Lectin precipitation using phytohemagglutinin-L(4) coupled to avidin-agarose for serological biomarker discovery in colorectal cancer.

Abstract

N-acetylglucosaminyltransferase V (GnT-V) has been reported to be upregulated in malignant cancer cells, and its targets have been sought after with regard to biomarker identification. The low capacity and high false positive rates of 2-DE gel-based lectin blots using phytohemagglutinin-L(4) (L-PHA) prompted us to develop a novel protocol for identifying GnT-V targets, in which serum proteins were subjected to immunodepletion, alkylation, and lectin precipitation using L-PHA coupled to avidin-agarose bead complexes, and tryptic digestion. Proteins captured by L-PHA conjugates were analyzed by a nano-LC-FT-ICR/LTQ MS. Here, we report 26 candidate biomarkers for colorectal cancer (CRC) that show 100% specificity and sensitivities of greater than 50%. Not only can these candidate proteins be used as analytes for validation, but the novel protocol described herein can be applied to biomarker discovery in nonCRCs.

Full text

Page 1

RAPIDCOMMUNICATION
Lectin precipitation using phytohemagglutinin-L4
coupled to avidin–agarose for serological biomarker
discovery in colorectal cancer
Yong-Sam Kim1*, Ok Lye Son1*, Ju Yeon Lee2, Sun Hee Kim1, Sejeong Oh3,
Yoon Suk Lee3, Cheorl-Ho Kim4, Jong Shin Yoo2, Jeong-Hwa Lee5, Eiji Miyoshi6,
Naoyuki Taniguchi6, Samir M. Hanash7, Hyang Sook Yoo1and Jeong Heon Ko1
1Daejeon-KRIBB-FHCRC Research Cooperation Center, KRIBB, 111 Gwahangno, Yuseong-gu, Deajeon, Korea
2Division of Proteome Research, Korea Basic Science Institute, Yuseong-gu, Daejeon, Korea
3Department of Surgery, Our Lady of Mercy Hospital, The Catholic University of Korea, Inchon, Korea
4Department of Biological Sciences, Sungkyunkwan University, Suwon City, Kyunggi-Do, Korea
5Department of Biochemistry, College of Medicine, The Catholic University of Korea, Banpo-Dong, Seoul, Korea
6Department of Biochemistry, Osaka University Medical School/Graduate School of Medicine, Suita, Japan
7Fred Hutchinson Cancer Research Center, Seattle, WA, USA
N-acetylglucosaminyltransferase V (GnT-V) has been reported to be upregulated in malignant
cancer cells, and its targets have been sought after with regard to biomarker identification. The
low capacity and high false positive rates of 2-DE gel-based lectin blots using phytohemaggluti-
nin-L4(L-PHA) prompted us to develop a novel protocol for identifying GnT-V targets, in which
serum proteins were subjected to immunodepletion, alkylation, and lectin precipitation using L-
PHA coupled to avidin–agarose bead complexes, and tryptic digestion. Proteins captured by L-
PHA conjugates were analyzed by a nano-LC-FT-ICR/LTQ MS. Here, we report 26 candidate
biomarkers for colorectal cancer (CRC) that show 100% specificity and sensitivities of greater
than 50%. Not only can these candidate proteins be used as analytes for validation, but the novel
protocol described herein can be applied to biomarker discovery in nonCRCs.
Received: January 14, 2008
Revised: March 24, 2008
Accepted: March 28, 2008
Keywords:
Biomarker / Colorectal cancer / GnT-V / L-PHA
Proteomics 2008, 8, 0000–0000
1
Nearly 800000 new CRC cases are believed to develop
each year globally, which accounts for approximately 10% of
all incident cancers. In addition, the mortality from CRC is
estimated to reach nearly 450000 deaths annually [1]. The
MLH1 and MSH genes are associated with hereditary non-
polyposis CRC [2], and the APC gene is associated with
familial adenomatous polyposis [3], but these factors fail to
accountforthe occurrenceof a wide range ofCRC. Moreover,
CRC is just one of many epithelium-derived cancers in
which circumstantial factors govern over hereditary genetic
factors. These attributes necessitate a clear marker that
serves as a tracer molecule for the efficacious treatment of
CRC, but unfortunately, a discretebiomarkerforCRC has yet
to be discovered.
Traditional approaches have stressed the dynamics of
protein expression levels associated with the biochemical
processes of cancer. In contrast, many lines of evidence have
demonstrated the role of various glycosyltransferases in the
pathogenicity of cancer cells, wherein changes in protein
glycosylation have been reported to be associated with the
Correspondence: Dr. Jeong Heon Ko, Daejeon-KRIBB-FHCRC Re-
search Cooperation Center, KRIBB, 111 Gwahangno, Yuseong-
gu, Deajeon 305-806, Korea
E-mail: jhko@kribb.re.kr
Fax: 182-42-879-8119
Abbreviations: AFP, a-fetoprotein; CRC, colorectal cancer; GlcNAc,
N-acetylglucosamine; GnT-V, N-acetylglucosaminyltransferase V;
ID,immunodepletion;L-PHA,phytohemagglutinin-L4
* Both authors contributed equally to this work.
DOI 10.1002/pmic.200800034
© 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.proteomics-journal.com

Page 2

2
Y.-S. Kim et al. Proteomics 2008, 8, 0000–0000
pathogenic processes of cells [4]. One of the best-character-
ized glycosyltransferases is N-acetylglucosaminyltransferase
V (GnT-V), which catalyzes the formation of a b1,6-N-acet-
ylglucosamine (GlcNAc) moiety in an N-linked core glycan.
An increase in b1,6-branching on N-linked glycans is asso-
ciated with the metastatic potential of cancer cells [5]. Several
target molecules for GnT-V have been proposed to be
involved in cancer progression, including matriptase [6], b1
integrin [7], N-cadherin [8], and TIMP-1 [9]. Although these
proteins were found to be relevant to the biological and
pathological processes in cancers, none of them has demon-
strated any use as a serum biomarker, which may be due to
the fact that molecular fluctuations observed at the cellular
level are buffered by the complexity of blood composition, or
that current proteomics technologies are unable to trace
changes in minute expression level of the proteins in blood.
Despite the strong implication of GnT-V in cancer
pathology, no targets of GnT-V have been identified in serum
as biomarker candidates. Thus, in this report, we attempted
to identify colorectal cancer (CRC)-related GnT-V targets in
serum using phytohemagglutinin-L4(L-PHA), a lectin that
recognizes b1,6-GlcNAc moiety attached to an N-linked core
glycan. Though L-PHA binds rather specifically to b1,6-
GlcNAc moiety of N-glycans, a method of detachment for L-
PHA-based lectin affinity chromatography is not available,
which has limited use ofthe lectin to the gel-based lectin blot
analysis for search for GnT-V targets. However, the gel-based
approach innately has several drawbacks: (i) the 2-DE gel has
limited protein-loading capacity, and (ii) despite the develop-
ment of cutting-edge, high-sensitivity mass spectrometers, a
ostensibly single protein spot can contain several scores of
proteins, in which case selection of a true positive is often
difficult. These two drawbacks make it even more difficult to
mine candidate proteins for GnT-V by lectin blot analysis in
serum that shows high dynamic range of proteins. In this
regard, we developed a simple and reliable protocol wherein
GnT-V targets are captured, precipitated, tryptic-digested,
and identified using an FT-ICR-LTQ mass spectrometer.
Here, we report our optimized protocol, as well as 26 candi-
date biomarkers for CRC discovered through this method.
Serum samples were prepared from CRC patients at Our
Lady of Mercy Hospital at The Catholic University of Korea
(Inchon, Korea) and from healthy volunteers at KRIBB
(Daejeon,Korea), with an agreement ofparticipation from all
subjects. Serum samples were diluted two to three-fold with
an appropriate buffer and subjected to partial purification
performedwith a ProteomPrep 20 plasma immunodepletion
(ID) LC column kit (Sigma) or by Con A-affinity column
chromatography (GE Healthcare). With a ProteomPrep 20
plasma ID LC column (Sigma), 20 highly abundant serum
proteins were removed according to the manufacturer’s
instructions. For Con A-affinity column chromatography,
diluted sera were loaded onto a Con A-agarose column pre-
equilibrated with 50 mM Tris-HCl (pH 7.4); the column was
then washed with equilibration buffer, and bound proteins
were eluted with 0.4 M b-D-methylmanno pyranoside in
50 mM Tris-HCl (pH 7.4). The partially purified samples
were reduced by treatment with 1% v/v b-mercaptoethanol
and alkylated with excess iodoacetamide at room tempera-
ture for 1 h. The modified samples were desalted on a
HiPrep 26/10 desalting column (GE Healthcare) and con-
centrated to a final volume of 1 mL. Protein samples were
precleared with avidin–agarose beads for 1 h at room tem-
perature, and the precleared proteins were allowed to bind to
L-PHA–avidin–agarose or L-PHA–agarose complexes over-
night at 47C. After extensive washing with PBS, the bound
proteins were separated from the bead complexes by adding
16SDS-PAGE denaturation buffer or 6 M urea. Protein
preparations that were denatured by addition ofdenaturation
buffer were in-gel digested as described previously [10]. Pro-
tein preparations denatured in 6 M urea were diluted to
0.6 M for tryptic digestion. The digested peptides were lyo-
philized in a SpeedVac system and analyzed by MS.
The peptide mixtures were loaded onto a C18 trap col-
umn (5 mm, 100mm, 300 mm id65 mm) by an autosampler
(Surveyor) at a flow of 20 mL/min for desalting and con-
centration. The trapped peptides were then back-flushed
and separated on a homemade column (length 100 mm)
packed with C18 resin (Aqua, 5 mm, Phenomenex) in
75 mm silica tubing (8 mm id orifice). The mobile phases A
and B were composed of 0 and 80% ACN, respectively, each
containing 0.5% acetic acid and 0.02% formic acid. The
gradient began at 5% B for 15 min; was ramped to 20% B
for 3 min, 50% for 47 min, and 95% for 2 min; and held at
95% for 5 min, then 5% for 2 min. The column was equili-
brated with 5% B for 6 min before the next run. The eluted
peptides were directly electrosprayed into a mass spectrom-
eter, which was controlled by Xcalibur software (Thermo-
Electron Corporation, Home Page Version 2.0 SR1). During
the gradient elution, three IT MS/MS spectra were acquired
per data-dependent cycle from a high-resolution (R set at
100000) FT-ICR master spectrum. Ions selected for MS/MS
were dynamically excluded for 60 s. Peak lists were gener-
ated, and the resulting .raw files were converted to .xml files
in Bioworks software (ThermoElectron, ver. 3.3). The.xml
files were used for protein identification using MASCOT
search engine version 2.0 (Matrix Science) against the IPI
human database 20070905 (67 524 sequences; 28 722 560
residues). MASCOT was used with monoisotopic mass
selected, a precursor mass tolerance of 61.5 Da, and a
fragment mass tolerance of 60.8 Da. Trypsin was selected
as the enzyme, with one potential missed cleavage. Oxi-
dized methionine, pyro-glutamate (N-term Q), propioamide
cystein and carbamidomethylated cystein was chosen as
variable modifications. With regard to acceptance criteria for
protein identification, proteins that were identified with
more than two peptides among which at least one peptide
shows MASCOT individual ion score more than 42
(p,0.05) were considered to be candidates. Criteria that
proteins must be a known glycoprotein or otherwise have at
least one N-glycosylation consensus motif (N-X-S/T, where
X is any amino acid except proline) were also considered.
© 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.proteomics-journal.com

Page 3

Proteomics 2008, 8, 0000–0000
Glycoproteomics
3
Our experiments constituted the following steps: partial
purificationofserumproteins,lectincaptureandprecipitation,
trypticdigestion,andproteinidentificationviaMSanddatabase
searches. As diagrammed in Fig. 1A, the overall workflow con-
sisted of two separate strategies: one for optimizing protocols
for L-PHA-based lectin capture and another for discovery of
CRC biomarkers based on the optimized protocol. Protocol
optimization included selection of enrichment or fractionation
methods, selection ofthe form of L-PHAconjugates tobe used
for capturing b1,6-GlcNAc-containing N-glycoproteins, and
determination of the steps for tryptic digestion. Each step that
improvedthe end-resultwas chosen and combinedto generate
an “optimized protocol” for biomarker discovery. A primary
reasonastowhytheuseofL-PHAisinappropriateforchroma-
tography originates from the unavailability of a nondenatura-
tion-basedelutionmethod,asopposedtootherwidelyusedlec-
tins such as Con A, Lens culinaris agglutinin, and wheat germ
agglutinin (WGA), which has necessitated of the development
ofamethodtodetachL-PHA-boundproteinsunderdenaturing
conditions.Tothisend,L-PHAboundproteinsweredenatured
ineither6 MureaorSDS-PAGEdenaturationbuffer.Intheso-
lutiondigestionmethod,proteinpreparationsdenaturedin6 M
urea were diluted ten-fold and subjected to tryptic digestion.
Meanwhile, protein preparations denatured in denaturation
buffer were resolved on an 8% SDS-PAGE gel, after which the
entire gel was cut into pieces and in-gel tryptic digestion was
performed. We established criteria to determine which diges-
tionmethodwasmoreeffectiveforproteinidentification,prior-
itized by the following criteria: (i) the number of proteins iden-
tifiedfromatleasttwononoverlappedpeptidesamongwhichat
leastonepeptidehasasignificantMASCOTscore(p,0.05),(ii)
sequence coverage, and (iii) total score values. As is seen in
Figs.1BandC,weidentified28and41proteinssatisfyingthese
criteriafromthein-gelandsolutiondigestionmethods,respec-
tively. In addition, proteins identified through the solution
digestion method had higher total scores and sequence cover-
age compared with those identified through in-gel digestion.
Thus, we chose the solution digestion method as described
aboveforsubsequent protocol optimizationandbiomarkerdis-
covery.
Figure 1. Strategy for protocol
optimization of biomarker dis-
covery using serum. (A) Partial
purification methods (Con A
chromatography or ID), L-PHA
bound forms (L-PHA agarose or
L-PHA-avidine-agarose),
trypticdigestion methods (in-gel
digestion or solution digestion)
werevaried to
effects on protein identification
(left panel). Methods that pro-
duced better results were select-
ed and combined for biomarker
discovery,asoutlinedintheright
panel. (B, C) Partial purification,
L-PHA bound form, and tryptic
digestion methods were varied,
and the effects on sequence cov-
erage (B) and MASCOTscore (C)
of the identified proteins from a
normal serum
investigated.Based
result, ID as a partial purification
method,L-PHA-avidine-agar-
ose,andsolution
methods (sol) were selected and
comprised the optimized proto-
col for biomarker discovery. A
combination of these methods
allowed for a ,2.8-fold increase
in the number of proteins identi-
fied with more than one signifi-
cant ion score (p,0.05). The
results were derived from an
experiment showing the best
result among three independent
experiments.
and
monitorthe
sample were
theon
digestion
© 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.proteomics-journal.com

Page 4

4
Y.-S. Kim et al. Proteomics 2008, 8, 0000–0000
Another issue related to optimization was deciding
which form of L-PHA conjugates – direct conjugation of L-
PHA to glycan-based beads such as agarose or sepharose, or
indirect conjugation of L-PHAto beads with an intermediate
protein – would yield better results. For direct conjugation of
L-PHA to beads, reactive group-activated conjugation of the
lectin to beads was possible. Instead of performing the con-
jugation in our laboratory, we purchased commercially
available L-PHA-bound agarose beads (L-PHA–agarose)
(Sigma) for consistency and reproducibility of results. Indi-
rect conjugation was performed by incubating biotin-labeled
L-PHA with avidin-coupled agarose beads (L-PHA–avidine–
agarose). Following enrichment of glycoproteins on a Con A-
agarose column and preclearing with agarose or avidin–
agarose beads, the partially purified serum samples were
subjected to lectin precipitation using either L-PHA–agarose
or L-PHA–avidine–agarose.The bound proteins were tryptic-
digested in solution, and peptide sequencing was performed
using an FT-ICR/LTQ mass analyzer. Figures 1B and C show
that L-PHA–avidine–agarose was much more effective in
capturing target proteins in sera. More proteins were cap-
tured by L-PHA–avidine–agarose compared with L-PHA–
agarose, most of which were identified by higher total scores
and sequence coverage. Steric hindrance and restricted flex-
ibility may be an obstacle for biomolecular interactions,
especially for ligand–receptor interactions. Thus, spacer
arm-appended matrices are often used to overcome these
restrictions in affinity chromatography. Avidin can act as a
spacer arm in L-PHA–avidine-agarose complexes and lessen
the hindrances that are presumably generated between agar-
ose beads and serum glycoproteins.
Biomarkerdiscoveryusing plasma orserum ishampered
primarily because of the high dynamic range of serological
proteins. Albumin accounts for ,45% of total serum pro-
teins, and when combined with Ig, accounts for ,60% of
them. In addition, since almost all highly abundant proteins
are glycoproteins, depletion of such proteins is demanding,
precluding the identification of low-abundance proteins with
high accuracy. To the best of our knowledge, the Proteom-
Prep 20 LC column (Sigma), a commercially available ID
column, removes the highest number of high-abundance
proteins. When run on the column, high-abundance pro-
teins were effectively resolved from the remaining serologi-
cal proteins (Fig. 2A). The pass-through fractions contained
mostly low-abundance proteins, from which most of the 20
high-abundance proteins were separated. When 50 mg of
proteins were loaded on an SDS-PAGE gel and stained with
CBB, albumin was notably visible together with Igs. How-
ever, equal amounts of the depleted proteins shows little
trace of the high-abundance proteins (Fig. 2B). More impor-
tantly, the depleted protein samples were found to be more
responsive to L-PHA compared with the undepleted batches
(Fig. 2C). Though the amounts of L-PHA-captured proteins
are so small that they are invisible in the coomassie stained
gel (Fig. 2D), several proteins were differentially displayed
between normal and cancer serum in an L-PHAblot analysis
Figure 2. ID of 20 high-abundance proteins from serum. (A)
Twenty high-abundance serum proteins were separated from
low-abundance proteins (unbound) on a ProteomPrep 20 plasma
ID LC column, and the unbound proteins were used for sub-
sequent procedures. (B) Depleted serum samples (1D) were
compared with the undepleted ones (2D) on an SDS-PAGE gel
following CBB staining. High-abundance proteins such as albu-
min (,67 kDa) and Ig (,56 kDa) were almost depleted. (C) Equal
amounts of serum samples (50 mg proteins) were separated on
an SDS-PAGE gel and blotted against L-PHA. After removal of
high-abundance proteins, the remaining proteins were more
responsive to L-PHA, suggesting that more expansive identifica-
tion of GnT-V targets is facilitated by ID. (D, E) The immunode-
pleted protein preparations of normal (N) and cancer (C) sera
were captured by L-PHA-aviding-agarose bead complexes, the
captured proteins were run on 10% SDS-PAGE gel, stained by
CBB (D) and lectin-blotted against L-PHA (E), in which an X-ray
film was exposed to chemiluminiscence for sufficient time.
(Fig. 2E). In addition, this ID step enabled us to identify
more proteins with higher total score values and sequence
coverages (Figs. 1B and C). Thus, we implemented ID as a
first step in our discovery protocol, which was compatible
© 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.proteomics-journal.com

Page 5

Proteomics 2008, 8, 0000–0000
Glycoproteomics
5
with the following step, obviating the need to insert an addi-
tional procedure such as buffer exchange that might cause
further loss of proteins. Taken together, our results rendered
an optimization protocol in which serum samples are
immunodepleted and then subjected to denaturation and
alkylation steps. These pretreated proteins are bound to L-
PHA–avidine–agarose conjugates, and the bound form pro-
teins are denatured in 6 M urea and subjected to tryptic
digestion and protein identification.
Clinical serum samples used in our study were grouped
into “normal” and “patient” samples. Normal blood samples
were taken from six volunteers (mean age, 44 years) who had
no previous record of occurrence of any cancer and showed
no abnormal signs in medical tests. Patient samples were
taken from ten colorectal patients whose demographics and
clinical parameters were as follows: mean age of 49 years,
regional lymph node metastasis (pN0: 40%, pN1: 50%, pN2:
10%), depth of invasion (pT2: 10%, pT3: 80%, pT4: 10%),
and TNM stage (stage I: 10%, stage II: 20%, stage III: 70%).
In this study, we adopted a semiquantitative approach: pro-
teins that were captured and detected in patient samples, but
not in normal ones, were screened using Microsoft Excel.
Among those proteins, proteins identified by only one pep-
tide were eliminated. More than a hundred of proteins satis-
fied both criteria, from which we narrowed down to 26 can-
didate biomarkers by screening (i) glycoproteins or proteins
that had at least one N-glycosylation consensus sequence (N-
X-S/T, where X is any amino acid except proline), and (ii)
proteins that were identified with a sensitivity of more than
50%. Proteins that passed these filters are listed in Table 1.
Because these proteins were not identified in any normal
samples, they showed a specificity of 100% in this study.
Oncological processes accompany bilateral changes of
biomolecules i.e., qualitative and quantitative. A quantitative
change refers to an alteration in expression, especially of
proteins, and accordingly, the majority of biomarker dis-
covery studies has relied on the expressional disturbance or
fluctuation of proteins throughout a pathological process.
Qualitativechangesinclude
phorylation, proteolytic cleavage, glycosylation, and other
phosphorylation/dephos-
Table 1. Lists of biomarker candidates for CRC
Accession
no.
IdentitiesMASCOT
scorea)
No. of
unique
peptides
Sequence
coverage
(%)
Sensitivity
(%)
Specificity
(%)
IPI00013835
IPI00166020
Isoform long of diacylglycerol kinase zeta
JmjC domain-containing histone demethylation
protein 2A
Isoform long of protocadherin a C2 precursor
Brain-rescue-factor-1
Host cell factor 2
Bifunctional heparan sulfate N-deacetylase/
N-sulfotransferase 4
Polyamine modulated factor 1-binding protein 1
Isoform 2 of carboxypeptidase B2 precursor
FLJ00268 protein (fragment)
Similartoa3typeVIcollagenisoform1precursor
Similar to B0416.5a
OTU domain containing 1
Complement C1r-like protein
Isoform 1 of A-kinase anchor protein 9
187 kDa protein
Rho-GTPase activating protein 10
Isoform 1 of WW domain-binding protein 7
183 kDa protein
Isoform 2 of kinesin-like motor protein C20orf23
361 kDa protein
Coiled–coil domain containing 132 isoform a
Similar to breast cancer antigen NY-BR-1.1
Isoform 1 of formin-1
Ephrin type-A receptor 8 precursor
Isoform 1 of aminopeptidase O
Isoform 2 of neural cell adhesion molecule
L1-like protein precursor
87–196
88–135
4–8
5–9
7.0–12.2
7.1–15.2
90
90
100
100
IPI00001516
IPI00718805
IPI00002469
IPI00021733
99–187
86–159
96–142
88–141
2–5
5–10
3–6
3–5
2.3–9.5
12.9–35.0
6.0–13.0
5.8–10.1
80
80
70
70
100
100
100
100
IPI00235481
IPI00293057
IPI00418544
IPI00739954
IPI00786924
IPI00787145
IPI00009793
IPI00019223
IPI00164623
IPI00169307
IPI00218823
IPI00291638
IPI00452247
IPI00477931
IPI00553067
IPI00746049
IPI00854651
IPI00021274
IPI00163493
IPI00299059
96–147
92–301
86–143
86–150
85–134
98–191
90–263
101–158
154–875
86–207
136–232
86–117
90–138
91–160
88–152
105–152
101–140
104–165
86–106
135–390
7–11
2–6
4–7
7–14
7–10
6–14
2–4
8–15
5–8
4–8
20–27
5–9
5–8
4–10
4–7
7–13
4–7
6–9
5–7
2–7
10.3–18.4
15.8–41.9
9.8–20.3
23.6–48.7
24.3–30.2
17.2–46.4
9.4–16.2
4.4–7.7
6.2–10.2
3.4–7.8
19.8–30.1
5.7–10.5
6.0–11.2
2.0–6.5
7.7–14.4
10.5–34.7
3.5–10.7
10.9–15.8
17.8–24.3
3.0–13.8
70
70
70
70
70
70
60
60
60
60
60
60
60
60
60
60
60
50
50
50
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
a) Score is 2106log (p), where p is the probability that the observed match is a random event.
© 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.proteomics-journal.com