Differences in metabolite profile between blood plasma and serum

Analytical Biochemistry 406 (2010) 105–112

Contents lists available at ScienceDirect

Analytical Biochemistry

journal homepage: www.elsevier .com/locate /yabio

Differences in metabolite profile between blood plasma and serum

Linsheng Liu a,1, Jiye Aa a,*,1, Guangji Wang a,*, Bei Yan a, Ying Zhang a, Xinwen Wang a, Chunyan Zhao a,Bei Cao a, Jian Shi a, Mengjie Li a, Tian Zheng a, Yuanting Zheng a, Gang Hao a, Fang Zhou a, Jianguo Sun a,Zimei Wu b

a Lab of Metabolomics, Key Laboratory of Drug Metabolism and Pharmacokinetics, China Pharmaceutical University, Nanjing 21009, Chinab School of Pharmacy, University of Auckland, Auckland 1142, New Zealand

a r t i c l e i n f o a b s t r a c t

Article history:Received 15 January 2010Received in revised form 24 May 2010Accepted 17 July 2010Available online 23 July 2010

Keywords:PlasmaSerumMetabolomicsGas chromatography/time-of-flight massspectrometryPrincipal components analysis

0003-2697/$ - see front matter � 2010 Elsevier Inc. Adoi:10.1016/j.ab.2010.07.015

* Corresponding author. Fax: +86 25 85306750.E-mail addresses: [email protected] (J. Aa), guang

Wang).1 Co-first author.2 Abbreviations used: LWC, low-molecular-weight

chromatography/time-of-flight mass spectrometry; PCysis; EDTA, ethylenediaminetetraacetic acid; SD, SprInstitute of Standards and Technology; PLS–DA, partilatent structures and discriminant analysis; TCA, tricar

In metabolomic research, blood plasma and serum have been considered to possess similar compositionsand properties. Their perceived equivalence has resulted in researchers choosing arbitrarily betweenserum and plasma for analysis. Here, routine serum and plasma were prepared and their low-molecu-lar-weight compounds were determined using gas chromatography/time-of-flight mass spectrometry.Principal components analysis was applied to process the acquired data, and marked differences inmetabolite profiles were observed between serum and plasma. Of the 72 identified compounds, 36(50%) discriminate serum from plasma, with 29 and 7 metabolites showing a significantly higher abun-dance (t test, P < 0.05) in serum and plasma, respectively. Incubation of blood had distinct effects on theanalyte peak areas, with the effects being more pronounced for plasma than for serum and more pro-nounced for a shorter incubation than for a longer incubation. These results highlight the importancein choosing serum or plasma as the analytical sample and in stipulating the incubation time. Becauseincubation affected the analyte peak areas less in serum than in plasma, we recommend serum as thesample of choice in metabolomic studies.

� 2010 Elsevier Inc. All rights reserved.

Plasma and serum are usually used to evaluate various biochem-ical parameters in medical science, in the study and diagnosis of dis-ease. As the output of the metabolic system, low-molecular-weightcompounds (LWCs)2 contribute to our understanding of metabolismin vivo and extend our knowledge in life sciences. Although blood ser-um and plasma are usually considered to have similar compositionsand properties, many analytes show differing concentrations in plas-ma and serum [1–7]. However, the majority of analytes targeted inprevious studies are clinical parameters such as proteins, enzymes,and metal ions. Only a small number of LWCs (e.g., glucose, lactate,creatinine, urate, urea, cholesterol, triglyceride) have been comparedin plasma and serum [1–3,6]. The differences in metabolite profile be-tween plasma and serum have yet to be thoroughly evaluated.

Because plasma and serum are prepared differently and are thecommonly analyzed samples in metabonomic/metabolomic stud-ies, the differences between them are of great interest. In addition,

ll rights reserved.

[email protected] (G.

compound; GC/TOFMS, gasA, principal components anal-ague–Dawley; NIST, Nationalal least squares projection toboxylic acid cycle.

the duration of the storage or incubation of blood specimens dur-ing the preparation of serum and plasma also varies among studies,and the effects of incubation on metabolite remain unclear. Rela-tive quantification analysis of the metabolites in plasma and serumsamples should provide basic data that can be used to determinewhether plasma or serum is the optimal sample for metabolomicresearch. In this study, we determined the profiles of the metabo-lites in plasma and serum using a well-developed metabolomicplatform based on the gas chromatography/time-of-flight massspectrometry (GC/TOFMS) technique coupled to multivariate sta-tistical analysis (principal components analysis [PCA]) [8–10].Methanol was used as the extraction solvent because the methanolextraction method is considered to be most efficient for the ana-lytes in blood plasma and serum [10,11]. We aimed to clarify thecompositional difference between plasma and serum and toquantitatively assess the variations in metabolite abundance withdifferent incubation times.

Materials and methods

Materials and reagents

All of the reference compounds were of analytical grade andpurchased from Sigma–Aldrich (St. Louis, MO, USA), Sigma–Aldrich

106 Metabolomic differences between blood plasma and serum / L. Liu et al. / Anal. Biochem. 406 (2010) 105–112

(Steinheim, Germany), Merck (Darmstadt, Germany), or Serva(Heidelberg, Germany). The stable isotope used as the internalstandard, [1,2-13C2]myristic acid (analytical grade), was obtainedfrom Sigma–Aldrich.

Collection of human blood samples and preparation of plasma andserum

In total, 15 healthy individuals (7 males and 8 females), 22 to43 years of age (M = 27.4 ± 5.7 years), were recruited. All of the vol-unteers gave their written informed consent to their participationin the study before the experiment began. The study was approvedby the human ethics committee of the China Pharmaceutical Uni-versity in accordance with the Declaration of Helsinki. On theday before the study, all of the volunteers consumed standardlow-fat meals. A sample (2 ml) of venous blood was collected earlyin the morning after an overnight fast. The fresh blood from eachindividual was immediately divided into two parts; one was addedto a blank tube and the other was added to a tube containing eth-ylenediaminetetraacetic acid (EDTA)–2Na. After the samples hadbeen incubated at 37 �C for 2 h, the blood serum and plasma wereisolated by centrifugation at 1600g for 10 min and then stored at�80 �C until analysis. To check the reproducibility of the separa-tion, we collected blood twice from 1 volunteer to produce plasmaand serum replicates.

Collection of rat blood samples and preparation of plasma and serum

To control for the individual human variation caused by con-founding factors, such as food preferences, sex, age, background,mood, and physical activity, blood plasma and serum from ratswere prepared and analyzed. Male Sprague–Dawley (SD) rats(n = 12, body weights = 180–200 g) were purchased from Wei TongLi Hua Animal Center (Beijing, China). Water and standard ratchow from the same source were available ad libitum. All of therats were housed individually in metabolic cages in a standard ani-mal laboratory with a 12-h light–dark cycle for 2 weeks. After anovernight fast, each rat was anesthetized with phenobarbital (i.p.30 mg/kg) and killed, with the immediate collection of 6 ml ofblood. The pooled blood (0.6 ml from each rat) was added to ablank tube and an EDTA–2Na-containing tube and incubated, andthe serum and plasma were prepared as described above. The ani-mal experiments were conducted with the approval of the animalethics committee of China Pharmaceutical University.

Incubation effects on analyte levels

To study the effects of incubation on analyte levels, 30 ml ofblood from a healthy male volunteer was divided into aliquots of0.5 ml, which were immediately added to (i) four blank tubesand incubated at 37 �C for 1, 2, 3, and 4 h or (ii) five EDTA–2Na-containing tubes and incubated at 37 �C for 0, 1, 2, 3, and 4 h. Sixreplicates were included in each group. The sera and plasma wereprepared immediately at each time point and then stored at �80 �Cuntil analysis.

Sample preparation, derivatization, and GC/TOFMS analysis

All frozen plasma and sera were thawed by incubation at 37 �Cfor 20 min and were vortexed before use. The same extraction,derivatization, and analysis procedures were applied as describedpreviously [10] with a few modifications. In brief, 50 ll of plasmaor serum was added to a tube containing 200 ll of methanol to ex-tract the metabolites and precipitate the protein. [1,2-13C2]Myristicacid (1 lg) was added to each tube as the internal standard.

With an Agilent 7683 autosampler, the derivatized sampleswere analyzed using an Agilent 6890N GC system (Agilent Tech-nologies, Atlanta, GA, USA) coupled with a Pegasus III mass spec-trometer (Leco, St. Joseph, MI, USA) in the same way as reportedpreviously [10]. To achieve a better separation of the peaks, thecolumn temperature was initially maintained at 70 �C for 2 minand then was increased at a rate of 35 �C/min from 70 to 305 �C,where it was held for 2 min.

Identification of the compounds

The retention index of each peak was calculated by comparingthe retention time of the peak with those of the alkane series C8

to C40. Each metabolite was identified by comparing its mass spec-trum and retention index with those of authentic reference stan-dards and those available in the National Institute of Standardsand Technology (NIST) library 2.0 (2005).

Data processing and statistics

ChromaTOF 3.25 software (Leco) was used to deconvolute themass spectra acquired. Chromatographic peaks with signal-to-noise ratios lower than 20 were excluded, and the peak areas wereobtained as reported previously [9]. Thus, a data matrix X of thepeak areas was constructed as described previously, with theobservations/samples in the first column and the responses/peaksas variables in the first row [10]. Then the GC/TOFMS data wereanalyzed using multivariate projection methods [8,12], such asPCA and partial least squares projection to latent structures (PLS)and discriminant analysis (DA), using SIMCA-P 11 (Umetrics,Umeå, Sweden). For PLS–DA modeling, samples were divided intodifferent groups (e.g., plasma and serum) as the qualitative‘‘dummy” variable, Y. The results of PCA and PLS–DA are given asscore plots that represent the metabolomic similarities of the sam-ples. Close clustering of the samples indicates their compositionalsimilarity, whereas distant clustering of the samples suggests theirdiverse metabolomic compositions. The purpose of PLS–DA was tocalculate models of the different groups and to identify the re-sponse variables that contribute most strongly to the model.

The goodness of fit for a model is evaluated using three quanti-tative parameters: R2X is the explained variation in X, R2Y is the ex-plained variation in Y, and Q2Y is the predicted variation in Y. Therange of these parameters is between 0 and 1, with valuesapproaching 1 indicating perfect fit of the model. Cross-validationwith seven cross-validation groups was used throughout to deter-mine the number of principal components [13], and the number ofprincipal components was determined once the Q2Y value de-creased continuously. Permutation tests were performed with aniteration of 100 for the validation of the model. Statistical analysisbetween groups was performed using a t test algorithm embeddedin Microsoft Excel with a significance level of 0.05 or 0.01.

Results and discussion

Differences in metabolite profile between human blood plasma andserum

Typical GC/TOFMS chromatograms of blood plasma and serumare shown in Fig. 1A and B. Visual inspection of the chromatogramsrevealed obvious differences between the plasma and serum. Afterdeconvolution of the GC/TOFMS profiles, 289 peaks were deter-mined and 72 of them were identified. They included low-molecu-lar-weight organic acids, amino acids, amines, carbohydrates, fattyacids, and lipids (see Table S-1 in Supplementary material). To gen-erate an overview of the data set, a PCA model was applied (of

Fig. 1. Typical GC/TOFMS chromatograms of blood plasma and serum from a healthy volunteer: (A) plasma; (B) serum. Visual inspection of the chromatograms revealedobvious differences between the plasma and serum. The compounds were identified as follows: 1, pyruvate; 2, alanine; 3, hydroxyacetate; 4, lactate; 5, b-hydroxybutyrate; 6,valine; 7, urea; 8, isoleucine, proline; 9, glycine; 10, serine; 11, threonine; 12, pyroglutamate; 13, creatinine; 14, ornithine; 15, phenylalanine; 16, glutamine, glutamate; 17,citrate; 18, glucose; 19, uric acid; 20, linoleic acid; 21, oleic acid; 22, tryptophan, stearate; 23, c-tocopherol; 24, cholesterol.

Metabolomic differences between blood plasma and serum / L. Liu et al. / Anal. Biochem. 406 (2010) 105–112 107

three principal components, R2X = 0.539 and Q2X = 0.368) (Fig. 2).As shown in Fig. 2, three outliers were identified, all in the plasma

Fig. 2. A PCA model of the GC/TOFMS data from human plasma and serum. Threeoutliers were identified in the plasma samples: the replicates PM7-1 and -2 andPF6. The close clustering of the replicates for both plasma (PM7-1 and -2) andserum (SM7-1 and -2) indicates good reproducibility in sample preparation and GC/TOFMS analysis.

samples, including the plasma replicates from the male subject(PM7-1 and -2 and PF6), but no serum outlier was observed. Theoutliers indicate their marked chemical differences from the othersamples. The close clustering of the replicates for both plasma(PM7-1 and -2) and serum (SM7-1 and -2) indicates good repro-ducibility in the sample preparation and GC/TOFMS analysis. Ingeneral, the plasma and serum samples clustered in two distinctgroups according to sample type (plasma or serum) far away fromeach other, with similarities within each group (except the outli-ers) and clear differences between the plasma and serum samples.The serum samples of the males and females tended to cluster clo-sely, whereas the plasma samples were loosely scattered aroundthe right part of the figure. The close clustering of the serum sam-ples indicates their high similarity in terms of their chemical com-positions and analyte abundances, whereas the loose clustering ofthe plasma samples suggests a distinct variance in the samples. Be-cause the body fluid (blood) of healthy individuals is always bal-anced at a steady state and the plasma and serum samples wereprepared from the same sources, the larger variation in the plasmasamples indicates the effects of some unknown factors.

To distinguish the differences between the serum and plasmasamples, all of the samples were divided into two groups (i.e., plas-ma and serum) and a PLS–DA model was calculated. After threecomponents, R2X, R2Y, and Q2Y amounted to 0.482, 0.993, and0.951, respectively, indicating a very good model to differentiatebetween the plasma and serum (Fig. S-1).

Table 1Peak areas of analytes showing statistically significant differences between blood plasma and serum from a human population (n = 15, 8 males and 7 females, paired t test).

Identified compounda Peak areas in plasma Peak areas in serum Fold change t TestMean ± SD Mean ± SD Plasma/serum P

Hypoxanthine 8.04E4 ± 3.25E4 4.53E5 ± 1.18E5 0.18 6.90E-10Glutamate 1.81E6 ± 8.38E5 2.79E6 ± 9.66E5 0.65 6.10E-03Cystine 8.90E5 ± 3.53E5 2.69E6 ± 5.71E5 0.33 8.90E-12Phenylalanine 3.64E6 ± 9.77E5 6.19E6 ± 9.41E5 0.59 2.20E-08Serine 5.23E6 ± 2.06E6 1.12E7 ± 2.72E6 0.47 8.50E-08Pyroglutamate 1.18E7 ± 2.81E6 1.66E7 ± 2.38E6 0.71 1.00E-05Glycine 5.36E6 ± 2.17E6 1.34E7 ± 4.73E6 0.40 3.90E-06Threonine 1.23E7 ± 4.81E6 2.14E7 ± 5.09E6 0.58 1.50E-05S-Methyl-cysteine 1.41E5 ± 7.29E4 2.83E5 ± 1.35E5 0.50 0.0012Ornithine 6.85E6 ± 3.20E6 1.28E7 ± 3.96E6 0.53 5.40E-05Proline 6.46E6 ± 2.72E6 9.97E6 ± 2.41E6 0.65 0.0005Methionine 1.63E6 ± 6.23E5 2.40E6 ± 4.31E5 0.68 0.0003Isoleucine 2.80E6 ± 9.82E5 4.12E6 ± 9.29E5 0.68 0.0005Valine 1.78E7 ± 5.73E6 2.49E7 ± 2.97E6 0.72 0.0002Tryptophan 9.88E6 ± 2.76E6 1.19E7 ± 1.56E6 0.83 0.0155Arginine 7.90E5 ± 2.45E5 1.53E6 ± 3.09E5 0.52 2.40E-08Glycerol-3-phosphate 1.03E6 ± 5.19E5 3.27E6 ± 6.23E5 0.32 4.50E-121-Monopalmitin 6.92E4 ± 2.52E4 1.99E5 ± 5.02E4 0.35 4.90E-092-Monooleoylglycerol 1.18E5 ± 6.79E4 3.25E5 ± 7.66E4 0.36 5.40E-09Glycerol 2.46E7 ± 7.06E6 3.14E7 ± 4.69E6 0.78 0.0032Arachidonic acid 1.11E6 ± 1.68E5 1.36E6 ± 1.93E5 0.81 0.0004b-Hydroxybutyrate 3.26E6 ± 4.80E6 7.06E6 ± 5.13E6 0.46 0.03842-Hydroxyvalerate 5.64E5 ± 3.70E5 1.04E6 ± 7.13E5 0.54 0.0278Aminomalonate 5.57E5 ± 8.05E5 1.38E6 ± 4.41E5 0.40 0.00162-Aminobutyrate 6.05E5 ± 3.95E5 1.62E6 ± 7.59E5 0.37 9.00E-05Ribose 1.83E5 ± 9.20E4 1.18E6 ± 5.16E5 0.15 1.10E-06Glucose 5.08E7 ± 2.13E7 9.35E7 ± 1.76E7 0.54 8.20E-07myo-Inositol 2.63E6 ± 9.67E5 3.81E6 ± 5.87E5 0.69 0.0002b-d-Methylglucopyranoside 3.27E6 ± 9.73E5 4.36E6 ± 8.07E5 0.75 0.0018CPU_ DB5_RI: 1872.9 1.55E5 ± 1.79E5 4.61E6 ± 1.77E6 0.03 4.20E-08CPU_ DB5_RI: 1230.4 5.54E6 ± 3.75E6 1.35E7 ± 2.28E6 0.41 4.20E-08Fumarate 4.69E5 ± 1.05E5 3.46E5 ± 1.31E5 1.36 0.0065Glycerate 1.14E5 ± 5.26E4 8.05E4 ± 2.47E4 1.42 0.0292Aspartate 3.10E6 ± 9.83E5 1.93E6 ± 1.27E6 1.60 0.0068Hydroxylamine 1.23E6 ± 3.13E5 7.96E5 ± 2.42E5 1.54 0.0001Urate 4.09E6 ± 9.20E5 2.58E6 ± 7.33E5 1.59 1.60E-05Pyruvate 3.95E6 ± 3.02E6 1.63E6 ± 1.16E6 2.42 0.0006Citrate 8.40E6 ± 1.90E6 2.82E6 ± 6.59E5 2.98 1.30E-09CPU_ DB5_RI: 2469.4 2.77E7 ± 7.18E6 7.39E6 ± 1.96E6 3.75 3.70E-09CPU_ DB5_RI: 1926.5 1.09E8 ± 3.60E7 1.03E7 ± 2.08E6 10.52 1.40E-08CPU_ DB5_RI: 2013.7 5.83E7 ± 1.76E7 3.14E6 ± 1.34E6 18.59 2.10E-09

Note. SD, standard deviation; CPU_ DB5_RI, unidentified compounds showing statistical significance between serum and plasma (CPU, China Pharmaceutical University; DB5,samples were analyzed using DB-5 capillary column; RI, retention index; x, retention index value).

a The identified compounds are MeOx and TMSi derivatives.

108 Metabolomic differences between blood plasma and serum / L. Liu et al. / Anal. Biochem. 406 (2010) 105–112

The differences between the samples from male and femalesubjects were also investigated and were consistent with the re-sults of a previous study [14]. Interestingly, based on the plasmadata, the calculated PLS–DA model (R2X = 0.303, R2Y = 0.963, andQ2Y = 0.469) showed better separation between the male and fe-male samples than the model based on the serum data(R2X = 0.135, R2Y = 0.878, and Q2Y = 0.347). However, the relativelylow values for the parameters suggest that the differences inmetabolite profile between the sexes were not very significant.

Analyte differences between human blood plasma and serum

The analytes that contributed most to the differentiation of theplasma and serum samples were identified by the model and fur-ther validated using a paired t test (Table 1). Among the 289 de-tected GC/TOFMS peaks/variables, 137 were statisticallysignificantly different in the plasma and serum. In total, 29 identi-fied metabolites were sharply reduced in plasma relative to theirpeak areas in serum, including most amino acids, carbohydrates,hypoxanthine, glycerol-3-phosphate, and b-hydroxybutyrate,whereas several metabolic products or intermediates occurred athigher peak areas in the plasma: pyruvate, citrate, glycerate, fuma-

rate, and the nitrogen metabolites urate and hydroxylamine. Inter-estingly, aspartate was the only amino acid to occur at significantlyhigher peak areas in the plasma. Because these serum and plasmasamples all were prepared after incubation for 2 h, the lower peakareas of nutrient compounds (amino acids and carbohydrates) inthe plasma samples indicate their greater use in the activities ofliving and in the supply of energy, and the higher peak areas ofmetabolic products provide further evidence that biochemicalmetabolism had occurred. The higher abundance of arachidonicacid in the serum was considered to have been released primarilyfrom thrombocytes [15] during the coagulation of the blood, whichdid not occur during the preparation of the plasma. The hypoxan-thine was considered to derive from slight hemolysis of the blood[16].

Differences in metabolite profile between rat blood plasma and serum

The human serum samples tended to cluster closely, whereasthe plasma samples showed larger positional variations. To mini-mize any possible interference among individuals caused by theirdiverse backgrounds, we applied the same analysis to SD rats.The typical GC/TOFMS chromatograms of the rat plasma and serum

Metabolomic differences between blood plasma and serum / L. Liu et al. / Anal. Biochem. 406 (2010) 105–112 109

showed analogous differences to those observed in the humanpopulation (see Fig. S-2 in Supplementary material). Interestingly,the plasma samples were loosely scattered, again showing largervariation than that observed in the serum samples (Fig. S-3). Theseresults confirm the finding that the plasma samples had a largervariance than the serum samples, in terms of their compositionsand analyte peak areas, and support the findings of another study[1].

Statistical analysis of the differences between the metabolitesin the serum and plasma from the rat blood showed the same ten-dencies as were observed in the plasma and serum from humanblood. Most of the metabolites that occurred at higher peak areasin human serum were also more abundant in the rat serum thanin the rat plasma and vice versa. The identified compounds thatdifferentiated the plasma from the serum were involved in the pri-mary metabolic pathways, such as glycolysis, and the tricarboxylicacid (TCA) and urea cycles. The consistent results from both humanand rat blood specimens strongly indicate that during incubationthe plasma and serum had been biochemically metabolized to dif-ferent extents.

Incubation effects on metabolomic patterns of plasma and serum

A PCA model of the data set showed no outliers, so a PLS–DAmodel was calculated based on the nine groups (plasma bloodand serum blood incubated for 0, 1, 2, 3, and 4 h) (Fig. 3). Consid-ering that there were nine groups for the calculation, this was agood PLS–DA model according to cross-validation (R2X = 0.577,R2Y = 0.421, and Q2Y = 0.377). The PLS–DA profile showed theclear clustering of the nine groups. As shown in two-dimensionalFig. 3, the sample types primarily contributed to the first princi-pal component on the x axis, and the incubation time dominatedthe second principal component on the y axis. For both the plas-ma and serum samples, there appeared to be a continual movingdown of the score plots in the figure as the incubation time in-creased from 1 to 4 h. This finding suggests that the incubation

Fig. 3. PLS–DA score plot of the GC/TOFMS data for rat plasma (P) and serum (S)after incubation. According to cross-validation, this is a three-component PLS–DAmodel (PC1: R2X = 0.472, R2Y = 0.239, and Q2Y = 0.224; PC2: R2X = 0.526,R2Y = 0.329, and Q2Y = 0.279; PC3: R2X = 0.577, R2Y = 0.421, and Q2Y = 0.332). Fromthe two-dimensional figure, a dynamic moving of the score plots was observed asthe incubation time increased from 1 to 4 h for all samples, and the incubation ofthe blood sample had the most marked impact on the metabolomic phenotype aftera short incubation (2 h). �, P0; h, P1; s, P2; e, P3; D, P4; j, S1; d, S2; �, S3; N, S4.

of the blood has a marked effect on the metabolite abundances.Although there was clear separation between the samples incu-bated for 0 and 1 h and those incubated for 1 and 2 h, the sam-ples incubated for 2, 3, and 4 h clustered closely. This indicatesthat the incubation of the blood from which the serum and plas-ma were derived had marked effects on the analyte peak areasduring a short incubation period (<2 h) but had much lesser ef-fects after a longer incubation.

Incubation effects on analyte peak areas in plasma and serum

By comparing the relative intensities of the GC/TOFMS re-sponses, we observed that the incubation of the blood resultedin steady fluctuations of many metabolites in both the serumand plasma (Table 2). Although a previous study reported a de-cline in glucose peak areas and a simultaneous increase in lactatein heparinized plasma and serum after the blood samples hadbeen kept at room temperature for several hours [1], we ob-served much more complex changes in the current study. In gen-eral, the incubation of the whole blood resulted in declines in thenutrient compounds, such as carbohydrates and amino acids, andincreases in metabolic products and TCA intermediates, such aspyruvate, glycerate, and a-ketoglutarate. Furthermore, themetabolite peak areas in the plasma changed more than thosein serum after incubation. After incubation for 1 h, the peak areasof glucose, ornithine, and cysteine had declined rapidly, whereasthose of pyruvate, glycerate, and a-ketoglutarate were markedlyelevated (Fig. 4). After incubation for 2 h, there was greatervariation in these compounds and the peak areas of more com-pounds were significantly different from their starting values,especially in the plasma. Thereafter, the changes in these peakareas became less affected for most of the metabolites (Fig. 3and Table 2).

It has been demonstrated that metabolites in erythrocytes showobvious time-dependent changes after incubation [17], especiallyfor nutrient substances and metabolic products. Consistent evi-dence suggests that the active metabolism of the blood cells con-sumes the nutrient substances and releases the metabolicproducts into the environment (i.e., the plasma and serum). We ob-served the gradual consumption of glucose and amino acids in boththe plasma and serum and the accumulation of their products,such as pyruvate, lactate, and TCA intermediates, which are consid-ered to be metabolized to provide energy.

In this study, the effects of incubation on the peak areas ofglucose and lactate in both the plasma and serum were consistentwith the results of a previous study [1], but the magnitude of thechanges observed in our study was greater. In our study, theblood specimens were incubated at 37 �C, whereas Boyantonand Blick conducted their experiments at room temperature [1].The incubation temperature (37 �C vs. room temperature) wasconsidered to be the major factor responsible for the discrepan-cies between these studies. To minimize the influence of incuba-tion on the analyte peak areas in plasma or serum, placing theblood samples at room temperature or at an even lower temper-ature seems to be a good idea. However, such methods are inef-ficient and take longer to produce serum. Although a special tubeis available for the rapid preparation of serum, the technique isclinically expensive and not applicable to small volumes (e.g.,<200 ll) of blood from small animals. The adaptation of thismethod for the rapid preparation of small amounts of serum isrequired.

Although incubation had similar effects on the serum and plas-ma metabolites, the serum metabolites were less affected. Incuba-tion significantly altered the peak areas of only pyruvate, glycerate,a-ketoglutarate, glucose, b-D-methylglucopyranoside, myo-inosi-tol, glycine, and glycerol-3-phosphate in serum. Peak areas of

Table 2Peak areas of analytes in serum and plasma showing significant variations after incubation of the blood for 1, 2, 3, and 4 h.

Identified compounda Peak areas in plasma Peak areas in serum

0 h 1 h 2 h 3 h 4 h 1 h 2 h 3 h 4 h

Pyruvate Mean ± SD 2.44E6 6.43E6b 8.58E6c 1.13E7c 1.36E7c 2.77E6e 4.44E6c,e 5.36E6c,e 7.68E6c,e

1.39E6 3.18E6 4.56E6 5.57E6 1.71E6 1.41E6 3.48E5 5.61E5 6.39E5Glycerate Mean ± SD 1.42E5 2.66E5c 3.45E5c 3.56E5c 3.52E5c 2.23E5c 3.25E5c 4.07E5c 4.95E5c,d

4.83E4 2.73E4 1.18E5 2.23E4 6.09E4 5.01E4 3.73E4 5.58E4 8.80E4a-Ketoglutarate Mean ± SD 2.80E5 4.52E5c 5.36E5c 7.23E5c 9.32E5c 2.01E5b,e 2.54E5e 3.70E5e 5.29E5c,e

7.05E4 5.43E4 5.70E4 6.53E4 1.70E4 7.89E4 4.75E4 4.66E4 1.07E5Aminomalonate Mean ± SD 3.53E6 3.81E6 4.02E6 4.38E6b 4.90E6b 4.04E6 4.02E6 3.56E6d 3.97E6d

7.23E5 5.02E5 1.11E6 6.01E5 1.17E6 2.52E5 3.49E5 6.51E5 4.20E5Fumarate Mean ± SD 4.93E5 5.34E5 4.74E5 5.51E5b 5.94E5b 5.27E5 4.84E5 5.29E5 4.90E5e

6.32E4 1.08E5 1.48E5 3.19E4 5.52E4 7.50E4 3.71E4 6.05E4 3.90E4Threonate Mean ± SD 8.75E5 1.01E6b 1.06E6b 1.07E6c 1.10E6b 8.74E5d 9.03E5 9.37E5 9.23E5

9.49E4 6.20E4 1.29E5 5.06E4 1.70E5 8.66E4 8.82E4 1.37E5 8.63E4Aspartate Mean ± SD 9.54E5 1.21E6 2.09E6c 2.27E6c 2.97E6c 4.97E5c,e 4.91E5c,e 3.69E5c,e 4.49E5c,e

4.94E5 2.57E5 2.49E5 5.31E5 1.10E6 6.90E4 6.04E4 1.10E5 1.08E5Arachidonic acid Mean ± SD 9.19E5 1.02E6 1.06E6 1.06E6 9.44E5 1.80E6c,e 2.15E6c,e 2.13E6c,e 2.21E6c,e

9.10E4 6.69E4 8.55E4 5.29E4 5.13E4 1.74E5 7.47E4 2.13E5 1.11E5Pyroglutamate Mean ± SD 7.96E6 8.46E6 9.23E6c 9.23E6c 9.43E6c 1.01E7c,e 1.00E7c,e 1.13E7c,e 1.19E7c,e

5.84E5 4.52E5 7.26E5 4.64E5 9.05E5 4.06E5 4.01E5 6.82E5 5.37E5Lactate Mean ± SD 3.36E7 6.69E7c 7.63E7c 5.04E7c 4.84E7 6.31E7c 5.43E7b,e 7.05E7c,e 4.91E7

2.37E7 1.43E7 1.97E7 2.07E7 2.20E7 4.06E7 1.99E7 3.09E7 4.28E6b-Hydroxybutyrate Mean ± SD 5.33E5 5.61E5 4.01E5 4.85E5 2.26E5b 2.31E6c,e 2.65E6c,e 2.76E6c,e 2.01E6c,e

3.87E5 2.61E5 1.69E5 2.63E5 1.70E5 9.62E5 1.89E5 1.81E5 7.52E5Alanine Mean ± SD 2.42E7 2.48E7 1.69E7b 7.28E6c 8.32E6c 2.00E7 1.99E7 2.08e 2.45E7e

5.51E6 5.47E6 5.71E5 9.44E5 1.06E6 3.97E6 2.95E6 2.63E6 6.17E6Valine Mean ± SD 8.65E6 9.58E6 7.79E6 7.24E6 6.62E6 1.33E7b,d 1.59E7c,e 1.54E7c,e 1.49E7c,e

2.41E6 1.87E6 1.48E6 1.02E6 1.54E6 4.17E6 2.08E6 1.71E6 2.70E6Proline Mean ± SD 1.17E7 1.08E7 9.05E6 9.01E6 7.32E6b 1.55E7c,e 1.53E7c,e 1.66E7c,e 1.51E7c,e

2.72E6 1.84E6 1.80E6 9.30E5 8.63E5 1.21E6 1.77E6 1.76E6 1.94E6Glycine Mean ± SD 3.63E6 3.86E6 2.89E6 2.88E6b 2.65E6b 4.17E6 5.04E6b,e 5.41E6c,e 6.15E6c,e

6.99E5 3.17E5 6.04E5 2.24E5 5.51E5 4.56E5 5.12E5 7.51E5 8.01E5Threonine Mean ± SD 6.30E6 6.33E6 5.21E6b 5.79E6 5.41E6 8.04E6b,d 8.53E6c,e 9.01E6c,e 9.37E6c,e

5.47E5 4.69E5 5.19E5 3.97E5 6.73E5 1.02E6 8.07E5 1.03E6 9.34E54-Hydroxyproline Mean ± SD 2.97E6 2.92E6 2.30E6b 2.32E6c 2.08E6c 3.57E6b,e 3.60E6c,e 3.69E6c,e 3.79E6c,e

3.86E5 2.00E5 4.00E5 1.88E5 3.38E5 1.62E5 1.19E5 1.97E5 1.72E5Cysteine Mean ± SD 1.96E6 1.56E6 1.16E6b 1.21E6b 1.09E6c 2.32E6e 2.46E6e 2.61E6b,e 3.19E6c,e

8.01E5 2.31E5 1.45E5 1.33E5 2.01E5 1.07E5 2.00E5 1.00E5 2.22E5Glutamine Mean ± SD 1.26E7 1.20E7 9.75E6 9.70E6 8.46E6c 1.04E7 8.66E6c 8.39E6c 8.37E6c

1.68E6 1.36E6 2.08E6 2.08E6 2.06E6 1.19E6 1.05E6 1.24E6 1.46E6Ornithine Mean ± SD 1.85E6 1.27E6c 9.39E5c 9.39E5c 7.40E5c 1.98E6e 1.71E6e 1.67E6e 1.76E6e

4.07E5 7.87E4 1.14E5 6.31E4 1.46E5 1.92E5 9.48E4 1.66E5 1.06E5Hypoxanthine Mean ± SD 9.37E4 1.92E5c 2.67E5c 2.37E5c 1.68E5c 9.65E5c,e 1.18E6c,e 1.10E6c,e 1.48E6c,e

1.27E4 1.21E4 1.21E5 4.74E4 1.15E4 1.66E5 4.87E4 1.10E5 7.55E4Tryptophan Mean ± SD 6.78E6 6.64E6 5.96E6b 6.20E6 5.65E6c 6.95E6 6.89E6e 6.97E6d 7.10E6e

3.93E5 2.76E5 5.49E5 5.54E5 5.01E5 4.92E5 3.67E5 4.04E5 3.29E5Cystine Mean ± SD 1.40E6 1.24E6 9.09E5 9.72E5 7.97E5 2.58E6c,e 2.24E6c,e 2.07E6c,e 1.73E6b,e

3.17E5 1.39E5 7.67E4 1.15E5 8.92E4 2.86E5 1.11E5 9.72E4 1.20E5Glucose Mean ± SD 5.37E7 4.10E7 3.33E7c 3.37E7c 2.97E7c 6.92E7b,e 5.47E7e 4.72E7e 4.38E7e

1.78E7 6.16E6 2.05E6 1.77E6 1.84E6 1.34E7 6.27E6 2.91E6 2.74E6b-D-Methyl-

glucopyranosideMean ± SD 2.20E6 1.85E6 1.02E6c 9.96E5c 6.77E5c 2.54E6e 1.97E6e 1.56E6c,e 1.35E6c,e

7.21E5 2.80E5 7.46E4 1.78E5 7.36E4 1.90E5 1.11E5 1.28E5 1.49E5Glycerol-3-phosphate Mean ± SD 7.69E5 9.00E5 1.01E6c 1.07E6c 1.00E6c 1.69E6c,e 2.52E6c,e 3.06E6c,e 3.47E6c,e

1.39E5 1.05E5 1.40E5 1.00E5 1.07E5 1.19E5 1.20E5 2.30E5 1.94E5myo-Inositol Mean ± SD 8.48E5 1.01E6 8.91E5 9.04E5 1.02E6 1.82E6c,e 2.07E6c,e 2.46E6c,e 2.55E6c,e

3.54E5 4.31E5 2.72E5 2.50E5 2.63E5 1.96E5 1.10E5 4.34E5 4.50E5Tyrosine Mean ± SD 4.78E6 5.58E6b 5.44E6 5.93E6c 5.68E6b 5.37E6 5.84E6c 6.67E6c,d 7.24E6c,e

5.90E5 2.70E5 6.04E5 5.21E5 6.01E5 3.82E5 2.82E5 3.99E5 3.33E5Phenylalanine Mean ± SD 2.43E6 2.64E6 2.33E6 2.50E6 2.37E6 3.08E6c,e 3.30E6c,e 3.62E6c,e 3.90E6c,e

2.28E5 1.48E5 3.43E5 1.14E5 2.52E5 1.53E5 1.11E5 1.00E5 1.56E5Citrate Mean ± SD 8.70E6 8.31E6 7.80E6 7.56E6 7.43E6 1.94E6c,e 1.89E6c,e 1.88E6c,e 1.88E6c,e

7.20E5 2.36E5 7.26E5 8.16E5 1.04E6 1.33E5 1.02E5 8.87E4 6.46E4Urate Mean ± SD 1.14E7 1.16E7 1.11E7 1.12E7 1.04E7 1.08E7 1.05E7 9.75E6c,e 9.95E6b

4.94E5 4.72E5 1.30E6 7.84E5 8.66E5 2.57E5 4.48E5 2.78E5 4.73E5

a The identified compounds are MeOx and TMSi derivatives.b,c Statistically significantly different from the value of plasma_0 h, Ps < 0.05 and 0.01 (t test).d,e Statistically significantly different from the value of plasma_(same time), Ps < 0.05 and 0.01 (t test).

110 Metabolomic differences between blood plasma and serum / L. Liu et al. / Anal. Biochem. 406 (2010) 105–112

many metabolites remained quite stable or changed very little dur-ing incubation (Table 2). These results indicate that serum metab-olites underwent less intensive changes than plasma metabolites.Interestingly, some compounds showed opposite changes in serumand plasma after incubation. For example, the peak areas of aspar-

tate, glycine, cysteine, and threonine decreased significantly inplasma but increased in serum (Table 2). These results indicatethat incubation is not the only factor affecting the analyte levels.The coagulating agent EDTA–2Na may be another factor warrant-ing further study.

Fig. 4. Changes in metabolite peak areas after incubation for 1, 2, 3, and 4 h: (A) pyruvate; (B) a-ketoglutarate; (C) glycerate; (D) glucose; (E) ornithine; (F) cysteine. Afterincubation, some metabolic products were significantly elevated, including pyruvate, a-ketoglutarate, and glycerate, whereas the peak areas of some nutrient materialsdeclined rapidly in the plasma, including glucose, ornithine, and cysteine. These results constitute direct evidence of the differences between plasma and serum and ofincubation-related metabolic variations (*P < 0.05 vs. the values for the plasma control [0 h]).

Metabolomic differences between blood plasma and serum / L. Liu et al. / Anal. Biochem. 406 (2010) 105–112 111

Conclusion

There are significant chemical differences between serum andplasma. The discriminatory compounds include carbohydratesand their metabolites, amino acids, and TCA intermediates. Incuba-tion of the blood during the preparation of plasma or serum haddistinct effects on the peak areas of analytes, most of which are in-volved in energy metabolism. Because incubation influences ana-lyte levels, the original data might be compromised in theanalysis of plasma or serum prepared from blood specimens incu-bated for prolonged periods. The incubation of blood specimens af-fects the analyte peak areas in serum less than those in plasma.Consequently, we recommend serum as the sample of choice inmetabolomic studies.

Acknowledgments

This study was financially supported by the National Key NewDrug Creation Special Programs (2009ZX09304-001 and2009ZX09502-004), the National Natural Science Foundation of

the People’s Republic of China (30630076), the Jiangsu Province So-cial Development Foundation (BE2008673), and the National 11th5-Year Technology Supporting Program of the People’s Republic ofChina (2006BAI08B04).

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.ab.2010.07.015.

References

[1] B.L. Boyanton Jr., K.E. Blick, Stability studies of twenty-four analytes in humanplasma and serum, Clin. Chem. 48 (2002) 2242–2247.

[2] R.R. Miles, R.F. Roberts, A.R. Putnam, W.L. Roberts, Comparison of serum andheparinized plasma samples for measurement of chemistry analytes, Clin.Chem. 50 (2004) 1704–1706.

[3] X.Y. Bi, J.S. Deng, X.L. Deng, W.J. Huang, Study on the biochemical marker ofserum and plasma, Mod. Med. Health 19 (2003) 1380–1381 (in Chinese).

[4] J.H. Ladenson, L.M. Tsai, J.M. Michael, G. Kessler, J.H. Joist, Serum versusheparinized plasma for eighteen common chemistry tests: is serum theappropriate specimen?, Am J. Clin. Pathol. 62 (1974) 545–552.

112 Metabolomic differences between blood plasma and serum / L. Liu et al. / Anal. Biochem. 406 (2010) 105–112

[5] R.E. Haymond, J.A. Knight, Venous serum, capillary serum, and capillaryplasma compared for use in determination of lactate dehydrogenase andaspartate aminotransferase activities, Clin. Chem. 21 (1975) 896–897.

[6] B.T. Doumas, L.L. Hause, D.M. Simuncak, D. Breitenfeld, Differences betweenvalues for plasma and serum in tests performed in the Ektachem 700 XRAnalyzer, and evaluation of plasma separator tubes (PST), Clin. Chem. 35(1989) 151–153.

[7] Y.J. Cheng, L. Feng, Comparison of heparinized plasma and serum samples formeasurement of ordinary chemistry analytes, J. Pract. Med. Technol. 14 (2007)4529–4531 (in Chinese).

[8] J. Trygg, E. Holmes, T. Lundstedt, Chemometrics in metabonomics, J. ProteomeRes. 6 (2007) 469–479.

[9] P. Jonsson, A.I. Johansson, J. Gullberg, J. Trygg, J. A, B. Grung, S. Marklund, M.Sjostrom, H. Antti, T. Moritz, High-throughput data analysis for detecting andidentifying differences between samples in GC/MS-based metabolomicanalyses, Anal. Chem. 77 (2005) 5635–5642.

[10] J. A, J. Trygg, J. Gullberg, A.I. Johansson, P. Jonsson, H. Antti, S.L. Marklund, T.Moritz, Extraction and GC/MS analysis of the human blood plasmametabolome, Anal. Chem. 77 (2005) 8086–8094.

[11] E.J. Want, G. O’Maille, C.A. Smith, T.R. Brandon, W. Uritboonthai, C. Qin, S.A.Trauger, G. Siuzdak, Solvent-dependent metabolite distribution, clustering,and protein extraction for serum profiling with mass spectrometry, Anal.Chem. 78 (2006) 743–752.

[12] L. Eriksson, E. Johansson, N. Kettaneh-Wold, S. Wold, Multi- and MegavariateData Analysis Principles and Applications, Umetrics, Umeå, Sweden, 2001.

[13] S. Wold, Cross-validatory estimation of the number of components in factorand principal components models, Technometrics 2 (1978) 397–405.

[14] U. Lutz, R.W. Lutz, W.K. Lutz, Metabolic profiling of glucuronides in humanurine by LC–MS/MS and partial least-squares discriminant analysis forclassification and prediction of gender, Anal. Chem. 78 (2006) 4564–4571.

[15] L. Wilhelmsen, Thrombocytes and coronary heart disease, Circulation 84(1991) 936–938.

[16] F. Niklasson, Simultaneous liquid-chromatographic determination ofhypoxanthine, xanthine, urate, and creatinine in cerebrospinal fluid, withdirect injection, Clin. Chem. 29 (1983) 1543–1546.

[17] Y. Zhang, J. A, G. Wang, Q. Huang, B. Yan, W. Zha, S. Gu, L. Liu, H. Ren, M. Ren, L.Sheng, Organic solvent extraction and metabonomic profiling of themetabolites in erythrocytes, J. Chromatogr. B 877 (2009) 1751–1757.