Thrombosis Research (2009) 123, 556–564

www.elsevier.com/locate/thromres

REGULAR ARTICLE

The potential biomarkers for thromboembolism

detected by SELDI-TOF-MS
a,b,⁎
Mei Hong a,b,1 , Xiaoping Zhang a,b,1 , Yu Hu , Huafang Wang a,b
,
Wenjuan He a,b , Heng Mei a,b , Jiekai Yu c ,
Tao Guo a,b , Shanjun Song a,b,⁎

a
Institute of Hematology, Affiliated Union Hospital of Tongji Medical College,
Huazhong University of Science and Technology, Wuhan, China
b
Key Laboratory of Biological Targeted Therapy of Hubei Province, Wuhan, China
c
Cancer Institute, Second Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou, China

Received 29 January 2008; received in revised form 19 April 2008; accepted 27 May 2008
Available online 11 July 2008

KEYWORDS Abstract
Acute myocardial infarction;
Biomarker; Introduction: Few studies were concerned about searching for specific biomarkers for
Deep vein thrombosis; thromboembolic (arterial and venous) diseases by the use of Surface-Enhanced Laser
SELDI-TOF-MS; Desorption/Ionization Time-of-Flight Mass Spectrometry (SELDI-TOF-MS).
Thromboembolism Materials and Methods: We screened for potential biomarkers in 69 plasma samples,
including samples from 20 patients with idiopathetic deep vein thrombosis (DVT), 20
patients with acute myocardial infarction (AMI), and 29 healthy controls without a
history of thromboembolism. Pretreated plasma samples were analyzed on the Protein
Biology System IIc plus SELDI-TOF-MS (Ciphergen Biosystems, Fremont, CA). Proteomic
spectra of mass to charge ratio (m/z) were generated by the application of plasma to
immobilized metal affinity capture (IMAC-3) ProteinChip arrays activated with copper.
Results: A pattern of three biomarkers (m/z: 2 667, 5 914, and 6 890 Da, respectively)
with a total accuracy of 100% was selected based on their collective contribution to
the optimal separation between patients with AMI and healthy controls. Another

Abbreviations: SELDI-TOF-MS, Surface-Enhanced Laser Desorption/Ionization Time-of-Flight Mass Spectrometry; DVT, deep vein
thrombosis; AMI, acute myocardial infarction; m/z, mass to charge ratio; IMAC-3, immobilized metal affinity capture; CUS,
compression ultrasonography; PGV, platelet glycoprotein V; ECG, electrocardiogram; PE, pulmonary embolism; PBS, Protein Biology
System; SPA, sinapinic acid; ZUCI-PDAS, Zhejiang University Cancer Institute-ProteinChip Data Analysis System; UDWT, undecimated
discrete wavelet transform; S/N, signal-to-noise ratio; SVM, support vector machine; CV, coefficient of variation; ROC, receiver
operator characteristic; CI, confidence interval; AUC, areas under the ROC curves; SVT, superficial venous thrombosis.
⁎ Corresponding authors. Institute of Hematology, Affiliated Union Hospital of Tongji Medical College, Huazhong University of Science
and Technology, Wuhan 430022, China.
E-mail address: huyu1964_04@yahoo.com.cn (Y. Hu).
1
These authors contributed equally to this work.

0049-3848/$ - see front matter © 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.thromres.2008.05.019
The potential biomarkers for thromboembolism detected by SELDI-TOF-MS 557

spattern consisting of only one biomarker (m/z: 5 914 Da) could totally discriminate
spattern with
patients consisting of only
DVT and one subjects.
control biomarkerFor(m/z: 5 914
further Da) could
analysis totallypatients
between discriminate
with
AMI and those with DVT, a pattern of four biomarkers (m/z: 3 418, 5 271, 33 378,with
patients with DVT and control subjects. For further analysis between patients and
AMI125
68 andDa,
those with DVT, awas
respectively) pattern of four
selected biomarkers
with (m/z: 3 of
a total accuracy 418, 5 271, 33 378, and
82.5%.
68 125 Da, respectively)
Conclusions: was selected
Plasma proteomic with a with
profiling total accuracy of 82.5%.
SELDI-TOF-MS and ProteinChip
Conclusions:provides
technologies Plasma highproteomic profiling
sensitivity with SELDI-TOF-MS
and specificity and patients
in discriminating ProteinChip
with
technologies provides high sensitivity and specificity in discriminating
thrombosis and healthy subjects. The discovered biomarkers might show great patients with
thrombosis
potential forand
earlyhealthy subjects.
diagnosis The discovered
of thromboembolic biomarkers might show great
diseases.
potential
© for early
2008 Elsevier Ltd.diagnosis
All rightsofreserved.
thromboembolic diseases.
© 2008 Elsevier Ltd. All rights reserved.

Introduction is a rapid and sensitive proteomic technique that

Acute myocardial infarction (AMI) and deep vein
thrombosis (DVT) are two major thromboembolic
diseases all over the world. The annual incidence of
AMI is estimated to be about 600 per 100 000 for
men and 200 per 100 000 for women [1]. The
incidence of DVT is about 50–67 per 100 000 [2,3].
Despite adequate thrombolytic therapy, the mor-
talities of AMI and DVT patients, in whom cardio-
genic shock or pulmonary embolism develops,
remain to be high [4–6]. For suspected DVT, venous
compression ultrasonography (CUS) is currently the
noninvasive imaging test. Full compressibility of a
venous segment excludes thrombosis with a pooled
sensitivity of 93.8% for proximal DVT and 56.8% for
distal DVT [7]. Among biomarkers for diagnosis of
thrombosis, D-dimer plasma levels below a certain
cut-off can be used for the exclusion of DVT [8].
Recently, a study about the D-dimer/fibrinogen
ratio to predict DVT indicated that the ratio was
significantly higher among patients with DVT com-
pared with those without (P b 0.0001) [9]. Some
studies also showed that concentrations of soluble
platelet glycoprotein V (PGV) were significantly
increased during the acute clinical course of
unstable angina pectoris and elevated in patients
with AMI [10,11]. However, the usefulness of these
markers needs to be considered with much caution.
Currently, the most reliable method of diagnosing
DVT is contrast venography [12], and the current
diagnosis of AMI is a clinical diagnosis based on
patient symptoms, electrocardiogram (ECG) changes
and highly sensitive biochemical markers, as well as
information collected from various imaging techni-
ques [13]. However, some of these methods cannot be
routinely used for screening purposes because of
their uncomfortable invasion and complex proce-
dures. Therefore, biomarkers with high sensitivity
and specificity are desirous of being identified for
screening and early diagnosis of thromboembolism.
Surface-Enhanced Laser Desorption/Ionization
Time-of-Flight Mass Spectrometry (SELDI-TOF-MS)
has been applied to identify biomarkers in various
diseases, including cancer and infectious diseases
[14–16]. In conjunction with a bioinformatics tool,
SELDI-TOF technology was used for the first time to
analyze proteomic patterns in a protein C-deficient
family with thrombotic episodes occurring before
age 40 [17]. Similarly, it has been used to analyze
the cardiac troponin I forms present in plasma from
patients with myocardial infarction [18].
The purpose of our study was to determine
whether there were certain biomarkers in plasma
useful for accurate diagnosis of thromboembolism.
We therefore applied the technology of SELDI-TOF-
MS with immobilized metal affinity capture (IMAC-3)
ProteinChip arrays activated with copper to the
identification of differentially expressed proteins in
plasma samples from patients with AMI or DVT and
from control subjects.
Materials and Methods
Cases
The thromboembolism group consisted of 20
patients with acute idiopathetic DVTand 20 patients
with AMI. All DVT patients included in the study
fulfilled the following criteria: (1) clinical manifes-
tations suggesting acute thrombosis, (2) CUS indi-
cated positive findings of deep veins of lower limb,
(3) detailed information about laboratory studies
and clinical history showed no any obvious induce-
ment, such as immobilization, surgery, trauma, and
malignancy. And all AMI patients included fulfilled
the following criteria: (1) presenting typical symp-
toms and positive findings of ECG, (2) abnormal
findings of coronary angiography, (3) distinctive
changes of serum myocardial enzymes. Patients
with a previous thromboembolic event were eligi-
ble. In DVT subgroup, 4 patients had a confirmed
venous thrombotic event (2 male and 2 female),
while 6 patients ever had a history of angina
558
pectoris (5 male and 1 female) in AMI subgroup. For
DVT cases, clinical assessment was evaluated on
admission by Wells score table [19]. Thirteen cases
scored 1 (low probability), five cases scored 2
(intermediate probability), and the other two
cases scored 3 (high probability). For AMI patients,
grade of heart function was assessed according to
the Killip classification. 3 patients were classified as
grade I, 13 patients as grade II, and the other 4
patients as grade III. All of these patients didn't
receive anticoagulation or thrombolytic therapy
within one month before inclusion in the study.
Control Subjects
The control group included 29 healthy individuals.
None had any of the following conditions within
3 months before admission: surgery, trauma, frac-
ture, malignancy, hematologic disorder, acute
infection, organ dysfunction, or use of drugs or
comorbidities, e.g. sudden deafness, known to have
great effects on the coagulation and anticoagula-
tion systems. Control subjects with a history of
heart attack, stroke or thromboembolism were
excluded.
Blood Samples
A total of 69 sodium citrate plasma samples were
collected with oral informed consent from all
subjects mentioned above. For all patient and
control specimens, a 4.5-ml blood sample was
drawn into a Vacutainer tube with 3.2% sodium
citrate (anticoagulant/blood ratio, 1:9, Gongdong,
Taizhou, China) from a fasting subject, centrifuged
at 3 500 rpm for 6 minutes at 4 °C within 30–
60 minutes of collection, aliquoted and then stored
at − 80 °C until analysis. Only samples that were
obtained before invasive operations and treat-
ments of anticoagulation and thrombolysis were
performed were included for proteomic analysis.
All plasma samples from the diseased and control
groups were randomized, and the investigator
was blinded to their identity. The demographic
information for all collected samples is shown in
Table 1.
Reagents and Instruments
ProteinChip Biosystems (Ciphergen Protein Biologi-
cal System IIc plus SELDI-TOF-MS) and IMAC-3 chips
were purchased from Ciphergen Biosystems, Inc.
(Fremont, CA). Sinapinic acid was purchased from
Fluka (Buchs, Switzerland). All other reagents were
purchased from Sigma (Saint Louis, MO).
M. Hong et al.
SELDI ProteinChip Array Binding
All plasma specimens were thawed in wet ice and
then centrifuged at 10 000 rpm at 4 °C for 2 minutes.
The supernatants were retained. 10μL of U9 buffer
(9 M Urea, 2% cholamidopropyl-dimethylammonio-
1-propanesulfonate, 1% ditheothreitol) was added
to 5 μL of each thawed plasma sample in a 96-well
cell culture plate, which was then agitated on a
platform shaker at 600 rpm at 4 °C for 30 minutes.
Eight-spot IMAC-3 ProteinChip arrays were activated
with 50 μL of 100 mM CuSO4, agitated at 600 rpm at
room temperature for 10 minutes, followed by a
deionized water rinse (200 μL of deionized water
was added), and then agitated at 600 rpm at room
temperature for 1 minute. After that, the chips
were performed by adding 200 μL of 100 mM sodium
acetate (pH4.0), agitated at 600 rpm at room
temperature for 5 minutes, followed by another
deionized water rinse, and then agitated at 600 rpm
at room temperature for 1 minute. The chips were
then washed with 200μL of PBS (pH7.2), agitated at
600 rpm at room temperature for 5 minutes,
inverted, and air dried. Repeat the whole washing
procedure once. Next, 185 μL of sodium acetate
(pH4.0) was added to the U9/plasma mixture and
was further agitated on a platform shaker at
600 rpm at 4 °C for 2 minutes. 100 μL of diluted
samples were applied to each spot of a bioprocessor
(Ciphergen Biosystems, Fremont, CA) that con-
tained the ProteinChip arrays. The bioprocessor
was then sealed and agitated on a platform shaker
at 600 rpm at 4 °C for 60 minutes. The excess of
plasma mixtures was discarded. The chips were
then washed three times with 200 μL of sodium
acetate and another two times with 200 μL of
deionized water. Finally, the chips were removed
Table 1 General information for patients with
thromboembolism and healthy controls
Variables Control DVT AMI
group
No. of samples 29 20 20
Mean (Range) age (y) 55.6 55.5 59.6
(48–79) (35–80)⁎ (41–77)⁎ #
No. of male sex (%) 16 (55) 11 (55)⁎ 14 (70)⁎ #
In-hospital mortality (n) 0 5(1)§ 0
No. of cases with 0 1 (5)& 0
concomitant PE (%)
⁎ When compared with the Control group, P N 0.05.
# When compared with the DVT group, P N 0.05.
§ The cause of death was brain hemorrhage.
& Low-risk PE (normal blood pressure without cardiac
insufficiency).
DVT, deep vein thrombosis; AMI, acute myocardial infarction;
PE, pulmonary embolism.
The potential biomarkers for thromboembolism detected by SELDI-TOF-MS 559

from the bioprocessor and air dried. Prior to the and 11 734 Da, respectively) which appeared in all of
SELDI-TOF-MS analysis, 1 μL of a saturated solution the selected spectra, the intensity scale was
of sinapinic acid (SPA) in 0.5% (v/v) trifluoroacetic adjusted. Then, the spectra were subjected to
acid and 50% (v/v) acetonitrile was applied onto baseline correction by aligning with a monotone
each chip twice with an interval of 5 minutes and local minimum curve and mass calibration. The peaks
then the chips were air dried. of proteomic profiling were detected and quantified
by an algorithm which takes the maximal height of
SELDI-TOF-MS Analysis every denoised, baseline-corrected, and calibrated
mass spectrum into account. Thirdly, we filtered out
Mass/charge (m/z) spectra of proteins with affinity peaks with a signal-to-noise ratio (S/N) of less than 3.
to the chelated metal surface (containing trypto- Finally, in order to match peaks across spectra, we
phan, cysteine, histidine, or phosphorylated amino pooled the detected peaks if the relative difference
acid residues) were generated in a Ciphergen Protein in their mass sizes was not more than 0.3%. The
Biology System (PBS) IIc plus TOF-MS Reader (Cipher- minimal percentage of each peak, appearing in all of
gen Biosystems). Data were collected by averaging the spectra, was specified to 10. The matched peak
the results of a total of 217 laser shots with an across spectra is defined as a peak cluster. If a
intensity of 210, a detector sensitivity of 9, a high spectrum does not have a peak within a given cluster,
mass to 100 kDa and an optimization range of 2– the maximal height within the cluster will be assigned
20 kDa. External calibration of the instrument was to its peak value. The identified peak clusters were
carried out using the manufacturer's standard (All- normalized together.
in-one Peptide molecular weight standard; Cipher- The preprocessed data were used to establish
gen Biosystems). All arrays binding and SELDI-TOF- models. In this experiment, we used a nonlinear
MS analysis were performed on the same day. support vector machine (SVM) classifier [20], firstly
developed by Vladimir Vapnik, with a radial based
Bioinformatics and Biostatistics function kernel, a parameter Gamma of 0.6, and a
cost of the constrain violation of 19, to discriminate
Analysis of the total experiment data was implemen- the different groups. The diagnostic models were
ted by the Zhejiang University Cancer Institute- evaluated and validated by leave-one-out cross-
ProteinChip Data Analysis System (ZUCI-PDAS, avail- validation approach. The principle of this approach
able at www.zlzx.net), which was designed by Dr. is to take out one sample each time as the test set
Jiekai Yu, including data preprocessing and model and keep the remaining samples as the training set,
building. Firstly, the original data were handled by and then the test is repeated until each sample has
using undecimated discrete wavelet transform been taken once as a test sample.
(UDWT) method and denoising the signals. Secondly, Each peak in the experiment data was estimated
according to three labeled peaks (m/z: 4 583, 6 640, by the P value of a Wilcoxon Rank Sum test. The top
ten peaks with the smallest p-values were selected
for further analysis. Combinations with the highest
Table 2 The descriptive statistics of the top ten accuracy in distinguishing different groups of data
peaks for further analysis based on the AMI and were selected as potential biomarkers. The SVM
healthy control groups model with the highest Youden's index was selected
m/z Intensity in AMI Intensity in P as the model for discriminating different groups.
(n = 20) Control (n = 29) This kind of model was designated as a differential
(mean ± SD) (mean ± SD) pattern.
5914⁎ 4541.66 ± 2008.33 15789.44 ± 3501.28 0.0000000039
The Union Hospital Ethics Committee approved
6121 879.50 ± 257.85 1985.72 ± 869.30 0.0000000743 the study protocol, and this study was completed as
5931 1989.38 ± 757.61 3958.53 ± 1139.23 0.0000002021 a retrospective case-control study with a limited
5957 1233.31 ± 286.37 2038.49 ± 512.06 0.0000002802 sample size.
8939 1518.91 ± 856.39 3703.70 ± 1658.27 0.0000004788
6136 862.66 ± 316.89 1383.45 ± 399.05 0.0000033497
2667⁎ 1760.87 ± 972.98 3499.08 ± 1221.85 0.0000040765 Results
6890⁎ 1879.65 ± 1054.55 5520.44 ± 3321.69 0.0000040765
5262 1025.99 ± 221.24 1638.32 ± 473.97 0.0000049533 Reproducibility of the Protein Profiling
6099 856.20 ± 217.90 1275.31 ± 317.71 0.0000049533
⁎Combinations of the three peaks (pattern 1) with the highest Reproducibility of the IMAC-3 chip with eight spots
accuracy in discriminating patients with AMI and healthy was estimated by applying 10 plasma samples from
subjects. AMI, acute myocardial infarction; m/z, mass to the same healthy subject to each chip in a random
charge ratio.
style. The inter-assay coefficients of variation (CVs)
560 M. Hong et al.

Figure 1 Representative spectra (left panel) and pseudo-gel views (right panel) of the candidate biomarkers from AMI (A 1–2)
vs. healthy subjects (H 1–2). X-axis represents the molecular mass calculation (m/z values), while y-axis represents relative
intensity. (a), (b), and (c) indicate different expressions of the plasma markers with m/z of 5 914, 2 667, and 6 890 Da,
respectively. AMI, acute myocardial infarction; m/z, mass to charge ratio.

for peak mass and normalized intensity of the selected
peak (m/z = 6 640 Da) were 0.03% and 18%, respec-
tively. These values were both consistent with the
reproducibility data for the PBS IIc TOF-MS reported by
the manufacturer (Ciphergen Biosystems).
The differential pattern of patients with AMI
versus healthy subjects (pattern 1)
pattern (named pattern 1), with optimal separation
between patients with AMI and healthy controls,
consisted of three potential biomarkers with m/z of 2
667, 5 914, and 6 890 Da, respectively. The three
peaks were all down-regulated in patients with AMI
when compared with healthy subjects (Fig. 1).
Pattern 1 had a sensitivity and specificity of both
100%, as evaluated by leave-one-out cross-validation.

After noise filtering and peak cluster identification, a
total of 203 qualified peaks were detected. These
peaks were ranked according to the P values of
Wilcoxon Rank Sum tests. The top ten peaks with the
smallest p-values were selected (shown in Table 2),
randomly combined, and fed into SVM. A differential
The differential pattern of patients with DVT
versus healthy subjects (pattern 2)
A total of 203 qualified peaks were detected based
on patients with DVT and healthy subjects. Accord-
ing to Wilcoxon Rank Sum tests, the top ten peaks
The potential biomarkers for thromboembolism detected by SELDI-TOF-MS 561

Table 3 The descriptive statistics of the top ten (shown in Table 4) according to Wilcoxon Rank Sum
peaks for further analysis based on the DVT and tests. For further analysis between cases with AMI
healthy control groups and DVT, another pattern (named pattern 3) was
m/z Intensity in Intensity in P established based on combinations of four potential
DVT (n = 20) Control (n = 29) biomarkers with m/z of 3 418, 5 271, 33 378, and 68
(mean ± SD) (mean ± SD) 125 Da, respectively. In this pattern, the peak of 5
271 Da was lowly expressed in cases with AMI, while
5914⁎ 4639.41 ± 1626.60 16353.65 ± 3696.63 0.0000000039
5931 1964.87 ± 520.38 4109.51 ± 1178.32 0.0000000130
the others were lowly expressed in cases with DVT
6121 891.00 ± 292.74 2060.64 ± 915.91 0.0000000334 (Fig. 3). Pattern 3 had a total accuracy of 82.5% (18
8938 1436.73 ± 901.32 3834.11 ± 1731.97 0.0000002514 cases with AMI and 15 cases with DVT were correctly
6557 869.23 ± 225.51 1993.17 ± 1334.28 0.0000005323 distinguished), as evaluated by leave-one-out cross-
6888 1665.60 ± 782.69 5700.40 ± 3422.78 0.0000006572 validation.
9021 455.17 ± 274.87 1070.00 ± 358.27 0.0000007298
8610 7988.77 ± 7298.30 21637.91 ± 5231.99 0.0000008101
4310 2236.88 ± 1662.21 6026.09 ± 2803.70 0.0000024875 The areas under the Receiver operator char-
6905 1540.77 ± 488.88 2748.49 ± 1142.02 0.0000030346 acteristic (ROC) curves for the candidate bio-
⁎ Pattern 2 consisting of only one peak of 5 914 Da with the markers in pattern 1
highest accuracy in discriminating patients with DVT and
healthy subjects. DVT, deep vein thrombosis; m/z, mass to For the three candidate biomarkers with m/z of 5
charge ratio.
914, 2 667, and 6 890 Da in pattern 1, the areas under
the ROC curves were 1.000 (95% confidence interval
[CI], 1.000–1.000), 0.947 (95% CI, 0.889–1.004), and
with the smallest p-values were selected (shown in 0.929 (95% CI, 0.857–1.002), respectively (Fig. 4).
Table 3). Similarly, a differential pattern (named
pattern 2), consisting of one peak at a m/z of 5
914 Da, was established to distinguish patients with Discussion
DVT from healthy subjects. The peak was lowly
expressed in patients with DVT when compared with Although laboratory studies on biochemical changes
healthy subjects (Fig. 2). Pattern 2 also had a in blood components may provide some useful in-
sensitivity and specificity of both 100%, as evalu- formation, the diagnosis for thromboembolic dis-
ated by leave-one-out cross-validation. eases (venous and arterial) is mainly based on
imaging tests [12,21]. Only the objective presence
The differential pattern to distinguish cases of a thrombus in blood vessels can be regarded as a
with AMI from those with DVT (pattern 3) positive finding by imaging techniques. However,
negative findings of routine imaging tests in persons
A total of 191 qualified peaks were detected based highly suspected with thrombosis couldn't exclude
on cases with AMI and those with DVT. The top ten the presence or forthcoming occurrence of a
peaks with the smallest p-values were selected thrombus. As we all know, plasma contains

Figure 2 Representative spectra (left panel) and pseudo-gel views (right panel) of the candidate biomarker with a m/z of 5
914 Da from DVT (D 1–2) vs. healthy subjects (H 1–2). X-axis represents the molecular mass calculation (m/z values), while y-
axis represents relative intensity. DVT, deep vein thrombosis; m/z, mass to charge ratio.
562 M. Hong et al.

Table 4 The descriptive statistics of the top ten search for specific biomarkers and seek new
peaks for further analysis based on the AMI and DVT diagnostic approaches for both venous and arterial
groups thrombosis.
m/z Intensity in AMI Intensity in DVT P SELDI-TOF-MS is a new proteomic approach that
(n = 20) (n = 20) allows a large number of samples to be analyzed
(mean ± SD) (mean ± SD) simultaneously in a relatively short period of time
[22,23]. It requires only small amounts of sample for
5271⁎ 994.28 ± 257.04 1590.64 ± 975.18 0.0005629036
33378⁎ 1593.64 ± 707.16 1256.65 ± 1124.50 0.0055604599
protein profiling without purification steps in
34457 365.77 ± 121.49 290.04 ± 182.57 0.0060403295 advance, which makes it particularly suitable for
66464 2518.60 ± 1103.08 1867.94 ± 1382.60 0.0065571926 clinical or basic studies. However, SELDI ProteinChip
33470 1491.99 ± 646.88 1197.58 ± 1006.41 0.0071134940 array technologies coupled with sophisticated bioin-
66563 2445.57 ± 1091.94 1861.26 ± 1336.82 0.0083548273 formatics tools are seldom used in the field of
5141 1120.64 ± 171.26 1580.07 ± 876.11 0.0097864867
3428 1340.26 ± 596.66 907.68 ± 361.23 0.0143638476
hemostasis and thrombosis to identify biomarkers
3418⁎ 1230.45 ± 609.24 788.50 ± 318.11 0.0192923797 and perform molecular classification [17]. The
68125⁎ 607.87 ± 220.84 470.57 ± 258.37 0.0192923797 pathogenesis of thrombosis in blood vessels is
⁎Combinations of the four peaks (pattern 3) with the highest complex, involving multifactorial roles [24,25]. In
accuracy in discriminating cases with AMI and those with DVT. the light of this multifactorial nature of thrombogen-
AMI, acute myocardial infarction; DVT, deep vein thrombosis; esis, it seems feasible that a combination of several
m/z, mass to charge ratio. biomarkers can be used to improve the early diagnosis
of thrombotic diseases. In this study, we applied
SELDI-TOF-MS with IMAC-3 ProteinChip arrays acti-
thousands of proteins or peptides that regulate a vated with copper to disclose plasma protein finger-
large number of physiologic functions and may be prints of thromboembolism, thereby, have
related to pathology. Identification of protein established new diagnostic patterns for AMI and
patterns in plasma could allow a valid clinical DVT, respectively.
diagnosis of thromboembolism to be made before Based on the difference of protein profiling
the onset of symptoms and positive findings of between patients with thromboembolism and healthy
imaging tests. Therefore, there is an urgent need to individuals, our study has identified three candidate

Figure 3 Representative spectra (left panel) and pseudo-gel views (right panel) of the candidate biomarkers from AMI (A 3–5)
vs. DVT cases (D 3–5). X-axis represents the molecular mass calculation (m/z values), while y-axis represents relative intensity.
(a), (b), and (c) indicate different expressions of the plasma markers with m/z of 5 271, 33 378, and 3 418 Da, respectively. AMI,
acute myocardial infarction; DVT, deep vein thrombosis; m/z, mass to charge ratio.
The potential biomarkers for thromboembolism detected by SELDI-TOF-MS
Figure 4 ROC curves of proteomic features at m/z of 5 914
(Red), 2 667 (Green), and 6 890 (Blue) Da for differentiating
patients with AMI from healthy subjects. AMI, acute
myocardial infarction; AUC, area under the curve; m/z,
mass to charge ratio.
biomarkers with m/z of 5 914, 2 667, and 6 890 Da,
respectively. A combination of the three biomarkers
(pattern 1, Fig. 1) was established for differentiating
patients with AMI from healthy subjects with a
sensitivity and specificity of both 100%. And the first
biomarker could entirely discriminate patients with
DVT and healthy controls (pattern 2, Fig. 2). The
result is different from another one which was
produced by Svensson AM et al [17]. In their study,
age of the oldest patient is less than 40, while age of
the youngest patient in our study is 35; Their cases
were from a type I protein C-deficient family and
patients with superficial venous thrombosis (SVT)
were also included in their study, but the cases in ours
were unrelated without other obvious risk factors
except a history of thrombosis; Moreover, the types of
ProteinChip arrays used for analysis between their
and ours were also different. These distinctions could
mainly account for the different results. Surprisingly,
all of the three peaks were down-regulated in the two
diseased groups when compared with healthy control
group. To our knowledge, the imbalance of regulatory
mechanisms in the coagulation, anticoagulation, and
fibrinolytic systems plays a critical role in thrombo-
genesis [26]. Therefore, we initially supposed that the
corresponding proteins or peptides should be consti-
tuents of the anticoagulation or fibrinolytic systems.
As we know, classic proteomics is mainly concerned
with searching for specific biomarkers with an
elevated level in blood samples. Our completely
different results indicated that we could explore the
proteomic patterns of AMI or DVT in other directions.
Further analyses of areas under the ROC curves
concerning the diagnostic values of these biomarkers
indicated that the peak with a m/z of 5 914 Da
563
presented a maximal potential for diagnosis of AMI
(Fig. 4). So, this candidate biomarker remains
interesting to be further investigated.
We also compared AMI with DVT cases to look for
their proteomic distinction which could help distin-
guish different types of thrombosis. A combination of
four biomarkers (pattern 3), with m/z of 5 271, 33
378, 3 418, and 68 125 Da, respectively, was
established with a total accuracy of 82.5% (2 cases
in AMI and 5 cases in DVTwere wrongly classified). The
first biomarker was down-regulated in AMI, while the
others were weakly expressed in DVT (Table 4, Fig. 3
Further research is needed to confirm our current
findings in larger cohorts of study samples. Further-
more, identification and quantitation of these plasma
proteins, and elucidation of their possible roles in
healthy blood circulation and in thromboembolic
diseases, remain to be important goals for research.
One of the challenges in the analysis of SELDI-TOF-
MS is about the reproducibility, which is critical for
reliable diagnosis and early detection. In order to
solve the problems in reproducibility and discrepancy,
we followed stringent experimental protocols, includ-
ing plasma samples collection, handling, shipping,
data preprocessing, and statistical analysis [27–29].
During the process of plasma samples collection, we
used citrate anticoagulant in which platelets are most
stable and the amount of blood sample was specified.
After venipuncture, samples were processed within
the first hour during which relatively few changes
occur [30]. To reduce false protein peaks, we used a
sophisticated bioinformatics tool, ZUCI-PDAS, to
analyze the data of spectra, including denoising with
the UDWT, baseline correction, peaks detection,
biomarker selection, and establishment and evalua-
tion of the SVM differential patterns. The algorithm is
likely to identify most of the true, reproducible peaks.
The SVM classification is specifically used to win the
optimal resolution with available information for
finite samples rather than for infinite samples
[31,32]. And leave-one-out cross-validation was uti-
lized to determine the accuracy of the SVM classifier.
All these steps ensured that the selection of candidate
biomarkers was not influenced by systematic bias.
One unique feature of this report is that we used
plasma to perform proteomic profiling. With the
development of new technologies, proteomic test-
ing will become available in routine clinical labora-
tory studies and peripheral blood plasma may
potentially provide more valuable information for
management of patients than serum.
However, one potential drawback of the study was
its relatively small sample size, which might reduce
the validity of generalized conclusions. Therefore,
larger patient populations will be important for
future research.
564
All together, the SELDI-TOF-MS ProteinChip tech-
nology combined with sophisticated bioinformatics
tools can demonstrate that biomarkers are present in
patients with thromboembolic diseases and help
establish differential patterns with high sensitivity
and specificity.
Acknowledgements
This study was supported by grants from the Ministry
of Health of P.R. China ([2007]353) and the National
Natural Science Foundation of P.R. China
(30700332), Beijing. And We thank the Departments
of Vascular Surgery and Cardiology, Union Hospital,
for samples collection.
References
[1] Mowatt G, Vale L, Brazzelli M, Hernandez R, Murray A, Scott N,
et al. Systematic review of the effectiveness and cost-
effectiveness, and economic evaluation, of myocardial
perfusion scintigraphy for the diagnosis and management of
angina and myocardial infarction. Health Technol Assess
2004;8:1–207.
[2] Fowkes FJI, Price JF, Fowkes FGR. Incidence of diagnosed
deep vein thrombosis in the general population: systematic
review. Eur J Vasc Endovasc Surg 2003;25:1–5.
[3] White RH. The epidemiology of venous thromboembolism.
Circulation 2003;1:I4–8.
[4] Hasdai D, Califf RM, Thompson TD, Hochman JS, Ohman EM,
Pfisterer M, et al. Predictors of cardiogenic shock after
thrombolytic therapy for acute myocardial infarction. J Am
Coll Cardiol 2000;35:136–43.
[5] Hirsh J, Bates SM. Prognosis in acute pulmonary embolism.
Lancet 1999;353:1375–6.
[6] Cushman M, Tsai AW, White RH, Heckbert SR, Rosamond WD,
Enright P, et al. Deep vein thrombosis and pulmonary
embolism in two cohorts: the longitudinal investigation of
thromboembolism etiology. Am J Med 2004;117:19–25.
[7] Goodacre S, Sampson F, Thomas S, van Beek E, Sutton A.
Systematic review and meta-analysis of the diagnostic
accuracy of ultrasonography for deep vein thrombosis. BMC
Med Imaging 2005;5:6.
[8] Keeling DM, Mackie IJ, Moody A, Watson HG. The diagnosis of
deep vein thrombosis in symptomatic outpatients and the
potential for clinical assessment and D-dimer assays to reduce
the need for diagnostic imaging. Br J Haematol 2004;124:15–25.
[9] Wuillemin WA, Korte W, Waser G, Lämmle B. Usefulness of
the D-dimer/fibrinogen ratio to predict deep venous
thrombosis. J Thromb Haemost 2005;3:385–7.
[10] Atalar E, Haznedaroglu IC, Kilic H, Ozer N, Coskun S, Ozturk E,
et al. Increased soluble glycoprotein V concentration during
the acute onset of unstable angina pectoris in association with
chronic cigarette smoking. Platelets 2005;16:329–33.
[11] Aleil B, Mossard JM, Wiesel ML, Lanza F, Cazenave JP.
Increased plasma levels of soluble platelet glycoprotein V in
patients with acute myocardial infarction. J Thromb Haemost
2003;1:1846–7.
[12] Kyrle PA, Eichinger S. Deep vein thrombosis. Lancet
2005;365:1163–74.
M. Hong et al.
[13] Alpert JS, Thygesen K, Antman E, Bassand JP. Myocardial
infarction redefined–a consensus document of The Joint
European Society of Cardiology/American College of Cardi-
ology Committee for the redefinition of myocardial infarc-
tion. J Am Coll Cardiol 2000;36:959–69.
[14] Yip TT, Lomas L. SELDI ProteinChip array in oncoproteomic
research. Technol Cancer Res Treat 2002;1:273–80.
[15] Poon TC, Yip TT, Chan AT, Yip C, Yip V, Mok TS, et al.
Comprehensive proteomic profiling identifies serum pro-
teomic signatures for detection of hepatocellular carci-
noma and its subtypes. Clin Chem 2003;49:752–60.
[16] Gravett MG, Novy MJ, Rosenfeld RG, Reddy AP, Jacob T,
Turner M, et al. Diagnosis of intra-amniotic infection by
proteomic profiling and identification of novel biomarkers.
JAMA 2004;292:462–9.
[17] Svensson AM, Whiteley GR, Callas PW, Bovill EG. SELDI-TOF
plasma profiles distinguish individuals in a protein C-
deficient family with thrombotic episodes occuring before
age 40. Thromb Haemost 2006;96:725–30.
[18] Peronnet E, Becquart L, Poirier F, Cubizolles M, Choquet-
Kastylevsky G, Jolivet-Reynaud C. SELDI-TOF MS analysis of
the Cardiac Troponin I forms present in plasma from patients
with myocardial infarction. Proteomics 2006;6:6288–99.
[19] Wells PS, Anderson DR, Bormanis J, Guy F, Mitchell M, Gray
L, et al. Application of a diagnostic clinical model for the
management of hospitalized patients with suspected deep-
vein thrombosis. Thromb Haemost 1999;81:493–7.
[20] Liu Y. Active learning with support vector machine applied
to gene expression data for cancer classification. J Chem Inf
Comput Sci 2004;44:1936–41.
[21] Nissen S. Coronary angiography and intravascular ultra-
sound. Am J Cardiol 2001;87:15A–20A.
[22] Pusch W, Flocco MT, Leung SM, Thiele H, Kostrzewa M. Mass
spectrometry-based clinical proteomics. Pharmacogenomics
2003;4:463–76.
[23] Tang N, Tornatore P, Weinberger SR. Current developments in
SELDI affinity technology. Mass Spectrom Rev 2004;23:34–44.
[24] Rosendaal FR. Venous Thrombosis: The Role of Genes,
Environment, and Behavior. Hematology 2005:1–12.
[25] Ajjan R, Grant PJ. Coagulation and atherothrombotic
disease. Atherosclerosis 2006;186:240–59.
[26] Amiral J, Fareed J. Thromboembolic diseases: biochemical
mechanisms and new possibilities of biological diagnosis.
Semin Thromb Hemost 1996;22(Suppl 1):41–8.
[27] Baggerly KA, Morris JS, Coombes KR. Reproducibility of SELDI-
TOF protein patterns in serum: comparing datasets from
different experiments. Bioinformatics 2004;20: 777–85.
[28] Seibert V, Ebert MP, Buschmann T. Advances in clinical
cancer proteomics: SELDI-TOF-mass spectrometry and
biomarker discovery. Brief Funct Genomic Proteomic
2005;4:16–26.
[29] Liotta LA, Lowenthal M, Mehta A, Conrads TP, Veenstra TD,
Fishman DA, et al. Importance of communication between
producers and consumers of publicly available experimental
data. J Natl Cancer Inst 2005;97:310–4.
[30] Banks RE, Stanley AJ, Cairns DA, Barrett JH, Clarke P,
Thompson D, et al. Influences of blood sample processing on
low-molecular-weight proteome identified by surface-
enhanced laser desorption/ionization mass spectrometry.
Clin Chem 2005;51:1637–49.
[31] Byvatov E, Schneider G. Support vector machine applica-
tions in bioinformatics. Appl Bioinformatics 2003;2:67–77.
[32] Jorissen RN, Gilson MK. Virtual screening of molecular
databases using a support vector machine. J Chem Inf
Model 2005;45:549–61.