A Comparative Study of Penta
fluorophenyl and Octadecylsilane Columns in
High-throughput Pro
filing of Biological Fluids
Yoong-Soon Yong ,
aEric Tzyy Jiann Chong,
bHsin-Chang Chen,
cPing-Chin Lee
band
Yee Soon Ling
a,d*
a
Biotechnology Research Institute, Universiti Malaysia Sabah, Kota Kinabalu, Sabah, 88400, Malaysia
b
Faculty of Science & Natural Resources, Universiti Malaysia Sabah, Kota Kinabalu, Sabah, 88400, Malaysia
c
Institute of Occupational Medicine and Industrial Hygiene, College of Public Health, National Taiwan University, Taipei 100, Taiwan
d
Water Research Unit, Universiti Malaysia Sabah, Kota Kinabalu, Sabah, 88400, Malaysia (Received: December 20, 2016; Accepted: March 25, 2017; DOI: 10.1002/jccs.201600873)
In high-throughput metabolomic profiling, chromatographic separation is crucial because a well-performed chromatographic separation may reduce signal suppression from complex biological matrices and improve the discoverability of low-abundance metabolites. We compared the perfor-mance of pentafluorophenyl (PFP)- and octadecylsilane (ODS)-based columns in profiling biologi-cal fluids. Peak resolutions and consistencies were acquired using several reversed-phase columns and were evaluated. Total and extracted ion chromatograms demonstrated that the PFP column achieved better analyte separations than the ODS column. Low relative standard deviations on peak areas and retention times (<10.2 and <0.9%, respectively) acquired using the PFP column evidenced the high reproducibility and consistency of this column. In our study, a PFP column was used for profiling metabolomes extracted from urine and serum samples. Metabolomic study revealed a metabolome difference in normal and overweight participants. In total, 26 lipid species were significantly perturbed and further identified. Choline-containing lipids were the most abun-dant perturbed lipidome in overweight participants, followed by sphingolipids and various phos-pholipids. We recommend the use of PFP columns in high-throughput metabolomic analysis to promote the development of basic biological and clinical research in the future.
Keywords: Biologicalfluids; Metabolomics; Octadecylsilane; Pentafluorophenyl; Partial least squares-discriminant analysis.
INTRODUCTION
Metabolomics is the study of endogenous low-weight metabolites (typically <1500 Da) for clarifying the functional information in biological system states.1,2 It is widely applied in the study of various diseases,3,4 biomarker discovery,1 drug efficacy and toxicity screening,5 nutrition6 and environmental exposure.7,8 Over the decades, researchers have attempted to explore specific low-weight molecules that can serve as diagnos-tic markers for monitoring and predicting disease pro-gression and therapeutic inspection.9,10 Up to 40% of the genes in the human genome are hypothetical and their functions remain unknown,11 of which some may be identified as orphan genes. Therefore, functional omics, including transcriptomics, proteomics, and
metabolomics, are gaining much attention in academia. Metabolomics is advantageous because it serves as a direct signature of the investigated biochemical activity, enabling the simple correlation of phenotypes and activ-ities. Through biofluid (blood and urine) metabolite profiling, metabolomics summarizes the physiological status of an individual and therefore has high potential in personalized medicine12 because different individuals metabolize compounds at different rates.
Nuclear magnetic resonance (NMR) and mass spectrometry (MS) are the most commonly used meth-odologies in metabolomics. Several studies have high-lighted the benefits of combining multiple platforms while analyzing biofluids and extracts. However, because of the limitations due to cost and instruments,
laboratories tend to use a single analytical platform. NMR-based metabolomics has the advantage of simple sample preparation methods and is nondestructive to the analytes. However, the low resolution of NMR lim-its lim-its applicability to nonpolar compounds, such as long-chain lipid detection and structural profiling and determination.1 Gas chromatography (GC) is widely applied in metabolomics because of its low equipment and maintenance costs and accessible databases. How-ever, GC has major drawbacks; in particular, the extracted metabolites have high tendency to undergo significant alterations with time at high temperatures,13 thus possibly yielding erroneous results.
Utilization of liquid chromatography in metabol-omics has grown rapidly over the last decade.14Among the liquid chromatography methods, hydrophilic inter-action liquid chromatography (HILIC) is used for ana-lyzing anionic, cationic, uncharged, and zwitterionic compounds.15 HILIC has been used extensively and optimized for analyzing targeted metabolites. However, its peak retention time drift renders untargeted metabo-lite profiling difficult.16 Moreover, it requires a longer equilibration time compared to reverse-phase liquid chromatography (RPLC).17 Therefore, untargeted pro-filing studies are performed mainly using a reverse-phase octadecylsilane (ODS) column because it gener-ates the reproducible data required for high-throughput studies.18 However, the major drawback of the ODS column, particularly the C18 column, is that polar com-pounds from the extracts tend to be eluted at or close to the void volume,19whereas pentafluorophenyl (PFP) column is able to separate polar isomeric compounds.20 In high-throughput metabolomics, the successful acquisition of metabolites within the sample depends on the quality of the peak shape of the analytes. Distorted peaks often jeopardize correct peak integration and reduce analyte sensitivities and resolutions,21 which may lead to inaccurate results. Several researchers have suggested multiple factors, such as mobile phase, ana-lyte concentration, and appropriate column selection, that affect the peak quality.21We compared the perfor-mance of three RPLC columns, namely the commonly used two ODS (including C18and T3) columns and one (PFP) column, in the chromatographic separation of polar and nonpolar metabolites extracted from the bio-logicalfluids of normal and overweight participants at positive and negative ionization modes.
RESULTS AND DISCUSSION
In high-throughput analysis, the analysis speed is critical because faster analyses may reduce the analysis cost and increase the throughput. Shortening the ana-lytical column reduces the anaana-lytical run but increases the risk of co-elution, which affects the resolution of the analyte peaks.22 Reducing the column particle size (to as low as 1.7–1.9 μm) improves the separation affin-ity because it increases the absorption surface while shortening the column length. This approach is chal-lenging because a column with smaller particles can eas-ily accumulate LC backpressure throughout; therefore, an LC system that strongly resists high backpressures is essential. Later, the development of core–shell technol-ogy strongly reduced the backpressure of the column while maintaining a high separation efficiency.23
C18 and T3 columns are commonly used for uni-versal metabolomic profiling of both polar and nonpo-lar compounds extracted from urine24and blood.25The efficacy of separating nonpolar and moderately polar analytes depends on the density of the C18alkyl chains, the accessibility of silanol, and the bonding groups between the silica beads and C18 alkyl groups within the ODS column.26 An organofluorinated-silica-based stationary phase was introduced for analyzing polar analytes over the last decade.27The PFP ring attached to silica in a propyl chain exhibits both reversed- and normal-phase retention and offers different selectivity for polar and nonpolar compounds compared to tradi-tional alkyl and HILIC phases.27,28 Moreover, Wong et al. reported the high efficiency of PFP in the chro-matographic separation and quantification of various chemical homologs.29However, the PFP phase has lim-ited applications in both qualitative and quantitative metabolomics.30 We compared the performance of commonly used ODS and PFP phase columns in polar and nonpolar high-throughput metabolite profiling. Chromatographic performance
Figures 1 and 2 present the total ion chromato-grams (TICs) obtained using different RP columns of the urine and serum extracts, respectively. The number of acquired metabolites from the urine samples profiled using F5 (Figure 1(a)) is comparable to that obtained using T3 (Figure 1(b)) and C18 (Figure 1(c)). The total compound chromatogram and spectra generated from the acquired TIC evidenced the higher efficiency of F5
in separating co-eluted metabolites (Figure 2); the TIC acquired from serum-extracted polar metabolites was similar (Figure 3). We deduced that the PFP phase exhibited both reverse- and normal-phase retention and offered higher selectivity for polar compounds
compared with ODS.28 In addition, dipole–dipole, hydrogen-bonding, π–π,27,31 and ion-exchange interac-tions29 in the PFP offer larger capacity factors for aro-matic and polycyclic aroaro-matic hydrocarbons because of the formation of a donor–acceptor complex. Moreover, the fluorinated phase inherent in the compound shape and size selectivity32enhances the separation of metab-olite matrices. Organofluorines of the PFP stationary phase show high electronegativity, low polarizability, and strong lipophobicity and hydrophobicity,29 offer-ring binding possibilities of various nonpolar com-pounds. These unique characteristics of the PFP column enable the good chromatographic separation of polar metabolite extracts. Figures 3(c) and (d) present the TIC acquired using F5 in the positive and negative ESI modes, respectively. The number of nonpolar Fig. 1. Total ion chromatogram acquired from urine
samples by using F5, T3, and C18 columns.
Asterisks (*) show the peaks that were further extracted from the total compound chromat-ogram (Figure 2).
Fig. 2. Spectra extracted from the selected peak (*) in Figure 1. Peaks were acquired using F5, T3, and C18 columns. Well-resolved peaks
can be observed from the peak acquired using F5; the spectra demonstrate less sup-pression from the matrices. Co-elution of analytes occurred in T3, and a slight separa-tion of the analytes was realized using the C18column.
Fig. 3. Total ion chromatogram profiled from polar (a and b) and nonpolar (c and d) extracts from serum samples by using F5, T3, and C18 columns at the positive (a and c) and
negative (b and d) ionization modes. a1, C16 sphinganine; a2, hydroxystearic acid; a3, luci-denic acid; a4, docosanedioic acid; a5, L-methionine; a6, unknown; b1, myoinositol; b2, linoleic acid; b3, N-palmitoyl proline; c1, monoglycerol (12:0); c2, sphingosine (t18:0); c3, ceramide (t28:0); c4, MG(P-20:0); c5, lysophosphotidylcholine (LPC) (18:2); c6, LPC (16:0); d1, fatty acid D1; d2, fatty acid D2; d3 lysophosphatidic acid (20:4).
metabolites acquired using the PFP column was compa-rable to that acquired using ODS C18 (Figures 3(c iii) and (d iii)).
Peak quality evaluation
We extracted the ion chromatogram of selected metabolites from the TIC. Table 1 lists some of the con-temporaneous m/z profiled from the serum sample by using the F5, T3, and C18columns. Choi and Row pro-posed that a good and well-resolved peak should have a near-zero asymmetry, narrow peak width, and positive kurtosis (leptokurtic features)33 (Table S1, Supporting information, peak shape category). We observed that metabolite peaks acquired using F5 generally exhibited satisfactory peak quality. However, several peaks (mostly from nonpolar extracts profiled in the negative ionization mode) presented with negative kurtosis. When zoomed in, these peaks were marginally flat-topped34 (platykurtic) without any peak tailing or broadening. These outcomes are attributable to multi-ple factors such as excessive sammulti-ple load, detector over-range, on-column analyte reaction, and mobile phases.35–37 The relative standard deviation (RSD, %) of the extracted ion chromatograms with a peak area of <10.2% and retention time acquired using F5 was rela-tively low compared to other columns (Table 2). This proved that F5 was highly stable and consistent in the intraday (n = 5) and interday (n = 15) analyses. Chro-matographic performance of the ODS column was dependent on the interaction of the analytes with the alkyl chains or the free silanol.22 Most peaks acquired using C18 or T3 columns were unsatisfactory, as they exhibited peak tailing or fronting because of their peak asymmetry features. In addition, their peak area RSD ranged from 0.4% to 25% (Table 2), suggesting incon-sistencies during the batchwise analysis of samples. Such results are not tolerable because high-throughput studies involve hundreds or thousands of samples, and inconsistencies can jeopardize the eventual outcomes and mislead the overall study.
Application of the F5 column for analyzing the metabolome perturbation in overweight participants
Obesity is defined as the medical condition in which excessive body fat accumulates to such an extent that it may negatively affect health.38 Growing inci-dence of obesity is a major public health concern
worldwide.39Obesity is one of the most crucial risk fac-tors contributing to the overall burden of diseases including heart diseases, type 2 diabetes mellitus, obstructive sleep apnea, cancer, and osteoarthritis.38 Using the body mass index (BMI) formula
weight kgð Þ height mð Þ × height mð Þ
½ 0, people are clinically classified as normal (BMI, 18.5–24.9), overweighed (25–29.5), and obese (≥30). Our study participants were healthy volun-teers (both sexes; age between 25 and 40) not under any medication. The participants were enrolled into two groups: normal (n = 6, BMI = 18.5–24.9) and over-weight (n = 6, BMI≥ 25). Blood and urine samples were collected from all participants after 4 h of fasting. The samples underwent preparation and extraction according to the methods described in this paper.
The biofluids of overweight and normal partici-pants were injected alternately during the analysis. One blank sample (8μL of methanol) was injected after every sixth sample during sequencing; no signi fi-cant carryover of polar or nonpolar extracts was observed. All acquired raw data were preprocessed using MZmine 2 and subjected to multivariate ana-lyses. Partial least squares-discriminant analysis (PLS-DA) was performed by using 12 samples from the overweight and normal groups (six biological replicates each). From the visualization data, the PLS-DA score plots enabled us to discriminate the properties of each sample: samples (symbols) that were close together shared similar properties, whereas those far apart were dissimilar.40
Figure 4 shows the results of the PLS-DA analysis of nonpolar serum extracts at both positive and nega-tive ionization modes. In total, 12 symbols were assigned into the normal and overweight groups along component 1 with the following established compo-nents: R2Y = 0.9869, Q2= 0.2627 and R2Y = 0.9913, Q2= 0.6436 for nonpolar extracts analyzed using posi-tive (Figure 4(a)) and negaposi-tive (Figure 4(b)) ionization modes, respectively. From these figures, the clustering of the normal participants was localized to the negative region of component 1, whereas that of the overweight participants was localized in the positive region of com-ponent 1. To inspect the contribution variables (nonpo-lar metabolites) and to distinguish between the groups in our PLS-DA model, variable importance projection (VIP) was conducted with a threshold of 1.5.
Subsequently, the VIP scores were cross-validated with statistical analysis by using the t-test; p < 0.05 was con-sidered significant. Thus, the nonpolar metabolites that led to clustering of both groups at different ionization modes were successfully identified (Table 3). A
multivariate analysis was conducted using the acquired dataset of urine samples. Using the established model, a negative Q2 (data not shown) was presented, which indicated that the model lacked a predictive value. Hence, the model was discarded.
Table 1. Peak quality evaluation on metabolites extracted from extracted ion chromatogram profiled using different RP columns
Fractio P ola r No n -P o la r on Ionization meM P o si ti ve Dihydro Tetradec C16 Sph Isovaler Unknow Docosan Unknow Chlorog Luciden Unknow Unknow Unknow Ne ga ti ve Tetradec Unknow Unknow N-palmi N-oleoy Po si ti v e Sph (14 Sph (16 Sph (d1 MG (12 Sph (t18 MG (18 Unknow LysoSM Cer (d26 MG (26 Unknow LPC (18 SM (d26 PC (36:3 Ne ga ti ve FA Cha Unknow Unknow LPA (10 LPA (12 LPA (16 LPA (20 LPA (20 PA (38: PI (26:1 PS (36:0 Matched tabolites[a] Width F5 T othymine canedioic acid hinganine rylglucuronide wn (318.3017) nedioic acid wn (429.3680) genin nic acid wn (485.4317) wn (679.5103) wn (737.5632) cyl isobutyrate wn (311.1706) wn (325.1853) itoyl proline yl glutamine :0) :0) 6:1) 2:0) 8:0) 8:0) wn (371.3159) M (t14:0) 6:0) 6:5) wn (485.4315) 8:2) 6:0) 3) ain length 10-16 wn (265.1479) wn (283.2646) 0:0) 2:0) 6:0) 0:4) 0:3) 9) ) 0) T3 C18 F5 T3 Asymmetry Kurtosis C18 F5 T3 C18 a
Each metabolite was matched with the available databases.
Each extracted peak was evaluated on the basis of the acquired peak quality, measured in terms of peak width, peak asymmetry, and kurtosis. The peak quality is color scaled:
Table 2. Intra-batch and inter-batch relative standard deviations of the metabolite peak area and retention time Fraction Polarity Metabolites RSD (%) area RSD (%) retention time Intra-batch (n = 5 ) Inter-batch (n = 15) Intra-batch (n = 5 ) Inter-batch (n = 15) F5 T3 C18 F5 T3 C18 F5 T3 C18 F5 T3 C18 Polar Positive Dihydrothymine 0.28 3.95 8.82 4.59 5.83 8.68 0.14 0.38 0.33 0.15 0.38 0.36 Tetradecanedioic acid 0.74 2.86 16.15 6.38 4.84 13.66 0.14 0.38 0.33 0.15 0.38 0.36 C16 Sphinganine 0.14 3.46 2.71 4.94 4.29 2.15 0.45 0.10 0.89 0.68 0.34 1.07 Isovalerylglucuronide 0.44 2.57 1.40 1.96 3.25 6.80 0.86 0.09 1.53 0.95 0.16 1.41 Unknown (31 8.3017) 0.50 3.52 4.02 3.96 5.00 13.22 0.43 0.09 0.83 0.64 0.35 1.03 Docosanedioic acid 5.52 13.82 19.91 8.33 10.0 0 17.88 0.14 0.39 0.28 0.15 0.38 0.36 Unknown (42 9.3680) 0.70 6.58 17.61 7.67 11.4 20.28 0.15 0.39 0.31 0.15 0.36 0.36 Chlorogenin 1.23 13.61 24.89 7.53 21.5 3 25.24 0.15 0.44 0.35 0.15 0.55 0.37 Lucidenic acid 0.56 10.28 12.97 4.02 17.3 5 18.22 0.12 0.23 0.19 0.15 0.26 0.33 Unknown (48 5.4317) 0.34 12.44 23.38 8.98 15.3 7 20.44 0.15 1.06 2.15 0.17 0.70 0.62 Unknown (67 9.5103) 0.23 2.55 1.47 8.04 6.38 4.79 0.91 0.04 2.54 0.73 0.09 1.80 Unknown (73 7.5632) 0.45 2.62 0.54 4.51 8.10 10.86 0.91 0.05 2.53 0.73 0.09 1.82 Negative Tetradecyl isobutyrate 4.00 5.08 10.21 8.50 13.3 0 7.06 0.40 0.26 0.06 0.27 0.26 0.19 Unknown (31 1.1706) 7.84 3.49 8.92 11.7 3 9.34 10.27 0.15 0.11 0.11 0.30 0.14 0.22 Unknown (32 5.1853) 3.00 7.22 4.31 3.58 7.39 5.08 0.23 0.10 0.10 0.19 0.16 0.21 N -Palmitoyl proline 0.84 7.25 8.04 5.84 6.07 6.52 0.35 0.33 0.14 0.23 0.28 0.19 N -Oleoyl glutamine 2.73 6.30 24.05 5.46 4.18 20.26 0.39 0.27 0.07 0.28 0.27 0.20 Nonpolar Positive Sph (14:0) 2.38 7.18 4.09 4.26 5.93 2.97 0.57 0.11 0.40 0.60 0.27 0.79 Sph (16:0) 4.03 6.06 6.58 7.10 12.1 0 11.90 0.18 0.45 0.27 0.16 0.38 0.25 Sph (d16:1) 1.64 7.53 11.15 3.31 5.92 6.10 0.49 0.10 0.24 0.55 0.25 0.74 MG (12:0) 2.89 5.49 6.03 5.39 4.58 3.74 0.59 0.12 0.45 0.62 0.26 0.82 Sph (t18:0) 4.90 6.64 9.55 6.11 9.64 6.51 0.56 0.16 0.40 0.59 0.24 0.80 MG (18:0) 4.62 12.39 11.92 7.87 10.5 3 14.93 0.19 0.46 0.29 0.17 0.37 0.24 Unknown (37 1.3159) 2.17 3.67 19.65 3.21 12.4 4 16.45 0.18 0.45 0.27 0.16 0.37 0.25 LysoSM (t14:0) 1.98 6.90 20.43 2.05 16.8 8 25.16 0.18 0.62 0.15 0.19 0.57 0.29 Cer (d26:0) 2.98 6.19 10.27 4.94 10.6 3 14.42 0.18 0.45 0.26 0.16 0.37 0.25 MG (26:5) 2.12 4.66 10.69 8.77 17.6 5 20.29 0.20 0.11 0.17 0.18 0.21 0.22 Unknown (48 5.4315) 2.48 9.05 24.95 7.47 7.14 25.30 0.18 0.58 1.45 0.19 0.56 0.32 LPC (18:2) 0.29 3.22 7.58 5.77 5.11 7.05 0.17 0.41 0.28 0.16 0.38 0.24 SM (d26:0) 2.26 10.18 9.27 4.25 10.9 1 6.15 0.20 0.05 0.17 0.19 0.18 0.25 PC (36:3) 2.57 1.72 8.12 4.43 5.80 8.19 0.19 0.46 0.29 0.17 0.38 0.24 Negative FA chain length 10-16 3.13 0.48 17.55 4.62 14.8 2 18.17 0.15 0.22 0.23 0.33 0.22 0.40 Unknown (26 5.1479) 4.77 1.61 2.84 6.57 7.12 3.11 0.61 0.09 0.43 0.90 0.09 0.45 Unknown (28 3.2646) 5.16 8.00 2.43 5.32 16.5 8 7.32 0.19 0.23 0.22 0.36 0.22 0.42 LPA (10:0) 0.48 3.99 3.89 3.11 6.00 11.08 0.19 0.32 0.39 0.39 0.24 0.37
Overweightness dysregulates lipid metabolism
Twenty-six significantly perturbed lipid species were observed in the overweight participants (Table 3). These lipids included choline-containing lipids (phos-phatidylcholine [PC] and sphingomyelin [SM]), sphingo-lipids (sphingosine and ceramides), glycerol (diacylglycerol and triacylglycerol), and polar phospho-lipids (phosphatidylglycerol [PG] and phosphatidylser-ine [PS]). Figure 5 shows the deduced interrelationship among the various classes of lipids in the overweight participants. Studies have highlighted that phosphorylcholine-containing lipids play a critical role in regulating cellular membranefluidity and antioxidant functioning and facilitating membrane fusion, energy storage, and release of polyunsaturated and saturated fatty acids.41Studies have also reported a strong correla-tion between the phosphorylcholine-containing lipids in obese patients42 via ceramides. Ceramide is formed fol-lowing the catalysis of sphingomyelin by sphingomyeli-nase40or a de novo synthesis following the condensation of serine and palmitoyl CoA, in which multiple enzymes are involved.43 Several proinflammatory cytokines are secreted and upregulated in the adipose tissues of obese individuals. Such upregulation potentially dysregulates sphingolipids (including sphingosine and ceramide), fur-ther triggering subsequent inflammatory reactions.43 In addition, lysophospholipid, a single-carbon-chain lipid molecule with a polar head group and a bioactive lipid molecule,44was perturbed in our study. These bioactive lipid molecules had been previously reported to inherit unknown functions in the development of various cancers45–47 and stroke.48 Thus, the high susceptibility of obese patients to cancer and stroke may be due to dysregulated lysophospholipids. Moreover, perturbation of triacylglycerides, which serve as key clinical diagnos-tic criteria for obesity,49 correlated positively with the participants’ blood biochemistry tests in this study. CONCLUSIONS
We found the performance of the PFP column superior to that of the ODS columns. Metabolites in the urine sample profiled using the PFP column exhib-ited high resolution (peak separations). The peak area acquired using the PFP column during intra-batch and inter-batch analysis (serum) evidenced that the RSD of the PFP column was relatively low compared to that of the ODS columns. This result indicates that the PFP
Table 2. Continued Fraction Polarity Metabolites RSD (%) area RSD (%) retention time Intra-batch (n = 5 ) Inter-batch (n = 15) Intra-batch (n = 5 ) Inter-batch (n = 15) F5 T3 C18 F5 T3 C18 F5 T3 C18 F5 T3 C18 LPA (12:0) 3.81 2.39 12.68 2.94 3.40 8.66 0.15 0.26 0.28 0.31 0.22 0.38 LPA (16:0) 4.35 3.72 7.57 6.51 5.54 7.51 0.22 0.27 0.22 0.38 0.24 0.44 LPA (20:4) 1.62 3.60 4.19 1.83 17.41 10.12 0.18 0.22 0.22 0.34 0.21 0.42 LPA (20:3) 4.16 3.87 5.87 9.44 15.64 4.19 0.46 0.11 0.52 0.68 0.17 0.55 PA (38 :9) 10.24 6.54 5.45 12.42 17.82 11.44 0.51 0.07 0.87 0.50 0.10 0.88 PI (26:1) 4.44 3.39 3.64 8.49 15.17 8.26 0.63 0.08 0.87 0.62 0.11 0.72 PS (36:0) 8.16 5.82 7.95 11.19 19.73 10.95 0.18 0.07 0.88 0.59 0.10 0.73
Table 3. Discriminating variables (metabolites) highlighted by the PLS-DA following variable importance projection (VIP) and cross-validation using the t-test
Ionization mode Detected m/z Subclass Matched metabolites VIP1 t-Test2(p-value) Positive 660.4963 Choline-containing lipid LPC (28:2) ••• *
756.5538 PC (34:3) •• ** 780.5538 PC (36:5) •• ** 838.6321 PC (38:5) •• * 806.5695 PC (38:6) ••• * 808.5851 PC (38:6) •• * 842.6634 PC (40:4) • * 866.6385 PC (42:4) ••• * 856.5851 PC (42:9) •• ** 892.6790 PC (44:5) ••• ** 992.7103 PC (52:11) •• * 1080.8355 PC (58:9) •• * 589.4340 SM (d26:2) •• ** 675.5436 SM (d32:1) ••• * 731.6062 SM (d36:1) •• ** 777.6480 SM (t38:0) •• ** 246.2428 Sphingolipids Sph (d14:0) •• * 298.2741 Sph (d18:2) ••• ** 432.2796 HexSph (d16:2) ••• * 658.4889 HexCer (t30:2) •• *
527.5034 Glycerol DAG (O-30:0) ••• *
651.5922 TAG (P-38:0) ••• *** Negative 745.5489 Phosphatidylglycerol PG (34:2) •• * 775.5494 PG (36:1) • * 826.4664 Phosphatidylserine PS (40:10) ••• * 854.4977 PS (42:10) •• * 1
VIP scores for the constructed PLS-DA model.
2*p < 0.05; **p < 0.01; ***p < 0.001 determined using the t-test.
Fig. 4. Partial least squares-discriminant analysis score plots for nonpolar extracts analyzed at (a) positive and (b) negative ion modes in samples extracted from overweight (Δ) and normal (+) participants. Eclipses are the mean standard deviation of the sample.
Fig. 5. Dysregulation and metabolic pathway of sphingomyelin (SM), ceramide (Cer), phos-photidylcholine/lysophosphotidylcholine (PC/ LPC), phosphatidylserine (PS), phosphatidyl-glycerol (PG), diacylphosphatidyl-glycerol (DAG), and triacylglycerol (TAG) in overweight participants.
column has higher reproducibility and enhances the dis-coverability of low-abundance metabolites. Through statistical analyses, we identified the differences between overweight and normal participants; 26 potential lipids with significant (p < 0.05) changes were identified in samples extracted from overweight participants, and the results concurred with those of previous studies. Therefore, we recommend the use of a PFP column in high-throughput metabolomics to promote the develop-ment of basic biological and clinical research in the future.
EXPERIMENTAL Chemicals and reagents
Betaine and hexakis(1H,1H,3H-per fluoropro-poxy)-phosphazene (HP-0921) (which served as the internal standard) were purchased from Apollo (Graham, NC, USA). American Chemical Society (ACS)-grade or LC-MS-grade methanol, acetonitrile, and chloroform were obtained from J. T. Baker (Philipsburg, NJ, USA). Sodium chloride (NaCl) was purchased from Merck (Darmstadt, Germany). The water used in the LC system was purified using a Milli-Q system (Millipore, Milford, MA, USA) at a resistiv-ity of >18.2 MΩcm. LC-MS grade ammonium acetate (NH4OAc) and formic acid (HCOOH) were acquired from Sigma–Aldrich (St. Louis, MO, USA).
Biological material and sample preparation
This study was approved by the Ethics Review Board of University Sabah Malaysia (REF: JKEtika 1/15 (7)). Blood and urine samples were obtained from deidentified donors after informed consent was obtained. Urine samples were collected and treated as described by Contrpois et al. but with a marginal modi-fication.15
Briefly, 200 μL of the urine sample was immediately desalted using 800μL of methanol and centrifuged at 15000 g for 10 min at 4C. The superna-tant was transferred and evaporated using a Speedvac concentrator system. The dried sample was stored at −80C. Before analysis, the dried urine sample was reconstituted using 400μL of methanol admixed with the internal standards.
Serum was prepared using whole blood coagulated at a low temperature (4C). The coagulated blood was centrifuged at 2500 g for 15 min at 4C. The superna-tant was collected for extraction. The extraction
protocol for serum was a modified version of the Bligh and Dyer extraction protocol.1,50 Briefly, the superna-tant was relocated into glass tubes and treated with a chloroform/methanol/NaCl (0.9%) in water (1:1:1 v/v/v) solvent mixture for polarity-based metabolite extrac-tion. The upper and lower layer of the mixtures were transferred to separate tubes and evaporated to dryness. The dry extracts were reconstituted with methanol, admixed with the internal standards, and stored at −80C until further analysis. These internal standards served to correct variations in the signal response dur-ing analysis and aided signal normalization to adjust the variations between each sample run.
LC-QTOF acquisition
The extracted urine and serum samples were ana-lyzed using an Agilent 1260 infinity HPLC system coupled with an Agilent 6520 Q-TOF MS system. A 1.5-μL sample was injected into the following three ana-lytical columns separately: Kinetex C18, Kinetex F5 (2.1 mm× 100 mm × 2.6 μm; Phenomenex, Torrance, CA, USA), and Acquity HSS T3 (2.1 mm× 100 mm × 3.0μm; Waters, Milford, MA, USA). All analytical columns used in this study were RP columns. There-fore, the same gradient program was applied for all col-umns. The columns were maintained at 35C at aflow rate of 350μL/min during analysis. The mobile phases were composed of solvent A (H2O–0.1% HCOOH–1% 10 mM NH4OAc) and solvent B (acetonitrile/methanol [6:4 v/v] – 1% of 0.1% HCOOH–1% 10 mM NH4OAc). The gradient elution program was initiated from 1% to 70% solvent B in 7 min, followed by 100% solvent B from 7.1 to 10 min and maintained for 3 min. Later, the columns were conditioned with the initial gradient for 3 min before the next sample was injected.
The data acquisition was set between an m/z of 100 and 1500. Positive and negative heated electrospray ionization (ESI) were deployed at 3500 and −3500 V, respectively. The ion source conditions were set as fol-lows: gas temperature of 325C, drying gasflow at 4 L/ min, and nebulizerflow at 45 psig. The mass spectrome-ter was calibrated with Tune Mix (Sigma–Aldrich, St Louis, MO, USA) before each batch analysis. Internal mass calibration standards betaine and hexakis (1H,1H,3H-tetrafluoropropoxy) phosphazine were introduced during the runs. In the positive ion mode, the internal mass calibration standards were m/z of
121.0509 and 922.0098, respectively, whereas in the negative ion mode the corresponding values were m/z of 112.9856 and 1033.9881 (TFA adducted HP-0921). Data analysis
Köfeler et al. considered every spectrum in a TIC to be a single event with unique matrix effects and sol-vent composition; this caused difficulties in standardiz-ing metabolite quantitation.51 The quality of the metabolite peaks acquired from the three RP columns was evaluated using the criteria proposed by Contrpois et al. with modifications. Briefly, metabolic features (characterized by a unique m/z and retention time) were extracted using the MassHunter Qualitative Analysis Software B 05.00 (Agilent Technologies, Santa Clara, CA, USA), followed by evaluation of the peaks that corresponded to peak quality, including peak width, asymmetries, kurtosis, and overall spectral signal inten-sities. Intraday precision and accuracy were measured for the selected metabolites by analyzing five extracted samples on the same day (n = 5), whereas the interday values were determined in triplicate per sample on three consecutive days (n = 15).
Metabolomics was conducted for comparing the metabolites extracted from participants with normal BMI and overweight participants. The metabolic fea-tures of the participants were extracted using MZmine 2 to compensate for retention times drift during each analysis run.52These preprocessed chromatograms were exported as a peak list table in the comma separated values format. The acquired datasets were further ana-lyzed through multivariate methods by using MetaboA-nalyst 3.0.53 Principal component analysis (PCA), an unsupervised–supervised pattern recognition technique, was carried out on all log-transformed and Pareto-scaled datasets to obtain an overview of the data distri-bution and to identify potential outliers.40 Through PCA, the direction of the variances in the dataset could be identified without relating to the class labels (i.e., it summarized the original variables to fewer variables on the basis of weighted averages).40 PLS-DA, a super-vised pattern recognition technique, was performed for inspecting the internal relationships between the matrix X variables and response matrix Y. The quality of the PLS-DA model was evaluated using residuals (R2), and the model predictability parameter (Q2) was determined using MetaboAnalyst 3.0. Throughout the analysis,
VIP was conducted for estimating the weightage of each variable in the projection used in the selected PLS-DA model, and the score threshold was set to 1.5; the vari-ables with values higher than this threshold were selected for additional analysis. The t-test was per-formed using Statistical Package for Social Sciences (SPSS; IBM corporation, Armonk, NY, USA). The putative features of the identified m/z were further frag-mented and matched against METLIN with a mass accuracy of5 ppm.
Each class of polar lipids exhibits a unique frag-mentation pattern. Throughout molecular fragmenta-tion, we could identify the lipids’ classes. During metabolite matching, molecular fragmentation was con-ducted on significantly (p < 0.05) perturbed metabolites (polar lipids), and the fragmentation patterns were matched with the Lipid Metabolites and Pathway Strat-egy (LIPIDMAPS) at a mass accuracy of5 ppm. ACKNOWLEDGMENTS
This work was supported by the Ministry of Higher Education, Malaysia (FRGS0428-SG-1/2015 and TRGS0006-SG-2/2014), and Universiti Malaysia Sabah (SBK0180-SG-2014).
Supporting information
Additional supporting information is available in the online version of this article.
REFERENCES
1. Y. S. Ling, H.-J. Liang, M.-H. Chung, M.-H. Lin, C.-Y. Lin, Mol. Biosyst. 2014, 10, 1918.
2. G. J. Patti, O. Yanes, G. Siuzdak, Nat. Rev. Mol. Cell Biol. 2012, 13, 263.
3. K. Kim, P. Aronov, S. O. Zakharkin, D. Anderson, B. Perroud, I. M. Thompson, R. H. Weiss, Mol. Cell. Proteomics 2009, 8, 558.
4. A. Sreekumar, L. M. Poisson, T. M. Rajendiran, A. P. Khan, Q. Cao, J. Yu, B. Laxman, R. Mehra, R. J. Lonigro, Y. Li, M. K. Nyati, A. Ahsan, S. Kalyana-Sundaram, B. Han, X. Cao, J. Byun, G. S. Omenn, D. Ghosh, S. Pennathur, D. C. Alexander, A. Berger, J. R. Shuster, J. T. Wei, S. Varambally, C. Beecher, A. M. Chinnaiyan, Nature 2009, 457, 910. 5. R. Kaddurah-Daouk, B. S. Kristal, R. M. Weinshilboum,
Annu. Rev. Pharmacol. Toxicol. 2008, 48, 653.
6. M. J. Gibney, M. Walsh, L. Brennan, H. M. Roche, B. German, B. van Ommen, Am. J. Clin. Nutr. 2005, 82, 497.
7. E. Pujos-Guillot, J. Hubert, J. F. Martin, B. Lyan, M. Quintana, S. Claude, B. Chabanas, J. A. Rothwell, C. Bennetau-Pelissero, A. Scalbert, B. Comte, S. Hercberg, C. Morand, P. Galan, C. Manach, J. Proteome Res. 2013, 12, 1645.
8. S. Tulipani, R. Llorach, O. Jáuregui, P. López-Uriarte, M. Garcia-Aloy, M. Bullo, J. Salas-Salvadó, C. Andrés-Lacueva, J. Proteome Res. 2011, 10, 5047.
9. Y. Chen, M. C. Lin, H. Wang, C. Y. Chan, L. Jiang, S. M. Ngai, J. Yu, M. L. He, P. C. Shaw, D. T. Yew, J. J. Sung, H. F. Kung, Proteomics 2007, 7, 3097. 10. A. P. Khan, T. M. Rajendiran, B. Ateeq, I. A. Asangani,
J. N. Athanikar, A. K. Yocum, R. Mehra, J. Siddiqui, G. Palapattu, J. T. Wei, G. Michailidis, A. Sreekumar, A. M. Chinnaiyan, Neoplasia 2013, 15, 491.
11. Y. Iwasaki, T. Sawada, K. Hatayama, A. Ohyagi, Y. Tsukuda, K. Namekawa, R. Ito, K. Saito, H. Nakazawa, Metabolites 2012, 2, 496.
12. R. Chen, G. I. Mias, J. Li-Pook-Than, L. Jiang, H. Y. Lam, R. Chen, E. Miriami, K. J. Karczewski, M. Hariharan, F. E. Dewey, Y. Cheng, M. J. Clark, H. Im, L. Habegger, S. Balasubramanian, M. O’Huallachain, J. T. Dudley, S. Hillenmeyer, R. Haraksingh, D. Sharon, G. Euskirchen, P. Lacroute, K. Bettinger, A. P. Boyle, M. Kasowski, F. Grubert, S. Seki, M. Garcia, M. Whirl-Carrillo, M. Gallardo, M. A. Blasco, P. L. Greenberg, P. Snyder, T. E. Klein, R. B. Altman, A. J. Butte, E. A. Ashley, M. Gerstein, K. C. Nadeau, H. Tang, M. Snyder, Cell 2012, 148, 1293.
13. M. Fang, J. Ivanisevic, H. P. Benton, C. H. Johnson, G. J. Patti, L. T. Hoang, W. Uritboonthai, M. E. Kurczy, G. Siuzdak, Anal. Chem. 2015, 87, 10935. 14. N. L. Kuehnbaum, P. Britz-McKibbin, Chem. Rev. 2013,
113, 2437.
15. K. Contrepois, L. Jiang, M. Snyder, Mol. Cell. Proteom-ics 2015, 14, 1684.
16. K. Spagou, H. Tsoukali, N. Raikos, H. Gika, I. D. Wilson, G. Theodoridis, J. Sep. Sci. 2010, 33, 716. 17. K. Dettmer, P. A. Aronov, B. D. Hammock, Mass
Spec-trom. Rev. 2007, 26, 51.
18. W. B. Dunn, D. Broadhurst, P. Begley, E. Zelena, S. Francis-McIntyre, N. Anderson, M. Brown, J. D. Knowles, A. Halsall, J. N. Haselden, A. W. Nicholls, I. D. Wilson, D. B. Kell, R. Goodacre, Nat. Protoc. 2011, 6, 1060.
19. A. Roux, Y. Xu, F. Heilier, M.-F. Olivier, E. Ezan, J.-C. Tabet, J.-C. Junot, Anal. Chem. 2012, 84, 6429.
20. T. Zhang, D. G. Watson, Analyst 2015, 140, 2907. 21. S. Keunchkarian, M. Reta, L. Romero, C. Castells,
J. Chromatogr. A 2006, 1119, 20.
22. L. Novakova, P. Solich, J. Chromatogr. A 2005, 1088, 24.
23. S. Fekete, E. Oláh, J. Fekete, J. Chromatogr. A 2012, 1228, 57.
24. M.-B. S. Andersen, Å. Rinnan, C. Manach, S. K. Poulsen, E. Pujos-Guillot, T. M. Larsen, A. Astrup, L. O. Dragsted, J. Proteome Res. 2014, 13, 1405.
25. W. Yang, Y. Chen, C. Xi, R. Zhang, Y. Song, Q. Zhan, X. Bi, Z. Abliz, Anal. Chem. 2013, 85, 2606.
26. O. Yanes, R. Tautenhahn, G. J. Patti, G. Siuzdak, Anal. Chem. 2011, 83, 2152.
27. S. Yang, M. Sadilek, M. E. Lidstrom, J. Chromatogr. A
2010, 1217, 7401.
28. D. S. Bell, A. D. Jones, J. Chromatogr. A 2005, 1073, 99. 29. Y. Foo Wong, A. Makahleh, B. Saad, M. N. M. Ibrahim,
A. Abdul Rahim, N. Brosse, Talanta 2014, 130, 299. 30. H. Yoshida, T. Mizukoshi, K. Hirayama, H. Miyano,
J. Agr. Food Chem. 2007, 55, 551.
31. W. Kiridena, C. DeKay, W. W. Koziol, Z. Ali, H. Ahmed, C. F. Poole, Chroma 2006, 63, 407.
32. F. M. Yamamoto, S. Rokushika, J. Chromatogr. A 2000, 898, 141.
33. D. Choi, K. Row, Biotechnol. Bioprocess Eng. 2004, 9, 495.
34. R. D. Caballero, M. C. Garcıa-Alvarez-Coque, J. J. Baeza-Baeza, J. Chromatogr. A 2002, 954, 59. 35. P. Chatrabhuji, C. Pandya, M. Patel, HPLC Method for
Determination of APIs in Pharmaceutical Formulation, Lulu Press, Morrisville, NC, 2015.
36. H. Engelhardt, In Handbook of HPLC, 2nd, Chapter 2, D. Corradini Ed., CRC Press, Boca Raton, FL, 2010, p. 47.
37. E. F. Rossomando, HPLC in Enzymatic Analysis, Wiley, New York, 2009.
38. D. W. Haslam, W. P. T. James, Lancet 2005, 366, 1197. 39. S. J. van Dijk, P. L. Molloy, H. Varinli, J. L. Morrison,
B. S. Muhlhausler, Members of EpiSCOPE, Int. J. Obes.
2015, 39, 85.
40. Y. S. Ling, H.-J. Liang, M.-H. Lin, C.-H. Tang, K.-Y. Wu, M.-L. Kuo, C. K.-Y. Lin, Biomed. Chromatogr.
2014, 28, 1284.
41. K. Gorgas, A. Teigler, D. Komljenovic, W. W. Just, Bio-chim. Biophys. Acta 2006, 1763, 1511.
42. A. Floegel, A. Wientzek, U. Bachlechner, S. Jacobs, D. Drogan, C. Prehn, J. Adamski, J. Krumsiek, M. B. Schulze, T. Pischon, H. Boeing, Int. J. Obes. 2014, 38, 1388.
43. S.-C. Kang, B.-R. Kim, S.-Y. Lee, T.-S. Park, Front. Endocrinol. 2013, 4, 67.
44. D. Heringdorf, In Encyclopedia of Molecular Pharmacol-ogy, Chapter 90, S. Offermanns, W. Rosenthal Eds., Springer, Berlin, 2008, p. 710. https://doi.org/10.1007/978-3-540-38918-7_90.
46. S. A. Kwee, A. A. Franke, L. J. Custer, X. Li, L. L. Wong, Cancer Res. 2015, 75, 1486.
47. H. Y. Kim, M. Kim, H. M. Park, J. Kim, E. J. Kim, C. H. Lee, J. H. Yoon Park, Nutrition 2014, 30, 1433. 48. H.-Y. J. Wang, C. B. Liu, H.-W. Wu, J. S. Kuo, Rapid
Commun. Mass Spectrom. 2010, 24, 2057.
49. T. McLaughlin, F. Abbasi, K. Cheal, J. Chu, C. Lamendola, G. Reaven, Ann. Intern. Med. 2003, 139, 802.
50. S. K. Jensen, Lipid Technol. 2008, 20, 280.
51. H. C. Köfeler, A. Fauland, G. N. Rechberger, M. Trötzmüller, Metabolites 2012, 2, 19.
52. T. Pluskal, S. Castillo, A. Villar-Briones, M. Oresic, BMC Bioinformatics 2010, 11, 395.
