Association of the Collagen Signature in the TumorMicroenvironment With Lymph Node Metastasis in EarlyGastric CancerDexin Chen, MD; Gang Chen, MD; Wei Jiang, MD, PhD; Meiting Fu, MD; Wenju Liu, MD; Jian Sui, MD; Shuoyu Xu, PhD; Zhangyuanzhu Liu, MD;Xiaoling Zheng, MD; Liangjie Chi, MD; Dajia Lin, MD; Kai Li, MD; Weisheng Chen, MD; Ning Zuo, PhD; Jianping Lu, MD; Jianxin Chen, PhD;Guoxin Li, MD, PhD; Shuangmu Zhuo, PhD; Jun Yan, MD, PhD
IMPORTANCE Lymph node status is the primary determinant in treatment decision making inearly gastric cancer (EGC). Current evaluation methods are not adequate for estimatinglymph node metastasis (LNM) in EGC.
OBJECTIVE To develop and validate a prediction model based on a fully quantitative collagensignature in the tumor microenvironment to estimate the individual risk of LNM in EGC.
DESIGN, SETTING, AND PARTICIPANTS This retrospective study was conducted from August 1,2016, to May 10, 2018, at 2 medical centers in China (Nanfang Hospital and Fujian ProvincialHospital). Participants included a primary cohort (n = 232) of consecutive patients withhistologically confirmed gastric cancer who underwent radical gastrectomy and received a T1gastric cancer diagnosis from January 1, 2008, to December 31, 2012. Patients withneoadjuvant radiotherapy, chemotherapy, or chemoradiotherapy were excluded. Anadditional consecutive cohort (n = 143) who received the same diagnosis from January 1,2011, to December 31, 2013, was enrolled to provide validation. Baseline clinicopathologicdata of each patient were collected. Collagen features were extracted in specimens usingmultiphoton imaging, and the collagen signature was constructed. An LNM prediction modelbased on the collagen signature was developed and was internally and externally validated.
MAIN OUTCOMES AND MEASURES The area under the receiver operating characteristic curve(AUROC) of the prediction model and decision curve were analyzed for estimating LNM.
RESULTS In total, 375 patients were included. The primary cohort comprised 232 consecutivepatients, in whom the LNM rate was 16.4% (n = 38; 25 men [65.8%] with a mean [SD] age of57.82 [10.17] years). The validation cohort consisted of 143 consecutive patients, in whom theLNM rate was 20.9% (n = 30; 20 men [66.7%] with a mean [SD] age of 54.10 [13.19] years).The collagen signature was statistically significantly associated with LNM (odds ratio, 5.470;95% CI, 3.315-9.026; P < .001). Multivariate analysis revealed that the depth of tumorinvasion, tumor differentiation, and the collagen signature were independent predictors ofLNM. These 3 predictors were incorporated into the new prediction model, and a nomogramwas established. The model showed good discrimination in the primary cohort (AUROC,0.955; 95% CI, 0.919-0.991) and validation cohort (AUROC, 0.938; 95% CI, 0.897-0.981). Anoptimal cutoff value was selected in the primary cohort, which had a sensitivity of 86.8%, aspecificity of 93.3%, an accuracy of 92.2%, a positive predictive value of 71.7%, and anegative predictive value of 97.3%. The validation cohort had a sensitivity of 90.0%, aspecificity of 90.3%, an accuracy of 90.2%, a positive predictive value of 71.1%, and anegative predictive value of 97.1%. Among the 375 patients, a sensitivity of 87.3%, aspecificity of 92.1%, an accuracy of 91.2%, a positive predictive value of 72.1%, and a negativepredictive value of 96.9% were found.
CONCLUSIONS AND RELEVANCE This study’s findings suggest that the collagen signature inthe tumor microenvironment is an independent indicator of LNM in EGC, and the predictionmodel based on this collagen signature may be useful in treatment decision making forpatients with EGC.JAMA Surg. 2019;154(3):e185249. doi:10.1001/jamasurg.2018.5249Published online January 30, 2019.
Invited Commentary
Supplemental content
Author Affiliations: Authoraffiliations are listed at the end of thisarticle.
Corresponding Authors: Jun Yan,MD, PhD, Department of GeneralSurgery, Nanfang Hospital, SouthernMedical University, Guangzhou,Guangdong 510515, People's Republicof China ([email protected]);Shuangmu Zhuo, PhD, KeyLaboratory of OptoElectronic Scienceand Technology for Medicine ofMinistry of Education, FujianProvincial Key Laboratory ofPhotonics Technology, Fujian NormalUniversity, Fuzhou, Fujian 350007,People's Republic of China([email protected]).
E arly gastric cancer (EGC) is defined as cancer limited tothe mucosa or submucosa, regardless of nodal status.1
Currently, endoscopic submucosal dissection (ESD) hasbecome more popular than surgical procedures in treating EGCbecause it is minimally invasive, preserves function, and re-sults in better quality of life.2-4 The principal indication for ESDis a tumor with a low risk of lymph node metastasis (LNM) thatcan undergo en bloc resection.3 The incidence of LNM is lessthan 3% when cancer is limited to the mucosa and increasesto approximately 20% after cancer invades the submucosa.5
Thus, the accurate assessment of nodal status in EGC is inte-gral to providing tailored surgical procedure.6,7 So far, the di-agnostic accuracy of endoscopic ultrasonography and com-puted tomography for the nodal status of EGC is limited.8,9
Early gastric cancer with undifferentiated histologic result, sub-mucosal invasion, and lymphovascular infiltration is deemeda high risk for LNM, and radical surgical procedure isconsidered,3,5,10 but a unanimous consensus has not beenreached. To estimate the likelihood of LNM for EGC, severalstudies have developed different prediction models.11-13 How-ever, these models focused on the clinical-pathologic charac-teristics, and the association of the tumor microenvironmentwith LNM was not investigated.
The extracellular matrix constitutes the scaffold of the tu-mor microenvironment, which regulates cancer behavior.14 Asthe main component of the extracellular matrix, collagen ac-counts for its major functions. The arrangement and orienta-tion of collagen were proven to be indicators of tumor metas-tasis in breast cancer,15,16 glioblastoma17 and prostate cancer.18
Nevertheless, the role of collagen in the process of LNM in EGCis still unclear.
Multiphoton imaging could provide detailed informationabout tissue architecture and cell morphology in specimensthrough a combination of 2-photon excitation fluorescencefrom cells and second harmonic generation from collagen.19
Because of the underlying physical origin, multiphotonimaging has emerged as a powerful modality for collagenimaging in diverse tissues.20,21 Moreover, multiphoton imagingcould be converted into high-dimensional and quantitativecomponents of collagen via automatic extraction of multiplefeatures. Collagen features analysis, including morphologic andtextural features extracted from multiphoton imaging, has beenapplied and demonstrated to be a powerful quantitative indi-cator for diagnosis in several diseases.22-24
Integrating multiple biomarkers into a single signature,rather than performing individual biomarker analysis, is apromising approach that would improve clinic almanagement.25,26 Currently, an appropriate method of inte-grating multiple collagen features into a single signature hasnot yet been developed. Hence, we propose the collagen sig-nature, deduced by multiple morphologic and textural fea-tures of collagen using multiphoton imaging. The aim of thisstudy was to develop and validate a prediction model basedon the collagen signature that can distinguish genuine high-risk EGC with LNM. To our knowledge, this is the first studyto investigate the role of collagen in EGC and to develop a pre-diction model for LNM based on the fully quantitative colla-gen signature.
Methods
The institutional review board at each participating center inChina (Nanfang Hospital, Guangzhou, Guangdong, People's Re-public of China and Fujian Provincial Hospital, Fuzhou, Fu-jian, People's Republic of China) approved this study. Patientinformed consent was waived by the institutional review boardbecause of the retrospective design of the study and patients'information was protected. The study was conducted from Au-gust 1, 2016, to May 10, 2018.
Patients and SpecimensThe primary cohort (n = 232) was retrospectively assembledusing the medical database of Nanfang Hospital. Consecutivepatients who received a diagnosis from January 1, 2008, to De-cember 31, 2012 (eFigure 1 in the Supplement) comprised thecohort. The inclusion criteria were patients with histologi-cally confirmed gastric cancer who underwent radical gastrec-tomy and received a T1 gastric cancer diagnosis after surgicalintervention. We excluded patients with neoadjuvant radio-therapy, chemotherapy, or chemoradiotherapy. An addi-tional consecutive cohort (n = 143) who received the same di-agnosis at the Fujian Provincial Hospital another hospital fromJanuary 1, 2011, to December 31, 2013, and who met the samecriteria as the primary cohort was enrolled to provide valida-tion. The formalin-fixed paraffin-embedded specimens of allpatients were used.
Baseline clinicopathologic data of each patient, includingsex, age at surgical intervention, macroscopic classification,tumor location, tumor size, tumor differentiation, lympho-vascular infiltration, and depth of invasion, were collected. Thetumor differentiation was divided into differentiated and un-differentiated types according to the 2014 Japanese gastric can-cer treatment guidelines (version 4).3
Selection of Regions of Interest, Multiphoton ImageAcquisition, and Collagen Feature ExtractionAll specimens were processed for hematoxylin-eosin stain-ing (original magnification ×200). Two of our independent pa-thologists (W. L. and J. L.), who were blinded to the nodal sta-tus, evaluated the region of the invasive margin of the EGC
Key PointsQuestion How can lymph node metastasis in early gastric cancerbe accurately assessed?
Finding In this study of 375 patients with early gastric cancer,collagen signature was statistically significantly associated withlymph node metastasis. A newly developed lymph nodemetastasis prediction model based on the collagen signatureshowed good discrimination in the primary cohort and wasexternally validated.
Meaning The new prediction model appears to be useful indecision making associated with tailored surgical interventions inpatients with early gastric cancer.
Research Original Investigation Association of the Collagen Signature With Lymph Node Metastasis in Early Gastric Cancer
2/9 JAMA Surgery March 2019 Volume 154, Number 3 (Reprinted) jamasurgery.com
using a microscope at ×200 magnification. The interrater re-liability was evaluated (κ = 0.437; 95% CI, 0.295-0.569) withapproximately 87.3% (95% CI, 84.3%-90.1%) agreement. Whenthe 2 pathologists differed in opinion, they consulted with thedirector (G.C.) of the Department of Pathology, Fujian Provin-cial Cancer Hospital, to make a decision. Five regions of inter-est with a field of view of 200 × 200 μm per specimen, whichwere equidistantly spread throughout the invasive margin,were selected to provide a realistic representation of each EGCsample.
Image acquisition for multiphoton imaging was per-formed with a 200× original magnification objective on an-other unstained serial section and then compared with hema-toxylin-eosin staining for histologic assessment. Themultiphoton imaging system used in this study has been de-scribed previously (eMethods in the Supplement).27
The extraction of collagen features was performed usingMATLAB 2015b (MathWorks) as previously reported.28,29 Atotal of 146 features, including 12 morphologic features and 134textural features, were extracted (eMethods and eTable 1 in theSupplement).
Feature Selection and Collagen Signature ConstructionThe LASSO (least-absolute shrinkage and selection operator)logistic regression, which has been broadly applied for high-dimensional data, was used to select the most predictive fea-tures in the primary cohort.30 The collagen signature construc-tion was calculated through a combination of selected features(eMethods in the Supplement).
Prediction Model Development and EvaluationBoth the 8 clinicopathologic variables and the collagen signa-ture were included in the univariate analysis to explore the as-sociation with LNM in the primary cohort, and variables withP < .05 were selected for the multivariate analysis. Backwardstepwise regression was applied to select the independent pre-dictors. The multicollinearity of the multivariate model wasassessed using the tolerance and variance inflation factor. Inaddition, the effect modification was evaluated. A nomo-gram was constructed according to independent predictors.For quantification of the discrimination of the nomogram, thearea under the receiver operating characteristic curve (AU-ROC) was measured. The calibration of the nomogram wasevaluated by the calibration curve to assess the goodness offit, accompanied by the Hosmer-Lemeshow test.
Prediction Model Internal and External ValidationThe bootstrap method was applied for internal validation, inwhich the random samples drawn with a replacement from theoriginal data set were the same size as the primary cohort.31
One thousand bootstrap repetitions were performed.The prediction model was applied in the validation co-
hort. Ultimately, the AUROC was calculated, and the calibra-tion curve was plotted.
Clinical ApplicationTo evaluate the clinical application of the nomogram, deci-sion curve analysis was used to assess the net benefits of the
prediction model at different threshold probabilities (eMethodsin the Supplement).32 The maximum Youden index was se-lected as the cutoff value to evaluate the sensitivity, specific-ity, accuracy, positive predictive value, and negative predic-tive value of the prediction model.
Statistical AnalysisAn independent-samples, unpaired, 2-tailed t test or Mann-Whitney H test, where appropriate, was used to assess the dif-ferences in continuous variables, and a χ2 test or Fisher exactprobability test was used to compare the differences betweencategorical variables. A multivariate logistic regression was per-formed to estimate the odds ratio (OR) with a 95% CI and toidentify the independent predictors for LNM. Statistical analy-sis was conducted with R software, version 3.4.2 (R Founda-tion for Statistical Computing). Differences with a 2-sidedP < .05 were considered statistically significant.
ResultsParticipantsThe primary cohort included 232 consecutive patients, inwhom the LNM rate was 16.4% (n = 38; 25 men [65.8%] witha mean [SD] age of 57.82 [10.17] years). The validation cohortincluded 143 consecutive patients, in whom the LNM rate was20.9% (n = 30; 20 men [66.7%] with a mean [SD] age of 54.10[13.19] years). Patient characteristics in the primary and vali-dation cohorts are given in Table 1. No statistically significantdifference in LNM prevalence was observed between the 2 co-horts (OR, 1.355; 95% CI, 0.796-2.307; P = .26). The clinico-pathologic characteristics were similar between the primaryand validation cohorts (eTable 2 in the Supplement).
Collagen Signature ConstructionThe construction framework of the collagen signature is pre-sented in Figure 1. All collagen features were reduced to the 6best potential predictors, using LASSO logistic regression (eFig-ures 2A and 2B in the Supplement; the 6 features are pre-sented in eAppendix 1 in the Supplement). A statistically sig-nificant difference in the collagen signature (median[interquartile range (IQR)]) was found between patients withLNM (0.284 [–0.836 to 0.872]) and patients without LNM(–2.856 [–3.630 to –2.088]) in the primary cohort (median dif-ference, 2.059; 95% CI, 1.413 to 2.757; P < .001). This findingwas consistent with the patients with LNM (–0.522 [–0.887 to–0.125]) and patients without LNM (–2.277 [–2.851 to –1.794])in the validation cohort (median difference, 1.793; 95% CI, 1.176to 2.361; P < .001) (Table 1). The collagen signature indicateda favorable prediction of LNM with an AUROC of 0.944 (95%CI, 0.905-0.982) in the primary cohort and 0.933 (95% CI,0.889-0.977) in the validation cohort (eFigures 2C and 2D inthe Supplement).
Prediction Model Development and EvaluationA univariate analysis was performed for each variable in theprimary cohort. A tumor size larger than 2 cm (OR, 3.249; 95%CI, 1.578-6.693; P = .001), undifferentiated tumor (OR, 2.956;
Association of the Collagen Signature With Lymph Node Metastasis in Early Gastric Cancer Original Investigation Research
jamasurgery.com (Reprinted) JAMA Surgery March 2019 Volume 154, Number 3 3/9
95% CI, 1.452-6.017; P = .002), lymphovascular infiltration (OR,7.341; 95% CI, 2.743-19.646; P < .001), submucosal invasion(OR, 9.231; 95% CI, 3.155-27.010; P < .001), and collagen sig-nature (OR, 5.470; 95% CI, 3.315-9.026; P < .001) were statis-tically significantly associated with LNM in EGC (Table 2). Fur-thermore, a multivariate analysis identified that tumordifferentiation (OR, 4.585; 95% CI, 1.310-16.041; P = .02), thedepth of tumor invasion (OR, 6.773; 95% CI, 1.636-28.039;P = .008), and the collagen signature (OR, 5.335; 95% CI, 3.042-9.358; P < .001) were independent predictors of LNM (Table 2).The variance inflation factor of each predictor was less than10, and the corresponding tolerance was more than 0.1; there-fore, no multicollinearity among these predictors was noted(eTable 3 in the Supplement).33 No effect modification wasfound in the prediction model (eTables 4 and 5 in the Supple-ment). The association between the collagen signature and therisk of LNM with different combinations of tumor differentia-tion states (differentiated or undifferentiated) and depths oftumor invasion (mucosa or submucosa) is presented in eFig-ure 3 in the Supplement. A nomogram was produced by in-corporating these 3 independent predictors (Figure 2).
The newly developed prediction model showed good dis-crimination with an AUROC of 0.955 (95% CI, 0.919-0.991), andthe calibration curve showed good agreement between the no-mogram-estimated probability of LNM and the actual LNM rate
in the primary cohort (eFigure 4A and B in the Supplement).The Hosmer-Lemeshow test demonstrated a P = .47, indicat-ing no departure from a good fit.
Internal and External Prediction Model ValidationFor internal validation, we used the bootstrap method with1000 bootstrap repetitions. The results remained largely un-changed between iterations, with a mean concordance indexof 0.911.
Good discrimination with an AUROC of 0.938 (95% CI,0.897-0.981) was externally validated, and the favorable cali-bration was also confirmed in the validation cohort (eFig-ure 4C and D in the Supplement). A Hosmer-Lemeshow testdemonstrated a nonsignificant P = .15.
Clinical ApplicationIn the decisive curve, the x-axis is a measure of patient or phy-sician preference, and the threshold probability indicates thatthe expected advantage of treatment is equal to the expectedadvantage of avoiding treatment.33 The decision curve re-vealed that if the threshold probability of a patient or physi-cian was greater than 5%, more advantages would be addedby using the nomogram to estimate LNM in EGC than the ad-vantage achieved in either the treat-all-patient scheme or thetreat-none scheme (eFigure 5 in the Supplement).
Table 1. Characteristics of Patients in the Primary and Validation Cohorts
Variable
Primary Cohort (n = 232)
P Value
Validation Cohort (n = 143)
P ValueWith LNM Without LNM With LNM Without LNMAge, mean (SD), y 57.82 (10.17) 58.82 (10.90) .60 54.10 (13.19) 58.64 (11.09) .06
In addition, in the primary cohort, the maximum Youdenindex of 0.301 was selected as the cutoff value, and the co-hort had a sensitivity of 86.8%, a specificity of 93.3%, an ac-curacy of 92.2%, a positive predictive value of 71.7%, and anegative predictive value of 97.3%. The validation cohort hada sensitivity of 90.0%, a specificity of 90.3%, an accuracy of90.2%, a positive predictive value of 71.1%, and a negative pre-dictive value of 97.1%. Among the 375 patients, a sensitivityof 87.3%, a specificity of 92.1%, an accuracy of 91.2%, a posi-tive predictive value of 72.1%, and a negative predictive valueof 96.9% were found (eTable 6 in the Supplement).
Comparison With the Traditional Prediction ModelTo elucidate the superiority of the model we built over the clini-copathologic characteristic-based model (ie, the traditionalmodel), we eliminated the collagen signature and developedthe traditional model on the basis of tumor differentiation (OR,2.576; 95% CI, 1.167-5.685; P = .02), lymphovascular infiltra-tion (OR, 3.333; 95% CI, 1.145-9.703; P = .03), and the depthof tumor invasion (OR, 9.923; 95% CI, 3.305-29.793; P < .001)(eTable 7 in the Supplement) after univariate and multivari-ate analyses. No multicollinearity in the traditional model wasfound (eTable 8 in the Supplement). Sex, age at surgical inter-vention, macroscopic classification, and tumor location were
Figure 1. Schematic Illustration of Collagen Signature Construction
Representative region of interestA
Computation frameworkB
Prim
ary
Coho
rtVa
lidat
ion
Coho
rt
Multiphoton Imaging
Multiphoton Imaging
Feature Extraction
MorphologicFeature
... ...TexturalFeature
... ...
Collagen Signature
Feature Extraction
MorphologicFeature
... ...TexturalFeature
... ...
Collagen Signature = ΣαiEi
H-E H-E TPEF/SHG SHG
–10 –8 –6 –4 –2
48 35 15 8 2
Coef
ficie
nt
Log(λ)
800
600
400
200
0
–200
Feature Selection
A, A representative region of interest with a field of view of 200 × 200 μm wasselected in the hematoxylin-eosin (H-E) stain (original magnification ×200). Thecorresponding multiphoton imaging, including 2-photon excitationfluorescence (TPEF) and second harmonic generation (SHG), was obtained, andthe SHG imaging was chosen for collagen feature extraction. B, A computationframework was used to establish the collagen signature. The SHG imaging of
multiphoton imaging was chosen for collagen feature extraction, includingmorphologic features and texture features. Next, the potential predictors wereselected using LASSO (least-absolute shrinkage and selection operator) logisticregression. The collagen signature can be calculated by these potentialpredictors.
Association of the Collagen Signature With Lymph Node Metastasis in Early Gastric Cancer Original Investigation Research
jamasurgery.com (Reprinted) JAMA Surgery March 2019 Volume 154, Number 3 5/9
chosen as variables, but these variables were not statisticallysignificant after univariate analysis. The performance of thetraditional model was similar to the performance previouslyreported, with an AUROC of 0.812 (95% CI, 0.752-0.872) in theprimary cohort and 0.768 (95% CI, 0.688-0.849) in the vali-dation cohort (eFigure 6 in the Supplement).11,12 Compared withthe traditional model, the new model based on the collagensignature showed a more robust ability to estimate the risk ofLNM in EGC in all 375 patients (AUROC comparison, 0.950 [95%CI, 0.923-0.977] vs 0.798 [95% CI, 0.749-0.847]; P < .001)(Figure 3).
Discussion
Accurate assessment of the nodal status in EGC is importantin the decision making for lymph node dissection. In this study,we developed and validated a nomogram for individual esti-mation of LNM in EGC, including the depth of tumor inva-sion, tumor differentiation, and the collagen signature.
Two key factors determine the construction of the colla-gen signature. The first is the use of a suitable imaging ap-proach to selectively visualize the collagen. In this study, mul-
Table 2. Univariate and Multivariate Logistic Regression of Lymph Node Metastasis in the Primary Cohort
The nomogram indicates the risk ofLNM in early gastric cancer. Forclinical use, tumor differentiation isdetermined by drawing a line straightup to the point axis to establish thescore associated with thedifferentiation. Next, this process isrepeated for the other 2 covariates(depth of tumor invasion andcollagen signature). The scores ofeach covariate are added, and thetotal score is located on the totalscore points axis. Last, a line is drawnstraight down to the risk of LNM axisto obtain the probability.
Research Original Investigation Association of the Collagen Signature With Lymph Node Metastasis in Early Gastric Cancer
6/9 JAMA Surgery March 2019 Volume 154, Number 3 (Reprinted) jamasurgery.com
tiphoton imaging was used because of its underlying physicalorigin.20,21 Previous research indicated that multiphotonimaging can distinguish between the mucosa and submucosaof cancerous gastric tissues, and collagen can be quantified bysecond harmonic generation in a stain-free section.34 Thus,multiphoton imaging is an ideal method for collagen imaging.The second factor is the quantitative analysis of collagen frommultiphoton imaging. For this purpose, we have establisheda stable framework for achieving precise quantification.28,29
After considering these 2 factors, we constructed the col-lagen signature. The collagen signature was substantially dif-ferent in EGC with and without LNM. To develop a clinicallypracticable prediction tool, we used other clinicopathologiccharacteristics. We also built a nomogram with good discrimi-nation and calibration. Our findings suggest that LNM is morelikely to appear in patients with an undifferentiated histo-logic result, submucosal invasion, and a high collagen signa-ture.
Compared with the traditional model based on tumor dif-ferentiation, the depth of tumor invasion, and lymphovascu-lar infiltration, the prediction model was more powerful in es-timating the risk of LNM in EGC. Although a tumor size largerthan 2 cm was statistically significantly associated with LNM,it was excluded after backward stepwise multivariate analy-sis in both the prediction model and the traditional model. Thereason for this exclusion was that the depth of tumor inva-sion was much more important than tumor size in the clinic.
Currently, endoscopic ultrasonography and computed to-mography are the 2 most common examination methods forN staging of gastric cancer. Endoscopic ultrasonography for Nstaging had a sensitivity of 83% and a specificity of 67%.35
Meanwhile, computed tomography for detecting LNM had a
sensitivity of 78% and a specificity of 62% in gastric cancer.36
In the prediction model, the sensitivity was 87.3% and thespecificity was 92.1%, with the cutoff value of the maximumYouden index. Therefore, the prediction model was ad-equate for base clinical decisions.
In this study, tumor differentiation and the depth of tu-mor invasion were categorical variables, and the collagen sig-nature was a continuous variable. The risk of LNM was al-ways contributed to by these 3 predictors. For example, for apatient without LNM with a collagen signature of –2.856, therisk of LNM was less than 1% for differentiated tumors that in-vaded only the mucosa. When the tumor was undifferenti-ated, the risk of LNM was approximately 2%, and if the tumoralso invaded the submucosa, the risk of LNM increased to 8%.Similarly, for a patient with LNM with a collagen signature of0.284, the risk of LNM was approximately 30% for differen-tiated tumors that invaded the mucosa. In the case of undif-ferentiated tumors, the risk of LNM increased to approxi-mately 70%. Once the tumors invaded the submucosa, the riskof LNM increased to approximately 90%. As tumor differen-tiation and the depth of tumor invasion are routinely as-sessed in endoscopic resection specimens, and the collagen sig-nature could be quantified using multiphoton imaging, theindividual risk of LNM could be conveniently estimated by thenomogram after ESD. For a low risk of LNM, the nomogram in-dicates that ESD is adequate. Inversely, for a high risk of LNM,additional lymph node dissection might be needed.
Collagen was identified as a component of cancer metas-tasis. Local collagen orientations have been shown to play animportant role in promoting cell breakage into the basementmembrane before entering the circulation systems.15 Kakkadet al24 reported that multiphoton imaging revealed that a sub-stantially increased density of collagen was associated withLNM in breast cancer. In our study, the collagen signature waspositively correlated with collagen straightness and cross-link density. This result indicated that the collagen arrange-ment was far straighter in the invasive margin of EGC with LNM.Straighter collagen in the tumor microenvironment could fa-cilitate invasion.37,38 Meanwhile, increased collagen cross-link density could stiffen the extracellular matrix, enhancegrowth factor signaling activity, and induce the invasion of anoncogene-initiated epithelium.39 Our data showed the asso-ciation between the collagen signature and LNM for EGC. Fu-ture studies should focus on the underlying molecular mecha-nisms.
The collagen features in this study were extracted frommultiphoton imaging. Because the components of the mul-tiphoton imaging system were fixed, pathologists could con-duct multiphoton imaging using a microscope. Finishing mul-tiphoton imaging took approximately 5 to 10 minutes.Multiphoton imaging was good at showing collagen and didnot change the tissue architecture and cell morphology. There-fore, pathologists could understand and analyze multipho-ton imaging after training. Meanwhile, the ESD of EGC has nospecial requirements, and specimens can be processed regu-larly, which would not affect the multiphoton imaging. Mul-tiphoton imaging is a promising method for realizing real-time in vivo optical biopsy, and several groups have reported
Figure 3. Comparison Between the Traditional Model and the NewCollagen Signature–Based Prediction Model
1.0
0.8
0.6
0.4
0.2
00 1.00.8
Sens
itivi
ty, %
Specificity, %0.60.40.2
New prediction modelTraditional model
The orange line represents the new model (area under the receiver operatingcharacteristic curve [AUROC], 0.950; 95% CI, 0.923-0.977), including thedepth of tumor invasion, tumor differentiation, and the collagen signature. Theblue line represents the traditional model (AUROC, 0.798; 95% CI,0.749-0.847), including the depth of tumor invasion, tumor differentiation, andlymphovascular infiltration.
Association of the Collagen Signature With Lymph Node Metastasis in Early Gastric Cancer Original Investigation Research
jamasurgery.com (Reprinted) JAMA Surgery March 2019 Volume 154, Number 3 7/9
the possible clinical applications in different organs.40-42 Weforesee that clinicians could obtain collagen signature in thenear future using multiphoton imaging. With the assistanceof the prediction model, EGC with a genuine high risk of LNMwould be distinguished, and more tailored surgical interven-tions could be performed.
LimitationsThis study has some limitations. First, because it was a retro-spective study, it might result in a potential selection bias. Thus,a multicenter prospective clinical trial is required to confirmthe prediction model we developed. We are comfortable withthe application of this technique in a clinical trial. Second, theclinicopathologic characteristics between the primary and vali-dation cohorts were similar, which made our validation lessrobust, and the distribution of clinicopathologic characteris-tics might be different in other countries. Therefore, cohortsfrom Western countries are needed to further validate our find-ings. Third, the sample-size calculation for logistic regres-sion analysis is still debated. We used 2 methods to calculatethe sample size: one requires at least 10 events per variable,43
and the other is based on the variance inflation factor and doesnot explicitly require knowledge of the number of variablesin the regression model.44 The sample size might not be ad-equate for the former method but was enough for the lattermethod (eAppendix 2 in the Supplement). Thus, we hope thatthis limitation will be solved in our upcoming clinical trial.Fourth, the weak interrater reliability between the 2 patholo-gists is also a limitation. One pathologist was a senior attend-ing pathologist, and the other was a junior attending patholo-gist. The weak interrater reliability was the result of thedifference in their experiences. In our next trial, we will re-quire 2 senior attending pathologists.
ConclusionsThe collagen signature in the tumor microenvironment is anindependent risk factor for LNM in EGC. The prediction modelwe developed and validated is useful for decision making intailored surgical intervention.
ARTICLE INFORMATION
Accepted for Publication: October 28, 2018.
Published Online: January 30, 2019.doi:10.1001/jamasurg.2018.5249
Author Affiliations: Department of GeneralSurgery, Nanfang Hospital, Southern MedicalUniversity, Guangzhou, Guangdong, People'sRepublic of China (D. Chen, Jiang, W. Liu, Sui, Xu,Z. Liu, Zheng, Chi, Lin, K. Li, W. Chen, G. Li, Yan);Key Laboratory of OptoElectronic Science andTechnology for Medicine of Ministry of Education,Fujian Provincial Key Laboratory of PhotonicsTechnology, Fujian Normal University, Fuzhou,Fujian, People's Republic of China (D. Chen, Zuo,J. Chen, Zhuo); Department of Pathology, FujianProvincial Cancer Hospital, Teaching Hospital ofFujian Medical University, Fuzhou, Fujian, People'sRepublic of China (G. Chen, W. Liu, Lu); Departmentof Gastroenterology, Nanfang Hospital, SouthernMedical University, Guangzhou, Guangdong,People's Republic of China (Fu); Department ofGastrointestinal Surgery, Fujian Provincial Hospital,Teaching Hospital of Fujian Medical University,Fuzhou, Fujian, People's Republic of China (Sui, Chi,Lin); Department of Radiology, Sun Yat-senUniversity Cancer Center, Guangzhou, Guangdong,People's Republic of China (Xu); Department ofEndoscopy Center, Fujian Provincial Hospital,Teaching Hospital of Fujian Medical University,Fuzhou, Fujian, People's Republic of China (Zheng).
Author Contributions: Drs D. Chen, G. Chen, Jiang,and Fu contributed equally to this article. Drs Yanand Zhuo jointly directed this work and contributedequally as corresponding authors. Dr Yan had fullaccess to all of the data in the study and takesresponsibility for the integrity of the data and theaccuracy of the data analysis.Concept and design: D. Chen, Chi, G. Li, Zhuo, Yan.Acquisition, analysis, or interpretation of data: D.Chen, G. Chen, Jiang, Fu, W. Liu, Sui, Xu, Z. Liu,Zheng, Lin, K. Li, W. Chen, Zuo, Lu, J. Chen, Yan.Drafting of the manuscript: D. Chen, Jiang, Fu, W.Liu, Sui, Xu, Z. Liu, Zheng, Lin, K. Li, W. Chen, Zuo,
Lu, J. Chen, Zhuo, Yan.Critical revision of the manuscript for importantintellectual content: D. Chen, G. Chen, Jiang, Xu,Chi, G. Li, Zhuo, Yan.Statistical analysis: D. Chen, Fu, Sui, Chi, Zhuo, Yan.Obtained funding: D. Chen, J. Chen, Zhuo, Yan.Administrative, technical, or material support: G.Chen, Xu, Zuo, J. Chen, G. Li, Zhuo, Yan.Supervision: D. Chen, G. Li, Zhuo.
Conflict of Interest Disclosures: None reported.
Funding/Support: This work was supported bygrants 81773117, 81771881, 81700576, and81672446 from the National Natural ScienceFoundation of China; grant 2015CB352006 fromthe National Key Basic Research Program of China;grant [2011]170 from the National Clinical KeySpecialty Construction Program; grants2017YFC0108300 and 2017YFC0108302 from theState’s Key Project of Research and DevelopmentPlan; grant 201402015 from the Research Fund ofPublic Welfare in the National Health and FamilyPlanning Commission of China; grant2014B020215002 from the Special Fund forGuangdong Province Public Research and CapacityBuilding; grant 2015A030308006 from the NaturalScience Foundation of Guangdong Province; grant2018J07004 from the Natural Science Foundationof Fujian Province; grant 2017L3009 from theSpecial Funds of the Central Government GuidingLocal Science and Technology Development; grant2014A020215014 from the Guangdong ProvincialScience and Technology Key Project; grant2012-CXB-7 from the Innovation Research of FujianHealth Bureau; grant IRT_15R10 from the Programfor Changjiang Scholars and Innovative ResearchTeam in University; grant 201704020062 from theGuangzhou Industry University ResearchCooperative Innovation Major Project; grant320.2710.1851 from the Special Fund from ClinicalResearch of Wu Jieping Medical Foundation; grantLC2016PY010 from the Clinical Research Project ofSouthern Medical University; grant 2014067 fromthe High-Level Research Matching Foundation ofNanfang Hospital; grant 201404280056 from the
Scientific Research Foundation for High-LevelTalents in Nanfang Hospital of Southern MedicalUniversity; grants pdjhb0100 and pdjh2017a0093from the Special Funds for the Cultivation ofGuangdong College Students’ Scientific andTechnological Innovation; grants 201612121008,201612121080, 201712121052, 201712121132,201712121149, 201812121265, and 201812121039Sfrom the Training Program for UndergraduateInnovation and Entrepreneurship; and grantB1000494 from the Scientific Enlightenment Planof Southern Medical University.
Role of the Funder/Sponsor: The funding sourceshad no role in the design and conduct of the study;collection, management, analysis, andinterpretation of the data; preparation, review, orapproval of the manuscript; and decision to submitthe manuscript for publication.
REFERENCES
1. Van Cutsem E, Sagaert X, Topal B, HaustermansK, Prenen H. Gastric cancer. Lancet. 2016;388(10060):2654-2664. doi:10.1016/S0140-6736(16)30354-3
2. Isomoto H, Shikuwa S, Yamaguchi N, et al.Endoscopic submucosal dissection for early gastriccancer: a large-scale feasibility study. Gut. 2009;58(3):331-336. doi:10.1136/gut.2008.165381
3. Japanese Gastric Cancer Association. Japanesegastric cancer treatment guidelines 2014 (ver. 4).Gastric Cancer. 2017;20(1):1-19. doi:10.1007/s10120-016-0622-4
4. Choi JH, Kim ES, Lee YJ, et al. Comparison ofquality of life and worry of cancer recurrencebetween endoscopic and surgical treatment forearly gastric cancer. Gastrointest Endosc. 2015;82(2):299-307. doi:10.1016/j.gie.2015.01.019
5. Hirasawa T, Gotoda T, Miyata S, et al. Incidenceof lymph node metastasis and the feasibility ofendoscopic resection for undifferentiated-typeearly gastric cancer. Gastric Cancer. 2009;12(3):148-152. doi:10.1007/s10120-009-0515-x
Research Original Investigation Association of the Collagen Signature With Lymph Node Metastasis in Early Gastric Cancer
8/9 JAMA Surgery March 2019 Volume 154, Number 3 (Reprinted) jamasurgery.com
6. An JY, Baik YH, Choi MG, Noh JH, Sohn TS, KimS. Predictive factors for lymph node metastasis inearly gastric cancer with submucosal invasion:analysis of a single institutional experience. Ann Surg.2007;246(5):749-753. doi:10.1097/SLA.0b013e31811f3fb7
7. Oh SY, Lee KG, Suh YS, et al. Lymph nodemetastasis in mucosal gastric cancer: reappraisal ofexpanded indication of endoscopic submucosaldissection. Ann Surg. 2017;265(1):137-142. doi:10.1097/SLA.0000000000001649
8. Spolverato G, Ejaz A, Kim Y, et al. Use ofendoscopic ultrasound in the preoperative stagingof gastric cancer: a multi-institutional study of theUS gastric cancer collaborative. J Am Coll Surg.2015;220(1):48-56. doi:10.1016/j.jamcollsurg.2014.06.023
9. Fujikawa H, Yoshikawa T, Hasegawa S, et al.Diagnostic value of computed tomography forstaging of clinical T1 gastric cancer. Ann Surg Oncol.2014;21(9):3002-3007. doi:10.1245/s10434-014-3667-9
10. Park JW, Ahn S, Lee H, et al. Predictive factorsfor lymph node metastasis in early gastric cancerwith lymphatic invasion after endoscopic resection.Surg Endosc. 2017;31(11):4419-4424. doi:10.1007/s00464-017-5490-4
11. Pyo JH, Shin CM, Lee H, et al; JHP and CMScontributed equally as the first authors of thisstudy. A risk-prediction model based onlymph-node metastasis for incorporation into atreatment algorithm for signet ring cell-typeintramucosal gastric cancer. Ann Surg. 2016;264(6):1038-1043. doi:10.1097/SLA.0000000000001602
12. Guo CG, Zhao DB, Liu Q, et al. A nomogram topredict lymph node metastasis in patients withearly gastric cancer. Oncotarget. 2017;8(7):12203-12210. doi:10.18632/oncotarget.14660
13. Jung DH, Huh CW, Kim JH, et al.Risk-stratification model based on lymph nodemetastasis after noncurative endoscopic resectionfor early gastric cancer. Ann Surg Oncol. 2017;24(6):1643-1649. doi:10.1245/s10434-017-5791-9
14. Buchheit CL, Weigel KJ, Schafer ZT. Cancer cellsurvival during detachment from the ECM: multiplebarriers to tumour progression. Nat Rev Cancer.2014;14(9):632-641. doi:10.1038/nrc3789
15. Han W, Chen S, Yuan W, et al. Oriented collagenfibers direct tumor cell intravasation. Proc Natl AcadSci U S A. 2016;113(40):11208-11213. doi:10.1073/pnas.1610347113
16. Conklin MW, Gangnon RE, Sprague BL, et al.Collagen alignment as a predictor of recurrenceafter ductal carcinoma in situ. Cancer EpidemiolBiomarkers Prev. 2018;27(2):138-145. doi:10.1158/1055-9965.EPI-17-0720
17. Pointer KB, Clark PA, Schroeder AB, SalamatMS, Eliceiri KW, Kuo JS. Association of collagenarchitecture with glioblastoma patient survival.J Neurosurg. 2017;126(6):1812-1821.
18. Penet MF, Kakkad S, Pathak AP, et al. Structureand function of a prostate cancerdissemination-permissive extracellular matrix. Clin
Cancer Res. 2017;23(9):2245-2254. doi:10.1158/1078-0432.CCR-16-1516
19. Zipfel WR, Williams RM, Christie R, Nikitin AY,Hyman BT, Webb WW. Live tissue intrinsic emissionmicroscopy using multiphoton-excited nativefluorescence and second harmonic generation. ProcNatl Acad Sci U S A. 2003;100(12):7075-7080. doi:10.1073/pnas.0832308100
20. Zipfel WR, Williams RM, Webb WW. Nonlinearmagic: multiphoton microscopy in the biosciences.Nat Biotechnol. 2003;21(11):1369-1377. doi:10.1038/nbt899
21. Campagnola P. Second harmonic generationimaging microscopy: applications to diseasesdiagnostics. Anal Chem. 2011;83(9):3224-3231. doi:10.1021/ac1032325
22. Cicchi R, Kapsokalyvas D, De Giorgi V, et al.Scoring of collagen organization in healthy anddiseased human dermis by multiphotonmicroscopy. J Biophotonics. 2010;3(1-2):34-43. doi:10.1002/jbio.200910062
23. Wen BL, Brewer MA, Nadiarnykh O, et al.Texture analysis applied to second harmonicgeneration image data for ovarian cancerclassification. J Biomed Opt. 2014;19(9):096007.doi:10.1117/1.JBO.19.9.096007
24. Kakkad SM, Solaiyappan M, Argani P, et al.Collagen I fiber density increases in lymph nodepositive breast cancers: pilot study. J Biomed Opt.2012;17(11):116017. doi:10.1117/1.JBO.17.11.116017
25. Huang YQ, Liang CH, He L, et al. Developmentand validation of a radiomics nomogram forpreoperative prediction of lymph node metastasisin colorectal cancer. J Clin Oncol. 2016;34(18):2157-2164. doi:10.1200/JCO.2015.65.9128
26. Agesen TH, Sveen A, Merok MA, et al.ColoGuideEx: a robust gene classifier specific forstage II colorectal cancer prognosis. Gut. 2012;61(11):1560-1567. doi:10.1136/gutjnl-2011-301179
27. Zhuo S, Chen J, Luo T, Zou D. Multimodenonlinear optical imaging of the dermis in ex vivohuman skin based on the combination ofmultichannel mode and Lambda mode. Opt Express.2006;14(17):7810-7820. doi:10.1364/OE.14.007810
28. Xu S, Wang Y, Tai DCS, et al. qFibrosis:a fully-quantitative innovative methodincorporating histological features to facilitateaccurate fibrosis scoring in animal model andchronic hepatitis B patients. J Hepatol. 2014;61(2):260-269. doi:10.1016/j.jhep.2014.02.015
29. Xu S, Kang CH, Gou X, et al. Quantification ofliver fibrosis via second harmonic imaging of theGlisson’s capsule from liver surface. J Biophotonics.2016;9(4):351-363. doi:10.1002/jbio.201500001
30. Sauerbrei W, Royston P, Binder H. Selection ofimportant variables and determination offunctional form for continuous predictors inmultivariable model building. Stat Med. 2007;26(30):5512-5528. doi:10.1002/sim.3148
31. Iasonos A, Schrag D, Raj GV, Panageas KS. Howto build and interpret a nomogram for cancerprognosis. J Clin Oncol. 2008;26(8):1364-1370. doi:10.1200/JCO.2007.12.9791
32. Kerr KF, Brown MD, Zhu K, Janes H. Assessingthe clinical impact of risk prediction models withdecision curves: guidance for correct interpretationand appropriate use. J Clin Oncol. 2016;34(21):2534-2540. doi:10.1200/JCO.2015.65.5654
33. Dormann CF, Elith J, Bacher S, et al. Collinearity:a review of methods to deal with it and a simulationstudy evaluating their performance. Ecography.2013;36(1):27-46. doi:10.1111/j.1600-0587.2012.07348.x
34. Chen J, Zhuo S, Chen G, et al. Establishingdiagnostic features for identifying the mucosa andsubmucosa of normal and cancerous gastric tissuesby multiphoton microscopy. Gastrointest Endosc.2011;73(4):802-807. doi:10.1016/j.gie.2010.12.016
35. Mocellin S, Pasquali S. Diagnostic accuracy ofendoscopic ultrasonography (EUS) for thepreoperative locoregional staging of primary gastriccancer. Cochrane Database Syst Rev. 2015;(2):CD009944.
36. Ajani JA, D’Amico TA, Almhanna K, et al. GastricCancer, Version 3.2016, NCCN Clinical PracticeGuidelines in Oncology. J Natl Compr Canc Netw.2016;14(10):1286-1312. doi:10.6004/jnccn.2016.0137
37. Provenzano PP, Eliceiri KW, Campbell JM,Inman DR, White JG, Keely PJ. Collagenreorganization at the tumor-stromal interfacefacilitates local invasion. BMC Med. 2006;4(1):38.doi:10.1186/1741-7015-4-38
38. Wyckoff JB, Wang Y, Lin EY, et al. Directvisualization of macrophage-assisted tumor cellintravasation in mammary tumors. Cancer Res.2007;67(6):2649-2656. doi:10.1158/0008-5472.CAN-06-1823
39. Levental KR, Yu H, Kass L, et al. Matrixcrosslinking forces tumor progression by enhancingintegrin signaling. Cell. 2009;139(5):891-906. doi:10.1016/j.cell.2009.10.027
40. Brown CM, Rivera DR, Pavlova I, et al. In vivoimaging of unstained tissues using a compact andflexible multiphoton microendoscope. J Biomed Opt.2012;17(4):040505. doi:10.1117/1.JBO.17.4.040505
41. Huland DM, Brown CM, Howard SS, et al. Invivo imaging of unstained tissues using longgradient index lens multiphoton endoscopicsystems. Biomed Opt Express. 2012;3(5):1077-1085.doi:10.1364/BOE.3.001077
42. Rivera DR, Brown CM, Ouzounov DG, et al.Compact and flexible raster scanning multiphotonendoscope capable of imaging unstained tissue.Proc Natl Acad Sci U S A. 2011;108(43):17598-17603.doi:10.1073/pnas.1114746108
43. Collins GS, Reitsma JB, Altman DG, Moons KG.Transparent Reporting of a multivariable predictionmodel for Individual Prognosis or Diagnosis(TRIPOD): the TRIPOD statement. Ann Intern Med.2015;162(1):55-63. doi:10.7326/M14-0697
44. Hsieh FY. Sample size tables for logisticregression. Stat Med. 1989;8(7):795-802. doi:10.1002/sim.4780080704
Association of the Collagen Signature With Lymph Node Metastasis in Early Gastric Cancer Original Investigation Research
jamasurgery.com (Reprinted) JAMA Surgery March 2019 Volume 154, Number 3 9/9
