Soluble c-Met protein as a susceptible biomarker for gastric cancer risk: A nested case-control study within the Korean Multicenter Cancer Cohort

Soluble c-Met protein as a susceptible biomarker for gastriccancer risk: A nested case-control study within the KoreanMulticenter Cancer Cohort

Jae Jeong Yang1,2, Ji Hyun Yang1,2, Jungkon Kim3, Seung Hyun Ma1,2, Lisa Y. Cho1,2, Kwang-Pil Ko4, Aesun Shin5,6,

Bo Youl Choi7, Hyun Ja Kim7, Dong Soo Han8, Chang Soo Eun8, Kyu Sang Song9, Yong Sung Kim10, Soung-Hoon Chang11,

Hai-Rim Shin6,12, Daehee Kang1,2,13, Keun-Young Yoo1 and Sue K. Park1,2,13

1 Department of Preventive Medicine, Seoul National University College of Medicine, Seoul, Korea2 Cancer Research Institute, Seoul National University, Seoul, Korea3 Risk Assessment Division, National Institute of Environmental Research, Incheon, Korea4 Department of Preventive Medicine, Graduate school of medicine, Gachon University, Incheon, Korea5Molecular Epidemiology Branch, Research Institute, National Cancer Center, Goyang, Korea6 Cancer Registration and Statistics Branch, National Cancer Control Institute, National Cancer Center, Goyang, Korea7 Department of Preventive Medicine, Hanyang University College of Medicine, Seoul, Korea8 Department of Internal Medicine, Hanyang University College of Medicine, Seoul, Korea9 Department of Pathology, Chungnam National University College of Medicine, Daejeon, Korea10Medical Genomics Research Center, Korea Research Institute of Bioscience and Biotechnology, Daejeon, Korea11 Department of Preventive Medicine, Konkuk University, Chungju, Korea12 Non Communicable Diseases and Health Promotion, World Health Organization, Western Pacific Regional Office, Manila, Philippines13 Department of Biomedical Sciences, Seoul National University Graduate School, Seoul, Korea

This study was conducted to evaluate the relevance of the soluble form of c-Met protein, a truncated form of the c-Met membrane

receptor involved in the CagA pathway, as a potential biomarker for gastric cancer. Among 290 gastric cancer case-control sets selected

from the Korean Multicenter Cancer Cohort, the plasma concentrations of soluble c-Met protein were measured with enzyme-linked

immunosorbent assays. Using analysis of variance and covariance models with age, sex, smoking, Helicobacter pylori infection, and

CagA seropositivity, the mean concentrations of soluble c-Met protein between cases and controls were compared. To evaluate the

association between gastric cancer and a c-Met protein level, odds ratios and 95% confidence intervals were estimated using

conditional logistic regression models. Interactions between CagA-related genes and the soluble c-Met protein concentration were also

investigated. The overall median plasma concentration of soluble c-Met among cases was significantly lower than those of controls

(1.390 vs. 1.610 ng/mL, p < 0.0001). Closer to the onset of gastric cancer, the soluble c-Met protein level decreased linearly in a

time-dependent manner (p for trend 5 0.0002). The combined effects between the CagA-related genes and the soluble c-Met protein

concentration significantly intensified risks for gastric cancer. Restricted analyses including cases that had been diagnosed within 1

year after entering the cohort had a fair degree of ability (area under the receiver operating characteristic curve of 0.73–0.77) to

discriminate gastric cancer cases from normal controls. Our findings demonstrate the potential of the soluble form of c-Met protein as a

novel biomarker for gastric cancer. The beneficial effects of a high soluble c-Met concentration in human plasma are strongly supported.

CagA-positive Helicobacter pylori infection is one of the mostpotent risk factor for gastric carcinogenesis. However, itsabsolute effect appears to be modified by genetic variantsinvolved in the CagA signal transduction pathway (i.e., SRC,c-Met, SHP2, CRK and CRKL), and some of those have beenreported to be a crucial initiation of gastric cancer.1–4 Amongthe CagA-interacting molecules, as a putative risk modifierand/or an individual contributor, the c-Met protein (encodedby c-Met gene), which is synonymous with the receptor forhepatocyte growth factor/scatter factor (HGF/SF), has beenintensively investigated due to its ambilaterality in cancertherapeutics. The normal function of c-Met is positively asso-ciated with embryonic development and damage repair,whereas deregulation of the protein induces tumorgenesis.5

In particular, c-Met/HGF signaling typically serves as a cru-cial initiation of an invasive cell phenotype such as a cancer.

Key words: soluble c-Met protein, gastric cancer, CagA-related genes,

biomarker

Additional Supporting Information may be found in the online

version of this article.

Grant sponsor: National Research Foundation of Korea funded by

the Ministry of Education, Science and Technology; Grant number:

NRF-2009-353-0066258; Grant sponsor: National R&D Program for

Cancer Control, Ministry of Health & Welfare, Republic of Korea;

Grant number: 0520140

DOI: 10.1002/ijc.27861

History: Received 26 Jun 2012; Accepted 5 Sep 2012; Online 24 Sep

2012

Correspondence to: Sue K. Park, M.D., Ph.D., Department of

Preventive Medicine, Seoul National University College of Medicine,

103 Yongon (Daehangno), Jongno-gu, Seoul, Republic of Korea, Tel.:

þ82-2-740-8338, Fax: þ82-2-747-4830, E-mail: [email protected]

Epidemiology

Int. J. Cancer: 132, 2148–2156 (2013) VC 2012 UICC

International Journal of Cancer

IJC

The c-Met protein, encoded by a met proto-oncogene, is atransmembrane protein derived from a 170-kDa precursor.6

In the mature c-Met protein with a disulfide-linked hetero-dimer a (50 kDa) and b (145 kDa) chain,7 the b chain spansboth the extracellular membrane domain to which HGFbinds and the cytoplasmic domain, which favors tyrosinekinase actions.8,9 Through a proteolysis process, an integralc-Met protein can be released from the endothelial cell mem-brane, facilitating the generation of soluble truncated c-Metprotein. This soluble truncated form of c-Met, consisting ofthe entire extracellular portion of the membrane domain,competitively binds to HGF instead of the integral c-Metwith reduced affinity.6 Alternative binding to soluble trun-cated c-Met indicates the possibility of the creation of oppo-site cellular signaling to interrupt tumorgenesis.

The integral c-Met protein is mainly expressed in humantissues and/or cells, whereas the soluble truncated form ofc-Met protein can be easily measured in human plasma. Thissuggests that the soluble truncated c-Met protein may (1) bea surrogate marker for the integral c-Met protein, (2) reflectintegral c-Met/HGF signaling in epithelial cells and (3) beused as a susceptible biomarker to detect the development ofgastric cancer. Considering that CagA-positive H. pylori caninterfere with c-Met signaling by itself,10 a soluble truncatedc-Met protein is also assumed to be regulated by CagA-related and/or H. pylori pathogen-related factors (i.e., geneticand environmental factors), thus modifying the developmentof gastric cancer in dependent and/or independent manners.

To evaluate the aforementioned hypotheses, we conductedintegrated research (1) to compare the plasma concentrationsof soluble c-Met protein between gastric cancer cases andnormal controls, (2) to examine the time-dependent variationof the protein level based on the onset point of gastric can-cer; (3) to investigate the combined effects between geneticvariants related to CagA signaling and soluble c-Met proteinlevels in the development of gastric cancer and (4) finally,to assess the potential of soluble c-Met protein as a novelbiomarker for gastric cancer.

MethodsStudy population

This study is a nested case-control study in which the studysubjects were recruited from the Korean Multicenter CancerCohort (KMCC). A detailed description of the KMCC studyprotocol is available in the literature.11,12 Briefly, from 1993

to 2004, a total of 19,688 healthy participants were recruitedfrom four urban and rural areas in Korea (Haman, Chungju,Uljin and Youngil). All participants voluntarily signed aninformed consent form before entering the study. Informa-tion on individual characteristics, including lifestyle, medicalhistory and environmental exposure factors, was collectedusing a detailed standardized questionnaire. Biological sam-ples (urine and blood) were also collected and stored understable conditions.

The study populations were passively followed-up throughrecord linkages to the national death certificate and health in-surance medical records databases and to the national cancerregistry. In December 2008, newly diagnosed gastric cancercases (N ¼ 304) were ascertained according to the Interna-tional Statistical Classification of Diseases and Related HealthProblems 10th Revision (ICD-10, C16). A total of 290 gastriccancer cases with plasma samples were included in this study.Gastric cancer cases were matched to a cancer-free controlby age (65 years), sex, residential district and enrollmentyear. Finally, 290 gastric cancer cases and 290 matched con-trols were defined.

Ethics Statement

The KMCC study protocols and the current nested case-con-trol study were approved by the Institutional Review Boardsof Seoul National University Hospital and the National Can-cer Center of Korea (H-0110-084-002, C-0910-049-297 andC-0907-044-286, respectively).

In vitro study for soluble c-Met protein expression

The plasma concentration of the soluble c-Met protein wasmeasured with a Human c-Met (soluble) enzyme-linked im-munosorbent assay (ELISA) kit according to the manufac-turer’s instructions (Invitrogen Corporation, USA). Fivemicroliter of each plasma sample was diluted with 495 lL ofStandard Diluent Buffer and was used for the assay. All sam-ples were assayed in duplicate. Standard curves were runwith each assay and produced by dilution of the 50 ng/mLstandard, which was a highly purified mouse myeloma-expressed recombinant protein. According to the absorbanceof the samples and standards at a wavelength of 450 nm in amicroplate reader, the plasma concentrations of the solublec-Met protein were determined. In terms of sensitivity, theminimum detectable dose was less than 0.5 ng/mL. This wasestablished by adding two standard deviations to the mean

What’s new?

Infection with Cag-A positive H. pylori is a major risk factor for gastric cancer. Certain genetic variants along the Cag-A

signaling pathway influence this risk, and one of these genes is the c-Met protein. This protein also has a soluble form that

can be easily measured in plasma. In this study, the authors investigated soluble c-Met as a biomarker for gastric cancer.

They compared the plasma concentration c-Met between 290 sets of gastric cancer cases and matched controls. Protein levels

were lower in cases than controls, and in fact, the levels decreased as the onset of cancer drew nearer, suggesting a

protective effect conferred by the circulating c-Met.

Epidemiology

Yang et al. 2149


optical density (O.D.) obtained when the zero standard wasassayed 30 times and then calculating the corresponding con-centration. Moreover, the ELISA kit used here had no cross-reactivity with concentrations from 4,000 to 50,000 pg/mLand other species.

Briefly, the assay procedure was as follows: plasma sam-ples were collected upon the baseline survey (conducted afterovernight fasting to reduce the measurement error) and fro-zen at �70�C until the analysis; 495 lL of Standard DiluentBuffer was added to tubes containing 5 lL of individualplasma (i.e., a 1:100 dilution). After incubating for 2 hr atroom temperature (RT), the plates were washed four timeswith Wash buffer. A total of 100 lL of biotinylated anti-Huc-Met solution was then added to the wells. The plates wereslowly shaken at RT for 1 hr. After washing the plates fourtimes, 100 lL of Streptavidin-HRP Working Solution wasadded to the wells. The plates were incubated for 30 min atRT and then washed four times. Then, 100 lL of StabilizedChromogen was added to each well and were all incubatedfor 30 min at RT and in the dark. After adding 100 lL ofStop Solution to each well and incubating within 30 min, anELISA was measured using a wavelength of 450 nm in amicroplate reader. The expected concentrations were esti-mated by multiplying the raw experimental values by thedilution factor of 100.

A pilot study for sample stability

To verify the feasibility and stability of the protein experi-ment, a pilot study was conducted using the epidemiologicalsamples from the KMCC.

First, the quadruplicated plasma samples kept at RT fordifferent durations were repetitively tested five times. Stand-ard curves were produced for each assay, and their r2 valuesranged from 0.9822 to 0.9996. Based on a statistical analysisusing repeated measures analysis of variance (ANOVA), nosignificant variations in the soluble c-Met protein levels inthe exposed plasma samples were observed when the sampleswere kept at RT 21�C for 24 hr (p for between-subjectseffects ¼ 0.9010 and p for different time exposures ¼0.4346). Between 1 and 5 hr later, the mean values for thesoluble c-Met protein were slightly higher than the baselinevalue of 1,330 ng/mL, whereas after 24 hr, the mean valueshowed a slight decrease compared with the baseline value(Supporting Information).

Due to the epidemiological features of the KMCC study,in which blood samples had been stored for several decades,a perfect storage condition without a freeze-thaw cycle and/or a temperature variation could not be assured. Thus,repeated experiments with eight freeze-thaw cycles were pre-liminarily conducted using four gastric cancer case-controlsets from the KMCC, and the stability of the soluble c-Metprotein in the plasma samples upon repeated freeze-thawcycles was investigated. Standard curves were generated foreach assay, and the r2 values ranged from 0.9408 to 0.9933.Based on a statistical analysis using repeated measures

ANOVA, the repeated freeze-thaw cycles did not appear toaffect the c-Met protein quantities significantly (p forbetween-subjects effects ¼ 0.6622 and p for within-subjectseffects on repeated cycles ¼ 0.1314) (SupportingInformation).

Finally, to confirm the potential bias due to a differentstorage time and/or condition from blood extraction to theprotein experiments, the protein variations for the total dura-tion after cohort enrollment were also checked. According tothe passage of time, all cancer-free controls (N ¼ 290) wereclassified into four groups (6–10, 11–13, 14–15 and 16–17years) and the mean value of each group was calculated.Regardless of the classifications of the time period fromblood extraction to the protein experiments, comparablemean values were observed and could be treated as a singlegroup (p ¼ 0.2241 and p for trend ¼ 0.2370) (SupportingInformation).

H. pylori infection and CagA seropositivity

H. pylori, deemed a group I human carcinogen by the Inter-national Agency for Research on Cancer, is a proven risk fac-tor for gastric cancer. To eliminate the potent effect of H.pylori and its virulent factors, CagA and VacA, we treatedthem as covariates and adjusted their values in the statisticalmodels. H. pylori infection and CagA seropositivity wereevaluated among all study subjects (N ¼ 580) using the im-munoblot assay of Helico Blot 2.1

TM

(MP Biomedicals AsiaPacific, Singapore). Helico Blot 2.1

TM

kits reportedly havehigh sensitivity (99% for both H. pylori and CagA seroposi-tivity) and specificity (98% for H. pylori and 90% for CagAseropositivity) among the Korean population.13

Statistical analysis

Using an analysis of variance and covariance (ANCOVA)with age, sex, smoking status (ever vs. never), H. pylori infec-tion (positive vs. negative) and CagA seropositivity (positivevs. negative) as potential risk factors for gastric cancer, themean concentrations of soluble c-Met protein between casesand controls were compared. According to an increased timeperiod from cohort enrollment to gastric cancer diagnosis, thetime-dependent serial change of the soluble c-Met proteinconcentration was evaluated in a linear regression model.Post-hoc analysis was conducted using the Student–Newman–Keuls procedures based on the Studentized range test.

Using conditional or unconditional logistic regressionmodels adjusted for the aforementioned covariates, the oddsratios (ORs) and 95% confidence intervals (CIs) were esti-mated to evaluate the association between gastric cancer riskand a single c-Met protein level or genotype–phenotype com-binations. Polytomous logistic regression analysis was alsoconducted to estimate the p values according to a nominalscale.

Based on the area under the receiver operating character-istic (AUC-ROC) curves as calculated from various logisticregression models using the C statistics, the optimal cutoff

Epidemiology

2150 Soluble c-Met proteins, gene and gastric cancer


level of soluble c-Met protein as a biomarker for gastric can-cer was determined. Sensitivity and specificity were graphi-cally represented by ROC curves that plotted the true-positiverate (sensitivity) against the false-positive rate (1-specificity).An AUC-ROC value of 0.90 to 1.00 indicates excellent, 0.80to 0.89 good, 0.70 to 0.79 fair, 0.60 to 0.69 poor, and 0.50 to0.59 no useful performance.14

Statistical analyses were performed using SAS softwareversion 9.2 (SAS Institute, Cary, North Carolina).

ResultsCases and controls were comparable in age, sex, H. pyloriinfection, CagA/VacA seropositivity, smoking/drinking status,and gastric ulcer history (p > 0.05). Significantly differentfeatures were not observed for any of the variables (data notshown).

The overall median plasma concentration of soluble c-Metamong cases was significantly lower than those of the con-trols (1,390 vs. 1,610 ng/mL, p < 0.0001). Although the me-dian follow-up time of the study population was 13 years(ranging from 8 to 17 years), variations attributable to thedifferent time periods from enrollment to the experimentswere not observed (p ¼ 0.2581 for cases and p ¼ 0.4564 forcontrols). In stratified analyses according to potential riskfactors (i.e., age, sex, H. pylori infection, CagA positivity andsmoking status) for gastric cancer, consistently lower concen-trations of soluble c-Met protein were observed among allcases regardless of the risk factor. Interestingly, significantlylower levels of the comparisons between the cases and thecontrols were presented among risk groups, specifically thoseof an older age (p < 0.0001), females (p ¼ 0.0825), thosewith positive experiences of H. pylori infection and CagA(p < 0.0001), and smokers (p ¼ 0.0003) (Table 1).

Based on a median value of 1,470 ng/mL, the absoluterisk for gastric cancer fell by one third (OR ¼ 0.29, 95% CI0.18–0.46). Compared with the reference group with the low-est tertile level, the highest level of soluble c-Met protein,�1,700 ng/mL, showed a significantly decreased risk for gas-tric cancer (OR ¼ 0.38, 95% CI 0.23–0.65). The highest riskgroup with a soluble c-Met protein level of �1,820 ng/mLshowed protective effects against gastric cancer (OR ¼ 0.23,95% CI 0.12–0.45 for the quartile classification). Regardlessof the classification of soluble c-Met protein, a continualdecrease in risk according to an increased level of soluble c-Met concentration was observed (p for trend ¼ 0.0003 fortertile; p for trend < 0.0001 for quartile) (Table 2).

If the soluble c-Met protein acts as a potential biomarkerfor gastric cancer, a change in the expression of the proteinmay directly or inversely correlate with the progression ofgastric cancer; thus, we examined the time-dependent varia-tion of the soluble c-Met protein level before gastric canceroccurrence. According to the time period from cohort enroll-ment to cancer ascertainment, cases were classified into fivegroups: cases diagnosed within 1 year, between 2 and 4 years,between 5 and 6 years, between 7 and 9 years and after 10years from the cohort enrollment time. Compared with themedian concentration of soluble c-Met protein among con-trols (1,610 ng/mL), cases who were diagnosed within 1 yearafter entering the cohort showed the lowest concentration ofthe protein (1,250 ng/mL), whereas cases diagnosed aftermore than 10 years presented a slightly lower level of solublec-Met protein (1,570 ng/mL) (p ¼ 0.0003). A post-hoc analy-sis revealed that the controls and cases diagnosed after morethan 10 years from the cohort enrollment time had nearlyidentical features and could be treated as a single group,whereas cases diagnosed within 1 year after enrolling in thecohort were wholly different from the controls or even other

Table 1. Plasma concentrations of soluble c-Met protein (ng/mL) among gastric cancer cases and their matched controls in a population-based nested case-control study

Cases Controls

N Median (IQR2) N Median (IQR2) P1

Soluble c-Met (ng/mL) 290 1,390 (1,120–1,750) 290 1,610 (1,220–1,870) <0.0001

Age3 �63 years old 148 1,440 (1,160–1,830) 152 1,620 (1,260–1,880) 0.0510

>63 years old 142 1,350 (1,050–1,680) 138 1,590 (1,220–1,860) <0.0001

Sex Male 196 1,420 (1,140–1,790) 196 1,620 (1,260–1,900) <0.0001

Female 94 1,350 (1,110–1,720) 94 1,590 (1,170–1,830) 0.0825

H. pylori infection Negative (�) 31 1,380 (1,060–1,820) 32 1,600 (1,170–1,890) 0.3286

Positive (þ) 259 1,390 (1,130–1,740) 258 1,610 (1,250–1,870) <0.0001

CagA Negative (�) 23 1,170 (1,000–1,750) 31 1,470 (1,090–1,650) 0.9152

Positive (þ) 267 1,410 (1,140–1,760) 259 1,630 (1,250–1,910) <0.0001

Smoking status Never 115 1,370 (1,117–1,740) 138 1,590 (1,220–1,830) 0.0440

Ever 175 1,410 (1,140–1,780) 152 1,630 (1,260–1,970) 0.0003

1Estimated by ANCOVA model adjusted for age, sex, smoking (never vs. ever), H. pylori infection (positive vs. negative) and CagA seropositivity(positive vs. negative). 2Interquartile range: from the lower quartile to the upper quartile. 3Divided into two groups by median age for the studypopulation ranging from 29 to 79 years of age.

Epidemiology

Yang et al. 2151


cases. Closer to the onset of gastric cancer, the soluble c-Metprotein level decreased linearly in a time-dependent manner(p for trend ¼ 0.0002) (Fig. 1).

According to the three sub-groups as determined by thepost-hoc analysis, the associations between the soluble c-Metprotein concentration and gastric cancer risk were furtherevaluated using a polytomous logistic regression model. Theresults were similarly replicated compared with those amongcontrols vs. all cases; but the absolute risk levels for gastriccancer among cases diagnosed within 1 year after enrollingin the cohort were much lower than those of other cases andthe controls (p polytomous < 0.0001 for median classifica-tion; p polytomous ¼ 0.0018 for tertile classification; p polyt-omous ¼ 0.0002 for quartile classification and p polytomous¼ 0.0011 for quintile classification) (Table 3).

In our two previous genetic studies,3,4 five SNPs in threeCagA-interacting genes (SRC rs6122566, rs6124914, c-Metrs41739, rs41737 and CRK rs7208768) and six SNPs in fivegenes on the ERK downstream signaling pathways activatedby CagA (ERK rs5999749, Dock180 rs4635002, C3Grs7853122, rs10901081, Rap1 rs530801 and SRC rs747182)were identified as significant during the primary scan andwere further investigated among large gastric cancer case-control sets that were nested in the KMCC. Among thosesubjects, 207 cases and 209 controls were also included inthis study, which is focused on c-Met protein experiments.To investigate whether the interaction between CagA-relatedgenes and soluble c-Met protein could modify the individuallevels of susceptibility to gastric cancer, subgroup analyseswere conducted. The highest risk groups composed the riskgenotypes of each individual gene and a lower concentrationof soluble c-Met protein (<1,470 ng/mL) showed up to afour-fold increased risk for gastric cancer, but the interactionsbetween these SNPs and soluble c-Met protein were not sig-

nificant (p interaction > 0.05) (data not shown). Another ge-notype–phenotype interaction focused on genetic counts cal-culated on the basis of the total number of risk alleles carriedby each person was also analyzed. The combined effectsbetween genetic counts higher than the median value of totalrisk alleles and a low c-Met protein level under 1,470 ng/mLsignificantly intensified the risk for gastric cancer (OR ¼2.97, 95% CI 1.72–5.15, p for trend < 0.0001 for CagA-inter-acting genes; OR ¼ 2.86, 95% CI 1.53–5.34, p for trend ¼

Table 2. Gastric cancer risk according to the plasma concentration level of soluble c-Met protein

Classification ofsoluble c-Met (ng/mL)1

Cases(N ¼ 290)

Controls(N ¼ 290) OR (95% CI)2 p trend

Median

[M1] < 1,470 169 114 1 (reference) <0.0001

[M2] � 1,470 121 176 0.29 (0.18–0.46)***

Tertile

[T1] < 1,280 111 82 1 (reference) 0.0003

[T2] 1,280–1,699 97 96 0.63 (0.39–1.00)*

[T3] � 1,700 82 112 0.38 (0.23–0.65)**

Quartile

[Q1] < 1,160 85 58 1 (reference) <0.0001

[Q2] 1,160–1,469 84 56 0.85 (0.50–1.44)

[Q3] 1,470–1,819 61 90 0.28 (0.15–0.51)***

[Q4] � 1,820 60 86 0.23 (0.12–0.45)***

***p <0.0001, **p <0.05, *0.05� p <0.10. 1Among total study population. 2Conditional logisticregression analysis adjusted for age, sex, smoking (never vs. ever), H. pylori infection (positive vs.negative) and CagA seropositivity (positive vs. negative).

Figure 1. Soluble c-Met protein concentrations among gastric

cancer cases according to the time period from cohort enrollment

to cancer ascertainment (onset time of cancer) and cancer-free

controls.

Epidemiology



0.0003 for CagA-interfering genes; and OR ¼ 3.96, 95% CI2.07–7.59, p for trend < 0.0001 for all CagA-related genes)(Table 4).

Best cutoff value of 1,450 ng/mL was determined by con-sidering the range of sensitivity and specificity with the maxi-mized AUC values calculated in sequential multivariate mod-els after adjusting for the potential contributing factors ofage, sex, smoking, H. pylori infection, CagA positivity andgenetic counts of the SNPs on CagA-related genes. Amongall of the gastric cancer cases and controls, the soluble c-Metprotein itself could not acquire sufficient predictive powerwith a low AUC value of 0.59 (95% CI 0.55–0.64), sensitivityof 56.2% and specificity of 61.4%. When smoking status,H. pylori infection and CagA positivity were included in amultivariate model, the predictive ability increased slightly to0.62 (95% CI 0.58–0.67); after adding genetic counts of theSNPs on CagA-interacting or CagA-interfering genes, theAUC values increased significantly to 0.66 (95% CI 0.60–0.71)and 0.67 (95% CI 0.61–0.73). The C statistics between the fivemodels were statistically different (p ¼ 0.004) (Fig. 2a).

Given that some cases were far from the onset of gastriccancer and should be treated as normal controls, the actualeffect of soluble c-Met protein as a precancerous biomarkermay be biased toward null in the primary analysis. Thus, wealso conducted a restricted analysis including cases diagnosedwithin 1 year after entering the cohort and controls. Sensitiv-ity improved to 73.4% and the soluble c-Met protein itselfindicated an improvement of the AUC value to 0.73 (95% CI0.66–0.81). Moreover, after adding genetic counts of theSNPs on CagA-related genes and the potential contributingfactors, the soluble c-Met protein level of 1,450 ng/mLshowed a fair amount of ability to discriminate gastric cancercases from normal controls (AUC ¼ 0.77, [95% CI 0.68–0.85] for CagA-interacting genes and AUC ¼ 0.77, [95% CI

0.67–0.86] for CagA-interfering genes). The C statisticsbetween the five models showed a marginal degree of signifi-cance (p ¼ 0.067) (Fig. 2b).

DiscussionIn this population-based nested case-controlled study, we elu-cidated the following scientific knowledge pertaining to gas-tric cancer susceptibility in terms of a soluble c-Met protein,CagA-related genes, and their interactive effects. The plasmaconcentrations of the soluble c-Met protein were significantlyassociated with the development of gastric cancer, and a highlevel of the soluble c-Met protein appeared to reduce theabsolute risk for gastric cancer. Moreover, a time-dependentdecrease in the soluble c-Met protein level closer to the onsetof gastric cancer implied that the soluble c-Met protein canbe used as a predictive biomarker for gastric cancer. Com-bined effects between the soluble c-Met protein and genesrelated to CagA signal transduction could play a crucial rolein gastric carcinogenesis; their synergistic effects intensifiedan individual’s absolute risk for gastric cancer.

One of the major CagA-interacting molecules, c-Met,appears to be a potential protein marker that predicts therisk of gastric cancer in the truncated form (i.e., soluble c-Met protein). This study revealed that the plasma concentra-tion of the soluble c-Met protein was significantly loweramong gastric cancer cases as well as time-dependent serialdecreases in the soluble c-Met concentration closer to theonset of the disease. Recent studies also reported that c-Met/HGF signaling induces aberrant cellular activities such as cellproliferation, motility, disassociation and morphologicalchange, which can be effectively blocked by the antagonisticfragment of HGF/SF (NK4)15–17 or by c-Met decoys, refer-ring to the soluble truncated form of c-Met.18–20 The solubletruncated form of c-Met competitively binds to HGF instead

Table 3. Changes in soluble c-Met concentrations and gastric cancer risk according to the comparison groups classified by post-hoc analysis

Classification of solublec-Met (ng/mL)1

Group 12

(N ¼ 321)Group 23

(N ¼ 210) OR (95% CI)Group 34

(N ¼ 49) OR (95% CI) p polytomous5

Median

[M1] < 1,470 128 (39.9) 119 (56.7) 1 (reference) 36 (73.5) 1 (reference) <0.0001

[M2] � 1,470 193 (60.1) 91 (43.3) 0.48 (0.34–0.69)*** 13 (26.5) 0.26 (0.13–0.51)***

Tertile

[T1] < 1,280 91 (28.4) 75 (35.7) 1 (reference) 27 (55.1) 1 (reference) 0.0018

[T2] 1,280–1,699 105 (32.7) 76 (36.2) 0.87 (0.56–1.33) 12 (24.5) 0.40 (0.19–0.85)**

[T3] � 1,700 125 (38.9) 59 (28.1) 0.54 (0.34–0.83)** 10 (20.4) 0.30 (0.14–0.65)**

Quartile

[Q1] < 1,160 65 (20.3) 59 (28.1) 1 (reference) 19 (38.8) 1 (reference) 0.0002

[Q2] 1,160–1,469 63 (19.6) 60 (28.6) 1.02 (0.62–1.70) 17 (34.7) 0.98 (0.46–2.10)

[Q3] 1,470–1,819 99 (30.8) 45 (21.4) 0.48 (0.29–0.80)** 7 (14.3) 0.27 (0.11–0.69)**

[Q4] � 1,820 94 (29.3) 46 (21.9) 0.49 (0.30–0.82)** 6 (12.2) 0.24 (0.09–0.63)**

***p < 0.0001, **p < 0.05, *0.05 � p < 0.10. 1Among total study population. 2290 cancer-free controls and 31 gastric cancer cases whodiagnosed after 10 years from cohort enrollment. 3Gastric cancer cases who diagnosed between 2 and 9 years from cohort enrollment. 4Gastriccancer cases who diagnosed within 1 year from cohort enrollment. 5Polytomous logistic regression by the outcome of nominal scale.

Epidemiology

Yang et al. 2153


of an integral c-Met; this alternative binding can frustratetumorgenesis. Other preclinical studies demonstrated that thegenerated soluble truncated c-Met can (1) inhibit c-Met acti-vation as induced in an HGF ligand-dependent as well as in-dependent manner; (2) interrupt c-Met/HGF signaling byrejecting the formation of the active homodimers requiredfor the signaling and (3) finally exhibit its potential as a ther-apeutic agent for anti-carcinogenic substances.6,18,20,21 Ourfindings add to the evidence that the soluble form of c-Metprotein naturally produced by proteolysis or amino acidtranslation in the human body has a consistently protectiveeffect against the development of gastric cancer, thus sup-porting the anti-tumor potential of the soluble truncated c-Met protein.

In this study, we attempted to evaluate the soluble c-Metprotein concentration in terms of its predictive value for thedevelopment of gastric cancer. Among all study subjects, theoptimal cutoff value of 1,450 ng/mL was not fully satisfactoryin most results including sensitivity, specificity and AUC-ROC values. This may have arisen because approximately49.7% of our cases were far from the onset of gastric cancer(more than 5 years). Particularly, some of those (i.e., cases

diagnosed after 10 years from cohort enrollment) were evensimilar to cancer-free controls. In other words, some casesshould be treated as different groups considering the durationbetween blood correction and the occurrence of gastric can-cer. Hence, we conducted a restricted analysis including onlycases that had been diagnosed within 1 year after enteringthe cohort and acquired advanced results for sensitivity andAUC-ROC values. Moreover, this restricted analysis consis-tently demonstrated that a soluble c-Met protein level of1,450 ng/mL after adding genetic counts related to CagA sig-naling and/or the potential contributing factors of age, sex,smoking status, H. pylori and CagA positivity had a fairdegree of ability (AUC-ROC 0.73–0.77) to discriminate gas-tric cancer cases from normal controls. These findings indi-cate that the soluble form of c-Met protein may be used as amore sensitive prognostic marker for an upcoming occur-rence of gastric cancer and cases closer to cancer onset.

Previous studies demonstrated that the genetic variantsinvolved in the CagA signaling transduction pathway(directly interacting with CagA) and involved in the ERKdownstream signaling pathways activated by CagA (secondar-ily affected by CagA) could be susceptible factors in gastric

Table 4. Risk for gastric cancer associated with the combined effects between CagA-related genotypes and the soluble c-Met proteinconcentrations

Genetic counts using gene markers1,2c-Met proteinconcentrations3 Cases No. (%) Controls No. (%) OR (95% CI)4 p trend

Using five SNPs in CagA-binding genes1,5

Low-risk gene markers Low risk c-Met 38 (18.7) 69 (33.8) 1 (reference) <0.0001

High-risk gene markers Low risk c-Met 45 (22.2) 51 (25.0) 1.68 (0.95–2.97)*

Low-risk gene markers High risk c-Met 45 (22.2) 34 (16.7) 2.48 (1.35–4.54)**

High-risk gene markers High risk c-Met 75 (36.9) 50 (24.5) 2.97 (1.72–5.15)**

Using six SNPs in CagA-interfering genes2,6

Low-risk gene markers Low risk c-Met 37 (21.0) 54 (33.8) 1 (reference) 0.0003

High-risk gene markers Low risk c-Met 39 (22.2) 47 (29.4) 1.19 (0.65–2.18)


High-risk gene markers High risk c-Met 55 (31.2) 28 (17.5) 2.86 (1.53–5.34)**

Using eleven SNPs in CagA-related genes1,2,7

Low-risk gene markers Low risk c-Met 23 (13.4) 47 (30.5) 1 (reference) <0.0001

High-risk gene markers Low risk c-Met 52 (30.2) 51 (33.1) 2.13 (1.13–4.04)**


High-risk gene markers High risk c-Met 70 (40.7) 37 (24.0) 3.96 (2.07–7.59)***

***p < 0.0001, **p < 0.05, *0.05 � p < 0.10. 1Previous genetic studies for 137 genetic polymorphisms coding CagA-binding proteins presentedan association between gastric cancer risk and five SNPs in three genes as follows: c-MET rs41737: [AG/AA genotype (high risk) vs. GG (low risk)];c-MET rs41739: [AG/GG (high risk) vs. AA (low risk)]; CRK rs7208768: [AG/AA (high risk) vs. GG (low risk)]; SRC rs6122566: [GG (high risk) vs. AA/AG(low risk)]; SRC rs6124914: [CT/CC (high risk) vs. TT (low risk)]. 2Previous genetic studies for 580 genetic polymorphisms involved in the ERKdownstream signaling pathways activated by CagA presented an association between gastric cancer risk and six SNPs in five genes as follows: ERKrs5999749: [CC/AC genotype (high risk) vs. AA (low risk)]; Dock180 rs4635002 [CC/AC (high risk) vs. AA (low risk)]; C3G rs7853122: [AA/AG (highrisk) vs. GG (low risk)]; C3G rs10901081: [GG/AG (high risk) vs. AA (low risk)]; Rap1 rs530801: [GG/AG (high risk) vs. AA (low risk)]; Src rs747182:[GG/AG (high risk) vs. AA (low risk)]. 3c-Met protein concentration according to median value: high-risk status (<1,470 ng/mL) vs. low-risk status(�1,470 ng/mL). 4Adjusted for smoking age, sex, smoking (never vs. ever), H. pylori infection (positive vs. negative) and CagA seropositivity(positive vs. negative). 5Genetic counts calculated on the basis of total number of risk alleles from five SNPs carried by each person: high-riskstatus (4–10 risky alleles) vs. low risk status (<4 risky alleles). 6Genetic counts calculated on the basis of total number of risk alleles from six SNPscarried by each person: high-risk status (8–12 risky alleles) vs. low-risk status (<8 risky alleles). 7Genetic counts calculated on the basis of totalnumber of risk alleles from eleven SNPs carried by each person: high-risk status (11–22 risky alleles) vs. low-risk status (<11 risky alleles).

Epidemiology



carcinogenesis in terms of how they control signals throughthe CagA-related pathways.3,4 Despite the inconclusive resultsnot assuring statistically significant p interaction in this study,gene–protein interactions between the soluble c-Met proteinand genetic variants related to CagA signaling also appearedto modify an individual’s susceptibility to gastric cancer. Inparticular, after adding genetic counts related to CagA signal-ing, the AUC-ROC values among all the study subjectssharply increased regardless how far the subject was from theonset of the disease; hence, it was presumed that genetic var-iants related to CagA signaling may be utilized in conjunc-tion with the soluble c-Met protein in explaining the etiologyof gastric cancer. Studies with a greater number of gastriccancer cases along with in-depth gene–protein interactionanalysis are required to elucidate the hidden mechanisms ingastric carcinogenesis.

To verify all results of this study, a small replication studybased on a hospital-based case-control design was conducted.One hundred eighty-nine pairs of gastric cancer case-controls

from two university hospitals in Korea (Chungnam Univer-sity Hospital and Hanyang University GURI Hospital) wereselected for the replication study. The detailed protocol of thehospital-based case-control study is available in the litera-ture.3 Comparable results were replicated as follows: (1) themedian plasma concentration of the soluble c-Met protein(1,470 ng/mL) among the cases was also significantly lowerthan those of the controls (p ¼ 0.0064); (2) the highest con-centrations of soluble c-Met protein (�1,700 ng/mL for ter-tile and �1,820 ng/mL for quartile) were significantly associ-ated with protective effects against gastric cancer comparedwith the lowest level of the protein (OR ¼ 0.23, 95% CI0.08–0.68; OR ¼ 0.07, 95% CI 0.01–0.54) and (3) continualdecreases in gastric cancer risk according to an increasedlevel of the soluble c-Met concentration were still observed (pfor trend < 0.0001). However, the overall values were slightlyhigher than the results of the KMCC study. To determine thebest clinical threshold of the soluble c-Met protein for gastriccancer prognosis and/or diagnosis, further studies includingvarious populations and clinical differences should beconducted.

Several limitations should be noted. First, the plasma con-centration of the soluble c-Met protein was measured once atthe time of cohort enrollment; thus, we could not considerindividual variations in the soluble c-Met protein levelaccording to any natural history of gastric cancer or healthstatus. Second, because the onset of gastric cancer wasdefined in terms of the diagnosis date from medical records,the actual onset point may be different in some cases. Thiscould introduce misclassification bias. Moreover, other gastric

Figure 2. The ROC curve at 1,450 ng/mL of soluble c-Met

concentration using all gastric cancer cases (N ¼ 290) and controls

(N ¼ 290) (a) and gastric cancer cases diagnosed within 1 year from

cohort enrollment (N ¼ 49) and controls (N ¼ 290) (b). (a) All gastric

cancer cases and controls (290 case-control sets). C-statistic

between Models 3 and 4 was marginally significant (p ¼ 0.06) but

those of other models were all statistically significant (p ¼ 0.04

between Model 1 and 2, p < 0.01 between Model 2 and 3, and p <

0.01 between Model 4 and 5, respectively). (b) Gastric cancer cases

diagnosed within 1 year from the cohort enrollment and controls (49

cases and 290 controls). C-statistic between Models 2 and 3 was

marginally significant (p ¼ 0.08) but those of other models were not

significant (p ¼ 0.37 between Model 1 and 2, p ¼ 0.97 between

Model 3 and 4, and p ¼ 0.65 between Model 4 and 5, respectively).

Both sensitivity and specificity was estimated in the model 3. Each

model is described below: black dot line (Model 1): AUC-ROC curve

based on the logistic regression model including c-Met protein, age

and sex; blue line (Model 2): Model 1 plus smoking status, CagA

positivity and H. pylori infection; green line (Model 3): Model 2 plus

CagA-interacting genetic counts calculated on the basis of total

number of risk alleles from five SNPs; red line (Model 4); model 3

plus CagA-interfering genetic counts involved in the ERK downstream

signaling pathways activated by CagA; gold dot line (Model5): Model

1 plus CagA-interacting genetic counts calculated on the basis of

total number of risk alleles from five SNPs. [Color figure can be

viewed in the online issue, which is available at

wileyonlinelibrary.com.]

Epidemiology

Yang et al. 2155


disorders were not concurrently considered in this study.Thus, if a large proportion of the study subjects contracted agastric disorder other than gastric cancer upon blood correc-tion, our study results would be biased and underestimated.Finally, due to the restriction of the passive follow-up meth-odology based on data linkages, detailed information on thediversity of gastric cancer such as histological subtypes (i.e.,diffuse type vs. intestinal type) and TNM staging have notbeen confirmed, and thus, further stratified analysis couldnot be conducted.

Despite these limitations, this is a population-based nestedcase-control study that is free of the latent biases common inretrospective designs. Basic confounders such as age, sex andenrollment information were also initially matched at the

study design stage, and other potential confounding factorswere mitigated by the use of multivariable models. Mostimportantly, as integrated research that includes both in vitroexperiments for soluble c-Met protein and genetic studiesfocused on CagA-related genetic variants was conducted, wewere able to formulate more conclusive evidence regardingindividuals0 risk levels for gastric cancer.

In conclusion, this study demonstrates that the potentialof the soluble form of c-Met protein as a novel biomarker forgastric cancer and the combined effects between the solublec-Met protein levels and genetic variants related to CagA sig-naling can play a role as susceptible factors in gastric carci-nogenesis. The beneficial effects of a high-soluble c-Met con-centration in human plasma are also strongly supported.

References

1. Goto Y, Ando T, Yamamoto K, et al. Associationbetween serum pepsinogens and polymorphismofPTPN11 encoding SHP-2 among Helicobacterpylori seropositive Japanese. Int J Cancer 2006;118:203–8.

2. Zhu F, Loh M, Hill J, et al. Genetic factorsassociated with intestinal metaplasia in a highrisk Singapore-Chinese population: a cohortstudy. BMC Gastroenterol 2009;9.

3. Yang JJ, Cho LY, Ma SH, et al. Oncogenic CagApromotes gastric cancer risk via activating ERKsignaling pathways: a nested case-control study.PLoS One 2011;6:e21155.

4. Yang JJ, Cho LY, Ma SH, et al. Geneticsusceptibility on CagA-interacting molecules andgene-environment interaction withphytoestrogens: a putative risk factor for gastriccancer. PLoS One 2012;7:e31020.

5. Porter J. Small molecule c-Met kinase inhibitors:a review of recent patents. Expert Opin Ther Pat2010;20:159–77.

6. Wajih N, Walter J, Sane DC. Vascular origin of asoluble truncated form of the hepatocyte growthfactor receptor (c-met). Circ Res 2002;90:46–52.

7. Giordano S, Di Renzo MF, Ferracini R, et al.p145, a protein with associated tyrosine kinaseactivity in a human gastric carcinoma cell line.Mol Cell Biol 1988;8:3510–17.

8. Bottaro DP, Rubin JS, Faletto DL, et al.Identification of the hepatocyte growth factorreceptor as the c-met proto-oncogene product.Science 1991;251:802–4.

9. Dean M, Park M, Le Beau MM, et al. Thehuman met oncogene is related to the tyrosinekinase oncogenes. Nature 1985;318:385–8.

10. Churin Y, Al-Ghoul L, Kepp O, et al.Helicobacter pylori CagA protein targets the c-Met receptor and enhances the motogenicresponse. J Cell Biol 2003;161:249–55.

11. Yoo KY, Shin HR, Chang SH, et al. KoreanMulti-center Cancer Cohort Study including aBiological Materials Bank (KMCC-I). Asian Pac JCancer Prev 2002;3:85–92.

12. Yoo KY, Shin HR, Chang SH, et al. Genomicepidemiology cohorts in Korea: present and thefuture. Asian Pac J Cancer Prev 2005;6:238–43.

13. Park CY, Cho YK, Kodama T, et al. Newserological assay for detection of putativeHelicobacter pylori virulence factors. J ClinMicrobiol 2002;40:4753–6.

14. Greiner M, Pfeiffer D, Smith RD. Principles andpractical application of the receiver-operatingcharacteristic analysis for diagnostic tests. PrevVet Med 2000;45:23–41.

15. Coxon A, Rex K, Meyer S, et al. Soluble c-Metreceptors inhibit phosphorylation of c-Met and

growth of hepatocyte growth factor: c-Met-dependent tumors in animal models. Mol CancerTher 2009;8:1119–25.

16. Date K, Matsumoto K, Shimura H, et al. HGF/NK4 is a specific antagonist for pleiotrophicactions of hepatocyte growth factor. FEBS Lett1997;420:1–6.

17. Tomioka D, Maehara N, Kuba K, et al. Inhibitionof growth, invasion, and metastasis of humanpancreatic carcinoma cells by NK4 in anorthotopic mouse model. Cancer Res 2001;61:7518–24.

18. Michieli P, Mazzone M, Basilico C, et al.Targeting the tumor and its microenvironmentby a dual-function decoy Met receptor. CancerCell 2004;6:61–73.

19. Kong-Beltran M, Stamos J, Wickramasinghe D.The Sema domain of Met is necessary forreceptor dimerization and activation. Cancer Cell2004;6:75–84.

20. Tiran Z, Oren A, Hermesh C, et al. A novelrecombinant soluble splice variant of Met is apotent antagonist of the hepatocyte growthfactor/scatter factor-Met pathway. Clin CancerRes 2008;14:4612–21.

21. Zhang YW, Graveel C, Shinomiya N, et al. Metdecoys: will cancer take the bait? Cancer Cell2004;6:5–6.

Epidemiology



Documents

Soluble c-Met protein as a susceptible biomarker for gastric cancer risk: A nested case-control study within the Korean Multicenter Cancer Cohort