolism in neonatal rats. Toxicology 312: 6-11.

II Elmsjö, A., Rosqvist, F., Engskog, M.K.R., Haglöf, J., Kullberg, J., Iggman, D., Johansson, L., Ahlström, H., Arvidsson, T., Risérus, U., Pettersson, C. (2015) NMR-based metabolic profiling in healthy individuals overfed different types of fat: links to changes in liver fat accumulation and lean tissue mass. Nutrition & Diabetes 5: e182.

III Elmsjö, A., Haglöf, J., Engskog, M.K.R., Nestor, M., Arvidsson, T., Pettersson, C. (2017) The co-feature ratio, a novel method for the measurement of chromatographic and signal selectivity in LC-MS-based metabolomics. Analytica Chimica Acta 956: 40-47.

IV Elmsjö, A., Haglöf, J., Engskog, M.K.R., Erngren, I., Nestor, M., Arvidsson, T., Pettersson, C. (2017) Selectivity evaluation using the co-feature ratio in LC-MS metabolomics: comparison of HILIC stationary phase’s performance for the analysis of plasma, urine and cell extracts. In manuscript.

Reprints were made with permission from the respective publishers.

Additional papers not included in this thesis:

McEwen, I., Elmsjö, A., Lehnström, A., Hakkarainen, B., Johansson, M. (2012) Screening of counterfeit corticosteroid in creams and ointments by NMR spectroscopy. Journal of Pharmaceutical and Biomedical Analysis 70: 245-250.

Aftab, O., Engskog, M.K.R., Haglöf, J., Elmsjö, A., Arvidsson, T., Pettersson, C., Hammerling, U., Gustafsson, M.G. (2014) NMR Spectroscopy-Based Metabolic Profiling of Drug-Induced Changes In Vitro Can Discriminate between Pharmacological Classes. Journal of Chemical Information and Modeling 54: 3251–3258.

Contents

Introduction ..................................................................................................... 9 

The metabolomics workflow ........................................................................ 11 Sample generation .................................................................................... 11 Sample preparation ................................................................................... 12 Data generation ........................................................................................ 13 Data analysis ............................................................................................ 14 Biological interpretation ........................................................................... 16 Validation ................................................................................................. 16 

AIM ............................................................................................................... 18 

Results and discussion .................................................................................. 19 Selectivity aspects of sample preparation ................................................ 19 Selectivity aspects of liquid chromatography .......................................... 23 The co-features ratio for method evaluation in metabolomics ................. 26 The biological aspect of study I and II ..................................................... 28 

Conclusions ................................................................................................... 30 

Populärvetenskaplig sammanfattning ........................................................... 31 

Acknowledgments......................................................................................... 33 

References ..................................................................................................... 35 

Abbreviations

ACN Acetonitrile BCAA Branched chained amino acids BMAA β-N-methylamino-L-alanine CFR Co-feature ratio CPMG Carr-Purcell-Meiboom-Gill EDTA Ethylenediaminetetraacetic acid ESI Electrospray ionization Feature A unique mass at a specific retention time HILIC Hydrophilic liquid chromatography HMDB Human metabolome database HRMS High resolution mass spectrometry LC Liquid chromatography LLE Liquid-liquid extraction LLE-Aq Aqueous phase of a liquid-liquid extraction LLE-Org Organic phase of a liquid-liquid extraction MS Mass spectrometry NMR Nuclear magnetic resonance PCA Principal component analysis PEG Polyethyleneglucol PLS Partial least square analysis PUFA Polyunsaturated fatty acids SFA Saturated fatty acids

9

Introduction

The modern concept of metabolomics was first introduced in the late 1990s by pioneers such as Oliver Fiehn and Jeremy Nicholson [1, 2]. Metabolomics can be defined as the scientific field aiming at a complete analysis of all low-weight substances (metabolites) in a given biological system [3]. The investigated system is often referred to as the metabolome and is considered as the downstream endpoint of all processes in that system, both environmental (diet, toxicology, drug, lifestyle) and genetic. Since its introduction, metabolomics has been fruitful within a wide range of scientific areas (e.g. nutrition and toxicology) [4-7] and has proven suitable in biomarker discovery as well as for hypothesis-generating studies [8].

A metabolome often consists of a complex mixture of thousands of metabolites from a variety of metabolite classes with a wide chemical diversity [9]. The metabolites vary not only in physicochemical properties such as molecular weight, hydrophobicity/hydrophilicity, acidity/basicity, solubility and boiling point, but also in their concentration magnitude. Due to this complexity, one of the major analytical challenges in metabolomics is to generate a representative biochemical presentation of the metabolome. The analysis is further complicated by numerous interfering matrix components that can have a strong impact on the results [10–12]. The choice of analytical method is therefore of fundamental importance in metabolomics, especially as metabolite coverage and data validity are highly dependent on the quality of the analytical methods in use [13].

Currently, there are two predominantly used techniques in metabolomics. Nuclear magnetic resonance (NMR), is characterized by its robustness and by its capacity to give both structural and (absolute) quantitative information for a large range of metabolites [14, 15]. The other frequently used technique is mass spectrometry (MS) (often hyphenated with liquid chromatography), which is known for a high sensitivity combined with a high selectivity for a large set of metabolites [16]. Even though these two techniques can, together, detect a large set of metabolites, they are not even close to unraveling the complete metabolome. One estimate is that a human metabolome contains around 8500 endogenous metabolites, and up to 40,000 additional exogenous metabolites (drugs, additives, toxins etc.) [17].

Considering that the two most capable techniques to date are not even close to being able to determine the complete metabolome, it is the author’s opinion that the field of metabolomics will be ‘under development’ for a long period.

10

It is also important to bear in mind that metabolomics investigations are seldom driven by analytical chemistry or by statistical questions, but by biology and biological questions. For these reasons, simple applicable strategies and tools will be constantly needed in order to evaluate and validate metabolomics methods. The author’s contribution to the metabolomics community has therefore been to highlight the importance of selectivity and to introduce a new concept for evaluation of selectivity in LC-MS metabolomics.

11

The metabolomics workflow

To put this thesis into its correct context, the general steps in all metabolomics investigations will be briefly described (Figure 1). The typical steps are sample generation and storage, sample preparation, data generation, data analysis, biological interpretation and validation. An overview of each step will be presented and the reader is referred to each individual paper for detailed information about the methods in use.

For clarification, this thesis will only use the term metabolomics, which could be considered an umbrella term for several different quantitative approaches used within the metabolomics community. Most of the terminologies and definitions are presented in more detail in a review by Dunn et al. [18].

Figure 1. An overview of the general steps often included in a metabolomics workflow. In all steps, analytical chemistry is incorporated in one way or another.

Sample generation Most metabolomics investigations can be likened to fishing expeditions, trying to catch the ancient pike. To ensure this is not an impossible task, it is very important to generate samples that can answer the research question being asked. The aim of the sample generation step, Figure 1, is therefore to acquire an unbiased metabolome capable of answering the current research question.

12

In this thesis, one animal study [18] and one clinical intervention [19] were included. The animal investigation was aimed at examining the metabolic response from rats following neonatal exposure to the neurodegenerative toxin β-N-methylamino-L-alanine (BMAA) [I]. In the clinical intervention, 39 human individuals were overfed for seven weeks with muffins containing saturated (SFA) or polyunsaturated fatty acids (PUFA) [II], to investigate the accumulation of fats in relation to insulin sensitivity. For papers III and IV in the thesis, blank and pooled samples were used, excluding variation associated with biological samples.

Choosing an appropriate sample type (urine, blood, saliva, etc.) in metabolomics is important and should relate to where the biological changes are expected to occur. In this thesis, four sample types were investigated: blood serum [I], blood plasma [II, IV], cells [III, IV], and urine [IV]. Whole blood is considered to be one of the most informative biological samples and is therefore one of the most popular sample types. To reduce perturbation in the metabolome during storage, whole blood is often turned into blood serum or blood plasma, by letting the blood coagulate or by adding anticoagulants (Ethylenediaminetetraacetic acid (EDTA), citrate or heparin) respectively. In some cases, serum can be preferable as it does not contain an anticoagulant that might disturb the analysis, but in general only small metabolite differences are observed between serum and blood plasma [20]. Urine is also a frequently used sample type in metabolomics, as it can be collected non-invasively. In paper IV, horse urine was used, which in many aspects is fundamentally different from human urine. Horses, which are herbivorous, have a well-developed cecum, so the excreted urine is heavily loaded with mucus [21]. Cells have also been used extensively in metabolomics investigations, as they are often used as in vitro models for a broad set of human diseases [22]. Within a lipid-rich cellular membrane, each cell contains an aqueous solution with a broad set of diverse biomolecules.

Sample collection, sample storage and sample handling are also important parameters affecting the generated biochemical presentation of the metabolome. Samples were collected through different sets of collaborations, each with their own scientific expertise. Sample storage and handling followed applicable recommendations [23–25].

Sample preparation After sample generation, a preparation step is often required to enable further analysis of the samples (Figure 1). In general, sample preparations in metabolomics are aimed at being as unselective and as comprehensive as possible. Hopefully this strategy enhances the probability of detecting as many metabolite perturbations as possible. Several publications compare and optimize different preparation procedures for metabolomics. The most

13

common strategy is to evaluate and optimize the methods, with the aim of detecting as many metabolites/features as possible [21,26–33]. A short summary of the sample preparation methods used in this thesis follows.

In papers I, II and IV, blood serum/plasma was precipitated with cold acetonitrile (ACN), enabling a physical removal of proteins, and lipids to some extent. The horse urine in paper IV was also precipitated with cold ACN. The sample preparation of cells is usually a two-step process, starting with a quenching followed by an extraction. The purpose of the quenching is to stop all cellular processes, thereby ‘freezing’ the metabolic profile. The aim of the extraction process is to extract as many metabolites as possible, while removing unwanted cellular debris such as lipids and proteins. In papers III and IV, cells were quenched and harvested using a standardized procedure [34, 35]. In paper III, four inherently different aliquots of cell extracts were acquired by using three different preparation techniques: dilution, solid phase extraction and liquid-liquid extraction (analyzing both the aqueous and organic phase). In paper IV, cells were prepared using a modified Folch-extraction [36], containing both water, chloroform and methanol [34, 37].

Data generation After sample preparation, the metabolite composition of each sample should be recorded (Figure 1). This is generally done either by assessing the intensities and resonance frequencies for all proton-containing metabolites or by assessing the intensities and the mass-to-charge ratios (m/z) for all ionizable metabolites. 1H-NMR is characterized by being an absolute quantitative technique, highly reproducible, and all metabolites in a given sample generally have the same response factor, dependent on the number of 1H in the molecule. Each recorded spectrum contains a lot of information (chemical shifts, intensities and split patterns) related to the chemical structures of the molecules in the sample. This information is highly usable for identifying the metabolites in the investigated samples.

Sensitivity and selectivity are much higher in MS than in NMR, but MS has some critical downsides. As the investigated samples physically interact with the instrument, because of contamination the response factor may vary over time if many samples are investigated one after the other [38]. Another factor, perhaps more severe, is that the response for a metabolite can depend on co-eluting species, an effect referred to as the matrix effect. As the matrix effect may not be constant between samples due to inter-individual variability in interfering components (e.g. lipids, proteins or inorganic salts), identical metabolite concentrations can show different response factors [39,40].

In papers I and II, precipitated blood plasma/serum was pulsed with the 1D NOESY-presat pulse sequence, a robust pulse sequence known for its capability to suppress large water signals [41,42]. In papers III and IV, cell

14

extracts, urine and blood plasma were analyzed on a UHPLC-ESI-Q-ToF system. Three HILIC stationary phases were used: amide [III, IV], bare silica [IV] and sulfobetaine [IV] [43,44]. The Q-ToF instrument was used to acquire accurate masses as well as fragmentation patterns through the collection of data-independent MSE spectra. Both accurate mass and MSE capability were needed to identify and correctly match the corresponding metabolites over different samples (see section on pre-processing).

Data analysis Metabolomics studies may include hundreds of observations (samples), where each sample often generates intensities of thousands of variables (bins or features). In addition, spectral data from NMR and MS are generally considered to be very complex, so this dense raw data needs to be processed and analyzed in several steps to make it interpretable (Figure 2). The description of data analysis has been divided into three sections: pre-processing, multivariate-/univariate analysis, and metabolite identification.

Data pre-processing Every step before the multivariate analysis can be considered as a pre-processing step (Figure 2). In simple terms, pre-processing is needed to convert raw data into usable variables for the multivariate analysis and to reduce any systematic variation in the samples. The first step in Figure 2 is aimed at reducing variation in the x-axis (e.g. chemical shift, retention times and m/z) and ensure that an identified metabolite/feature corresponds to the same metabolite/feature over all samples. For the NMR data [I, II], narrow spectra segments (‘bins’) were generated. For the LC-MS data [III, IV], each chromatographic peak was linked to its response by creating ‘features’ (a unique mass at a specific retention time). The normalization step focuses on reducing the systematic variation in the acquired intensities (the y-axis), which can differ due to, for example, instrumental imperfection or extraction yields. It is important that each metabolite, regardless of abundance, has a similar impact on the statistical models, so each feature is scaled. Pareto scaling has been used in all papers in this thesis, Figure 2. There are currently a number of different pre-processing procedures for NMR and LC-MS, all with benefits and disadvantages [27,45–49].

15

Figure 2. A general workflow for the data analysis in this thesis.

Multivariate and univariate analysis After pre-processing, bins or features were analyzed and interpreted by multivariate data analysis (Figure 2). Principal component analysis (PCA) and partial least squares analysis (PLS) are by far the most popular unsupervised and supervised models in the metabolomics community. In this thesis, PCA has been used [I-IV] to extract and display systematic variation in the data to provide an overview of the data [50]. The PLS models [I-IV], where all variance unrelated to the predefined y-variable (e.g. dose group, paper I) is filtered out, have been used to locate specific x-variables (bins or features) that contribute to group classifications. As PLS models are powerful tools for locating x-variables with high impact on the predictive components, there are risks of overfitting the model and locating random associations that fit the correlation at hand. Univariate models have been used in all four papers [I-IV] to visually demonstrate (biological) difference. Furthermore, as the univariate analysis has been performed by excluding a vast number of variables and manually reintegrating the metabolites that contribute to group separation, this could also be seen as a validation step.

16

Metabolite identification After important bins or features have been extracted by the multivariate/univariate analysis, the underlying metabolite needs to be identified (Figure 2). Even though thousands of metabolites have been identified in the past two decades, metabolite identification is still considered a major bottleneck in metabolomics. In this thesis [I-IV], metabolites have been primarily identified using the Human Metabolome Database [51] and the Metlin database [52], by comparing chemical shifts, split patterns, integrals for NMR spectra and accurate mass, MSE fragments, and isotope patterns for MS spectra. Assignments have been done according the Metabolomics Standards Initiative [53], and most metabolites have putative identities.

Biological interpretation It is important to remember that metabolomics investigations are often biological- or biochemical-driven and the primary aim is often to put the results of an investigation into its biological context. This could be done by the biologist or the biochemist alone, but it is important not exclude the analytical chemist in this process. Multivariate results must be interpreted with caution as patterns can occur due to other factors than biology, so biological interpretation should therefore be done in close collaboration with the principal investigator and the analytical chemist. The biological interpretations of the studies included in this thesis were performed in close collaboration with experts within their particular field.

Validation If not controlled, biological and experimental variations can induce systematic or randomized errors, resulting in erroneous biological interpretations (Figures 1 and 2). Validation therefore aims at ensuring that each step in the metabolomics workflow performs as intended. Each step described above has been validated in one way or another and, below, the most important strategies are described.

Multivariate analyses were used to systematically search for co-founders in the study designs (gender, age, sampling, run order, etc.). Use of extraction blanks and standards while monitoring the CV values has ensured the reproducibility of the sample preparations. Data acquisition was evaluated by using standards and a systematic injection of QC-samples (LC-MS) [54]. Multivariate models were primarily evaluated by cross validation, and metabolites important for group classifications in the clinical studies were re-integrated and subjected to t-tests. All biological interpretations were

17

performed in close collaboration with experts within their particular field. The combined expertise should be able to draw valid biological conclusions, and the process could also be considered as a biological validation step.

18

AIM

In metabolomics, there is often a trade-off between sample throughput, metabolite coverage and selectivity. Unfortunately, most metabolomics methods are solely evaluated on the basis of how fast they are and how many metabolites they acquire, even though selectivity might give a better measurement of data quality. The aims of this thesis were to demonstrate the importance of selectivity in metabolomics and to introduce a method for determining selectivity in LC-MS metabolomics.

The underlying aims of each paper were to: Determine the initial changes of major intermediary metabolites in

rat serum following neonatal exposure to BMAA by NMR spectroscopy-based metabolomics (I).

Identify non-lipid metabolites that may differentiate between a diet high in saturated (SFA) or polyunsaturated fatty acids (PUFA) (II).

Investigate the importance of selectivity in sample preparations (I-III).

Introduce a novel method for determining selectivity in metabolomics (III).

Apply the co-feature ratio for determination of the selectivity differences regarding three chromatographic methods (IV).

19

Results and discussion

Within the metabolomics community, there is no consensus on how to evaluate and validate methods applied in the studies. To date, most protocols have been optimized towards sample throughput and metabolite coverage, parameters that reflect the aim to acquire as many potential biomarkers as possible. Sample throughput and metabolite coverage could also be regarded as simpler to evaluate (due to the large number of analytes in metabolomics) than common bioanalytical validation parameters, such as accuracy, selectivity and linearity. From an analytical perspective, this is a concern, as there is often a trade-off between sample throughput, metabolite coverage and selectivity, which could ultimately be reflected in the quality of the data. Poor selectivity could potentially affect both the linearity and the accuracy, resulting in concealed or induced variations that could, in the worst-case scenario, affect the biological output of the metabolomics investigation in question [55,56].

From a quality assurance perspective, a detection technique with an absolute selectivity capable of measuring each metabolite individually without interference would be preferable in metabolomics. Unfortunately, there are no such techniques, and both NMR and MS are prone to suffer when selectivity is insufficient. It is therefore surprising that few researchers in the interdisciplinary field of metabolomics strive towards maximizing selectivity. This could be because, to the author’s knowledge, there are no good methods for measuring selectivity in metabolomics.

This thesis is based on four papers, each of which considers aspects of selectivity. Papers I and II discuss the importance of reducing the influence of macromolecules in NMR metabolomics; paper III introduces the novel co-feature ratio usable for evaluating selectivity in LC-MS based metabolomics; and paper IV demonstrates how the co-feature ratio could be used for evaluation of chromatographic methods.

Selectivity aspects of sample preparation Sample preparations in metabolomics are generally optimized towards being as unselective as possible. Generic methods such as dilution or precipitation are therefore often preferred over more selective procedures like SPE or LLE. However, using unselective methods can have drastic consequence where

20

matrix components might disturb the analysis for a broad set of analytes. Biological samples (e.g. cell extracts, urine or plasma) are very complex, and contain both low and high concentrations of interfering components such as proteins, inorganic salts and lipids.

In NMR metabolomics, salts and macro-molecules from the sample matrix can cause problems with baseline distortion and variation in the chemical shifts [57]. In papers I and II, lipids and proteins were physically removed through a precipitation step instead of the well-known Carr-Purcell-Meiboom-Gill (CPMG) pulse sequence, where the background is removed through spectral filtering, utilizing the differences in relaxation times between large and small molecules [58].

During the method development stage, sample precipitation with ACN had proved highly reproducible while inducing the selectivity for a number of metabolites, e.g. ketone bodies and branched chained amino acids (BCAA), compared to using the CPMG pulse (Figure 3). A precipitation step might also improve the quantitative signal for metabolites known to bind to proteins, e.g. ketone bodies, as these substances might have shorter relaxation times than expected, due to the binding [59]. The CPMG pulse would have been a more time-efficient approach, as no sample preparation is needed. However, as both ketone bodies and BCAA have been reported to correlate with insulin resistance [60–62], these metabolites were considered important in paper II. Figure 3 also shows that saturated and unsaturated protons in the carbon chains have different chemical shifts, meaning that a procedure including lipids might have missed the effects on small molecules in favor of the differences in the administered lipids types in paper II. It is also important to stress that there was a greater risk that the signals from lipids and proteins would considerably overlap both the signals from ketone bodies and the essential amino acids (Figure 3). For these reasons, a method with efficient removal of lipids and proteins was prioritized, and the use of a general metabolomics protocol using the CPMG pulse [58] was considered unwise. In addition, the introduced re-solvation step might have led to a homogenization of the samples (pH and ion strength) and thereby improve the reproducibility in comparison to the simple dilution commonly applied when using the CPMG pulse.

21

Figure 3. Differences in selectivity using sample preparation or the CPMG pulse sequence [unpublished data].

Matrix components such as inorganic salts, lipids and proteins may also disturb the analysis in LC-MS metabolomics [10,11,63]. However due to the complexity and multidimensional aspects of LC-MS data, visual comparisons like the one in Figure 3, might not be the most suitable way of comparing selectivity differences of LC-MS methods. Therefore, a novel numerical method for measuring selectivity was introduced in paper III to enable comparisons between sample preparations for LC-MS metabolomics. By dividing the intensity of each feature with the total intensity of its neighboring features (± 5s retention window), a ratio referred to as the co-feature ratio was obtained (Figure 4).

22

Figure 4. Ion-chromatograms illustrating the concept of the co-feature ratio [III]. The co-feature ratio (CFR) is calculated by dividing the intensity of a feature by the summarized total intensity of the co-eluting features (co-features). Features with an intensity apex within ± 5s of the selected feature’s apex are considered co-features. The ion chromatogram for an [M-H]- ion (black, solid line) and the ion chromatograms of five co-feature ions [adducts and co-eluting molecules] (grey, dashed lines) are illustrated in the figure.

The co-feature ratio is affected by the amount of co-elution as well as the isotope pattern, the adduct formation and the in-source fragmentation. The co-feature ratio could therefore be considered as a measurement of the combined chromatographic and the MS signal selectivity. A co-feature ratio value close to 1 is therefore indicative of high chromatographic and signal selectivity while a low ratio is due to poor chromatographic and/or signal selectivity.

The novel co-feature ratio was used in paper III to compare the selectivity of four extraction procedures. Large differences between the preparation methods were observed and several metabolites showed a biased quantitative signal due to poor chromatographic selectivity (e.g. due to co-elution) and/or poor signal selectivity (e.g. differences in adduct formation and in-source fragmentation during ionization). Tryptophan was one of the metabolites with the biggest difference in co-feature ratio between the four extraction procedures. This difference could be related to the amount of co-eluting sodium formate. The concentration of sodium was higher in the polar fractions (dilution and LLE-Aq) compared to the nonpolar fractions (SPE and LLE-org). The consequences for the intensities are shown in Figure 5. An overall

23

improved selectivity, without a drastic loss of features, was acquired when using LLE-Aq compared with a simple dilution [III]. It could therefore be argued that improved selectivity can be obtained without loss of coverage. Increased selectivity through a proper choice of the sample preparation may reduce the false positive and false negative discovery rate, thereby increasing the validity of any metabolomics investigation.

Fig. 5. Illustrations of ion suppression effects on L-tryptophan [III]. The left diagram shows extracted ion chromatograms for L-tryptophan (solid, black line) and one co-eluting feature (dashed, grey line) in DDI (A) and LLE-org (B) (see sample preparation section 2.8). The right diagram (C) shows the intensity of L-tryptophan in all samples by run order.

Selectivity aspects of liquid chromatography It is often stated that one of the most positive aspects of a separation step in LC-MS metabolomics, compared to NMR or direct infusion MS, is the improved selectivity. It is therefore interesting that there are very few metabolomics investigations evaluating the chromatographic systems in terms of their selectivity performance [64]. In paper IV, the co-feature ratio approach was used to evaluate three HILIC stationary phases (amide, bare silica and sufobetaine) for analyses of blood plasma, urine and cell extracts.

The multivariate analysis of the co-feature ratios demonstrated two clear trends. The first, illustrated in Figure 6A, related to the sample types. This trend was expected, as both type, number and quantity of metabolites normally differs between these types of samples, as shown in Figure 6C. The second trend, illustrated in Figure 6B, related to column type.

24

Figure 6. Overview of the trends relating to sample and column type [IV]. (A) Principal components 1 and 2 showed a clear trend based on sample type: cell extracts (black), blood plasma (white) and urine (grey). (B) Principal components 3 and 4 on the other hand separate the data based on column type: bare silica (■), amide (▲) and sulfobetaine (●). (C) Total ion current chromatograms of cell extracts, urine and blood plasma separated on the amide column. PCA models and TICs derive from the positive ionization.

In the study, the stationary phases had been selected based on their orthogonal properties (neutral, charged and zwitterion). It can be assumed that they would cause a large retention difference, which should have a great impact on the selectivity and the observed co-feature ratios. Most of the metabolites exhibiting the most distinct differences in co-feature ratios between the three stationary phases did co-elute with either of the three matrix components: inorganic salt, glycerophospholipids or polyethylene glycols (PEGs). In many cases, metabolites co-eluting with either of these matrix components showed changes in ionization efficiency and/or adduct formation, influencing their quantitative signals. Propionyl carnitine, Figure 7, illustrates the drastic

25

consequence that can occur when a metabolite co-elutes with an inorganic salt cluster.

Figure 7. The effect on signal intensity and its relation to co-elution with salt clusters [IV]. The sodium formate cluster (red) strongly suppresses the [M+H] of propionyl carnitine (black) when a co-elution of these two occurs (top ion chromatogram, sulfobetaine column).

Even though the consequences of co-elution with matrix components are well reported [63,65,66], the effects are often neglected in metabolomics. This is surprising, as matrix components such as salt clusters, e.g. salt concentrations in human urine, can demonstrate huge inter-individual variability [67]. To minimize such matrix effects, it is the author’s opinion that HILIC phases displaying distinct, narrow and well-separated salt clusters, as well as a narrow overall spread in the retention range of lipids and PEGs, should be preferred in metabolomics, Figure 8. However, it is important to stress that column choice should depend on the research question, as each metabolome/metabolomics investigation is unique.

26

Figure 8. Ion chromatograms presenting the overall spread in retention of glycerophosphocholines (A) and PEGs (B) for the three HILIC columns [IV]. The ion chromatograms of the primary fragment (184 m/z) of glycerophosphocholines and eight PEG oligomers here illustrate the overall retention range of these compounds over the three HILIC columns.

The co-features ratio for method evaluation in metabolomics As there are no golden standards or guidelines on how to perform method evaluation in metabolomics, each research group often has its own approach on how to evaluate and validate metabolomics methods. The predominantly used strategies for measuring a method’s performance regarding metabolite coverage include using a set of standards or counting the number of detected features [68–74]. However, combining a single set of standards without any prior knowledge of the biochemical composition of the investigated metabolome is questionable, especially as the matrix effects are known to be both analyte and matrix dependent [75,76]. Furthermore, counting the number of detected features can also be disputed, as the quantity of peaks seldom relates to the quality of the detected peaks. As an example, methods based on feature counting will automatically favor metabolites that produce numerous features (due to isotopes, adducts or in-source fragmentation) and disfavor metabolites that that only produce a single feature.

There are however two other strategies of note. These strategies attempt to ensure the quality of each feature, by either only including features that display a linear response (often multiple dilutions of a QC-sample) [77,78] or only including features that have been successfully annotated [21,77]. Unfortunately, these strategies either disfavor metabolites in low concentrations or could be impractical due to the extensive data analysis.

27

A feature’s co-feature ratio, Figure 4, is affected by the amount of co-elution (chromatographic selectivity), as well as the isotopic pattern, the adduct formation, and the in-source fragmentation (MS-signal selectivity). In a metabolomics study, acquired co-feature ratios can be used for an untargeted evaluation of the overall selectivity between methods, for comparing the methods’ selectivity towards a set of putatively annotated metabolites (e.g. belonging to a specific pathway), or as a method for identifying the difference between interfering matrix components. The co-feature ratio approach has been successfully used for evaluating sample preparation methods [III] and for comparing chromatographic setups [IV], Figure 9, but the author points out that the co-feature ratio has potential to be used for selectivity evaluation in many of the other stages generally included in metabolomics investigations.

Figure 9. Examples of steps where a co-feature ratio approach could be applicable. The co-feature ratio approach can be used for a multivariate evaluation of a number of steps in the metabolomics workflow, e.g. sample preparation, analytical separation and the MS-acquisition [III, IV].

As a method’s selectivity often relates to the risk of matrix effects, one of the more essential aspects of the co-features ratio approach is its relation to the signal/feature quality. This aspect can be compared to counting the number of detected features, which in some cases can be a biased evaluation parameter. In paper IV, for example, it is demonstrated that when different salt clusters (e.g. sodium and potassium, Figure 10) co-elute, hundreds of extract features are formed, an unwanted side effect that would be stimulated if the number of features alone were used as an evaluation parameter. Furthermore, neither use of standards nor counting the number of detected features consider the consequences of co-eluting interferences from the sample matrix. This is critical, as each metabolome is often unique and interfering matrix components can demonstrate huge inter-individual variability. Other positive aspects of using the co-feature ratio approach are that standards are not needed, the procedure does not disfavor low-abundant metabolites, and the approach requires no extra laboratory work.

28

Figure 10. Specific m/z-patterns arising when the potassium and sodium clusters co-elute [IV]. When co-elution occurs, combinations of salt clusters are formed, containing both sodium and potassium [CHOO(NaxKyCHOO)n].

The biological aspect of paper I and II This thesis includes two metabolic profiling studies, I and II. Briefly, β-N-methylamino-L-alanine (BMAA) has long been considered a potential health threat because of its assumed role in neurodegenerative diseases. In paper I, serum from rats was collected following neonatal exposure to BMAA [18]. A non-lipid metabolomics approach identified five main metabolites important for group classification. Integration and t-testing indicated significant altered levels of acetate, glucose, lactate, 3-hydroxybutyrate and creatine for the lowest-dose group (40 mg/kg of BMAA) [I]. This study is the only study so far that has observed significant biochemical effects after a low short-term dose of BMAA. This non-lipid NMR-based metabolomics approach could reveal up and down regulation of metabolites that could be related to both amino acids and energy metabolism.

Studies have suggested that dietary fat composition could play a key role in ectopic fat deposition. In paper II, plasma was collected from 39 individuals who had been overfed for seven weeks with muffins containing saturated (SFA) or polyunsaturated fatty acids (PUFA) [19]. Acquired metabolomics data suggested that several metabolites were significantly different (p˂0.1)

29

between the two diet groups. The PUFA group showed higher levels of lactate and lower levels of acetate and 3-hydroxybutyrate compared to the SFA group. Three fat depots (liver fat, total body fat, and VAT) showed similar correlation to ketone bodies, lactate, alanine, and the branched amino acids. For lean tissue, a reverse relationship (decreasing levels with increase in lean tissue mass) was found for the above metabolites [II]. Using a non-lipid NMR-based metabolomics approach allowed us to demonstrate that SFA and PUFA have different effects on the intermediary metabolism during hypercaloric conditions. These changes in metabolic profile might reflect a depot-specific fat storage and insulin resistance, something that needs further investigation.

30

Conclusions

One of the most difficult and critical aspects in metabolomics is to acquire a representative biochemical presentation of the metabolome. This relates both to the complexity of the metabolome and the fact that each step in the metabolomics workflow has multiple options that, in one way or another, will affect the outcome. Strategies and tools for comparing and evaluating analytical procedures in metabolomics are therefore needed to enable selection of the most suitable method. One important evaluation parameter often overlooked in metabolomics is selectivity, often in favor of other parameters such as metabolite coverage and sample throughput.

By comparing different sample preparations and using different HILIC stationary phases, this thesis demonstrates the importance of acquiring sufficient selectivity in NMR and LC-MS metabolomics. In many cases, poor selectivity affected the observed signal intensity where, for example, lipids and proteins produced broad and abundant peaks in NMR and causing matrix effects in MS. Both scenarios resulted in affected quantitative signals for a broad set of metabolites. This is critical in metabolomics, as matrix components such as proteins, inorganic salt, and lipids can show great inter-individual variability, with drastic consequences for a metabolomics investigation.

This thesis introduces a novel and straightforward method for obtaining a measurement of the combined chromatographic (i.e. amount of co-elution) and signal selectivity (e.g. adduct formation). The procedure, referred to as the co-feature ratio, can in an untargeted manner evaluate the selectivity for LC-MS metabolomics procedures and potentially identify metabolites that experience matrix-associated effects. The co-feature ratio approach has been used to study selectivity by applying different sample preparation and chromatographic methods.

As each metabolomics investigation is unique and the matrix effects relate to the sample composition, generic methods should be avoided to ensure that results are valid. Each step in the metabolomics workflow should be optimized to the specific research question, and selectivity should be an important optimization parameter to ensure valid results in metabolomics. Improved selectivity could reduce the risks of accounting matrix effects, thereby improving the overall quality and validity of metabolomics investigations.

31

Populärvetenskaplig sammanfattning

En människas metabolom består av en rad olika småmolekyler, s.k. metaboliter. Exempel på metaboliter kan vara aminosyror, vitaminer och sockerarter. Sammansättningen av en människas metabolom påverkas dels av det genetiska arvet, men också av allt en kropp kan tänkas utsättas för såsom mat, dryck, gifter och sjukdomar. Genom att studera förändringar i metabolomet, s.k. metabolomik, kan man få en uppfattning om vad som försiggår i kroppen. Informationen från en metabolomikstudie kan således användas för att till exempel försöka förstå mekanismen bakom en sjukdom eller för att få en inblick i hur en viss kost påverkar kroppen. För att öka chansen att upptäcka en eventuell förändring i metabolomet har de flesta mätmetoder inom metabolomik utvecklats för att kunna bestämma så många metaboliter som möjligt. Tyvärr har denna prioritering oftast gjorts utan reflektion över vilken kvalité varje enskild metabolit mäts med.

Selektivitet är en parameter som är av stor vikt när det kommer till mätkvalitet. Selektivitet skulle kunna liknas vid ett mått på andelen störande ämnen (interferenser) vid bestämning av en metabolit. Trots denna viktiga aspekt negligeras oftast selektivitet inom metabolomikforskning. Problematiken med att åsidosätta selektivitet skulle kunna liknas vid att man var ute efter att väga en gammelgädda i en sjö, men att man inte reflekterar över att man vid invägningen av gäddan samtidigt har en brax eller mört på vågen. Denna avhandling avser att belysa betydelsen av selektivitet samt att utveckla en metod för att kunna mäta selektivitet i metabolomik.

För att minimera effekten av interferenser på mätningen av metaboliter är det brukligt att prover förbehandlas innan själva mätningen utförs. Eftersom de flesta metabolomikmetoder har utvecklats för att detektera så många metaboliter som möjligt används oftast väldigt generella förbehandlingar av provet. Det leder till att en förhållandevis stor andel av interferenser är kvar i provet vid mätning. I avhandlingen studeras flera olika förberedelsertekniker och dess effekt på selektivitet. Genomgående upptäcktes en rad interferenser störa bestämningen av metaboliter som ett resultat av en alltför dålig selektivitet. Denna effekt kan ha stor betydelse i en metabolomikstudie då en observerad mätförändring hos en molekyl de facto kan vara en sekundär effekt av en förändring i en interferens. Till exempel visade det sig att salthalten i proverna hade stor inverkan på mätningarna och effekten har visats kunna ge

32

en upp till 30 ggr förändring i respons hos metaboliter. Detta kan i en metabolomikstudie vara förödande då det sedan länge är känt att salthalten i t.ex. urin tenderar att fluktuera både mellan individer och tid på dygnet.

En metabolomikstudie består av en rad steg där följande oftast är inkluderade: generering av prover, provupparbetning, mätning, dataanalys och tolkning av resultat. Varje steg i flödet kan utföras på olika sätt, vilket i sin tur kommer att ha mer eller mindre påverkan på metabolitpresentationen av metabolomet. Därför behövs det bra tillvägagångsätt för att kunna utvärdera olika metoder för varje enskilt steg. I avhandlingen presenteras ett nytt och unikt tillvägagångsätt för att bestämma selektivitet i metabolomikstudier. Tillvägagångsättet går ut på att det för varje enskild metabolit beräknas en kvot som anger hur stor andel den aktuella metabolitens respons utgör av den totala responsen vid just den mätningen (dvs. en fisks viktsandel av den totala vikten på vågen). Ett värde på 1 betyder hög selektivitet då det bara är en metabolit som mäts vid just den tidpunkten. Vid ett värde nära 0 så gäller det omvända och metaboliten i fråga mäts uppenbarligen med en väldigt låg selektivitet. Detta värde skulle därmed kunna liknas vid ett mått på hur hög kvalitet varje enskild metabolit kan mätas med. Tillvägagångsättet kan därför vara användbart vid jämförelser mellan metoder och hjälpa till att välja den mest lämpliga.

Eftersom varje metabolomikstudie oftast är unik kan man anta att interferenserna i stor utsträckning också är unika. Det är därför författarens uppfattning att generiska metoder bör undvikas och varje steg i metabolomikflödet bör optimeras med avseende på nyckelmetaboliter och förväntade interferenser i det studerade provet. Selektivitet bör betraktas som en viktig parameter i metabolomikstudier eftersom dess länk till mätkvalitet är tydligare än när man t.ex. mäter antalet detekterbara metaboliter. En allmänt förbättrad selektivitet inom metabolomikfältet bör leda till ökad kvalitet inom fältet. Detta skulle kunna leda till att färre falska positiva/negativa resultat genereras vilket i sin tur skulle spara både tid och resurser.

33

Acknowledgments

The work of this thesis was carried out at the Division of Analytical Pharmaceutical Chemistry at Uppsala University, Sweden. I would like to express my sincere gratitude and appreciation to the following persons, I wouldn’t have managed without you;

First of all I would like to thank my main supervisor prof. Curt Pettersson for recruiting and guiding me throughout these years. I am very grateful for your straightforward manner and for always having time for even the smallest things. Even though I’ve been part of teams throughout my entire life, it is not until now I understand the true concept of being part of a team. I would also like to thank prof. Torbjörn Arvidsson, my co-supervisor, for always staying positive and for always trying to see things from another perspective.

All former and present colleagues:

Mikael Engskog and Jakob Haglöf, thanks for all the stimulating discussions. It has been a real pleasure and a privilege being part of the same research group for more than 6 years. Ylva Hedeland, thank you for all advices, both technology and teaching related; I wouldn’t have managed without them. Mikael Hedeland, thank you for being a role model when it comes to critical thinking, a true inspiration source. The present members of the A-team; Alfred and Annelie, thanks for the friendship during these years, it wouldn’t have been the same without you guys. I have truly valued our relaxed discussions, both the scientific and the non-scientific ones, and I’m really looking forward to our next trip. The former members of the A-team; Alexander and Axel, thanks for climbing the ‘teaching ladder’ together. I still remember the first course we supervised together, it was insane. Victoria, thanks for being an inspiration source when it comes to orderliness and, even though I try, I will never reach your caliber. Olle, thanks for all tips and chats over these years, I have enjoyed them all. Ida, thanks for valid tips during our group meetings and special thanks for not participating in the ‘Blodomloppet 2016’, I really needed to ‘win’ something that year. Kristian, my roommate, thanks for pleasant company, nice chats and for the quick fixes with my computer and Excel. To all the other current and former colleagues at the division, thanks for making this trip enjoyable and for all the help I have

34

received. Also big thanks to all other colleagues at the department, for assistance and for the lunch and coffee breaks during these years.

This thesis would not have been possible without the expertise and contribution from my co-authors, all of whom I greatly thank. I would also like to thank the current and former employees at the Medical product agency that I unfortunate don’t meet every day. Torgny, Kalle, Birgit, Andreas, Ahmad and Monika, thanks for all the tips and for always staying positive. Special thanks to Ian McEwen, for inspiring me to go into research and introducing me to NMR spectroscopy.

Smålands nation and The Swedish Academy of Pharmaceutical Sciences (Apotekarsocieteten), thanks for traveling grants and scholarships enabling me to participate in international courses and conferences.

Friends and family;

Åke, thanks for all weird topics we have discussed during these years and for being a good friend. I still enjoy your company, even though we have swapped the bouldering hall, our playground, to a playground for children. Jonas, thanks for once in a while reminding me the importance of having a ‘semla’ or ‘smågodis’. Big thanks to all friends outside the university, helping me to relax and enjoy life. Lars-Göran and Berit Nilsson, thanks for all support during these years and for welcoming me into your family with open arms. My parents Lisa and Lars, and my siblings Nina, Josefin and Freja, thanks for support and for all wide discussions at the kitchen table; they have enlarged my perspective and built my character.

Finally I would like to thank the most important ones in my life, Ellinor, Myra and Ture. Thanks for making my worst days feel much better and for giving me the energy to pull through. Ellinor, it’s an understatement when I say that I wouldn't have managed without you. Our family is the best reason for looking forward and I love you with all of my heart.

Albert,

Uppsala, April 2017

35

References

1. J.K. Nicholson, J.C. Lindon, E. Holmes, “Metabonomics”: understanding the metabolic responses of living systems to pathophysiological stimuli via multivariate statistical analysis of biological NMR spectroscopic data. Xenobiotica. 29 (1999) 1181–1189.

2. O. Fiehn, Metabolomics--the link between genotypes and phenotypes. Plant Mol. Biol. 48 (2002) 155–171.

3. S.G. Oliver, M.K. Winson, D.B. Kell, F. Baganz, Systematic functional analysis of the yeast genome. Trends Biotechnol. 16 (1998) 373–378.

4. M. Coen, E. Holmes, J.C. Lindon, J.K. Nicholson, NMR-based metabolic profiling and metabonomic approaches to problems in molecular toxicology. Chem. Res. Toxicol. 21 (2008) 9–27.

5. J.L. Griffin, R.A. Kauppinen, Tumour metabolomics in animal models of human cancer, J. Proteome Res. 6 (2007) 498–505.

6. S. Rezzi, Z. Ramadan, L.B. Fay, S. Kochhar, Nutritional metabonomics: applications and perspectives. J. Proteome Res. 6 (2007) 513–525.

7. J.K. Nicholson, J. Connelly, J.C. Lindon, E. Holmes, Metabonomics: a platform for studying drug toxicity and gene function. Nat. Rev. Drug Discov. 1 (2002) 153–161.

8. A. Roux, D. Lison, C. Junot, J.F. Heilier, Applications of liquid chromatography coupled to mass spectrometry-based metabolomics in clinical chemistry and toxicology: A review, Clin. Biochem. 44 (2011) 119–135.

9. D. Vuckovic, Current trends and challenges in sample preparation for global metabolomics using liquid chromatography-mass spectrometry, Anal. Bioanal. Chem. 403 (2012) 1523–48.

10. F. Calbiani, M. Careri, L. Elviri, A. Mangia, I. Zagnoni, Matrix effects on accurate mass measurements of low-molecular weight compounds using liquid chromatography-electrospray-quadrupole time-of-flight mass spectrometry, J. Mass Spectrom. 41 (2006) 289–294.

11. Y.F. Xu, W. Lu, J.D. Rabinowitz, Avoiding misannotation of in-source fragmentation products as cellular metabolites in liquid chromatography-mass spectrometry-based metabolomics, Anal. Chem. 87 (2015) 2273–2281.

12. P.A. Guy, I. Tavazzi, S.J. Bruce, Z. Ramadan, S. Kochhar, Global metabolic profiling analysis on human urine by UPLC – TOFMS : Issues and method validation in nutritional metabolomics, 871 (2008) 253–260.

13. G.J. Patti, Separation strategies for untargeted metabolomics, J. Sep. Sci. 34 (2011) 3460–3469.

14. C. Stella, B. Beckwith-hall, O. Cloarec, E. Holmes, J.C. Lindon, J. Powell, F. Van Der Ouderaa, S. Bingham, A.J. Cross, J.K. Nicholson, Susceptibility of Human Metabolic Phenotypes to Dietary Modulation research articles, (2006) 2780–2788.

15. E.M. Lenz, I.D. Wilson, Analytical strategies in metabonomics, J. Proteome Res. 6 (2007) 443–458.

36

16. H.G. Gika, G.A. Theodoridis, R.S. Plumb, I.D. Wilson, Current practice of liquid chromatography–mass spectrometry in metabolomics and metabonomics, J. Pharm. Biomed. Anal. 87 (2014) 12–25.

17. D.S. Wishart, Advances in metabolite identification, Bioanalysis. 3 (2011) 1769–1782.

18. O. Karlsson, K. Kultima, H. Wadensten, A. Nilsson, E. Roman, P.E. Andrén, E.B. Brittebo, Correction for Neurotoxin-Induced Neuropeptide Perturbations in Striatum of Neonatal Rats, J. Proteome Res. 12 (2013) 2321–2321.

19. F. Rosqvist, D. Iggman, J. Kullberg, J. Cedernaes, H.E. Johansson, A. Larsson, L. Johansson, H. Ahlström, P. Arner, I. Dahlman, U. Risérus, Overfeeding polyunsaturated and saturated fat causes distinct effects on liver and visceral fat accumulation in humans, Diabetes. 63 (2014) 2356–2368.

20. J.R. Denery, A.A.K. Nunes, T.J. Dickerson, Characterization of Differences between Blood Sample Matrices in Untargeted Metabolomics, Anal. Chem. 83 (2011) 1040–1047.

21. N. Stojiljkovic, A. Paris, P. Garcia, M.A. Popot, Y. Bonnaire, J.C. Tabet, C. Junot, Evaluation of horse urine sample preparation methods for metabolomics using LC coupled to HRMS, Bioanalysis. 6 (2014) 785–803.

22. W.B. Dunn, D.I. Broadhurst, H.J. Atherton, R. Goodacre, J.L. Griffin, Systems level studies of mammalian metabolomes: the roles of mass spectrometry and nuclear magnetic resonance spectroscopy, Chem. Soc. Rev. 40 (2011) 387–426.

23. J.L. Luque-Garcia, T. a Neubert, Sample preparation for serum/plasma profiling and biomarker identification by mass spectrometry, J. Chromatogr. A. 1153 (2007) 259–276.

24. B. Jørgenrud, S. Jäntti, I. Mattila, P. Pöhö, K.S. Rønningen, H. Yki-Järvinen, M. Orešič, T. Hyötyläinen, The influence of sample collection methodology and sample preprocessing on the blood metabolic profile, Bioanalysis. 7 (2015) 991–1006.

25. P. Yin, R. Lehmann, G. Xu, Effects of pre-analytical processes on blood samples used in metabolomics studies, Anal. Bioanal. Chem. 407 (2015) 4879–4892.

26. C.A Daykin, P.J.D. Foxall, S.C. Connor, J.C. Lindon, J.K. Nicholson, The comparison of plasma deproteinization methods for the detection of low-molecular-weight metabolites by (1)H nuclear magnetic resonance spectroscopy, Anal. Biochem. 304 (2002) 220–230.

27. H.J. Issaq, Q.N. Van, T.J. Waybright, G.M. Muschik, T.D. Veenstra, Analytical and statistical approaches to metabolomics research., J. Sep. Sci. 32 (2009) 2183–2199.

28. D.G. Sitnikov, C.S. Monnin, D. Vuckovic, Systematic Assessment of Seven Solvent and Solid-Phase Extraction Methods for Metabolomics Analysis of Human Plasma by LC-MS, Sci. Rep. 6 (2016) 38885.

29. N.G. Mahieu, X. Huang, Y.J. Chen, G.J. Patti, Credentialing features: A platform to benchmark and optimize untargeted metabolomic methods, Anal. Chem. 86 (2014) 9583–9589.

30. M. Calderón-Santiago, F. Priego-Capote, M.D.L. de Castro, Enhancing detection coverage in untargeted metabolomics analysis by solid-phase extraction on-line coupled to LC-MS/MS, Electrophoresis. 36 (2015) 2179–2187.

31. P. Masson, A.C. Alves, T.M.D. Ebbels, J.K. Nicholson, E.J. Want, Optimization and evaluation of metabolite extraction protocols for untargeted metabolic profiling of liver samples by UPLC-MS, Anal. Chem. 82 (2010) 7779–7786.

32. F.M. Geier, E.J. Want, A.M. Leroi, J.G. Bundy, Cross-platform comparison of caenorhabditis elegans tissue extraction strategies for comprehensive metabolome coverage, Anal. Chem. 83 (2011) 3730–3736.

37

33. S. Tufi, M.H. Lamoree, J. De Boer, P.E.G. Leonards, Cross-platform metabolic profiling: Application to the aquatic model organism Lymnaea stagnalis, Anal. Bioanal. Chem. 407 (2015) 1901–1912.

34. Z. León, J.C. García-Cañaveras, M.T. Donato, A. Lahoz, Mammalian cell metabolomics: Experimental design and sample preparation., Electrophoresis. (2013) 1–29.

35. M. Engskog, M. Björklund, J. Haglöf, T. Arvidsson, M. Shoshan, C. Pettersson, Metabolic profiling of epithelial ovarian cancer cell lines : evaluation of harvesting protocols for profiling using NMR spectroscopy, Bioanalysis. 7 (2015) 157–166.

36. J. Folch, M. Lees, G.H. Sloane, A simple method for the isolation and purification of total lipides from animal tissues, J. Biol. Chem. 226 (1956) 497–509.

37. K. Dettmer, N. Nürnberger, H. Kaspar, M. a Gruber, M.F. Almstetter, P.J. Oefner, Metabolite extraction from adherently growing mammalian cells for metabolomics studies: optimization of harvesting and extraction protocols., Anal. Bioanal. Chem. 399 (2011) 1127–1139.

38. E. Zelena, W.B. Dunn, D. Broadhurst, S. Francis-McIntyre, K.M. Carroll, P. Begley, S. O’Hagan, J.D. Knowles, A. Halsall, I.D. Wilson, D.B. Kell, D.B. Kell, Development of a Robust and Repeatable UPLC−MS Method for the Long-Term Metabolomic Study of Human Serum, Anal. Chem. 81 (2009) 1357–1364.

39. I. Kohler, A. Verhoeven, R.J. Derks, M. Giera, Analytical pitfalls and challenges in clinical metabolomics, Bioanalysis. 8 (2016) 1509–1532.

40. A. Van Eeckhaut, K. Lanckmans, S. Sarre, I. Smolders, Y. Michotte, Validation of bioanalytical LC–MS/MS assays: Evaluation of matrix effects, J. Chromatogr. B. 877 (2009) 2198–2207.

41. J.K. Nicholson, P.J. Foxall, M. Spraul, R.D. Farrant, J.C. Lindon, 750 MHz 1H and 1H-13C NMR spectroscopy of human blood plasma, Anal. Chem. 67 (1995) 793–811.

42. B.M.C. Potts, A.J. Deese, G.J. Stevens, M.D. Reily, D.G. Robertson, J. Theiss, NMR of biofluids and pattern recognition: assessing the impact of NMR parameters on the principal component analysis of urine from rat and mouse, J. Pharm. Biomed. Anal. 26 (2001) 463–476.

43. A.J. Alpert, M. Shukla, A.K. Shukla, L.R. Zieske, S.W. Yuen, M.A.J. Ferguson, A. Mehlert, M. Pauly, R. Orlando, Hydrophilic-interaction chromatography of complex carbohydrates, J. Chromatogr. A. 676 (1994) 191–202.

44. P. Jandera, Stationary and mobile phases in hydrophilic interaction chromatography: a review, Anal. Chim. Acta. 692 (2011) 1–25.

45. A. Craig, O. Cloarec, E. Holmes, J.K. Nicholson, J.C. Lindon, Scaling and normalization effects in NMR spectroscopic metabonomic data sets, Anal. Chem. 78 (2006) 2262–2267.

46. B.A. Ejigu, D. Valkenborg, G. Baggerman, M. Vanaerschot, E. Witters, J.C. Dujardin, T. Burzykowski, M. Berg, Evaluation of normalization methods to pave the way towards large-scale LC-MS-based metabolomics profiling experiments, OMICS. 17 (2013) 473–485.

47. R. Smith, D. Ventura, J.T. Prince, LC-MS alignment in theory and practice: A comprehensive algorithmic review, Brief. Bioinform. 16 (2013) 104–117.

48. N.G. Mahieu, J.L. Genenbacher, G.J. Patti, A roadmap for the XCMS family of software solutions in metabolomics, Curr. Opin. Chem. Biol. 30 (2016) 87–93.

49. Y. Wu, L. Li, Sample normalization methods in quantitative metabolomics, J. Chromatogr. A. 1430 (2016) 80–95.

50. S. Wold, K. Esbensen, P. Geladi, Principal component analysis, Chemom. Intell. Lab. Syst. 2 (1987) 37–52.

38

51. D.S. Wishart, D. Tzur, C. Knox, R. Eisner, A.C. Guo, N. Young, D. Cheng, K. Jewell, D. Arndt, S. Sawhney, C. Fung, L. Nikolai, M. Lewis, M.-A. Coutouly, I. Forsythe, P. Tang, S. Shrivastava, K. Jeroncic, P. Stothard, G. Amegbey, D. Block, D.D. Hau, J. Wagner, J. Miniaci, M. Clements, M. Gebremedhin, N. Guo, Y. Zhang, G.E. Duggan, G.D. Macinnis, A.M. Weljie, R. Dowlatabadi, F. Bamforth, D. Clive, R. Greiner, L. Li, T. Marrie, B.D. Sykes, H.J. Vogel, L. Querengesser, HMDB: the Human Metabolome Database, Nucleic Acids Res. 35 (2007) D521-6.

52. C. A Smith, G. O’Maille, E.J. Want, C. Qin, S. A Trauger, T.R. Brandon, D.E. Custodio, R. Abagyan, G. Siuzdak, METLIN: a metabolite mass spectral database, Ther Drug Monit. 27 (2005) 747–751.

53. L.W. Sumner, T. Samuel, R. Noble, S.D. Gmbh, D. Barrett, M.H. Beale, N. Hardy, Proposed minimum reporting standards for chemical analysis Chemical Analysis Working Group (CAWG) Metabolomics Standards Initiative (MSI), Metabolomics. 3 (2007) 211–221.

54. W.B. Dunn, The importance of experimental design and QC samples in large-scale and MS-driven untargeted metabolomic studies of humans, Bioanalysis 4 (2012) 2249–2264.

55. S. Tulipani, R. Llorach, M. Urpi-Sarda, C. Andres-Lacueva, Comparative analysis of sample preparation methods to handle the complexity of the blood fluid metabolome: When less is more, Anal. Chem. 85 (2013) 341–348.

56. C. Bylda, R. Thiele, U. Kobold, D. A Volmer, Recent advances in sample preparation techniques to overcome difficulties encountered during quantitative analysis of small molecules from biofluids using LC-MS/MS., Analyst. 139 (2014) 2265–2276.

57. S. Zhang, G. a Nagana Gowda, T. Ye, D. Raftery, Advances in NMR-based biofluid analysis and metabolite profiling, Analyst. 135 (2010) 1490–1498.

58. O. Beckonert, H.C. Keun, T.M.D. Ebbels, J. Bundy, E. Holmes, J.C. Lindon, J.K. Nicholson, Metabolic profiling, metabolomic and metabonomic procedures for NMR spectroscopy of urine, plasma, serum and tissue extracts, Nat. Protoc. 2 (2007) 2692–2703.

59. J.D. Bell, J.C. Brown, G. Kubal, P.J. Sadler, NMR-invisible lactate in blood plasma, FEBS Lett. 235 (1988) 81–86.

60. C.B. Newgard, J. An, J.R. Bain, M.J. Muehlbauer, R.D. Stevens, L.F. Lien, A.M. Haqq, S.H. Shah, M. Arlotto, C. a Slentz, J. Rochon, D. Gallup, O. Ilkayeva, B.R. Wenner, W.S. Yancy, H. Eisenson, G. Musante, R.S. Surwit, D.S. Millington, M.D. Butler, L.P. Svetkey, A branched-chain amino acid-related metabolic signature that differentiates obese and lean humans and contributes to insulin resistance, Cell Metab. 9 (2009) 311–326.

61. B. Laferrère, D. Reilly, S. Arias, N. Swerdlow, P. Gorroochurn, B. Bawa, M. Bose, J. Teixeira, R.D. Stevens, B.R. Wenner, J.R. Bain, M.J. Muehlbauer, A. Haqq, L. Lien, S.H. Shah, L.P. Svetkey, C.B. Newgard, Differential metabolic impact of gastric bypass surgery versus dietary intervention in obese diabetic subjects despite identical weight loss, Sci. Transl. Med. 3 (2011) 80re2.

62. A.C. Shin, M. Fasshauer, N. Filatova, L. a Grundell, E. Zielinski, J.Y. Zhou, T. Scherer, C. Lindtner, P.J. White, A.L. Lapworth, O. Ilkayeva, U. Knippschild, A.M. Wolf, L. Scheja, K.L. Grove, R.D. Smith, W.J. Qian, C.J. Lynch, C.B. Newgard, C. Buettner, Brain Insulin Lowers Circulating BCAA Levels by Inducing Hepatic BCAA Catabolism, Cell Metab. 20 (2014) 898–909.

63. A. Ekdahl, M.C. Johansson, M. Ahnoff, Tracing and separating plasma components causing matrix effects in hydrophilic interaction chromatography–electrospray ionization mass spectrometry, J. Chromatogr. B. 923 (2013) 83–91.

39

64. K. Ortmayr, T.J. Causon, S. Hann, G. Koellensperger, Increasing selectivity and coverage in LC-MS based metabolome analysis, TrAC - Trends Anal. Chem. 82 (2016) 358–366.

65. X. Guo, E. Lankmayr, Phospholipid-based matrix effects in LC – MS bioanalysis, Bioanalysis. 3 (2011) 349–352.

66. R. Weaver, R.J. Riley, Identification and reduction of ion suppression effects on pharmacokinetic parameters by polyethylene glycol 400, Rapid Commun. Mass Spectrom. 20 (2006) 2559–2564.

67. S. Bouatra, F. Aziat, R. Mandal, A.C. Guo, M.R. Wilson, C. Knox, T.C. Bjorndahl, R. Krishnamurthy, F. Saleem, P. Liu, Z.T. Dame, J. Poelzer, J. Huynh, F.S. Yallou, N. Psychogios, E. Dong, R. Bogumil, C. Roehring, D.S. Wishart, The Human Urine Metabolome, PLoS One. 8 (2013) e73076.

68. S.U. Bajad, W. Lu, E.H. Kimball, J. Yuan, C. Peterson, J.D. Rabinowitz, Separation and quantitation of water soluble cellular metabolites by hydrophilic interaction chromatography-tandem mass spectrometry, J. Chromatogr. A. 1125 (2006) 76–88.

69. K. Contrepois, L. Jiang, M. Snyder, Optimized Analytical Procedures for the Untargeted Metabolomic Profiling of Human Urine and Plasma by Combining Hydrophilic Interaction (HILIC) and Reverse-Phase Liquid Chromatography (RPLC)–Mass Spectrometry, Mol. Cell. Proteomics. 14 (2015) 1684–1695.

70. V. V Tolstikov, O. Fiehn, Analysis of Highly Polar Compounds of Plant Origin: Combination of Hydrophilic Interaction Chromatography and Electrospray Ion Trap Mass Spectrometry, Anal. Biochem. 301 (2002) 298–307.

71. S. Wernisch, S. Pennathur, Evaluation of coverage, retention patterns, and selectivity of seven liquid chromatographic methods for metabolomics, Anal. Bioanal. Chem. 408 (2016) 6079–6091.

72. R. Zhang, D.G. Watson, L. Wang, G.D. Westrop, G.H. Coombs, T. Zhang, Evaluation of mobile phase characteristics on three zwitterionic columns in hydrophilic interaction liquid chromatography mode for liquid chromatography-high resolution mass spectrometry based untargeted metabolite profiling of Leishmania parasites, J. Chromatogr. A. 1362 (2014) 168–179.

73. D.P. Kloos, H. Lingeman, W.M.A. Niessen, A.M. Deelder, M. Giera, O.A. Mayboroda, Evaluation of different column chemistries for fast urinary metabolic profiling, J. Chromatogr. B. 927 (2013) 90–96.

74. T. Zhang, D.J. Creek, M.P. Barrett, G. Blackburn, D.G. Watson, Evaluation of Coupling Reversed Phase, Aqueous Normal Phase, and Hydrophilic Interaction Liquid Chromatography with Orbitrap Mass Spectrometry for Metabolomic Studies of Human Urine, Anal. Chem. 84 (2012) 1994–2001.

75. R. Dams, M.A. Huestis, W.E. Lambert, C.M. Murphy, Matrix effect in bio-analysis of illicit drugs with LC-MS/MS: Influence of ionization type, sample preparation, and biofluid, J. Am. Soc. Mass Spectrom. 14 (2003) 1290–1294.

76. O.A. Ismaiel, T. Zhang, R.G. Jenkins, H.T. Karnes, Investigation of endogenous blood plasma phospholipids, cholesterol and glycerides that contribute to matrix effects in bioanalysis by liquid chromatography/mass spectrometry, J. Chromatogr. B Anal. Technol. Biomed. Life Sci. 878 (2010) 3303–3316.

77. P.A. Vorkas, G. Isaac, M.A. Anwar, A.H. Davies, E.J. Want, J.K. Nicholson, E. Holmes, Untargeted UPLC-MS profiling pipeline to expand tissue metabolome coverage: Application to cardiovascular disease, Anal. Chem. 87 (2015) 4184–4193.

40

78. M. Eliasson, S. Rännar, R. Madsen, M.A. Donten, E. Marsden-Edwards, T. Moritz, J.P. Shockcor, E. Johansson, J. Trygg, Strategy for Optimizing LC-MS Data Processing in Metabolomics: A Design of Experiments Approach, Anal. Chem. 84 (2012) 6869–6876.

Acta Universitatis UpsaliensisDigital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Pharmacy 232

Editor: The Dean of the Faculty of Pharmacy

A doctoral dissertation from the Faculty of Pharmacy, UppsalaUniversity, is usually a summary of a number of papers. A fewcopies of the complete dissertation are kept at major Swedishresearch libraries, while the summary alone is distributedinternationally through the series Digital ComprehensiveSummaries of Uppsala Dissertations from the Faculty ofPharmacy. (Prior to January, 2005, the series was publishedunder the title “Comprehensive Summaries of UppsalaDissertations from the Faculty of Pharmacy”.)

Distribution: publications.uu.seurn:nbn:se:uu:diva-318296

ACTAUNIVERSITATIS

UPSALIENSISUPPSALA

2017