Sie sind auf Seite 1von 11

Clinical Chemistry 59:9 Endocrinology and Metabolism

1338–1348 (2013)

Fasting Serum Lipid and Dehydroepiandrosterone Sulfate


as Important Metabolites for Detecting Isolated
Postchallenge Diabetes:
Serum Metabolomics via Ultra-High-Performance LC-MS
Liyan Liu,1 Maoqing Wang,1 Xue Yang,1 Mingxin Bi,1 Lixin Na,1 Yucun Niu,1 Ying Li,1* Changhao Sun1*

BACKGROUND: Isolated postchallenge diabetes (IPD), a acid, oleic acid, and DHEA-S in the validation study
subtype of type 2 diabetes mellitus (T2DM) defined as was 0.849 for the IPD group.
2-h postprandial plasma glucose ⱖ200 mg/dL (ⱖ11.1
mmol/L) and fasting plasma glucose (FPG) ⬍108 CONCLUSIONS: The current study provides useful infor-
mg/dL (⬍6.0 mmol/L), is often overlooked during mation to bridge the gaps in our understanding of the
screening for diabetes on the basis of FPG concentra- metabolic alterations associated with IPD and might
tions. A key challenge is early identification of IPD by facilitate the characterization of patients with IPD by
the use of fasting serum, which is critical for large-scale the use of fasting serum.
diabetes screening. © 2013 American Association for Clinical Chemistry

METHODS: We applied a nontargeted metabolomic


approach using ultra-high-performance liquid Isolated postchallenge diabetes (IPD),2 characterized
chromatography– quadrupole TOF–mass spectrometry by 2-h postprandial plasma glucose (2-h PG) ⱖ200
(UPLC-QTOF-MS) to analyze serum samples from 51 mg/dL (ⱖ11.1 mmol/L) and fasting plasma glucose
patients with IPD, 52 with newly diagnosed T2DM, (FPG) ⬍108 mg/dL (⬍6.0 mmol/L), is a subtype of
and 49 healthy individuals. We processed metabolite type 2 diabetes mellitus (T2DM). The earlier Diabetes
profiles by multivariate analysis to identify potential Epidemiology: Collaborative Analysis of Diagnostic
metabolites, which were further confirmed by tandem Criteria in Europe (DECODE) study reported that
MS (MS/MS). We also used GC-MS and ELISA meth- 50% of patients with newly diagnosed T2DM and 2-h
ods to detect potentially important metabolites. A PG ⱖ200 mg/dL (ⱖ11.1 mmol/L) have FPG ⬍108
number of independent samples were selected to vali- mg/dL (⬍6.0 mmol/L) (1 ). Similarly, in the Third Na-
date the identified candidates. tional Health and Nutrition Examination Survey, 41%
of patients with previously undiagnosed T2DM were
RESULTS: We selected 15 metabolites with a view to dis- identified as having IPD (2 ). FPG is generally found
tinguishing patients with IPD, whereas 11 were identi- during routine physical examination or large-scale
fied with an authentic standard. The selected metabo- screening for diabetes. However, the use of FPG alone
lites included linoleic acid, oleic acid, phospholipids, fails to distinguish IPD, often leading to missing the
and dehydroepiandrosterone sulfate (DHEA-S). In diagnosis. Indeed, a key challenge is early diagnosis
IPD samples, significantly higher linoleic and oleic acid with IPD by the use of fasting serum samples, which is
(P ⬍ 0.001) and lower DHEA-S (P ⬍ 0.001) concen- critical for large-scale screening of diabetes. With the
trations were observed, compared with controls. The advent of new measurement technologies, several stud-
area under the curve from a combination of linoleic ies on T2DM diagnosis have been performed by the use

1
Department of Nutrition and Food Hygiene, Public Health College, Harbin Diagnostic Criteria in Europe; UPLC-MS, ultra-high-performance LC-MS; QTOF,
Medical University, Harbin, P. R. China. quadrupole TOF; FFA, free fatty acid; Hb A1c, glycosylated hemoglobin; ESI,
* Address correspondence to these authors at: Department of Nutrition and Food electrospray ionization; PCA, principal component analysis; OPLS-DA, orthogo-
Hygiene, Public Health College, Harbin Medical University, 157 Baojian Road, nal partial least-squares-discriminant analysis; VIP, variable importance in the
Nangang District, Harbin, P. R. China, 150086. Fax: ⫹86-(0)451-87502885; projection; DHEA-S dehydroepiandrosterone sulfate; ANCOVA, analysis of co-
e-mail changhao2002sun@yahoo.com or liying_helen@163.com. variance; BMI, body mass index; TC, total cholesterol; AUC, area under the
Received November 29, 2012; accepted March 29, 2013. curve; HOMA-IS, homeostasis model of assessment for insulin sensitivity;
Previously published online at DOI: 10.1373/clinchem.2012.200527 HOMA-IR, HOMA for insulin resistance; MS/MS, tandem MS; lysoPC, lyso-
2
Nonstandard abbreviations: IPD, isolated postchallenge diabetes; 2-h PG, 2-h phosphatidylcholine; lysoPE, lysophosphatidyl ethanolamine; DRF, diabetes
postprandial plasma glucose: FPG, fasting plasma glucose; T2DM, type 2 risk factor.
diabetes mellitus; DECODE, Diabetes Epidemiology: Collaborative Analysis of

1338
Serum Metabolomics in IPD Patients

of high-content methods of analysis, such as genomics, with newly diagnosed T2DM, 51 with IPD, and 49
proteomics, and metabolomics. In the current investi- healthy individuals were included in the study. T2DM
gation, we focused on using metabolomics for the de- was diagnosed according to the 1997 American Diabe-
tection of isolated postchallenge diabetes. tes Association criteria (14 ), defined as FPG ⬎126
The field of metabolomics provides a dynamic mg/dL (⬎7.0 mmol/L) and 2 h-PG ⬎200 mg/dL
portrait of metabolic status. Metabolic profiles are par- (⬎11.1 mmol/L). The cutoff values for FPG and 2
ticularly effective indicators of the phenotype or tissue h-PG used to diagnose IPD were 108 mg/dL (6.0
physiology of an organism (3 ) and provide informa- mmol/L) and 200 mg/dL (11.1 mmol/L), respectively.
tion that is not accessible through other “omics” ap- The patient samples in the current study were different
proaches. To date, metabolomics has been successfully from those used previously (13 ). Enrolled patients
applied to several fields, including disease diagnosis were not on any medication before sample collection.
(4 ), biomarker screening (5, 6 ), characterization of A 10-mL blood sample was collected into a serum sep-
biological pathways (7 ), and nutrition research (8 ). arator tube (red-top Vacutainer) in the morning before
Recently, the ultra-high-performance LC-MS (UPLC- breakfast. These fasting blood samples were immedi-
MS)– based metabolomics approach has been increas- ately centrifuged at 3000g for 10 min at room temper-
ingly applied to identify biomarkers for distinguishing ature and stored at ⫺80 °C until analysis.
variants of T2DM (9 –12 ). Until now, however, the We further validated our findings in independent
majority of studies have focused on individuals with samples (400 individuals). Individuals in the latter
FPG ⬎126 mg/dL (⬎7.0 mmol/L), and limited atten- sample were recruited in 2012 from the Hexing and
tion has been paid to IPD patients. On the basis of the Yixing districts in Harbin to establish the prevalence
findings of earlier studies, we hypothesized that con- and risk factors of chronic noncommunicable diseases.
siderable metabolic variation exists in the fasting serum Mean (SD) values for age and glycosylated hemoglobin
glucose concentrations of patients with IPD, support- (Hb A1c) in the different groups were as follows: IPD,
ing the possibility of identifying clinically significant n ⫽ 133; age ⫽ 50.5 (5.5) years, Hb A1c ⫽ 6.4% (0.5%);
metabolites to facilitate IPD diagnosis. In keeping with newly diagnosed T2DM, n ⫽ 137, age ⫽ 49.2 (5.4)
this theory, our group recently showed that free fatty years, Hb A1c ⫽ 7.9% (1.2%); and healthy controls,
acid (FFA) profiles in fasting serum of patients with n ⫽ 130; age ⫽ 51.1 (5.4) years, Hb A1c ⫽ 5.4% (0.3%).
diabetes are useful for IPD diagnosis in large popula-
tions (13 ). Notably, FFA profiles represent only a small SAMPLE PREPARATION
proportion of the metabolite fingerprint. The specific- A 200-␮L aliquot of each serum sample was used for
ity of FFAs was further addressed in the current study, metabolite extraction before UPLC-QTOF-MS analy-
whereby we determined the metabolite fingerprint and sis. We carried out the extraction procedure after addi-
comprehensively characterized patients with IPD using tion of 1.0 mL methanol (chromatographic grade,
a metabolomics approach. Honeywell Burdick & Jackson) to serum. The mixture
We used UPLC and quadrupole TOF (QTOF)/ was vortex-mixed for 1 min and incubated at room
MS– based metabolomic platforms to profile metabo- temperature for 10 min, followed by centrifugation at
lites in fasting serum from IPD and T2DM patients as 14 000g for 10 min at 4 °C. The supernatant was evap-
well as healthy individuals. Initially, the repeatability of orated to dryness under N2, and 200 ␮L of a mixture of
FFAs was validated. Subsequently, we attempted to acetonitrile (chromatographic grade, Honeywell
identify the metabolites significantly associated with Burdick & Jackson) and water (3:1) was added to each
IPD that may be effectively used for distinguishing IPD tube. The solution was filtered through a syringe filter
in the clinic. (0.22 ␮m) and placed into a 2-mL glass vial pending
UPLC-QTOF-MS analysis.
Materials and Methods
UPLC-QTOF-MS ANALYSIS
STUDY SAMPLES UPLC-MS analysis was performed with a Waters Ac-
The study was approved by the Ethics Committee of quity UPLC system coupled to a Waters Micromass
Harbin Medical University and conducted in accor- Q-TOF micro™ mass spectrometer with electrospray
dance with the Declaration of Helsinki. Written in- ionization (ESI) in the positive and negative modes. A
formed consent was obtained from each participant. 2-␮L aliquot of the sample solution was injected into
All subjects participated in a population survey to an Acquity UPLC™ BEH C18 column (50 ⫻ 2.1 mm;
establish the prevalence and risk factors of diabetes in i.d. 1.7 ␮m; Waters Corp.). The flow rate of the mobile
Harbin (PRFD-Harbin). Subjects were recruited in phase was 350 ␮L/min. Analytes were eluted from the
2008 from communities in Harbin from the Heilongji- column under a gradient (solvent A, 0.1% formic acid
ang Province in northern China. Fifty-two patients in water, and solvent B, acetonitrile). The optimal con-

Clinical Chemistry 59:9 (2013) 1339


ditions for UPLC separation and ESI-TOF-MS detec- ting, supervised models were validated with a per-
tion are shown in online Supplemental Table 1, which mutation test. A description of the multivariate sta-
accompanies the online version of this article at http:// tistical analysis can be found in the online
www.clinchem.org/content/vol59/issue9. All analyses Supplemental Methods.
were acquired with a lock spray to ensure accuracy and
reproducibility. A lock mass of leucine enkephalin IDENTIFICATION OF SIGNIFICANT METABOLITES
for positive ([M⫹H]⫹ ⫽ 556.2771) and negative We selected the serum differential variables associated
([M⫹H]⫺ ⫽ 554.2615) ion modes was applied via a with IPD and T2DM on the basis of a threshold of variable
lock spray interface. Lock spray frequency was set at importance in the projection (VIP) value (VIP ⬎1.0)
0.48 s, and lock mass data were averaged over 10 scans from the OPLS-DA model (15 ). The VIP values reflect the
for correction. importance of terms in the model both with respect to Y
We used QC samples to assess reproducibility and (its correlation to all the responses), and with respect to X
reliability of the UPLC-MS system. The QC samples (the projection). VIP values are computed, by default,
were prepared by mixing equal volumes of different from all extracted components. One can compare the VIP
individual serum samples (10 healthy controls, 10 pa- of 1 term to the others. Terms with large VIP, ⬎1, are the
tients with IPD, and 10 with T2DM). One QC sample most relevant for explaining Y (15 ).
was injected at the start of the analytical batch, followed In parallel, we validated these differential metabo-
by analysis of 1 QC sample at every fifth sample injec- lites from the OPLS-DA model at a univariate level
tion throughout the analytical workflow. The repro- with the Student t-test. The critical P value of the test
ducibility of the method was determined with principal was set to 0.05. Metabolites were identified on the basis
component analysis (PCA), and reliability was assessed of the accurate mass, retention time, and matching MS
by extracting 6 ions from chromatographic peaks of the spectra of the unknown with standard model com-
QC samples. pounds. To support the identification process, we used
PCA performed on QC and other groups revealed
Metlin (http://metlin.scripps.edu/) and a human me-
that QC samples were clustered in the PCA scores plot
tabolite database (http://www.hmdb.ca/). Finally, me-
(see online Supplemental Fig. 1). Six ions extracted
tabolites were confirmed by comparison of retention time
from the chromatographic peaks (m/z 100.07, 330.33,
and fragmentation pattern with authentic standards.
496.34, 524.37, 566.43, and 792.59 in the positive ion
mode) were selected for method reliability. The relative
QUANTITATIVE ANALYSIS OF 3 METABOLITES
standard deviations (CV) of peak intensity, retention
Three important metabolites identified in UPLC-
times, and m/z were estimated as 3.13%–15.24%, 0%–
QTOF-MS experiments were further quantified for
0.051%, and 0%– 0.005%, respectively. These results
collectively indicated good repeatability, reliability, confirmation purposes with the addition method.
and stability of this method for metabolite analysis. Quantification of linoleic acid and oleic acid was per-
UPLC-QTOF-MS raw data were analyzed with formed with GC-MS, as previously described (13 ). De-
MarkerLynx Application Manager 4.1 (Waters Corp.). hydroepiandrosterone sulfate (DHEA-S) was mea-
The online Supplemental Methods contains a descrip- sured with a commercial ELISA kit (IBL Corp.).
tion of the data analysis. The matrix from UPLC-
STATISTICAL ANALYSIS
QTOF-MS was introduced into SIMCA-P 11.0 soft-
ware (Umetrics) and standardized to a mean of 0 and Serum biochemical indicators were expressed as mean
variance of 1, according to the formula [X – mean(X)]/ (SD). We analyzed the differences between the 2
std(X), for multivariate statistical analysis. Initially, we groups with independent sample t-tests. The signifi-
used unsupervised PCA to demonstrate general sepa- cance of linoleic acid, oleic acid, and DHEA-S was as-
ration in all samples. Next, we performed orthogonal sessed with analysis of covariance (ANCOVA) by ad-
partial least-squares discriminant analysis (OPLS-DA). justing for potential covariates [body mass index
The parameters of the modeling (PCA and OPLS-DA), (BMI), TG, total cholesterol (TC), 2-h PG, blood pres-
R2 and Q2, were used to evaluated the model. R2 was sure, and insulin]. Data were considered statistically
the fraction of the variation of the variables (X or Y) significant at P ⬍ 0.05. We performed ROC curve anal-
explained by the model, and Q2 represented the frac- ysis (16 ) on the basis of a logistic regression model to
tion of the variation of the variables (X or Y) predicted determine the area under the curve (AUC) as a mea-
by the model. Furthermore, R2cum and Q2cum repre- sure for comparing the predictive ability of metabolites
sented the accumulated values of R2 and Q2 from (linoleic acid, oleic acid, and DHEA-S) and known risk
several principal components in the model. When factors for diabetes (age, sex, weight, height, BMI, TG,
the values of R2cum and Q2cum were ⬎0.5, the model TC, waist circumference, and blood pressure). We per-
was considered successful. To avoid model overfit- formed t-tests and ANCOVA with SPSS software (ver-

1340 Clinical Chemistry 59:9 (2013)


Serum Metabolomics in IPD Patients

Table 1. Demographic and clinical chemistry characteristics of participants in the PRFD-Harbin study.a

Control IPD T2DM

n 49 51 52
Age, years 52.3 (6.8) 53.7 (7.4) 52.2 (6.5)
Sex, F/M 35/24 31/20 31/21
2 b
BMI, kg/m 22.0 (1.4) 26.9 (3.0) 26.6 (3.3)b
b
Diastolic blood pressure, mmHg 68.3 (6.3) 86.2 (13.7) 85.0 (10.5)b
b
Systolic blood pressure, mmHg 112.8 (6.3) 142.1 (25.7) 142.8 (24.0)b
FPG, mg/dL 77.4 (9.5) 91.7 (11.4) 194.8 (50.6)b
b
2-h PG, mg/dL 81.2 (16.4) 255.7 (81.4) 338.6 (85.1)b
b
TG, mg/dL 80 (27) 177 (97) 213 (97)b
b
TC, mg/dL 162.4 (27.1) 201.1 (50.3) 209.9 (39.6)
Insulin, mU/L 6.04 (2.26) 6.72 (3.73)b 8.27 (4.47)b
b
Hb A1c, % 5.3 (0.4) 6.2 (0.5) 7.6 (1.2)b
b
HOMA-IS 226.65 (136.17) 92.59 (47.62) 19.09 (9.69)b
b
HOMA-IR 1.25 (0.56) 1.85 (0.77) 4.01 (1.70)b
a
Data are mean (SD). To convert glucose concentrations in mg/dL to mmol/L, multiply by 0.05551; to convert total cholesterol concentrations in mg/dL to mmol/L,
multiply by 0.02586; to convert triglycerides concentrations in mg/dL to mmol/L, multiply by 0.01129.
b
P ⬍0.05 compared with healthy controls.

sion 13.0; SPSS) and ROC with SigmaPlot 11.0 soft- sults obtained indicated clear separations in the 3 sub-
ware (Systat Software). comparison groups (Fig. 1).
To further identify the significant serum metabo-
Results lites that effectively distinguish IPD from the other
groups, 3 subcomparisons were performed by use of
BIOCHEMICAL CHARACTERISTICS OPLS-DA analyses, which revealed distinct separation
In comparison with the control group, significant dif- between the 3 groups (see online Supplemental Fig. 3).
ferences in insulin and homeostasis model assessment On the basis of the permutation test, all R2cum and
for insulin sensitivity (HOMA-IS) and for insulin re- Q2cum values were lower than the original values in the
sistance (HOMA-IR) values were observed in the IPD validation plot (see online Supplemental Fig. 4).
and T2DM groups (P ⬍ 0.05) (Table 1). Notably, BMI, The subcomparison results confirmed the validity of
2-h PG, TG, TC, and systolic and diastolic blood pres- the OPLS-DA model.
sure were not found to be significantly different be-
tween the IPD and T2DM groups. IDENTIFICATION OF POTENTIAL METABOLITES
We calculated VIP values in the OPLS-DA model. The
UPLC-QTOF-MS DATA ANALYSIS metabolite ions with VIP values ⬎1.0 were initially se-
The PCA scores plot from all samples in ESI⫹ revealed lected as distinguishing factors, and the most signifi-
separation between T2DM patients and the 2 other cant variables contributing to class separation were se-
groups, whereas IPD samples were dispersed among lected (Table 2). We identified 11 metabolites using
those of the control group (see online Supplemental commercial standards. Four metabolites were identi-
Fig. 2A). In ESI⫺, we observed a separation between fied by use of tandem MS (MS/MS) analysis or com-
control and other groups (see online Supplemental Fig. parison of the extract mass. The concentrations of spe-
2B), whereby IPD samples were dispersed among those cific metabolites were significantly higher in IPD and
of T2DM. In general, there may be an overlap in classes T2DM patients, compared with the control group, in-
owing to Q2 values (PCA) ⬍0.4 – 0.6, and the model cluding linoleic acid, oleic acid, docosanoic acid,
may lead to multiple misclassifications. To manage this cholesteryl-␤-D-glucoside, and 1,2-distearoyl phos-
issue, 3 subcomparisons were performed by use of PCA phatidyl serine, whereas other metabolites, such as lyso-
to compare control and IPD patients, control and phosphatidylcholine (lysoPC), lysophosphatidyl ethanol-
T2DM patients, and IPD and T2DM patients. The re- amine (lysoPE), DHEA-S, and 5-hydroxykynurenine,

Clinical Chemistry 59:9 (2013) 1341


Fig. 1. PCA score plots between 3 subgroups.
(A), Healthy control and IPD groups (ESI⫹, 5 components, R2Xcum ⫽ 0.578, Q2cum ⫽ 0.609). (B), Healthy control and T2DM
groups (ESI⫹, 5 components, R2Xcum ⫽ 0.561, Q2cum ⫽ 0.639). (C), IPD and T2DM groups (ESI⫹, 5 components, R2Xcum ⫽
0.501, Q2cum ⫽ 0.598). (D), Healthy control and IPD groups (ESI⫺, 5 components, R2Xcum ⫽ 0.512, Q2cum ⫽ 0.607). (E),
Healthy control and T2DM groups (ESI⫺, 5 components, R2Xcum ⫽ 0.743, Q2cum ⫽ 0.612). (F), Healthy control and IPD groups
(ESI⫺, 5 components, R2Xcum ⫽ 0.853, Q2cum ⫽ 0.650). Œ, Healthy control; Œ, IPD; , T2DM. t[1], t[component 1] (the scores
of the first component); t[2], t[component 2] (the scores of the second component).

were significantly lower in these groups. Bar plots based ated an OPLS-DA model using UPLC-QTOF-MS data
on the relative intensities of differential metabolites as the X-matrix and relative intensity of several metab-
[lysoPC(16:0), lysoPC(18:0), linoleic acid, oleic acid, olites in the differential metabolic pathways as the
DHEA-S, and 5-hydroxykynurenine] from different met- Y-matrix. The t-predicted scores plot (see online Sup-
abolic pathways are depicted in Fig. 2. plemental Fig. 5) indicates that linoleic acid, oleic acid,
and DHEA-S are more significant than lysoPC, since
PREDICTION OF CRITICAL METABOLITES FOR IPD more individuals in the control group are localized be-
To establish the optimal serum metabolites that can be low the baseline, whereas the majority of patients are
effectively used to discriminate IPD patients, we gener- correctly assigned to the IPD group.

1342 Clinical Chemistry 59:9 (2013)


Table 2. Important metabolites in UPLC/MS positive and negative ion modes (ESIⴙ and ESIⴚ).

IPD vs control T2DM vs control IPD vs T2DM


Ion
a b c
Compound name mode m/z VIP P Trend VIP P Trend VIP P Trend Related metabolism
d ⫹ ⫺6 ⫺23 ⫺15
LysoPC(16:0) ESI 496.3445 22.47 1.31 ⫻ 10 2 29.79 2.53 ⫻ 10 2 5.69 1.75 ⫻ 10 2 Phospholipid metabolism
Serum Metabolomics in IPD Patients

LysoPC(18:2)d ESI⫹ 520.3426 19.64 3.12 ⫻ 10⫺6 2 27.56 4.16 ⫻ 10⫺23 2 5.04 4.45 ⫻ 10⫺5 2 Phospholipid metabolism
LysoPC(18:0)d ESI⫹ 524.3774 18.88 2.67 ⫻ 10⫺6 2 25.99 2.61 ⫻ 10⫺32 2 4.86 3.06 ⫻ 10⫺5 2 Phospholipid metabolism
d ⫹ ⫺15 ⫺31 ⫺11
Docosanoic acid ESI 358.3689 8.27 4.12 ⫻ 10 1 6.43 4.36 ⫻ 10 1 2.17 1.67 ⫻ 10 1 FFA biosynthesis
Cholesteryl-␤-D-glucosidee ESI⫹ 566.4370 4.56 0.0023 1 19.09 1.26 ⫻ 10⫺18 1 1.24 2.08 ⫻ 10⫺16 1 Cholesterol metabolism
Cholesteryl-␤-D-glucoside ESI⫹ 341.2445 3.05 2.91 ⫻ 10⫺8 1 3.52 4.75 ⫻ 10⫺23 1 1.19 1.37 ⫻ 10⫺4 1 Cholesterol metabolism
fragmente
1,2-Distearoyl phosphatidyl ESI⫹ 792.5965 2.53 2.23 ⫻ 10⫺4 1 3.67 4.38 ⫻ 10⫺18 1 2.08 3.61 ⫻ 10⫺5 1 Glycine, serine, and
serinee threonine metabolism
Linoleic acidd ESI⫺ 279.2358 44.41 2.65 ⫻ 10⫺8 1 41.95 1.93 ⫻ 10⫺8 1 6.75 0.047 1 FFA metabolism
d ⫺ ⫺5
Oleic acid ESI 281.2491 9.02 3.10 ⫻ 10 1 12.55 1.71 ⫻ 10⫺7 1 5.14 0.0017 1 FFA metabolism
d ⫺
DHEA-S ESI 367.1608 14.44 0.042 2 20.05 0.0078 2 6.17 0.014 2 Steroid hormone
biosynthesis
LysoPE(20:1/0:0)d ESI⫺ 506.3279 1.08 5.17 ⫻ 10⫺4 2 2.34 9.63 ⫻ 10⫺18 2 1.45 1.18 ⫻ 10⫺6 2 Phospholipid metabolism
d ⫺ ⫺11
LysoPE(20:2/0:0) ESI 504.3194 1.22 0.0025 2 2.29 8.27 ⫻ 10 2 2.34 2.17 ⫻ 10⫺5 2 Phospholipids metabolism
d ⫺ ⫺5 ⫺14 ⫺12
LysoPE(20:0/0:0) ESI 508.3342 1.24 7.55 ⫻ 10 2 1.88 5.14 ⫻ 10 2 1.58 3.37 ⫻ 10 2 Phospholipid metabolism
LysoPC(18:0)d ESI⫺ 568.3619 2.07 0.0083 2 2.89 1.81 ⫻ 10⫺7 2 2.34 4.59 ⫻ 10⫺4 2 Phospholipid metabolism
5-Hydroxykynureninee ESI⫺ 223.0672 1.32 0.0088 2 1.73 5.43 ⫻ 10⫺4 2 1.24 0.0038 2 Tryptophan catabolism
a
Variable importance in the projection (VIP) was obtained from OPLS-DA with a threshold of 1.0.
b
P values calculated from Student t-test.
c
Fold change calculated from the arithmetic mean values of each group. Trend 1, higher than control levels; 2, lower than control levels.
d
Identified based on retention time and MSn spectrum of an authentic standard.
e
Identified by the accurate mass and observed MSn fragments acquired from UFLC-QTOF /MS analysis and the online database.

Clinical Chemistry 59:9 (2013) 1343


Fig. 2. Bar plots of the relative intensity of typical differential metabolites obtained from serum samples in the
PRFD-Harbin study.
(A), LysoPC(16:0). (B), LysoPC(18:0). (C), Linoleic acid. (D), Oleic acid. (E), DHEA-S. (F), 5-Hydroxykynurenine. *P ⬍ 0.05, **P ⬍
0.01, ***P ⬍ 0.001. P values were calculated by comparing IPD or T2DM samples with controls.

QUALIFICATION OF THE 3 METABOLITES AND ROC ANALYSIS tively. AUCs of the combined diabetes risk factors (DRFs)
We measured the absolute concentrations of linoleic were 0.881 between control and IPD patients and 0.886
acid, oleic acid, and DHEA-S in the 3 groups (Figs. between control and T2DM patients (see online Supple-
3A–C). Compared with the control group, DHEA-S mental Table 2). The diagnostic sensitivity and specificity
concentrations were significantly lower in IPD (P ⬍ of prediction from the ROC curve was ⬎72%.
0.001) and T2DM groups (P ⬍ 0.001), whereas linoleic
and oleic acid concentrations were significantly higher VALIDATION OF THE METABOLITES FOR IPD
in the 2 patient groups (P ⬍ 0.001). Moreover, we ob- We quantitatively measured the 3 metabolites in inde-
served significantly higher amounts of DHEA-S (P ⫽ pendent samples from IPD and non-IPD subjects (in-
0.024) and lower concentrations of linoleic acid (P ⫽ cluding controls and T2DM patients), as shown in Fig.
0.047) and oleic acid (P ⬍ 0.01) in the IPD group, 3, D–F. Trends similar to those obtained with the me-
compared with the T2DM group. tabolite study were observed for all 3 compounds.
AUCs for linoleic acid, oleic acid, and DHEA-S are In ROC analysis (Fig. 4), AUC values for the 3
shown in online Supplemental Table 2. AUCs (95% CI) metabolites were 0.821, 0.780, and 0.739 for IPD and
for the 3 parameters were 0.987, 0.926, and 0.772 be- non-IPD subjects, and that of the combined DRFs was
tween control and IPD patients, and 0.990, 0.924, and 0.746. To optimize the diagnostic ability of the 3 me-
0.876 between control and T2DM patients, respec- tabolites, we performed ROC analysis of the 3 com-

1344 Clinical Chemistry 59:9 (2013)


Serum Metabolomics in IPD Patients

Fig. 3. Quantitative analysis of 3 metabolites in 2 studies.


(A–C), PRFD-Harbin study. (D–F), Validation study. (A) and (D), Linoleic acid. (B) and (E), Oleic acid. (C) and (F), DHEA-S.
ANCOVA analysis by adjusting for BMI, TG, TC, 2-h PG, blood pressure. and insulin. ***P ⬍ 0.001. P values were calculated
by comparing IPD or T2DM samples with controls. ˆP ⬍ 0.05, ˆP ⬍ 0.01, IPD compared with T2DM samples.

bined metabolites. AUC for discrimination between Discussion


IPD and non-IPD subjects was 0.849 (Fig. 4). Compar-
ison of ROC curve areas disclosed statistically signifi- Our results suggest distinct ESI⫹ and ESI⫺ modes be-
cant differences between the combined metabolites tween the IPD and T2DM groups (Fig. 1). There are
and DRFs (P ⫽ 0.0382), and no differences among li- different pathogenic mechanisms underlying IPD and
noleic acid, oleic acid, DHEA-S, and DRFs. Further- T2DM (FPG ⱖ126 mg/dL [ⱖ7.0 mmol/L] and 2-h PG
more, we observed significant differences between ⱖ200 mg/dL [ⱖ11.1 mmol/L]) (17 ). IPD is character-
combined metabolites and DHEA-S (P ⫽ 0.0349), but ized by increased peripheral insulin resistance and re-
not among linoleic acid, oleic acid, and combined me- duced first-phase insulin secretion, and T2DM by pro-
tabolites (P ⬎ 0.05). These results indicate that a com- gressive increase in peripheral and hepatic insulin
bination of the three metabolites can be effectively used resistance and reduced basal and first-phase insulin se-
for distinguishing IPD in patients. cretion (17 ), signifying that T2DM is a more serious

Clinical Chemistry 59:9 (2013) 1345


with this finding, Wang et al. (20 ) reported that
branched-chain amino acids and aromatic amino acids
could be effectively used to predict T2DM in a prospec-
tive cohort study. In our experiments, 15 metabolites
involved in different biochemical metabolic pathways,
such as phospholipids, FFAs, and steroid hormones,
were distinguishable between the healthy control and 2
patient groups. Earlier studies have demonstrated that di-
abetes is intimately associated with metabolic disorders of
lipids, such as phospholipids (21, 22 ). Phospholipids are
important constituents of biomembranes that participate
in various biological pathways (23, 24 ) and are therefore a
particular focus of attention in metabolomic studies. Our
results suggest that 7 metabolites belong to the phospho-
lipid group, mainly lysoPC and lysoPE.
Notably, lysoPC(16:0), lysoPC(18:2), and lyso-
PC(18:0) were found to be important phospholipid
metabolites, owing to their high VIP values. LysoPC, an
endogenous phospholipid, contributes to pathophysi-
Fig. 4. ROC analysis for discrimination of IPD and
ologic changes in humans (25 ). Various species of ly-
non-IPD groups by DRFs, linoleic acid, oleic acid,
soPC, with different physical and biological properties,
DHEA-S, and the combination of the 3 metabolites.
have been identified on the basis of FFA chain length
and degree of saturation (26 ). Our research showed
DRFs: ROC analysis of the combined known risk factors of
that 3 lysoPC species are downregulated in the IPD and
diabetes including age, sex, weight, height, BMI, TG, TC,
T2DM groups, possibly owing to low activity of phos-
waist circumference, and blood pressure. †P ⬍ 0.05 com-
pholipase A2, which catalyzes PC hydrolysis to lysoPC.
pared with the combined 3 metabolites.
Alternatively, these changes are related to deficiency of
lecithin-cholesterol acyltransferase (27 ). Consistent
with our findings, recent studies have shown that de-
metabolic disturbance than IPD. We detected higher creased concentrations of lysoPC(18:2) are associated
insulin concentrations in T2DM than IPD cases (Table with increased risk of T2DM (12, 19 ). Furthermore,
1), which may be the basis for more serious disorders of lysoPC(18:2) was significantly altered in patients with
lipid, steroid hormone, and glucose metabolism in impaired glucose tolerance (IGT) and was identified as
these patients. Conversely, metabolic disorders (e.g., an IGT-specific marker (18 ). IPD is a more serious
imbalances in FFAs and DHEA-S) may lead to in- form of diabetes than IGT in terms of onset and devel-
creased islet damage that contributes to IPD and opment. Therefore, low concentrations of lysoPC
T2DM development. Our experiments revealed signif- (18:2) may predict the risk of IPD. However, imbalance
icantly higher DHEA-S and lower linoleic and oleic of lysoPC appears to be involved in several other met-
acid concentrations in the IPD group, compared with abolic disorders, and 3 lysoPC metabolites have been
T2DM. The high DHEA-S concentration should aid in identified as markers of disorders such as chronic renal
improving insulin resistance, and lower linoleic acid and failure and Alzheimer disease (28, 29 ). Hence, despite our
oleic acid concentrations in reducing islet damage. These finding that 3 lysoPC metabolites are critical for identify-
results collectively demonstrate metabolic differences be- ing IPD patients, the specificity of these metabolites may
tween IPD and T2DM, among which 3 metabolites limit application in the clinical diagnosis of IPD.
(DHEA-S, linoleic acid, and oleic acid) are of significant Another important lipid change in IPD and T2DM
benefit in distinguishing between IPD and T2DM. patients is the dramatic increase in FFAs, which leads to
Recently, metabolic profiling technology has been accumulation of FFAs in blood. Although an acute in-
used to analyze significant metabolic variations in the crease of blood FFAs is necessary for insulin secretion,
sera of patients with prediabetes and diabetes. Wang- chronic exposure to high concentrations leads to apo-
Sattler et al. (18 ) reported marked alterations in the ptosis of pancreatic islet ␤-cells and aggravation of in-
concentrations of lysoPC (18:2) and acetylcarnitine in sulin resistance (30, 31 ), as well as presenting a major
patients with prediabetes. Furthermore, Floegel et al. risk factor for cardiovascular disease and sudden death
(19 ) reported that glucose metabolites, amino acids, in patients with insulin resistance (32 ). In our study,
and choline-containing phospholipids are associated the concentrations of 3 different FFAs (linoleic, oleic,
with higher risk of T2DM in the early stages. Consistent and docosanoic acids) were increased in the IPD and

1346 Clinical Chemistry 59:9 (2013)


Serum Metabolomics in IPD Patients

T2DM groups. By use of GC-MS and UPLC-MS, lino- nose IPD, specificity of diagnosis was not high. It must
leic and oleic acids were identified as the crucial metab- be noted that these metabolites are present in other
olites for distinguishing IPD. In a previous study by our diseases and do not demonstrate a comprehensive re-
group (13 ), linoleic and oleic acids have been identified sponse to metabolic alterations in the body. However, a
as important metabolites of IPD via detection of FFA combination of the metabolites facilitated more accu-
profiles in patients by use of GC-MS. The utility of rate assessment of IPD, with a high AUC value (Fig. 4).
linoleic acid and oleic acids was validated, as both ana- Therefore, a combination of the linoleic acid, oleic
lytical platforms revealed the same pattern of alteration acid, and DHEA-S metabolites in fasting serum might
(upregulation) in different studies. Validation by use be helpful for application to the clinical identification
of different instruments and populations facilitates of patients with IPD.
greater confidence of measurements for metabolite In summary, data obtained with our metabolo-
biomarker identification. Our ROC results confirmed mics approach reveal significant metabolic differences
that linoleic and oleic acids are effective in predicting between IPD patients and healthy controls. From a
IPD. Linoleic acid is an essential FFA in the body af- panel, several significant metabolites in IPD were iden-
fected by diet. In agreement with our results, Li et al. tified, including linoleic and oleic acids and DHEA-S,
(33 ) reported that linoleic acid is an important metab- indicative of lipid and steroid hormone metabolic dis-
olite in diabetes. However, those earlier studies did not orders in IPD patients. In particular, linoleic acid, oleic
consider the influence of diet. Therefore, it is not clear acid, and DHEA-S were associated with greater confi-
whether changes in linoleic acid are the result or cause dence in the specificity and accuracy of IPD identifica-
of IPD, and the utility and specificity of linoleic acid in tion in patients. On the basis of the collective findings,
the diagnosis of IPD require further confirmation. In- we propose that measurement of these metabolites in
terestingly, a combination of metabolites (Fig. 4) ap- fasting serum effectively aids in the clinical identifica-
pears more beneficial in improving specificity and abil- tion of patients with IPD.
ity of IPD diagnosis.
One highlight of our study is the discovery of the
previously unidentified metabolite, DHEA-S, present
at low concentrations in IPD and T2DM patients.
Author Contributions: All authors confirmed they have contributed to
DHEA-S, which is converted to the active form, dehy-
the intellectual content of this paper and have met the following 3 re-
droepiandrosterone (DHEA) in a linear manner, rep- quirements: (a) significant contributions to the conception and design,
resents the circulating hormonal pool of DHEA and acquisition of data, or analysis and interpretation of data; (b) drafting
presents a marker for DHEA availability. DHEA and or revising the article for intellectual content; and (c) final approval of
DHEA-S, the most abundant circulating adrenal ste- the published article.
roids in humans, have been shown to increase glucose Authors’ Disclosures or Potential Conflicts of Interest: Upon man-
uptake rates and affect glucose transporters in cell ex- uscript submission, all authors completed the author disclosure form.
periments (34 ). Epidemiologic studies have shown Disclosures and/or potential conflicts of interest:
that low serum DHEA-S concentrations contribute to Employment or Leadership: None declared.
insulin resistance (35 ), whereas DHEA supplementa- Consultant or Advisory Role: None declared.
tion appears to improve insulin sensitivity and retard Stock Ownership: None declared.
T2DM progression (36 ). Moreover, high insulin con- Honoraria: None declared.
Research Funding: L. Liu, Natural Science Foundation of China (no.
centrations in IPD and T2DM patients may suppress
81202186), Youth Seed Fund of Public Health College of Harbin
the concentration of DHEA-S by inhibiting production Medical University, as a contribution from Department of Nutrition
and promoting clearance (37 ). In the current report, se- and Food Hygiene at Harbin Medical University; L. Na, National
rum DHEA-S was identified as an important metabolite Natural Science Foundation of China (no. 81202188); Y. Li, grant
in IPD patients for the first time. The t-predicted scores NCET-10 – 0148 from the Department of Nutrition and Food Hy-
plot indicated that IPD can be predicted with DHEA-S. giene at Harbin Medical University; C. Sun, Natural Science Foun-
dation of China (no. 81130049), National High Technology Research
To facilitate the clinical application of DHEA-S, its abso- and the National 12th Five-Year Scientific and Technical Support
lute concentration in fasting serum was detected by use of Program of China (2012BAI02B00).
ELISA. The results indicate lower concentrations of Expert Testimony: None declared.
DHEA-S in IPD and T2DM patients than healthy con- Patents: None declared.
trols, consistent with metabolomic data. Moreover, ROC Role of Sponsor: The funding organizations played a direct role in
analyses in 2 separate studies confirmed the utility of the review and interpretation of data, and the preparation and ap-
DHEA-S in fasting serum for predicting IPD. proval of manuscript.
Repeated quantitative analyses of the 3 metabo- Acknowledgments: The authors gratefully acknowledge the skillful
lites were additionally performed. Interestingly, al- technical assistance of Guo-wang Xu and thank International Sci-
though the separate metabolites could be used to diag- ence Editing for editorial assistance.

Clinical Chemistry 59:9 (2013) 1347


References
1. The DECODE-study group. European Diabetes Ep- phy/electrospray mass spectrometry and multi- folding the pathophysiological role of bioactive
idemiology Group. Diabetes Epidemiology: Col- variate statistical analysis. Anal Chem 2005;77: lysophospholipids. Curr Drug Targets Immune En-
laborative Analysis of Diagnostic Criteria in Eu- 4108 –16. docr Metabol Disord 2003;3:23–32.
rope. Is fasting glucose sufficient to define 13. Liu LY, Li Y, Guan C M, Li K, Wang C, Feng RN, 26. Kougias P, Chai H, Lin PH, Lumsden AB, Yao Q,
diabetes? Epidemiological data from 20 European et al. Free fatty acid metabolic profile and bio- Chen C. Lysophosphatidylcholine and secretory
studies. Diabetologia 1999;42:647–54. markers of isolated post-challenge diabetes and phospholipase A2 in vascular disease: mediators
2. Resnick HE, Harris MI, Brock DB, Harris TB. Amer- type 2 diabetes mellitus based on GC-MS and of endothelial dysfunction and atherosclerosis.
ican Diabetes Association diabetes diagnostic cri- multivariate statistical analysis. J Chromatogr B Med Sci Monit 2006;12:RA5–16.
teria, advancing age, and cardiovascular disease 2010;878:2817–25. 27. Gillett MPT, Obineche EN, Lakhani MS, Abdulle
risk profiles: results from the Third National 14. Expert Committee on the Diagnosis and Classifi- AM, Amirlak I, Rukhaimi MA, Suleiman MN. Lev-
Health and Nutrition Examination Survey. Diabe- cation of Diabetes Mellitus. Report of the Expert els of cholesteryl esters and other lipids in the
tes Care 2000;23:176 – 80. Committee on the Diagnosis and Classification of plasma of patients with end-stage renal failure.
3. Nicholson JK, Lindon JC, Holmes E. ‘Meta- Diabetes Mellitus. Diabetes Care 1997;20:1183–
Ann Saudi Med 2001;21:283– 6.
bonomics’: understanding the metabolic re- 97.
28. Jia LW, Chen J, Yin PY, Lu X, Xu GW. Serum
sponses of living systems to pathophysiological 15. Eriksson L, Johansson E, Kettaneh-Wold N, Wold
metabonomics study of chronic renal failure by
stimuli via multivariate statistical analysis of bi- S. Multi- and megavariate data analysis. Princi-
ultra performance liquid chromatography coupled
ological NMR spectroscopic data. Xenobiotica ples and applications. Umeå (Sweden): Umetrics
with Q-TOF mass spectrometry. Metabolomics
1999;29:1181–9. Academy; 2001.
2008;4:183–9.
4. Wikoff WR, Gangoiti JA, Barshop BA, Siuzdak G. 16. Greiner M, Pfeiffer D, Smith RD. Principles and
Metabolomics identifies perturbations in human practical application of the receiver-operating 29. Li NJ, Liu WT, Li W, Li SQ, Chen XH, Bi KS, He P.
disorders of propionate metabolism. Clin Chem characteristic analysis for diagnostic tests. Prev Plasma metabolic profiling of Alzheimer’s disease
2007;53:2169 –76. Vet Med 2000;45:23– 41. by liquid chromatography/ mass spectrometry.
5. Xue R, Lin Z, Deng C, Dong L, Liu T, Wang J, Shen 17. Melchionda N, Forlani G, Marchesini G, Baraldi L, Clin Biochem 2010;43:992–7.
X. A serum metabolomic investigation on hepa- Natale S. WHO and ADA criteria for the diagnosis 30. Cnop M, Hannaert JC, Hoorens A, Eizirik DL,
tocellular carcinoma patients by chemical deriva- of diabetes mellitus in relation to body mass Pipeleers DG. Inverse relationship between cyto-
tization followed by gas chromatography/mass index. Insulin sensitivity and secretion in resulting toxicity of free fatty acids in pancreatic islet cells
spectrometry. Rapid Commun Mass Spectrom subcategories of glucose tolerance. Int J Obes and cellular triglyceride accumulation. Diabetes
2008;22:3061– 8. Relat Metab Disord 2002;26:90 – 6. 2001;50:1771–7.
6. Bogdanov M, Matson WR, Wang L, Matson T, 18. Wang-Sattler R, Yu ZH, Herder C, Messias AC, 31. Weinberg JM. Lipotoxicity. Kidney Int 2006;70:
Saunders-Pullman R, Bressman SS, Flint Beal M. Floegel A, He Y, Heim K, et al. Novel biomarkers 1560 – 6.
Metabolomic profiling to develop blood biomark- for pre-diabetes identified by metabolomics. Mol 32. Wyne KL. Free fatty acids and type 2 diabetes
ers for Parkinson’s disease. Brain 2008;131:389 – Syst Biol 2012;8:615. mellitus. Am J Med 2003;115:29S-36S.
96. 19. Floegel A, Stefan N, Yu ZH, Mühlenbruch K, 33. Li X, Xu ZL, Lu X, Yang XH, Yin PY, Kong HW,
7. Nicholson JK, Connelly J, Lindon JC, Holmes E. Drogan D, Joost HG, Fritsche A, et al. Identifica- Ying Yu, et al. Comprehensive two-dimensional
Metabonomics: a platform for studying drug tox- tion of serum metabolites associated with risk of gas chromatography/time-of-flight mass spec-
icity and gene function. Nat Rev Drug Discov type 2 diabetes using a targeted metabolomic trometry for metabonomics: biomarker discovery
2002;1:153– 61. approach. Diabetes 2013;62:639 – 48. for diabetes mellitus. Anal Chim Acta 2009;633:
8. Zeisel S. Nutrigenomics and metabolomics will 20. Wang TJ, Larson MG, Vasan RS, Cheng SS, Rhee 257– 62.
change clinical nutrition and public health EP, McCabe E, Lewis GD, et al. Metabolite pro- 34. Perrini S, Natalicchio A, Laviola L, Belsanti G,
practice: insights from studies on dietary require- files and the risk of developing diabetes. Nat Med Montrone C, Cignarelli A, Minielli V, et al. Dehy-
ments for choline. Am J Clin Nutr 2007;86:542– 8. 2011;17:448 –53. droepiandrosterone stimulates glucose uptake in
9. Zhao XJ, Fritsche J, Wang JS, Chen J, Rittig K, 21. Perassolo MS, Almeida JC, Pra RL, Mello VD,
human and murine adipocytes by inducing GLUT1
Schmitt-Kopplin P, Fritsche A, et al. Metabonomic Moulin CC, Zelmanovitz T, Azevedo MJ, et al.
and GLUT4 translocation to the plasma mem-
fingerprints of fasting plasma and spot urine re- Fatty acid composition of serum lipid fractions in
brane. Diabetes 2004;53:41–52.
veal human pre-diabetic metabolic traits. type 2 diabetic patients with microalbuminuria.
35. Haffner SM, Valdez RA, Mykkanen L, Stern MP,
Metabolomics 2010;6:362–74. Diabetes Care 2003;26:613– 8.
Katz MS. Decreased testosterone and dehydroepi-
10. Zhang J, Yan LJ, Chen WG, Lin L, Song XY, Yan 22. Hsu FF, Bohrer A, Wohltmann M, Ramanadham S,
androsterone sulfate concentrations are associ-
XM, Hang W, et al. Metabonomics research of Ma Z, Yarashesk K, Truk J. Electrospray ionization
diabetic nephropathy and type 2 diabetes melli- mass spectrometric analyses of changes in tissue ated with increased insulin and glucose concen-
tus based on UPLC-oaTOF-MS system. Anal Chim phospholipid molecular species during the evolu- trations in nondiabetic men. Metabolism 1994;
Acta 2009;650:16 –22. tion of hyperlipidemia and hyperglycemia in 43:599 – 603.
11. Huo TG, Cai S, Lu XM, Sha Y, Yu MY, Li FM. Zucker diabetic fatty rats. Lipids 2000;35:839 – 36. Kawano H, Yasue H, Kitagawa A, Hirai N, Yo-
Metabonomic study of biochemical changes in 54. shida T, Soejima H, Miyamoto S, et al. Dehydro-
the serum of type 2 diabetes mellitus patients 23. Berridge MJ. Inositol trisphosphate and calcium epiandrosterone supplementation improves en-
after the treatment of metformin hydrochloride. signaling. Nature 1993;361:315–25. dothelial function and insulin sensitivity in men.
J Pharm Biomed Anal 2009;49:976 – 82. 24. Exton JH. Phosphatidylcholine breakdown and J Clin Endocrinol Metab 2003;88:3190 –5.
12. Wang C, Kong HW, Guan YF, Yang J, Gu JR, Yang signal transduction. Biochim Biophys Acta 1994; 37. Nestler JE, Clore JN, Blackard WG.
SL, Xu GW. Plasma phospholipid metabolic pro- 1212:26 – 42. Dehydroepiandrosterone: the “missing link” be-
filing and biomarkers of type 2 diabetes mellitus 25. Xu Y, Xiao YJ, Zhu K, Baudhuin L M, Lu J, Hong tween hyperinsulinemia and atherosclerosis?
based on high-performance liquid chromatogra- G, Kim KS, Cristina KL, Song L, William FS. Un- FASEB J 1992;6:3073–5.

1348 Clinical Chemistry 59:9 (2013)