immunomarkersupportvectormachineclassiﬁer for prediction ... · clinical stage and similar...

Precision Medicine and Imaging

Immunomarker Support Vector Machine Classifierfor Prediction of Gastric Cancer Survival andAdjuvant Chemotherapeutic BenefitYuming Jiang1,2, Jingjing Xie3, Zhen Han1,Wei Liu2,4, Sujuan Xi2,5, Lei Huang6,Weicai Huang1, Tian Lin1, Liying Zhao1, Yanfeng Hu1, Jiang Yu1, Qi Zhang2,4,Tuanjie Li1,2, Shirong Cai7, and Guoxin Li1

Abstract

Purpose: Current tumor–node–metastasis (TNM) stagingsystem cannot provide adequate information for prediction ofprognosis and chemotherapeutic benefits. We constructed aclassifier to predict prognosis and identify a subset of patientswho can benefit from adjuvant chemotherapy.

Experimental Design:We detected expression of 15 immu-nohistochemistry (IHC) features in tumors from 251 gastriccancer (GC) patients and evaluated the association of theirexpression level with overall survival (OS) and disease-free survival (DFS). Then, integrating multiple clinicopatho-logic features and IHC features, we used support vectormachine (SVM)–based methods to develop a prognostic clas-sifier (GC-SVM classifier) with features. Further validation oftheGC-SVMclassifierwas performed in two validation cohortsof 535 patients.

Results: The GC-SVM classifier integrated patient sex,carcinoembryonic antigen, lymph node metastasis, and

the protein expression level of eight features, includingCD3invasive margin (IM), CD3center of tumor (CT), CD8IM,CD45ROCT, CD57IM, CD66bIM, CD68CT, and CD34. Sig-nificant differences were found between the high- and low-GC-SVM patients in 5-year OS and DFS in training andvalidation cohorts. Multivariate analysis revealed that theGC-SVM classifier was an independent prognostic factor.The classifier had higher predictive accuracy for OS andDFS than TNM stage and can complement the prognosticvalue of the TNM staging system. Further analysis revealedthat stage II and III GC patients with high-GC-SVM werelikely to benefit from adjuvant chemotherapy.

Conclusions: The newly developedGC-SVM classifier was apowerful predictor of OS and DFS. Moreover, the GC-SVMclassifier could predict which patients with stage II and IIIGCbenefit fromadjuvant chemotherapy.ClinCancer Res; 24(22);5574–84. �2018 AACR.

IntroductionGastric cancer (GC) is one of the most common malignancies

and the second leading cause of cancer-related deaths worldwide(1). Surgical resection is the main curative method for GC, but ahigh rate of relapse in patients with advanced GC makes itimportant to consider adjuvant treatments (2, 3). Currently, thetumor–node–metastasis (TNM) staging system and histologicclassification are used for routine prognostication and treatmentamong patients with GC, but neither provides substantialpredictive value (2–4). Given that GC patients with the sameclinical stage and similar treatment regimens often undergosubstantially different clinical courses, a new GC classificationsystem is needed for more precise prediction of prognosis, thusenabling a more tailored therapeutic approach with improvedoutcomes for GC patients.

Extensive studies have suggested tumor-infiltrating immunecells and tumor angiogenesis in cancers were correlated withprognosis (5–10). Galon and colleagues showed the type,density, and location of immune cells in colorectal cancers hada prognostic value that was superior to and independent of thoseof the TNM stage (11–13). Based on the numeration of lympho-cyte and/or myeloid cell populations in the center of tumor (CT)and the invasive margin (IM), immune score could predictsurvival and/or treatment response (5–7, 11, 13–18). An immu-noscore of colon cancer, derived from a measure of CD3-positiveand CD8-positive cell densities in the CT and IM, had a larger

1Department of General Surgery, NanfangHospital, SouthernMedical University,Guangzhou, China. 2Guangdong Key Laboratory of Liver Disease Research, theThird AffiliatedHospital of Sun Yat-sen University, Guangzhou, China. 3ResearchCenter for Clinical Pharmacology, Nanfang Hospital, Southern MedicalUniversity, Guangzhou, China. 4Biotherapy Center, the Third Affiliated Hospitalof SunYat-senUniversity, Guangzhou, China. 5Department of InfectiousDisease,the Third Affiliated Hospital of Sun Yat-sen University, Guangzhou, China.6German Cancer Research Center (DKFZ), Heidelberg, Germany. 7Departmentof Gastrointestinal Surgery, the First Affiliated Hospital, Sun Yat-sen University,Guangzhou, Guangdong, China.

Note: Supplementary data for this article are available at Clinical CancerResearch Online (http://clincancerres.aacrjournals.org/).

Y. Jiang, J. Xie, Z. Han, and W. Liu contributed equally to this article.

Corresponding Authors: Guoxin Li, Nanfang Hospital, Southern MedicalUniversity, 1838 North Guangzhou Avenue, Guangzhou 510515, China. Phone:86-20-6164-1682; Fax: 86-20-6164-1683; E-mail: [email protected]; ShirongCai, Department of Gastrointestinal Surgery, the First Affiliated Hospital, SunYat-sen University, Guangzhou 510700, Guangdong, China. E-mail:[email protected]; Tuanjie Li, Department of General Surgery, NanfangHospital, Southern Medical University, 1838 North Guangzhou Avenue,Guangzhou 510515, China. Phone: 86-20-6278-7170; E-mail:[email protected];andQiZhang,GuangdongKeyLaboratoryof LiverDiseaseResearch, the Third Affiliated Hospital of Sun Yat-sen University, Guangzhou510630, China. Phone: 86-20-85253106; Fax: 86-20-5252276; E-mail:[email protected].

doi: 10.1158/1078-0432.CCR-18-0848

�2018 American Association for Cancer Research.

ClinicalCancerResearch

Clin Cancer Res; 24(22) November 15, 20185574

on February 26, 2021. © 2018 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst July 24, 2018; DOI: 10.1158/1078-0432.CCR-18-0848

http://crossmark.crossref.org/dialog/?doi=10.1158/1078-0432.CCR-18-0848&domain=pdf&date_stamp=2018-10-24

http://clincancerres.aacrjournals.org/

relative prognostic value thanpT stage, pN stage, lympho-vascularinvasion, tumor differentiation, and microsatellite instability(MSI) status (16). Two immunoscores of GC also showed greatprognostic values, which were derived from 5 features (CD3 IM,CD3 CT, CD8IM, CD45RO CT, and CD66b IM) and 11 types ofimmune cell fraction, respectively (17, 18). Our previous studiesalso revealed that tumor-infiltrating lymphocytes, myeloid cells,and angiogenesis play critical roles in GC progression, and com-bining multiple immune biomarkers would substantiallyimprove the prognostic value (10, 17, 19). Furthermore, a survivalprediction model based on specific tumor and patient clinico-pathologic characteristics could be used to predict survival benefitfrom adjuvant chemotherapy for patients with stage II or stage IIIGC (20). On the basis of these findings, we hypothesize thatcombiningmultiple clinicopathologic features and immunomar-kers of immune cells and angiogenesis in tumor can improveoverall prediction of GC outcome.

Recently, several supervised learningmethods, such as decisiontrees, have been applied to the analysis of cDNA or tissue micro-arrays to refine prognosis in breast cancer, nasopharyngeal carci-noma, and non–small cell lung cancer (21). State-of-the-artclassification algorithms such as support vector machines (SVM)can be used to select a small subset of highly discriminatingmarkers and patients or disease attributes to build reliable cancerclassifiers (22, 23).

Therefore, in this study, integrating multiple clinicopathologicfeatures and immunomarkers, we developed an SVM-based GCprognostic classifier (GC-SVM) to predict overall survival (OS)and disease-free survival (DFS), and explored whether GC-SVMclassifier could identify the patients with stage II and III GC whomight benefit more from postoperative adjuvant chemotherapy.

Materials and MethodsPatients and tissue samples

The study enrolled three independent cohorts of patients withGC. The training cohort and internal validation cohort thatcomprised 251 consecutive patients and 248 consecutive patientswith total or partial radical gastrectomy were obtained fromNanfang Hospital of Southern Medical University (Guangzhou,

China) between January 2005 and August 2007, September 2007and August 2009, respectively. The external validation cohortcomprising 287 consecutive patients was obtained from the FirstAffiliated Hospital of Sun Yat-sen University (SYSU) betweenJanuary 2005 and December 2007 with same enrollment criteria.Use of human tissues with informed consent from patients wasapproved by the Clinical Research Ethics Committee of eachhospital. Clinical baseline data were retrospectively collectedfor each patient. The clinical sources of the 786 patients withGC are listed in Table 1. Inclusion criteria were availability ofhematoxylin and eosin slides with invasive tumor components,availability of follow-up data and clinicopathologic characteris-tics, no history of cancer treatment, and appropriate patient-informed consent. We excluded patients if formalin-fixedparaffin-embedded (FFPE) tumor (CT and IM) and normal sam-ples from the initial diagnosis were unavailable or if they hadreceived previous treatment with any anticancer therapy. Twoindependent pathologists reassessed all these samples. This studywas conducted in accordance with the Declaration of Helsinki.Written-informed consent was obtained from all patients,and this study was approved by the Review Boards at NanfangHospital of Southern Medical University and the First AffiliatedHospital of SYSU. Data were analyzed from July 21, 2017, toDecember 2, 2017.

Baseline information for each patients with GC, including age,gender, American Society of Anesthesiologists score, EasternCooperative Oncology performance status, Charlson comorbid-ity index, tumor location, tumor size, differentiation, Lauren type(17), carcinoembryonic antigen (CEA), cancer antigen 19-9(CA19-9), TNM staging at surgery, postsurgical chemotherapy,and follow-up data (follow-up duration and survival), was docu-mented. The TNM staging was reclassified according to theseventh edition of the American Joint Committee on Cancer(AJCC) Cancer Staging Manual of the American Joint Committeeon Cancer/International Union Against Cancer (24). The tumorsize was defined according to the longest diameters of the sam-ples. Patients diagnosed at advanced stage or early-stage tumorsthat have excessive lymph node metastasis were candidates forreceiving postsurgical chemotherapy. Collectively, there were 106(40.5%), 157 (63.3%), and 138 (48.1%) patients who received5-fluorouracil–based postsurgical chemotherapy in the 3 cohorts,respectively. Of the 401 patients treated with postoperative che-motherapy, 126 (31.4%) patients received the XELOX (capecita-bine–oxaliplatin) regimen, 271 (67.6%) patients received theFOLFOX (fluorouracil–folinic acid–oxaliplatin) regimen, andonly 4 (1.0%) patients received 5-FU treatment alone (Supple-mentary Table S1). Follow-up data were collected from hospitalrecords for patients who were lost to follow-up. The follow-updurationwasmeasured from the time of surgery to the last follow-up date, and information regarding the survival status at the lastfollow-up was collected.

Immunohistochemistry and image analysisOn the basis of previous study findings, we selected eight

molecular markers involved in different aspects of GC develop-ment and metastasis in the present study, including sevenimmune cell biomarkers [CD3 (pan T cells), CD8 (cytotoxicT cells), CD45RO (memory T cells), CD45RA (na€�ve T cells),CD57 (natural killer cells), CD68 (macrophages), CD66b(neutrophils)] and amicrovascularmarker (CD34). FFPE sampleswere cut into 4-mm sections, which were then processed for

Translational Relevance

Tumor–node–metastasis (TNM) staging system of gastriccancer (GC) is not adequate for definition of prognosis andcannot predict the candidates who are likely to benefit fromchemotherapy. In this research, we constructed an SVM-basedGC prognostic classifier (GC-SVM) integrating 3 clinicopath-ologic features and 8 immunohistochemistry features in thetraining cohort of 251 patients. And further validation of theGC-SVM classifier was performed in 2 validation cohorts of535 patients. Multivariate analysis revealed that the GC-SVMclassifier was an independent prognostic factor. Furthermore,the classifier had higher predictive accuracy for overall survivaland disease-free survival than TNM stage and can add prog-nostic value to the TNM staging system. Moreover, theGC-SVM classifier might be able to predict which patients willbenefit fromadjuvant chemotherapy. Thus, the classifier couldfacilitate patient counseling and individualized management.

Immunomarker SVM–Based Predictive Classifier

www.aacrjournals.org Clin Cancer Res; 24(22) November 15, 2018 5575




immunohistochemistry (IHC) as previously described (10, 17).Detailed information was provided in the SupplementaryMateri-als. Every staining run contained a slide treated with PBS buffer inplace of the primary antibody as a negative control (25, 26). Slidesections of lymph nodes were employed as positive control forimmune cell staining. Slide sections of GC tumor tissues that werepreviously verified with CD34 overexpression were chosen as

positive controls for CD34 staining. Every staining run containeda slide of positive control. Prior to staining, the sections weresubjected to endogenous peroxidase blocking in 1% H2O2

solution diluted in methanol for 10 minutes and then heated ina microwave for 30 minutes with 10 mmol/L citrate buffer(pH 6.0). Serum blocking was performed using 10% normalrabbit serum for 30 minutes. Furthermore, all slides were stainedwith the same concentrations of primary antibody for eachantibody and incubated with monoclonal primary antibodyovernight at 4�C. The reaction was visualized using diaminoben-zidine (DAB) þ chromogen, and nucleus was counterstainedusing hematoxylin. And all slides were stained with DAB dyeingfor the same time for each antibody (Supplementary Table S2).

The IHC results were evaluated by two independent gastroen-terology pathologists whowere blinded to the clinical data. At lowpower (100), the tissue sections were screened using an invertedresearch microscope (model DM IRB; Leica Germany), and thefive most representative fields were selected. Thereafter, to eval-uate the density of stained immune cells, the two respective areasof CT and IM were measured at 200� magnification. The nucle-ated stained cells in each areawere quantified and expressed as thenumber of cells per field. For immune cells and microvessels, thestaining intensity was evaluated by two independent gastroen-terology pathologists with great experience. The pathologistsdetermined whether each immune cell or microvessel was pos-itive or not (10, 17, 19, 27, 28). Themicrovessel density (MVD) inGC tumor tissueswas evaluatedby staining forCD34.Anydiscretecluster or single cell stained for CD34 was counted as onemicrovessel (10, 27). Five representative fields were quantified,and the average number of microvessels per field (�200) waspresented as theMVD. Two pathologists independently scored allsamples blindly with regard to clinical characteristics and prog-nosis. Their results were in complete agreement in 90% of thecases. A third pathologist was consulted when different opinionsarose between the two primary pathologists. If the third pathol-ogist agreed with one of them, then that value was selected. Ifthe conclusion by the third pathologist was completely different,then the three of them would work collaboratively to find acommon answer. We selected the optimum cutoff score forevery feature using X-tile software version 3.6.1 (Yale UniversitySchool of Medicine, New Haven, CT) based on the associationwith the patients' OS.

Prognosis prediction using SVM-based methodsSVM was introduced by Vapnik (29) for data classification and

function approximation. An SVM is a binary classifier trained on aset of labeled patterns called training samples. The purpose oftraining an SVM is to find a hyperplane that divides these samplesinto two sides so that all the points with the same label will be onthe same side of the hyperplane (22, 29–32). In this study, SVMwas used to predict whether a patient died within 5 years. Weadopted the SVM-recursive feature elimination algorithm to selectand rank useful features (22). To investigate the possibility ofidentifying different prognostic subsets of patients based ontheir clinicopathologic features and immunomarkers usingSVM, we performed a set of experiments in the training cohortof 251 patients; the developed GC-SVM classifier was furthervalidated in 535 patients from 2 independent cohorts. In thetraining cohort, patients on the side of the hyperplane who hadbetter survival were classified into high GC-SVM group. The SVMdata processing methods were conducted as previously described

Table 1. Patient and tumor characteristics in the training and validation cohorts

Trainingcohort(N ¼ 251)

Internalvalidationcohort(N ¼ 248)

Externalvalidationcohort(N ¼ 287)

Variable N (%) N (%) N (%)

Age (years), median (IQR) 57 (50–64) 55 (46–64) 58 (49–67)GenderMale 180 (71.7) 151 (60.9) 184 (64.1)Female 71 (28.3) 97 (39.1) 103 (35.9)

ASA scoreI 125 (49.8) 127 (51.2) 143 (49.8)II 121 (48.2) 117 (47.2) 138 (48.1)III 5 (2.0) 4 (1.6) 6 (2.1)

ECOG PS0 178 (70.9) 178 (71.8) 198 (69.0)1 69 (27.5) 67 (27.0) 84 (29.3)2 4 (1.6) 3 (1.2) 5 (1.7)

Charlson comorbidity index0 169 (67.3) 165 (66.5) 190 (66.2)1 60 (23.9) 64 (25.8) 71 (24.7)2 18 (7.2) 15 (6.0) 21 (7.3)3 4 (1.6) 4 (1.6) 5 (1.7)

Tumor size (cm)<4 126 (50.2) 152 (61.3) 145 (50.5)�4 125 (49.8) 96 (38.7) 142 (49.5)

Tumor locationCardia of stomach 49 (19.5) 43 (17.3) 78 (27.2)Body of stomach 40 (15.9) 47 (19.0) 82 (28.6)Antrum of stomach 127 (50.6) 121 (48.8) 113 (39.4)Whole 35 (13.9) 37 (14.9) 14 (4.9)

Differentiation statusWell þ moderate 130 (51.8) 94 (37.9) 81 (28.2)Poor and undifferentiated 121 (48.2) 154 (62.1) 206 (71.8)

Lauren typeIntestinal type 202 (80.5) 169 (68.1) 187 (65.2)Diffuse or mixed 49 (19.5) 79 (31.9) 100 (34.8)

CEAElevated 69 (27.5) 50 (20.2) 44 (15.3)Normal 182 (72.5) 198 (79.8) 243 (84.7)

CA199Elevated 74 (29.5) 62 (24.9) 75 (26.1)Normal 177 (70.5) 187 (75.1) 212 (73.9)

Depth of invasionT1 þ T2 57 (22.7) 52 (21.0) 62 (21.6)T3 þ T4 194 (77.3) 196 (79.0) 225 (78.4)

Lymph node metastasisN0 73 (29.1) 71 (28.6) 100 (34.8)N1 42 (16.7) 55 (22.2) 114 (39.7)N2 83 (33.1) 64 (25.8) 45 (15.7)N3 53 (21.1) 58 (23.4) 28 (9.8)

TNM stageI 31 (12.4) 41 (16.5) 50 (17.4)II 56 (22.3) 38 (15.3) 52 (18.1)III 139 (55.4) 135 (54.4) 141 (49.1)IV 25 (10.0) 34 (13.7) 44 (15.3)

ChemotherapyNo 145 (59.5) 91 (36.7) 149 (51.9)Yes 106 (40.5) 157 (63.3) 138 (48.1)

Abbreviation: ECOG PS, Eastern Cooperative Oncology performance status.

Jiang et al.

Clin Cancer Res; 24(22) November 15, 2018 Clinical Cancer Research5576




(22, 31, 32). The programs were coded using R software; scriptsare available on request.

Statistical analysisWe compared two groups using the t test for continuous

variables and c2 test for categorical variables. Survival curves weredepicted according to the Kaplan–Meier method and comparedusing the log-rank test. In univariable analysis, survival curves fordifferent variable values were generated using the Kaplan–Meiermethod andwere compared using the log-rank test. Variables thatreached significance with P < 0.05 were entered into the multi-variable analyses using the Cox regression model. Interactionsbetween the classifier and treatment were detected bymean of theCox model as well. All statistical analyses were performed usingR software (version 3.0.1) and SPSS software (version 19.0).Statistical significance was set at 2-sided P < 0.05.

ResultsPatient characteristics and components of the GC-SVMclassifier

The clinicopathologic characteristics for the training cohort(n ¼ 251), internal (n ¼ 248), and external (n ¼ 287) validationcohorts were listed in Table 1. Of the 786 patients included in thestudy, 515 (65.5%) were men, and the median [interquartilerange (IQR)] age of all patients was 57 (49–65) years. In thetraining cohort, the median (IQR) survival time for DFS and OSwere 26 (7–73) and 36 (15–76) months, respectively. In theinternal and external validation cohorts, the median (IQR) sur-vival time for DFS and OS were 32 (9–75) and 51 (15–75)months, 33 (11–80) and 40 (15–82) months, respectively.

Weused X-tile plots to generate the optimum cutoff score for all15 IHC features in the training cohort, and Supplementary TableS3 showed the results of the univariate analysis between each ofthe 15 features and the survival in the training cohort.On the basisof the SVM analysis of the training data, the GC-SVM classifierintegrated sex, CEA, lymph node metastasis, and expression ofeight features, including CD3IM, CD3CT, CD8IM, CD45ROCT,CD57IM, CD68CT, CD66bIM, and CD34, as critical factors. Uni-variate associations of the GC-SVM classifier, clinicopathologicparameters, and the expression of each of the evaluated featureswith OS and DFS in the training and validation cohorts wereshown in Supplementary Tables S3 to S5. Representative IHCstaining of all the features in tumor tissues was shown in Sup-plementary Fig. S1.

The ROC curves for traditional clinicopathologic prognosticfactors, including age, sex, tumor size, differentiation, lauren type,CEA, CA19-9, and TNMstage aswell as each immune features andthe overall GC-SVM classifier illustrated the point with the max-imum area under the curve (AUC) for each factor. In the threecohorts, the AUCs of the GC-SVM classifier for 5-year OS and DFS(training cohort: 0.796, 0.805; internal validation cohort: 0.809,0.813; external validation cohort: 0.834, 0.828; respectively;Supplementary Fig. S2; Supplementary Tables S6 and S7) weresignificantly greater than the AUCs for all other prognostic factorsconsidered (next largest AUC for TNM stage, training cohort:0.649, 0.659; internal validation cohort: 0.746, 0.678; externalvalidation cohort: 0.745, 0.737, respectively). SupplementaryTable S8 lists the relationships between the GC-SVM classifierand clinicopathologic characteristics in the training, internal,and external validation cohorts. The GC-SVM classifier was

significantly associated with TNM stage, tumor size, CA199, andlymph node metastasis of GC (Supplementary Table S8).

GC-SVM classifier and GC survivalIn the training cohort, we defined 172 patients as low GC-SVM

and 79 patients as high GC-SVM. The 5-year OS and DFS were15.7%and 10.5%, respectively, for the low-GC-SVMpatients, and78.5% and 68.4%, respectively, for the high-GC-SVM patients[HR 0.113 (0.065–0.197) and 0.138 (0.084–0.224), respectively;all P < 0.0001; Fig. 1A]. We performed the same analyses inthe internal validation cohort. The 5-year OS and DFS were19.0% and 15.3%, respectively, for the low-GC-SVM patients,and 82.0% and 77.5%, respectively, for the high-GC-SVMpatients [HR 0.166 (0.110–0.250) and 0.186 (0.126–0.274),respectively; all P < 0.0001; Fig. 1B]. To confirm that the GC-SVMclassifier had an excellent prognostic value in different popula-tions, we further applied it to the external validation cohortsand found similar results (Fig. 1C). The GC-SVM classifier alsoremained a clinically and statistically significant predictorof prognosis after stratification by clinicopathologic factors(Supplementary Figs. S3–S6).

In univariable analysis, low-GC-SVM patients were associatedwith significantly poorer OS and DFS (Supplementary Tables S3–S5). Variables demonstrating a significant effect on OS and DFSwere included in the multivariable analysis. Multivariate Coxregression analysis after adjustment for clinicopathologic vari-ables andTNMstage revealed that theGC-SVMclassifier remaineda powerful and independent prognostic factor for OS and DFS inthe training, internal, and external validation cohorts (Table 2).

We performed stratified analyses of GC patients with stage I,II, III, and IV disease in the combined internal cohort andexternal validation cohort. High-GC-SVM patients with stage I,II, III, or IV disease had a longer OS and DFS than patients withlow-GC-SVM did both in internal and external cohorts (Fig. 2).Furthermore, the GC-SVM classifier exhibited a higher prog-nostic accuracy than TNM stage, any clinicopathologic riskfactor, or single IHC feature alone (Supplementary Fig. S2;Supplementary Tables S6 and S7).

GC-SVM classifier and adjuvant chemotherapy benefitFurthermore, we investigated whether low- or high-GC-

SVM patients with stage II or III GC could benefit from postop-erative adjuvant chemotherapy. A test for an interaction betweenGC-SVM and adjuvant chemotherapy indicated that, either instage II or III disease, the benefit fromadjuvant chemotherapywassuperior among patients with high GC-SVM [stage II: OS, HR0.156 (0.044–0.554), 0.004; DFS, 0.280 (0.106-0.741), 0.010;stage III: OS, HR 0.472 (0.242–0.919), 0.027; DFS, 0.448 (0.240–0.836), 0.012; all P < 0.0001 for interaction; Table 3] than amongthose with low GC-SVM. The corresponding Kaplan–Meier sur-vival curves for patients with stage II or stage III disease, whichcomprehensively compared lowwith high GC-SVMby treatment,are shown in Fig. 3. The results from the subset analysis using GC-SVM classifier revealed that adjuvant chemotherapy significantlyincreased OS and DFS in the high-GC-SVM group (stage II: P ¼0.001 and P ¼ 0.006; stage III: P ¼ 0.023 and P ¼ 0.009,respectively), but had no significant effect in the low-GC-SVMgroup (stage II: P¼ 0.139 and P¼ 0.395; stage III: P¼ 0.347 andP ¼ 0.394, respectively; Fig. 3). Consequently, these resultssuggest that stage II and III patients with high GC-SVM couldbenefit from adjuvant chemotherapy.






Figure 1.

Kaplan–Meier analysis of DFS andOS according to GC-SVM classifier inpatients with GC. Left plot, OS; rightplot, DFS. A, Training cohort (n¼ 251).B, Internal validation cohort (n¼ 248).C, External validation cohort(n ¼ 287).

Jiang et al.





DiscussionPrognostic assessment is crucial for formation of appropriate

treatment choices. Because GC is a clinically heterogeneous dis-ease, with large variations in the clinical outcomes even amongpatients with the same stage (33, 34), we sought to improve theprediction of GC prognosis by developing a novel 11-featureGC-SVM classifier to categorize patients into low- and high-GC-SVM groups with large differences in 5-year OS and DFS. Coxregression analysis showed the GC-SVM classifier was an inde-pendent prognostic factor for OS and DFS, even after adjustmentfor TNM stage and clinicopathologic characteristics. In addition,ROC analysis suggested that the survival predictive ability of GC-SVM classifier was better than TNM stage and clinicopathologiccharacteristics. Moreover, in stratified analyses with TNM stage,GC-SVM classifier can distinguish each stage patients into low-and high-risk groups with significant differences inOS andDFS inthe internal and external cohorts, supporting the prognostic valueof the classifier and allowing clinicians to potentially identifycandidates for systemic approaches with greater effectiveness toimprove treatment outcomes. Thus, theGC-SVMclassifier can addprognostic value to TNM staging system. Therefore, the GC-SVMclassifier provides clinicians with a valid and reliable tool forbetter prediction of GC prognosis. Ultimately, patients classifiedwith the same TNM stage might be able to be stratified intodifferent risk groups on the basis of the GC-SVM classifier, andthus treated with systemic approaches of different intensities toimprove outcomes.

Adjuvant chemotherapy has been recommended as a stan-dard component of therapies for patients with stage II and IIIGC and improves their outcomes (2, 3). However, not everyonecould benefit from adjuvant chemotherapy and the criterion forthe selection of candidates is still controversial (2, 3, 20, 33). Itis important to identify patients whose tumor will not only besensitive to chemotherapy, but also have overall better out-comes which would prevent excessive toxicities. Assignment oftreatment based in part on tumor molecular characteristics is anincreasingly promising approach (17, 23, 35). Previous studieshave shown that tumor-infiltrated immune cells were associ-ated with chemotherapeutic response in other types of cancer(6, 36–39). In this study, we assessed the association betweenGC-SVM classifier and clinical outcomes in stage II and IIIpatients receiving adjuvant chemotherapy. The resultssuggested that patients with high GC-SVM were easier to obtaina better survival benefit from adjuvant chemotherapy com-pared with those with low GC-SVM, indicating that GC-SVM

classifier could be an important factor for predicting the effi-ciency of chemotherapy. This will be useful for better selectionand management of patients who would receive adjuvantchemotherapy.

Substantial efforts have been made toward identification ofmolecular signatures to predict survival in GC patients, includinggene signatures, miRNAs, and epigenetic biomarkers (18, 40–42).However, these gene-based signatures have not been widelyintroduced into clinical practice as initially expected due to thevariability of measurements in microarray assays, inconsistenciesin assay platforms, and the requirement for analytical expertise(41, 43, 44). Cheong and colleagues described a predictive test forprognosis and response to adjuvant chemotherapy in patientswith localized, resectableGC (40). The test is based on a four-genereal-time RT-PCR assay, whichmeasures gene-expression levels inFFPE tumor tissues, but use of FFPE tumor tissues may decreasethe reliability of RNA quantitation. IHC not only provides asemiquantitative assessment of protein abundance but alsodefines the cellular localization of their expression (40). In thiscase, identification of immunobiomarkers with IHC, which hasbeen widely applied in clinical diagnosis, is found to serve aspromising alternative strategies for the molecular profiling oftumors (31, 45).

In recent years, immune profiling studies have reached aforefront position in research of solid tumors, including GC.Several studies showed that tumor-infiltrating CD3þ, CD4þ,CD8þ, CD57þ, CD45ROþ, and CD45RAþ cells were associatedwith better survival in patients with GC, whereas CD66bþ andCD68þ cellswere associatedwith significantly aworse outcome inpatients with GC (8, 10, 17, 19, 46). A recent meta-analysissummarized the impact of immune cells, includingB cells, naturalkiller (NK) cells, myeloid-derived suppressor cells, macrophages,and all subsets of T cells on clinical outcome frommore than 120published articles (9). Importantly, the beneficial impact of theimmune infiltration with cytotoxic and memory T-cell pheno-types has been demonstrated in cancers of diverse anatomicalsites, including not only GC but also malignant melanoma, lung,colorectal, esophageal, breast, and bladder cancers (9). Ourprevious studies also showed that tumor-infiltrating NK cellpredicted a good prognosis and infiltrated neutrophils wereinversely correlated with survival in patients with GC (10, 17,19), which was consistent with the results of other studies (8, 18).In this study, the GC-SVM classifier, including eight features(CD3 IM, CD3 CT, CD8 IM, CD45RO CT, CD57 IM, CD68 CT,CD66b IM, and CD34), could effectively predict survival andcomplemented the prognostic value of the TNM staging system.

Table 2. Multivariable cox regression analysis of the GC-SVM classifier, TNM stage, and survival in the training, internal, and external validation cohorts

DFS OSVariable HR (95% CI) P value HR (95% CI) P value

Training cohort (N ¼ 251)GC-SVM (high vs. low) 0.167 (0.102–0.275) <0.0001 0.137 (0.078–0.242) <0.0001Stage (IIIþIV vs. IþII) 1.399 (1.227–1.595) <0.0001 1.357 (1.180–1.562) <0.0001

Internal validation cohort (N ¼ 248)GC-SVM (high vs. low) 0.209 (0.141–0.309) <0.0001 0.188 (0.124–0.286) <0.0001Stage (IIIþIV vs. IþII) 1.320 (1.144–1.523) <0.0001 1.339 (1.148–1.561) <0.0001CA199 (high vs. low) 1.467 (1.027–2.096) 0.035 1.671 (1.165–2.398) 0.005Differentiation (low vs. wellþmoderate) / / 1.536 (1.059–2.227) 0.024

External validation cohort (N ¼ 287)GC-SVM (high vs. low) 0.173 (0.115–0.260) <0.0001 0.159 (0.104–0.244) <0.0001Stage (IIIþIV vs. IþII) 1.232 (1.085–1.399) 0.001 1.258 (1.103–1.436) 0.001CA199 (high vs. low) 1.618 (1.171–2.234) 0.003 1.495 (1.076–2.078) 0.017






Tumor progression is the product of evolving crosstalk betweenmalignant cells and various stromal and immune cell subsets ofthe surroundingmicroenvironment (47, 48). During this process,

the tumor cells interact with their microenvironment, which iscomplex and composed of stromal and immune cells that pen-etrate the tumor site via blood vessels and lymphoid capillaries

Figure 2.

Kaplan–Meier survival analysis of OSand DFS according to TNM stageand the GC-SVM classifier in patientswith GC in the internal and externalvalidation cohorts. A, Internalvalidation cohort: stage I (n ¼ 41);stage II (n ¼ 38); stage III (n ¼ 135);stage IV (n ¼ 34). B, Externalvalidation cohort: stage I (n ¼ 50);stage II (n ¼ 52); stage III (n ¼ 141);stage IV (n ¼ 44).

Jiang et al.





(49). All subsets of immune cells can be found in tumors, but theirrespective density, functionality, and organization may vary indifferent spatial locations of tumor (10, 28, 49–51). It is knownthat immune cells are scattered in the CTwithin the tumor stromaand the tumor glands, in the IM, and in organized lymphoidfollicles distant from the tumor. A statistically significant corre-lation between immune cell density in either tumor region (CT orIM) and patient outcome has been shown in gastric and colorectalcancers (10–12, 17). Given the major clinical importance ofdistinct tumor regions, it is appropriate to conduct immune cellinfiltration evaluation systematically in the two separate areas, theCT and the IM (11, 12, 17). After ROC curve–based optimization,although the AUC values of some markers in the IM and CT weresimilar in this study,mostmarkers in IMandCT still had relativelylarge differences of predictive abilities, such as CD66b (AUCvalues for 5-year death: IM, 0.657; CT, 0.582; in the trainingcohort) and CD45RO (AUC values for 5-year death: IM, 0.620;CT, 0.654; in the training cohort). Besides, Tumeh and colleaguesshowed that preexisting CD8þ T cells distinctly located at theinvasive tumor margin were associated with the expression of thePD-1/PD-L1 immune inhibitory axis and might predict responseto therapy (52). Tumor-infiltrating lymphocyte densities at the IMof liver metastases could predict response to chemotherapy inmetastatic colorectal cancer (36). Our previous studies also foundthat the infiltrating neutrophils in the IMof tumor aremuchmorethan those in the CT and have a higher prognostic value (10).Thus, it could be considered that there are biological significancesof the spatial information and further research is needed.

Compared with other machine learning algorithms, SVM isbetter suited to manage classification based on high-dimensionaldata with a limited number of training samples to select the mostefficient of all available features (29, 30). Previous studies haveshown that single biomarker has limited prognostic value for GC(8, 17, 23). To improve the prognostic predictive value of indi-vidualmarkers, SVM can combine clinicopathologic features withindependently informative markers to predict disease outcome.Our GC-SVM classifier integrates patient sex, CEA, lymph nodemetastasis along with eight features, including CD3IM, CD3CT,CD8IM, CD45ROCT, CD57IM, CD68CT, CD66bIM, and CD34.Furthermore, the GC-SVM classifier was substantially morestrongly associated with OS and DFS than any individual com-ponent. We adopted the SVM-recursive feature elimination algo-rithm to select the features and developed the GC-SVM classifier,including 11 features. When integrating these 11 features, theaccuracy of the model was the highest in the training cohort. TheSVM model also accounted for the interactions of all features(29, 30, 53). Though not all these features had the highestprognostic value, the accuracy of the model was the best. How-ever, the underlying biological reasons are not very clear, andfurther research should explore the potential biological reasons.

Comparedwith the immunoscores andother prognosticmodel ofGC reported by previous studies (11, 12, 17, 54, 55), our GC-SVMclassifier comprehensively integrated information of tumor-infiltrating lymphocyte features, myeloid cell features, and clin-icopathologic features using SVM algorithms that could signifi-cantly improve its predictive accuracy. Thus, our results indicatedthat the GC-SVM classifier was able to select themost informativefactors that contributed independently and collectively to theprediction of prognosis.

In The Cancer Genome Atlas project, GC was divided into foursubtypes based on themolecular classification: Epstein–Barr virus(EBV)-positive, MSI, genomically stable, and chromosomal insta-bility (56). Subsequently, various hypotheses were proposed todescribe the impact of molecular processes on the intensity andnature of the host immune response (57, 58). Immune cells ofmany different types are frequently observed in primary GC, andcertain molecular subtypes of GC, such as with EBV-positive andwithMSI, arewell known tobe associatedwith ahigh lymphocyticinfiltrate (57–59). About 10% of the tumors are EBV-positive,which display recurrent PIK3CA mutations, extreme DNA hyper-methylation, and amplification of JAK2, PD-L1, and PD-L2 (56).It is noteworthy that aberrant methylation can be induced byinfectious agents such as Helicobacter pylori or EBV infection(42, 60, 61). According to a phase II study, mismatch repairdeficiency renders different solid tumors highly sensitive toimmune checkpoint blockade with the PD-1 inhibitor pembro-lizumab, and these tumors contain prominent immune infiltrates(62, 63). Therefore, future studies should investigate the associ-ation between these immunobiomarkers included in the classifierand molecular classification based on GC causes, and explorewhether the classifier can predict the responses of patients withGC to immunotherapy.

Our study has several limitations. First, it was retrospective innature and all specimens were obtained frompatients in southernChina. The full chemotherapy details might not be available forthe entire cohorts. Therefore, our results need aprospective, larger,multicentered randomized trial to validate. Furthermore, themechanism for the predictive value of multifeature classifierpredicting is not very clear, and further investigationmay providemore information for better understanding of the roles of thesefeatures in the development and progression of GC and provideadditional information and strategies for treatment (38, 39,64, 65). Another limitation is that the constrained number ofbiomarkers screened in the training cohort, which in turn resultedin a smaller panel of features integrated into theGC-SVMclassifierthan in some gene expression profiling studies by cDNA array(40–42). Although the GC-SVM classifier was a highly accuratepredictor of OS andDFS, we are aware that other biomarkers mayextend the precision and predictive value of the classifier, and newmarkers are being found and new techniques developed every

Table 3. Treatment interaction with GC-SVM classifier for DFS and OS in patients with stage II and III disease

DFS OS

GC-SVM CT No CT CT vs. No CT, HR (95% CI) PP value forinteraction CT vs. No CT, HR (95% CI) P

P value forinteraction

Stage II (n ¼ 146)GC-SVM high 46 35 0.280 (0.106–0.741) 0.010 <0.0001 0.156 (0.044–0.554) 0.004 <0.0001GC-SVM low 34 31 0.780 (0.436–1.395) 0.402 0.641 (0.352–1.166) 0.145

Stage III (n ¼ 415)GC-SVM high 74 51 0.448 (0.240–0.836) 0.012 <0.0001 0.472 (0.242–0.919) 0.027 <0.0001GC-SVM low 134 156 0.900 (0.701–1.155) 0.406 0.887 (0.688–1.144) 0.356

Abbreviation: CT, chemotherapy.






Figure 3.

Association between GC-SVM classifier and survival benefit from chemotherapy. A and B, In TNM stage II disease, GC-SVM–high patients (n¼ 81) could significantlybenefit from chemotherapy (OS, P ¼ 0.001; DFS, P ¼ 0.006), whereas no significant difference was observed with respect to OS and DFS in GC-SVM–lowpatients (n ¼ 65) when they were given adjuvant chemotherapy (n ¼ 34) or not (n ¼ 31). C and D, In TNM stage III disease, GC-SVM–high patients (n ¼ 125) couldsignificantly benefit from chemotherapy (OS,P¼0.023; DFS,P¼0.009), whereas no significant differencewas observedwith respect toOS andDFS inGC-SVM–lowpatients (n ¼ 290) when they were given adjuvant chemotherapy (n ¼ 134) or not (n ¼ 156).

Jiang et al.





year. Thus, the GC-SVM classifier may be further improved byincluding additional markers.

In conclusion, the study demonstrates that the GC-SVM clas-sifier can accurately distinguish GC patients with substantiallydifferent OS and DFS. Furthermore, the GC-SVM classifier couldidentify a subgroup of patients with stage II and III disease whocould benefit from adjuvant chemotherapy. Thus, the GC-SVMclassifier might facilitate patient counseling, decision-makingregarding individualized therapy, and follow-up scheduling.

Disclosure of Potential Conflicts of InterestNo potential conflicts of interest were disclosed.

Authors' ContributionsConception and design: L. Huang, Y. Hu, Q. Zhang, T. Li, S. Cai, G. LiDevelopment of methodology: J. Xie, W. Liu, L. Huang, T. Li, G. LiAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): Y. Jiang, J. Xie, Z. Han, W. Liu, S. Xi, L. Huang,W. Huang, T. Lin, L. Zhao, Y. Hu, J. Yu, T. Li, S. Cai, G. LiAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): Y. Jiang, J. Xie, Z. Han, W. Liu, S. Xi, L. Huang,W. Huang, L. Zhao, Y. Hu, J. Yu, Q. Zhang, S. Cai, G. Li

Writing, review, and/or revisionof themanuscript:Y. Jiang, J. Xie, Z.Han, S. Xi,L. Huang, W. Huang, L. Zhao, Y. Hu, J. Yu, Q. Zhang, T. Li, S. Cai, G. LiAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): Y. Jiang, J. Xie, Z. Han, W. Liu, S. Xi, L. Huang,W. Huang, T. Lin, L. Zhao, Y. Hu, J. Yu, Q. Zhang, T. Li, S. Cai, G. LiStudy supervision: Q. Zhang, T. Li, S. Cai, G. Li

AcknowledgmentsThis work was supported by grants from the National Natural Science

Foundation of China (81672446, 81600510, 81370575, and 81570593),Key Clinical Specialty Discipline Construction Program (2017YFC0108300),the National Key Research and Development Program of China(2017YFC0108300), Natural Science Foundation of Guangdong Province(2014A030313131), Science and Technology Planning Project of Guangzhou(2014B020228003, 2014B030301041, and 2015A030312013), and Director'sFoundation of Nanfang Hospital (2016B010).

The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely to indicatethis fact.

Received March 14, 2018; revised June 6, 2018; accepted July 17, 2018;published first July 24, 2018.

References1. Torre LA, Bray F, Siegel RL, Ferlay J, Lortet-Tieulent J, Jemal A. Global cancer

statistics, 2012. Cancer J Clin 2015;65:87–108.2. Noh SH, Park SR, Yang HK, Chung HC, Chung IJ, Kim SW, et al. Adjuvant

capecitabine plus oxaliplatin for gastric cancer after D2 gastrectomy(CLASSIC): 5-year follow-up of an open-label, randomised phase 3 trial.Lancet Oncol 2014;15:1389–96.

3. Nishida T. Adjuvant therapy for gastric cancer after D2 gastrectomy. Lancet2012;379:291–2.

4. Sasako M, Inoue M, Lin JT, Khor C, Yang HK, Ohtsu A. Gastric cancerworking group report. Japanese J Clin Oncol 2010;40:i28–37.

5. Galon J, Pages F, Marincola FM, Angell HK, ThurinM, Lugli A, et al. Cancerclassification using the Immunoscore: a worldwide task force. J Transl Med2012;10:205.

6. Zhou L, Xu L, Chen L, Fu Q, Liu Z, Chang Y, et al. Tumor-infiltratingneutrophils predict benefit from adjuvant chemotherapy in patients withmuscle invasive bladder cancer. Oncoimmunology 2017;6:e1293211.

7. Powell DR, Huttenlocher A. Neutrophils in the tumor microenvironment.Trends Immunol 2016;37:41–52.

8. Liu K, Yang K, Wu B, Chen H, Chen X, Chen X, et al. Tumor-infiltratingimmune cells are associated with prognosis of gastric cancer. Medicine(Baltimore) 2015;94:e1631.

9. FridmanWH, Pages F, Sautes-Fridman C, Galon J. The immune contexturein human tumours: impact on clinical outcome. Nat Rev Cancer 2012;12:298–306.

10. Li TJ, Jiang YM, Hu YF, Huang L, Yu J, Zhao LY, et al. Interleukin-17-producing neutrophils link inflammatory stimuli to disease progression bypromoting angiogenesis in gastric cancer. Clin Cancer Res 2017;23:1575–85.

11. Galon J,Mlecnik B, BindeaG, AngellHK, Berger A, Lagorce C, et al. Towardsthe introduction of the 'Immunoscore' in the classification of malignanttumours. J Pathol 2014;232:199–209.

12. Galon J, Costes A, Sanchez-Cabo F, Kirilovsky A, Mlecnik B, Lagorce-PagesC, et al. Type, density, and location of immune cells within humancolorectal tumors predict clinical outcome. Science 2006;313:1960–4.

13. Galon J, Pages F, Marincola FM, Thurin M, Trinchieri G, Fox BA, et al.The immune score as a new possible approach for the classification ofcancer. J Transl Med 2012;10:1.

14. Kirilovsky A, Marliot F, El Sissy C, Haicheur N, Galon J, Pages F. Rationalbases for the use of the Immunoscore in routine clinical settings as aprognostic and predictive biomarker in cancer patients. Int Immunol2016;28:373–82.

15. Angell H, Galon J. From the immune contexture to the Immunoscore: therole of prognostic and predictive immune markers in cancer. Curr OpinImmunol 2013;25:261–7.

16. Pages F, Mlecnik B, Marliot F, Bindea G, Ou FS, Bifulco C, et al.International validation of the consensus Immunoscore for the classi-fication of colon cancer: a prognostic and accuracy study. Lancet2018;391:2128–39.

17. Jiang Y, Zhang Q, Hu Y, Li T, Yu J, Zhao L, et al. ImmunoScore signature:a prognostic and predictive tool in gastric cancer. Ann Surg 2018;267:504–13.

18. Zeng D, Zhou R, Yu Y, Luo Y, Zhang J, Sun H, et al. Gene expressionprofiles for a prognostic immunoscore in gastric cancer. Br J Surg 2018Apr 25. [Epub ahead of print].

19. Li T, Zhang Q, Jiang Y, Yu J, Hu Y, Mou T, et al. Gastric cancer cells inhibitnatural killer cell proliferation and induce apoptosis via prostaglandin E2.Oncoimmunology 2016;5:e1069936.

20. Jiang YM, Li TJ, Liang XL, Hu YF, Huang L, Liao ZC, et al. Association ofadjuvant chemotherapy with survival in patients with stage II or III gastriccancer. JAMA Surg 2017;152:e171087.

21. Chen HY, Yu SL, Chen CH, Chang GC, Chen CY, Yuan A, et al. A five-genesignature and clinical outcome in non-small-cell lung cancer. N Engl J Med2007;356:11–20.

22. Wang HY, Sun BY, Zhu ZH, Chang ET, To KF, Hwang JS, et al. Eight-signature classifier for prediction of nasopharyngeal [corrected] carcinomasurvival. J Clin Oncol 2011;29:4516–25.

23. Jiang Y, Liu W, Li T, Hu Y, Chen S, Xi S, et al. Prognostic and predictivevalue of p21-activated Kinase 6 associated support vector machineclassifier in gastric cancer treated by 5-fluorouracil/oxaliplatin chemo-therapy. EBioMedicine 2017;22:78–88.

24. Dindo D, Demartines N, Clavien PA. Classification of surgical complica-tions: a new proposal with evaluation in a cohort of 6336 patients andresults of a survey. Ann Surg 2004;240:205–13.

25. Khabaz MN, Abdelrahman A, Butt N, Damnhory L, Elshal M, AldahlawiAM, et al. Immunohistochemical staining of leptin is associatedwith grade,stage, lymph node involvement, recurrence, and hormone receptor phe-notypes in breast cancer. Bmc Womens Health 2017;17:105.

26. Lin H,Wong RP, MartinkaM, Li G. Loss of SNF5 expression correlates withpoor patient survival in melanoma. Clin Cancer Res 2009;15:6404–11.

27. Fang JH, ZhouHC, ZengCX, Yang J, Liu YL,HuangXZ, et al.MicroRNA-29bsuppresses tumor angiogenesis, invasion, and metastasis by regulatingmatrix metalloproteinase 2 expression. Hepatology 2011;54:1729–40.

28. Kuang DM, Zhao Q, Wu Y, Peng C, Wang J, Xu Z, et al. Peritumoralneutrophils link inflammatory response to disease progression byfostering angiogenesis in hepatocellular carcinoma. J Hepatol2011;54:948–55.

29. Vapnik VN. An overview of statistical learning theory. IEEE transactions onneural networks. IEEE Neural Networks Council 1999;10:988–99.






30. Choi H, Yeo D, Kwon S, Kim Y. Gene selection and prediction for cancerclassification using support vector machines with a reject option. ComputStat Data An 2011;55:1897–908.

31. Xu J, Ding T, He Q, Yu XJ, Wu WC, Jia WH, et al. An in situ molecularsignature to predict early recurrence in hepatitis B virus-related hepatocel-lular carcinoma. J Hepatol 2012;57:313–21.

32. Zhu ZH, Sun BY, Ma Y, Shao JY, Long H, Zhang X, et al. Three immuno-marker support vector machines-based prognostic classifiers for stage IBnon-small-cell lung cancer. J Clin Oncol 2009;27:1091–9.

33. Razzak M. Genetics: new molecular classification of gastric adenocarcino-ma proposed by the cancer genome atlas. Nat RevClinOncol 2014;11:499.

34. McLeanMH,El-Omar EM.Genetics of gastric cancer.Nat RevGastroenterolHepatol 2014;11:664–74.

35. Zhu CQ, Ding K, Strumpf D, Weir BA, Meyerson M, Pennell N, et al.Prognostic and predictive gene signature for adjuvant chemotherapy inresected non-small-cell lung cancer. J Clin Oncol 2010;28:4417–24.

36. Halama N, Michel S, Kloor M, Zoernig I, Benner A, Spille A, et al.Localization and density of immune cells in the invasive margin of humancolorectal cancer liver metastases are prognostic for response to chemo-therapy. Cancer Res 2011;71:5670–7.

37. Andre F, Berrada N, Desmedt C. Implication of tumor microenvironmentin the resistance to chemotherapy in breast cancer patients. Curr OpinOncol 2010;22:547–51.

38. Zitvogel L, Apetoh L, Ghiringhelli F, Andre F, Tesniere A, Kroemer G. Theanticancer immune response: indispensable for therapeutic success? J ClinInvest 2008;118:1991–2001.

39. Galluzzi L, Buque A, KeppO, Zitvogel L, KroemerG. Immunological effectsof conventional chemotherapy and targeted anticancer agents. Cancer Cell2015;28:690–714.

40. Cheong JH, YangHK, KimH, KimWH, Kim YW, KookMC, et al. Predictivetest for chemotherapy response in resectable gastric cancer: a multi-cohort,retrospective analysis. Lancet Oncol 2018;19:629–38.

41. Ueda T, Volinia S, Okumura H, Shimizu M, Taccioli C, Rossi S, et al.Relation between microRNA expression and progression and prognosis ofgastric cancer: a microRNA expression analysis. Lancet Oncol2010;11:136–46.

42. Chia NY, Tan P. Molecular classification of gastric cancer. Ann Oncol2016;27:763–9.

43. Zhu ZH, Sun BY, Ma Y, Shao JY, Long H, Zhang X, et al. Three immuno-marker support vector machines-based prognostic classifiers for stage IBnon-small-cell lung cancer. J Clin Oncol 2009;27:1091–9.

44. Widschwendter M, Jones A, Evans I, Reisel D, Dillner J, Sundstrom K, et al.Epigenome-based cancer risk prediction: rationale, opportunities andchallenges. Nat Rev Clin Oncol 2018;15:292–309.

45. Mlecnik B, Tosolini M, Kirilovsky A, Berger A, Bindea G, Meatchi T, et al.Histopathologic-based prognostic factors of colorectal cancers are associ-ated with the state of the local immune reaction. J Clin Oncol2011;29:610–8.

46. Wang B, Xu D, Yu X, Ding T, Rao H, Zhan Y, et al. Association of intra-tumoral infiltrating macrophages and regulatory T cells is an independentprognostic factor in gastric cancer after radical resection. Ann Surg Oncol2011;18:2585–93.

47. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell2011;144:646–74.

48. Huang L, Xu AM, Liu S, Liu W, Li TJ. Cancer-associated fibroblasts indigestive tumors. World J Gastroenterol 2014;20:17804–18.

49. Fridman WH, Remark R, Goc J, Giraldo NA, Becht E, Hammond SA, et al.The immunemicroenvironment: amajor player in human cancers. Int ArchAllergy Immunol 2014;164:13–26.

50. Mlecnik B, Bindea G, Angell HK, SassoMS,Obenauf AC, Fredriksen T, et al.Functional network pipeline reveals genetic determinants associated within situ lymphocyte proliferation and survival of cancer patients. Sci TranslMed 2014;6:228ra37.

51. Kim J, Bae JS. Tumor-associated macrophages and neutrophils in tumormicroenvironment. Mediators Inflamm 2016;2016:6058147.

52. Tumeh PC, Harview CL, Yearley JH, Shintaku IP, Taylor EJ, Robert L, et al.PD-1 blockade induces responses by inhibiting adaptive immune resis-tance. Nature 2014;515:568–71.

53. Xu AM, Huang L, Zhu L, Wei ZJ. Significance of peripheral neutrophil-lymphocyte ratio among gastric cancer patients and construction of atreatment-predictive model: a study based on 1131 cases. Am J CancerRes 2014;4:189–95.

54. Woo Y, Son T, Song K, OkumuraN, Hu Y, ChoGS, et al. A novel predictionmodel of prognosis after gastrectomy for gastric carcinoma: developmentand validation using Asian databases. Ann Surg 2016;264:114–20.

55. Han DS, Suh YS, Kong SH, Lee HJ, Choi Y, Aikou S, et al. Nomogrampredicting long-term survival after D2 gastrectomy for gastric cancer. J ClinOncol 2012;30:3834–40.

56. Cancer Genome Atlas Research N. Comprehensive molecular characteri-zation of gastric adenocarcinoma. Nature 2014;513:202–9.

57. Tan P, Yeoh KG. Genetics and molecular pathogenesis of gastric adeno-carcinoma. Gastroenterology 2015;149:1153–62 e3.

58. StrongMJ, XuG, Coco J, Baribault C, Vinay DS, LaceyMR, et al. Differencesin gastric carcinomamicroenvironment stratify according to EBV infectionintensity: implications for possible immune adjuvant therapy. PLoSPathog 2013;9:e1003341.

59. Wang W, Wang K, Chen Z, Chen L, Guo W, Liao P, et al. Immuno-classification characterized by CD8 and PD-L1 expression is associatedwith the clinical outcome of gastric cancer patients. Oncotarget 2018;9:12164–73.

60. Maekita T, Nakazawa K, MiharaM, Nakajima T, Yanaoka K, Iguchi M, et al.High levels of aberrant DNA methylation in Helicobacter pylori-infectedgastric mucosae and its possible association with gastric cancer risk. ClinCancer Res 2006;12:989–95.

61. Kaneda A, Matsusaka K, Aburatani H, Fukayama M. Epstein-Barr virusinfection as an epigenetic driver of tumorigenesis. Cancer Res2012;72:3445–50.

62. Le DT, Uram JN,WangH, Bartlett BR, Kemberling H, Eyring AD, et al. PD-1blockade in tumors with mismatch-repair deficiency. N Engl J Med2015;372:2509–20.

63. Lee V, Murphy A, Le DT, Diaz LA Jr.Mismatch repair deficiencyand response to immune checkpoint blockade. Oncologist 2016;21:1200–11.

64. Coffelt SB, de Visser KE. Immune-mediated mechanisms influencing theefficacy of anticancer therapies. Trends Immunol 2015;36:198–216.

65. Zitvogel L, Galluzzi L, Smyth MJ, Kroemer G. Mechanism of action ofconventional and targeted anticancer therapies: reinstating immunosur-veillance. Immunity 2013;39:74–88.


Jiang et al.




2018;24:5574-5584. Published OnlineFirst July 24, 2018.Clin Cancer Res Yuming Jiang, Jingjing Xie, Zhen Han, et al. Gastric Cancer Survival and Adjuvant Chemotherapeutic BenefitImmunomarker Support Vector Machine Classifier for Prediction of

Updated version

10.1158/1078-0432.CCR-18-0848doi:

Access the most recent version of this article at:

Material

Supplementary

http://clincancerres.aacrjournals.org/content/suppl/2018/07/24/1078-0432.CCR-18-0848.DC1

Access the most recent supplemental material at:

Cited articles

http://clincancerres.aacrjournals.org/content/24/22/5574.full#ref-list-1

This article cites 64 articles, 14 of which you can access for free at:

Citing articles

http://clincancerres.aacrjournals.org/content/24/22/5574.full#related-urls

This article has been cited by 3 HighWire-hosted articles. Access the articles at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected]

To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at

Permissions

Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://clincancerres.aacrjournals.org/content/24/22/5574To request permission to re-use all or part of this article, use this link



http://clincancerres.aacrjournals.org/lookup/doi/10.1158/1078-0432.CCR-18-0848

http://clincancerres.aacrjournals.org/content/suppl/2018/07/24/1078-0432.CCR-18-0848.DC1

http://clincancerres.aacrjournals.org/content/24/22/5574.full#ref-list-1

http://clincancerres.aacrjournals.org/content/24/22/5574.full#related-urls

http://clincancerres.aacrjournals.org/cgi/alerts

mailto:[email protected]

http://clincancerres.aacrjournals.org/content/24/22/5574


immunomarkersupportvectormachineclassiﬁer for prediction ... · clinical stage and similar...

Documents