A Monotonic and Prognostic Genomic Signature from Fibroblasts … · Genomics A Monotonic and...

Genomics

A Monotonic and Prognostic Genomic Signature fromFibroblasts for Colorectal Cancer Initiation, Progression,and Metastasis

Mireia Berdiel-Acer1,6, Daniel Cuadras4, Natalia Guill�en Díaz-Maroto1, Xavier Sanjuan5, Teresa Serrano5,Antoni Berenguer2, Victor Moreno2, Samuel Goncalves-Ribeiro1, Ramon Salazar3, Alberto Villanueva1, andDavid G. Molleví1

AbstractThe differential gene expression patterns between normal colonic fibroblasts (NCF), carcinoma-associated

fibroblasts from primary tumors (CAF-PT), and CAFs from hepatic metastasis (CAF-LM) are hypothesized to beuseful for predicting relapse in primary tumors. A transcriptomic profile of NCF (n ¼ 9), CAF-PT (n ¼ 14), andCAF-LM (n¼ 11) was derived. Prediction Analysis of Microarrays (PAM) was used to obtain molecular details foreach fibroblast class, and differentially expressed transcripts were used to classify patients according to recurrencestatus. A number of transcripts (n¼ 277) were common to all three types of fibroblasts and whose expression levelwas sequentially deregulated according to the transition: NCF!CAF-PT!CAF-LM. Importantly, the genesignature was able to accurately classify patients with primary tumors according to their prognosis. This capacity wasexploited to obtain a refined 19-gene classifier that predicted recurrence with high accuracy in two independentdatasets of patients with colorectal cancer and correlates with fibroblast migratory potential. The prognostic powerof this genomic signature is strong evidence of the link between the tumor-stroma microenvironment and cancerprogression. Furthermore, the 19-gene classifier was able to identify low-risk patients very accurately, which is ofparticular importance for stage II patients, whowould benefit from the omission of chemotherapy, especially T4N0patients, who are clinically classified as being at high risk.

Implications: A defined stromal gene expression signature predicts relapse in patients with colorectal cancer.Mol Cancer Res; 12(9); 1254–66. �2014 AACR.

IntroductionThe huge outcome heterogeneity among patients with

cancer, which hinders their clinical management, cannot beexplained solely by the genetic background of the cancercells. Some of this variation is due to intratumoral hetero-geneity, whereby tumors contain multiple cell populations,each with distinct characteristics of karyotype, receptors,antigenicity, etc. The relative abundance of each cell typevaries considerably between tumors and even between parts

of the same tumor (1). However, the predominant cell typein desmoplastic tumors is the carcinoma-associated fibro-blast (CAF; ref. 2).To date, several genetic classifiers of colorectal cancer have

been identified (3–6) but none of these is of relevance inrelation to the stroma. In this article, we provide informationabout the importance of the transcriptomic status of fibro-blasts to colorectal cancer prognosis.Fibroblasts are ubiquitous mesenchymal cells with differ-

ent vital functions during development, wound healing, anddisease. Depending on their anatomic location, fibroblastsshould be considered as different cell types as they haveparticular gene expression programs and different featuresaccording to their tissue specialization (7). However, astereotyped gene expression program in response to serumwas identified in fibroblasts from 10 different anatomic sites,revealing that in response to a wound many of the normaldifferentiated particularities of the fibroblasts in the wound-ed tissue are temporarily set aside in favor of a commonresponse (8). In addition, this common serum response geneexpression program was identified in many human tumorsand was a powerful predictor of metastasis in differentcarcinomas. The transcriptomic program of serum responsegenes was used as a prototype of the fibroblasts' activation

Authors' Affiliations: 1Translational Research Laboratory, 2Biomarkersand Susceptibility Unit, 3Medical Oncology Department, Institut Catal�ad'Oncologia-ICO; 4Statistics and Bioinformatics Unit, 5Pathology Depart-ment, Hospital Universitari de Bellvitge-IDIBELL, IDIBELL, Barcelona; and6Department of Medicine, Autonomous University of Barcelona, Bellaterra,Barcelona, Catalonia, Spain

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

Corresponding Author: David G. Mollevi, Translational Research Labo-ratory, Institut Catal�a d'Oncologia-ICO, IDIBELL, Av. Gran Via 199-203,08908 L'Hospitalet de Llobregat, Barcelona, Catalonia, Spain. Phone: 34-93-2607370; Fax: 34-93-2607466; E-mail: [email protected]

doi: 10.1158/1541-7786.MCR-14-0121

�2014 American Association for Cancer Research.

MolecularCancer

Research

Mol Cancer Res; 12(9) September 20141254

on November 18, 2020. © 2014 American Association for Cancer Research. mcr.aacrjournals.org Downloaded from

Published OnlineFirst May 14, 2014; DOI: 10.1158/1541-7786.MCR-14-0121

http://mcr.aacrjournals.org/

pattern, and the correlation of this expression in fibroblastswith a centroid of the differential expression in tumorsamples displayed a powerful prognostic information, gen-erating a continuous score that can be scaled (9).It is already known that as tumor evolves, malignant

cells educate surrounding stroma to create an adequatemicroenvironment for their convenience (10). This edu-cation can be understood through the degree of activationof fibroblasts as a surrogate marker. In other words, as thefibroblasts interact with other cells in the tumor, fibro-blasts progressively evolve, and this evolution could pro-vide new biomarkers and valuable information forpatient's outcome.Thus, according to the concept that there is a common

serum response program in different fibroblasts, despitebeing considered different cells, we hypothesize that sincethe earliest start of tumorigenic process, when an oncogenicstress transforms an epithelial cell until the cell metastasizes,there is a common andmonotonic response of the surround-ing fibroblasts that can be measured and associated toprognosis.We call this response as the fibroblasts progressiongenes (FPG).We used this common response signature (277genes) to develop a 25-gene classifier that characterized eachfibroblast type with respect to location and that was ofprognostic value for classifying patients with primarytumors. We also defined and validated a 19-gene classifier,being the best from the 277 genes predicting recurrence inpatients with colorectal cancer. This simple signature scorehas remarkable predictive capabilities that confirm thebiologic relevance of CAFs and their gene programs tocancer progression.

Materials and MethodsFibroblast isolationWe obtained fibroblast cultures from fresh surgical speci-

mens resected from patients with primary colorectal carci-noma: normal colonic fibroblasts (NCF, 9) from the normalcolonic mucosa (NCM) at least 5 to 10 cm from the surgicalmargin, carcinoma-associated fibroblasts from the primarytumor (CAF-PT, 14), and carcinoma-associated fibroblasts(CAF-LM, 11) from fresh surgical specimens of liver metas-tases. All samples were collected under the supervision of theEthics Committee of the Hospital Universitari de Bellvitge.These 34 samples comprise theMBdataset. Clinical baselinedata of patients whose fibroblasts were isolated are inSupplementary Table S1. To establish primary cultures, wehomogenized surgical specimens with collagenase IV anddispase in a MACS Separation Unit (Miltenyi Biotec),depleted homogenates from epithelial cells using anti-EPCAM Dynabeads (Invitrogen), and positively selectedfibroblasts using anti-fibroblast microbeads (reference 130-050-601 Miltenyi Biotec). Cells were then cultured inDMEMF12 þ 10% FBS (Gibco) with added penicillin/streptomycin.To confirm that cultures derived from tumor specimens

were pure fibroblast cultures, we used a large panel ofmarkers (vinculin, a-smooth muscle actin, vimentin,

E-cadherin, N-cadherin, and VE-cadherin) to exclude epi-thelial and endothelial cell contamination (SupplementaryFig. S1).

Microarray procedures and signature definitionRNA from fibroblast cultures in passage #3 was used to

hybridize with the Affymetrix GeneChip Human Gene 1.0ST Array. The microarray data were read, normalized, andtransformed to numerical expression values using thejustRMA function in the simpleaffy package (11). Differen-tially expressed genes were identified using the multi-classSignificance Analysis of Microarrays (SAM) technique,available in the samr package (12).We selected those probes that simultaneously displayed

a robust multi-array average (RMA) normalized expressionvalue of >4 in the three fibroblast populations (MBdataset). This yielded 10,779 probes common to NCF,CAF-PT, and CAF-LM. This rule was intended to excludegenes that had not normalized expression in at least onecategory, as well as background noise (Supplementary Fig.S2). The signature was defined as follows: fibroblastprogression genes (FPG; NCF>CAF-PT>CAF-LM andNCF<CAF-PT<CAF-LM) contained genes with a log n-fold change (logFC) NCF versus CAF-PT of >1.5 andCAF-PT versus CAF-LM of >1.5 (179 genes), and CAF-PT versus NCF of >1.5 and CAF-LM versus CAF-PT of>1.5 (98 genes); q < 0.05. Therefore, the FPG signaturecomprises genes whose expression level increased ordecreased during the NCM–primary tumor–liver metas-tasis sequence.

Definition of FPG class predictionThe Prediction Analysis of Microarrays (PAM; ref. 13)

nearest shrunken centroid classifier was applied to the MBdataset to refine the 277 differentially expressed genes(FPGs) to derive a classifier that better characterized eachclass (normal colonic fibroblasts, NCFs; CAFs from theprimary tumor, CAF-PTs; and CAFs from the liver metas-tasis, CAF-LMs). PAM estimates the predicted error ratebased on misclassification error, which is calculated byaveraging the errors from each of the 10-fold cross-valida-tions (CV).The classifier was then validated in external datasets.

GSE31279 (14) comprised 24 samples obtained by lasercapture microscopy (LCM): 10 normal colonic stroma, 10carcinoma-associated stroma, and 4 metastasis-stroma;GSE22598 (15) comprised 17 whole-tissue samples ofNCM and 17 paired colorectal carcinomas; the LP datasetcomprised 19 colorectal carcinomas and 21 paired livermetastases from our own institution. Public datasets fromGEO are detailed in Supplementary Table S2.Finally, we computed a signature score for the genes of the

classifier to assess its power in predicting patient survival intwo independent datasets, GSE14333 (ref. 6; 31, stage I; 64,stage II; and 71, stage III patients) and GSE33133 (16). Thesignature score was obtained by summarizing centered geneexpression means and computing a z-score for each patient.The expression score for each patient was classified as high or

Carcinoma-Associated Fibroblasts and Prognosis

www.aacrjournals.org Mol Cancer Res; 12(9) September 2014 1255




low according to the cutoff provided by Euclidian distance inthe ROC curve analysis.

Recurrence-prediction classifier developmentWe developed a recurrence signature from the 277 genes

comprising the FPG signature. PAM analysis identifiedrelevant genes for the classifier from GSE14333 (RMA-normalized) defining the recurrent and nonrecurrent classes.Internal validation was done by the leave-one-out CVmethod. Validation was carried out with GSE17538 (ref. 5;RMA-normalized).We also computed the z-score (signature score normalized

by SD) and classified patients as high or low expressionaccording to the cutoff obtained by Euclidean distance in theROC curve analysis for recurrence prediction (cutoff, 0.187).Patients with a score >0.187 were classified as high risk andscore <0.187 as low-risk. The same cutoff value (0.187) wasapplied to the validation datasets. Results were validated in theindependent datasetsGSE17538 (144patients; 33, stage I; 50,stage II; and 61, stage III) and GSE33113 (ref. 16; 90 stage IIpatients; RMA-normalized) taking disease-free survival (DFS)as a primary endpoint. GSE14333 and GSE17538 cohortspartially overlapped. Cases duplicated in GSE14333 andGSE17538 were excluded from the validation.We chose these datasets for training and validation

because data came from whole-tumor samples without anyenrichment in tumor cells, ensuring a minimum stromalpercentage between 30% and 50%.

Wound-healing assayFibroblasts were grown on P60 dishes until confluence.

We made a small scratch with a yellow pipette tip. Afterseveral PBS washes to remove floating cells, we began theassay by adding fresh medium. We obtained images of apredefined area of the scratch under a Leicamicroscope every2 hours. The experiment was done twice, with three repli-cates on each occasion.

Ontology analysisThe Database for Annotation, Visualization and Integrat-

ed Discovery (DAVID) v6.7 was used to determine likelybiologic processes in which differentially expressed genes(FPGs) were involved. Gene ontology (GO) attributes ofbiologic processes were listed, ordered by their FDR-adjust-ed q value of <0.05.

Determination of the 19-gene signature cell-typespecificityWe used data from GSE39396 (17) to check the stromal

specificity of the genes in our CAF signature. GSE39396consisted of four cell type populations isolated by FACSfrom six different colorectal tumors: EPCAMþ epithelialcells, CD45þ inflammatory cells, CD31þ endothelial cells,and FAPaþ CAFs.

Internal microarray validationWe internally validated the genes comprising the 25-gene

signature by means of RT-PCR, confirming the results by

Western blotting. The antibodies used for this purpose wererabbit anti-human SLC7A2 (HPA009169; Sigma-Aldrich),rabbit anti-human TGFB2 (V) (sc-90; Santa Cruz Biotech-nology, Inc.), and rabbit anti-human ARHGDIB(Ab15198; Abcam).

Statistical analysisKaplan–Meier and Cox regression analysis estimates for

DFS and disease-specific survival (DSS) were calculated andcompared by the likelihood ratio tests. Recurrence and deathfrom disease were considered as events.Positive and negative posterior probabilities were chosen

to measure the power of the classifier when consideringclinical decision making.

ResultsDifferential gene expression of fibroblasts along cancerprogressionAccording to the overall design (Supplementary Fig. S2),

we isolated fibroblasts (n ¼ 34) from the three microenvir-onments in which colorectal epithelial cells reside (NCM,primary tumor, and liver metastasis). Because fibroblastsfrom different locations seem to have particular transcrip-tomic programs but share a stereotyped gene expressionpattern in response to serum (8) and this response can bescaled to predict prognosis (9), we aimed to establishwhetherthere is a set of fibroblastic genes whose expression arisesgradually according to their anatomic demarcation involvedin the steps of colorectal cancer progression (NCF, CAF-PT,and CAF-LM). After filtering the data, 10,779 probes metthe characteristics imposed (cartoon, Supplementary Fig.S3A), most of which corresponded to development-relatedontologies (Table 1). Interestingly, overexpressed genes inthe sequence NCF-to-CAF-LM (low in NCF and highin CAF-LM) were also associated with wound-healingresponse, whereas in the opposite direction the biologicprocesses seem to be enriched in inflammatory response.From these probes, we selected those that fulfilled thecriterion of a >1.5-fold change (normalized log2 value) inexpression at each step at the same time (as illustrated in thecartoon, Supplementary Fig. S3B). We refer to this asmonotonic gene expression during cancer progression,meaning that gene expression follows a decreasing or increas-ingmonotonic function. By thismeans, we identified 98 and179 genes that were upregulated and downregulated, respec-tively, from NCF to CAF-LM (Supplementary Table S3).Together, these 277 genes were designated as the FPGsignature (fibroblastic genes relevant to cancer progression;Supplementary Fig. S3C).

Refining the FPG signature: developing a classifierWe aimed to refine the FPG signature to determine the

minimum number of genes whose expression levels charac-terize each class of fibroblast involved in colorectal cancerprogression. Our 34 samples of fibroblasts (MB Dataset)were subjected to PAM using the nearest shrunken centroidmethod to define the classifier. Ten-fold cross-validation was

Berdiel-Acer et al.

Mol Cancer Res; 12(9) September 2014 Molecular Cancer Research1256




used, randomly dividing the samples into 10 approximatelyequally sized parts. All parts were balanced to ensure that thethree classes (NCF, CAF-PT, and CAF-LM) were distrib-uted proportionally among each of the 10 parts. Aminimumof 25 estimated genes predicted the lowest misclassificationerror (Supplementary Fig. S3D and S3E). Genes are listedin Table 2 and Supplementary Fig. S4A. Nine of these geneshave been validated by means of RT-PCR (SupplementaryFig. S4B). Eleven of the 25 genes were downregulated, mostof them [EYA4 (18), FBLN1 (19), (20), FBN2 (21–23),ADH1B (24), CYP39A1 (25), CHRDL1 (26), and TCF21(27, 28)] by methylation in tumorigenic processes. Theother 14, most of which are already known to be involvedin cancer (TGFB2, NTF3 (29), UACA (30), SFRP4 (31),and ULBP2 (32), were overexpressed. Another gene likeARHGDIB has been associated with prognosis (33, 34) andcisplatin resistance (35). Selected genes were validated at theprotein level (Supplementary Fig. S3F), and were found tohave a greater level of expression in fibroblasts as the cancerprogressed.

The overall error rate after 10-fold cross-validation was0.174 in the training samples (Table 2). The CAF-PT classwas the biggest contributor to the misclassification, somesamples beingmistaken for CAF-LM and forNCF (Fig. 1A).We validated the 25-gene signature in three independentdatasets, two whole-tumor samples (Dataset LP andGSE22598) and one microdissected sample (GSE31279).The probability of correctly classifying a sample was veryhigh (Fig. 1B, Table 2), enabling us to conclude that these 25genes were those whose expression levels best characterizeeach fibroblast class. Therefore, depending on the averagegene expression value we could discriminate whether a givensample was an NCM, a primary tumor, or a liver metastasis.

Gene expression associated with risk of progressionTheCAF-PT class exhibited the greatest variation of those

considered. We attempted to check that misclassified sam-ples could be associated with prognosis, on the assumptionthat CAF-PTs with a higher signature score (reaching a valuetypical of CAF-LM) were at greater risk (i.e., had a poorer

Table 1. Gene ontology

Overexpressed along the sequence NCF, CAF-PT, CAF-LM

Category Term Count PFoldenrichment FDR

GOTERM_BP_ALL GO:0009611�response to wounding 31 4.86E�07 2.841 8.42E�04GOTERM_BP_ALL GO:0042060�wound healing 17 2.04E�06 4.319 3.54E�03GOTERM_BP_ALL GO:0009605�response to external stimulus 41 4.62E�06 2.173 8.00E�03GOTERM_BP_ALL GO:0008544�epidermis development 16 5.95E�06 4.222 1.03E�02GOTERM_BP_ALL GO:0007398�ectoderm development 16 1.36E�05 3.94 2.36E�02GOTERM_BP_ALL GO:0048856�anatomical structure development 82 1.59E�05 1.568 2.75E�02GOTERM_BP_ALL GO:0048513�organ development 62 1.80E�05 1.723 3.11E�02GOTERM_BP_ALL GO:0048731�system development 77 1.81E�05 1.595 3.13E�02GOTERM_BP_ALL GO:0006928�cell motion 26 1.81E�05 2.64 3.14E�02GOTERM_BP_ALL GO:0050878�regulation of body fluid levels 13 3.24E�05 4.491 5.61E�02GOTERM_BP_ALL GO:0001568�blood vessel development 17 5.09E�05 3.346 8.81E�02

Infraexpressed along the sequence NCF, CAF-PT, CAF-LM

Category Term Count PFoldenrichment FDR

GOTERM_BP_ALL GO:0050727�regulation of inflammatory response 12 9.41E�07 7.021 1.64E�03GOTERM_BP_ALL GO:0051270�regulation of cell motion 18 1.62E�06 4.147 2.83E�03GOTERM_BP_ALL GO:0042127�regulation of cell proliferation 40 2.51E�06 2.263 4.38E�03GOTERM_BP_ALL GO:0009611�response to wounding 30 7.01E�06 2.546 1.22E�02GOTERM_BP_ALL GO:0002526�acute inflammatory response 12 9.74E�06 5.559 1.70E�02GOTERM_BP_ALL GO:0045595�regulation of cell differentiation 28 1.60E�05 2.541 2.80E�02GOTERM_BP_ALL GO:0045597�positive regulation of cell differentiation 18 1.62E�05 3.495 2.82E�02GOTERM_BP_ALL GO:0030334�regulation of cell migration 15 2.72E�05 3.947 4.75E�02GOTERM_BP_ALL GO:0040012�regulation of locomotion 16 2.82E�05 3.706 4.92E�02GOTERM_BP_ALL GO:0009605�response to external stimulus 41 2.93E�05 2.012 5.10E�02GOTERM_BP_ALL GO:0006950�response to stress 63 3.49E�05 1.682 6.10E�02GOTERM_BP_ALL GO:0006954�inflammatory response 21 3.98E�05 2.900 6.94E�02GOTERM_BP_ALL GO:0050793�regulation of developmental process 33 4.75E�05 2.184 8.29E�02GOTERM_BP_ALL GO:0032101�regulation of response to external stimulus 14 5.65E�05 3.940 9.85E�02






Table 2. Twenty-five–gene signature and prediction in training dataset MB and validation datasets LP,GSE22598, and GSE31279

Average expression (log ratio)

Gene symbol Group NCF CAF-PT CAF-LM

ARHGDIB UP �0.0179 0 0.9268C5orf46 UP 0 0 0.8287TMEM176B DW 0 0 �0.5635TSPAN2 DW 0 0 �0.5229EYA4 DW 0 0 �0.4271CYP39A1 DW 0 0 �0.327ADH1B DW 0.3203 0 0SRGN UP 0 0 0.3089CHRDL1 DW 0 0 �0.304PTX3 UP 0 0 0.1956SLC7A2 UP 0 0 0.1849TGFB2 UP �0.172 0 0SFRP4 UP 0 0 0.1665FBLN1 DW 0 0 �0.1621ULBP2 UP �0.1598 0 0.0049FBN2 DW 0 0 �0.1571CCDC81 UP 0 0 0.1485GPR1 UP 0 0 0.1256TNFSF4 UP �0.0732 0 0CHL1 DW 0 0 �0.0609UACA UP 0 0 0.056TCF21 DW 0 0 �0.0406NTF3 UP �0.0191 0 0HECW2 UP 0 0 0.0072SLC38A4 DW 0 0 �0.0004

Dataset MBPrediction

Real status NCF CAF-PT CAF-LM Class error rate

NCF 9 0 0 0CAF-PT 3 8 3 0.428CAF-LM 0 0 11 0Overall error rate ¼ 0.174 (cross-validation)

Validation

Dataset LPPrediction

Real status Primary tumor Liver metastasis Class error rate

Primary tumor 16 3 0.158Liver metastasis 0 21 0Overall error rate ¼ 0.075 (cross-validation)

GSE22598Prediction

Real status Normal mucosa Primary tumor Class error rate

Normal mucosa 15 2 0.118Primary tumor 0 17 0Overall error rate ¼ 0.059 (cross-validation)

(Continued on the following page)

Berdiel-Acer et al.





prognosis) than CAF-PTs with a lower signature score (Fig.1C). We evaluated the prognostic power of the 25-geneclassifier in GSE14333, which consisted of 166 colorectalprimary tumors (stages I to III) and GSE33113 (90 patients,stage II). Assigning patients to high or low expression score(according to the 50th percentile of the 25-gene scoresobtained from isolated fibroblasts; cutoff, �0.43), ROCcurve (Fig. 1D), and DFS Kaplan–Meier analysis (Fig. 1E)indicated that these genes are associated with prognosis (HR,3.5; P ¼ 0.00029). An additional validation in GSE33113(90 patients stage II) confirmed the tendency associated toprognosis (Fig. 1F and G). Therefore, this 25-gene score notonly classified each fibroblast sample correctly but alsoclassified patients with primary tumors according to theirprognosis. The FPG and the 25-gene signature wereobtained regardless of the gene expression data of the tumorspecimens and patients' clinical data, so the prognosticpower of this signature is strong evidence of the link betweenthe tumor stroma and cancer progression.

Best genes associated with recurrence in patients withcolorectal cancerThe 25-gene classifier was defined as the best genes whose

monotonic expression characterized each fibroblast type.However, it was not specifically designed for predictingoutcome. Nevertheless, given the prognostic power of the25-gene classifier obtained from the FPG stereotyped geneexpression pattern of fibroblasts, we aimed to define the bestsignature (from the 277 deregulated genes) for classifyingpatients with colorectal cancer, using recurrence as a primaryendpoint. A PAM-optimized process yielded a 19-genesignature from GSE14333 (Supplementary Table S4). Thescore for each gene derived by PAM was associated with thestatus of the patient (recurrent or nonrecurrent) and the level

of expression of the gene during cancer progression (UP orDW in Supplementary Table S4). Validation of PAM in theindependent dataset GSE17538 showed a specificity of79.7% and sensitivity of 54.9% after cross-validation. Ofthese 19 genes, 15 were overexpressed in the sequenceNCF<CAF-PT<CAF-LM, and were thereby overexpressedduring cancer progression; four genes were underexpressed.To analyze the cell type–specific expression of the 19-

gene signature further, we used data from GSE39396(17), consisting of FACS-purified cell types from 6patients with colorectal cancer. Relative levels of celltype–specific marker genes confirmed that most of the19 genes are representative of the microenvironmentcompartment, especially from FAPaþ CAFs, with theexception of CFTR, which was expressed by EPCAMþ

cells (Fig. 2A). In addition, because recruitment of fibro-blasts is one of the characteristics of desmoplastic tumors,we performed a wound-healing assay to compare themigration of fibroblasts with high and low 19-gene sig-nature scores. Fibroblasts with high signature scores hadbetter migratory abilities, as illustrated in Fig. 2B.To rank patients by their level of expression of the 19

genes, and assuming equal biologic relevance for each of the19 genes of the signature, we derived a score by calculatingthe average of the z-scores, applying a weight of þ1 tooverexpressed genes and of�1 to underexpressed genes. Forpatient's categorization, we used the Euclidian distancebetween sensitivity and specificity (ROC curve analysis;AUC, 0.79; cutoff, 0.187; Fig. 3A and B). The recurrenceclassifier identified 97 low-expression (58.4%) and 69 high-expression patients (41.6%; OR, 3.47; Fig. 3C). Low-riskand high-risk patients had significantly different survivaloutcomes, measured as DFS [HR, 5.49; 95% confidenceintervals (CI), 2.86–10.53; P < 0.0001; Fig. 3D]. An

Table 2. Twenty-five–gene signature and prediction in training dataset MB and validation datasets LP,GSE22598, and GSE31279 (Cont'd )

GSE31279Prediction

Real status Normal stroma Primary tumor stroma Liver metastasis Class error rate

Normal stroma 9 1 0 0.10Primary tumor stroma 3 7 0 0.30Liver metastasis 1 0 3 0.25Overall error rate ¼ 0.203 (cross-validation)

NOTE: Shrunken centroids for each class of fibroblast (n ¼ 34). The column "group" describes the direction of the expression in thesequence NCF-to-CAF-LM.The method computes a standardized centroid for each class. Centroids are the average gene expression for each gene in each classdivided by the within-class SD for that gene. Nearest centroid classification takes the gene expression profile of a new sample, andcompares it with each of these class centroids. The classwhose centroid that it is closest to, in squared distance, is the predicted classfor that new sample. Nearest shrunken centroid classification shrinks each of the class centroids toward the overall centroid for allclasses by a threshold. This shrinkage consists of moving the centroid toward zero by threshold, setting it equal to zero if it hits zero.Methylation status of these 25 genes has been evaluated in 258 samples of primary tumors from the TCGA (TheCancerGenomeAtlas).Results are depicted in Supplementary Table S5.Abbreviations: UP, CAF-LM > CAF-PT > NCF; DW, NCF > CAF-PT > CAF-LM.






Figure1. A, estimatedprobabilities for the fibroblast training set. Samples are partitionedby the true (top) andpredicted (bottom) classes.NCFandCAF-LMarealways well classified and some samples of CAF-PT have higher probabilities of being misclassified. B, validation of PAM analysis in the GSE22598 datasetconsisting of 17 pairs of homogenized samples of whole-tumor or whole normal mucosa. (Continued on the following page.)

Berdiel-Acer et al.





independent cohort of 141 patients (GSE17538)was used toevaluate the performance of the signature score of the 19-gene classifier. In the validation set, 55 patients were definedas high risk (39%) and 86 as low risk (61%; Fig. 3E; AUC,0.81). The prevalence of this validation dataset was 24.1%and the positive and negative posterior probabilities were50% (95% CI, 36%–64%) and 8% (95% CI, 3.5%–16.4%), respectively. Stratifying patients by stage, the prev-alence for stage II patients was 18%, with positive and

negative posterior probabilities of 37% (95% CI, 17%–61%) and 6% (95% CI, 1%–22%), respectively. Interest-ingly, the clinical decision margin was wider for stage III,with a prevalence of 39% and positive and negative posteriorprobabilities of 61% (95% CI, 42%–77%) and 16% (95%CI, 6%–34%), respectively, thus with a difference of 45points. Univariant Cox regression analysis revealed thathigh-risk patients had an HR of 7.43 (P < 0.0001) for DFS(Fig. 3G; statistically significant also stratifying by

(Continued.) The 25-gene signature can correctly classify normal mucosa samples and tumor samples with an accuracy of 94.1% after cross-validation.C, 25-gene signature score for the 34 samples of fibroblasts. Green, NCFs; orange, CAF-PT; red, CAF-LM. CAF-PT samples are those with a widestrange of values, some samples having values similar to those of the CAF-LM and some being closer to those of the NCF. Red dotted line, 50th percentileof the signature score. We used this cutoff to classify patients in F and G. D, ROC curve for recurrence prediction according to the 25-gene signaturescore of the dataset GSE14333 (166 cases of colorectal cancer patients; AUC, 0.69). E, DFS Kaplan–Meier curves for the performance of the 25-genesignature in the GSE14333 dataset, categorizing the signature expression as high or low according to the cutoff obtained from the isolated fibroblasts(HR, 3.5; P ¼ 0.00029). F, ROC curve for recurrence prediction according to the 25-gene signature score of the dataset GSE33113 (90 cases ofpatients with stage II colorectal cancer; AUC, 0.63). G, DFS Kaplan–Meier curves for the performance of the 25-gene signature in the GSE33113 dataset,categorizing the signature expression as high or low according to the cutoff obtained from the isolated fibroblasts (HR, 2.35; P ¼ 0.12).

Color keyand histogramA

B

010

2030

40

–2 –1 0 1 2Value

Cou

nt

19-gene signature low fibroblasts

0 h 24 h 0 h 24 h

19-gene signature high fibroblasts

Time

Replicate 1

Replicate 2

Replicate 3

Figure 2. A, unsupervisedhierarchical clustering of the 19-gene signature with data fromGSE39396, consisting of four celltype populations isolated by FACSfrom six different colorectaltumors: EPCAMþ epithelial cells(green), CD45þ inflammatory cells(gray), CD31þ endothelial cells(yellow), and FAPaþ CAFs(orange). The heatmap shows thatCAFs (orange) cluster in a separategroup with respect to theexpression values of the 19-genesignature. With the exception ofCFTR, which is mainly expressedby epithelial cells, the other 18genes are basically of stromalorigin. B, representative wound-healing assay images of replicatesof 19-gene signature low-scorefibroblasts and 19-gene signaturehigh-score fibroblasts. Themicrographs show that highsignature score fibroblasts (right)have a higher migratory capacitythan those with a low signaturescore (left).






Figure 3. A, ROC curve for the 19-gene signature performance in the GSE14333 dataset (AUC, 0.79). The AUC corresponds to the probability of the 19-genesignature ranking a randomly selected positive example higher than a randomly selected negative example, where 1 is the perfect test value andthe dotted diagonal line shows the performance of a random predictor. B, histogram for 19-gene classifier score values for the patients of the trainingset. The red line corresponds to the cutoff (0.187) obtained from the Euclidian distance between sensitivity and specificity for recurrence prediction. The samecutoff value (0.187) is applied to the validation datasets. C, waterfall plot for recurrence risk of 166 patients from GSE14333 according to the 19-genesignature score (OR, 3.47). D, DFS Kaplan–Meier curves for the performance of the 19-gene signature in the GSE14333 dataset for signatures classifiedas high (red) or low (black) expression. HR, 5.49; likelihood ratio test, P < 0.0001. Five-year DFS rates were 36% (high-expression patients) and83% (low-expression patients). E, ROC curve for the 19-gene signature performance in the GSE17538 validation dataset (AUC, 0.81). F, waterfall plot forrecurrence risk of 141 patients from validation dataset GSE17538 according to the 19-gene signature score (OR, 3.29; sensitivity, 78.8%; and specificity,73.6%). G, DFS Kaplan–Meier curves for the performance of the 19-gene signature in the GSE17538 dataset for scores classified as high (red) or low (black)expression. HR, 7.43; likelihood ratio test, P < 0.0001. Five-year DFS rates were 32% (high-expression patients) and 83% (low-expression patients). H and I,stratifying for stages, patientswith highexpression scores havea risk of shorterDFS5.71-times higher than low-expressionpatients for stage II (P¼0.015) and4.71 times for stage III. J, DSS Kaplan–Meier curves for the performance of the 19-gene signature in the GSE17538 validation dataset for scores classified ashigh (red) or low (black) expression. HR, 7.44; P < 0.0001. Five-year DSS rates were 41% (high-expression patients) and 87% (low-expression patients).

Berdiel-Acer et al.





stages; Fig. 3H and I) and of 7.44 (P < 0.0001) for DSS(Fig. 3J). In multivariate analysis, after adjusting for signif-icant variables in univariate analysis (stage and grade) theclassifier score remained statistically significant for DFS(HR, 5.52; 95% CI, 2.23–13.65; P < 0.0001; Table 3).The 19-gene signature was also independently associatedwith DSS (P ¼ 0.004; Table 3). In addition, the signaturescore as a continuous variable was also significant, increasingthe risk of relapse (HR) 1.93 times per unit increase of thescore (P < 0.0001; 95% CI, 1.47–2.53).To confirm the prognostic value in stage II patients, we

used a third independent cohort of 90 patients (GSE33113).Using the cutoff obtained from the training dataset (Fig. 4A),the classifier also performed well in this third validationcohort (AUC, 0.71; Fig. 4B and C). A high level ofexpression of the 19-gene signature was clearly associatedwith poor prognosis in this cohort (HR, 3.84; P ¼0.0034; Fig. 4D). Interestingly, the prevalence in this cohortwas 21% and the positive and negative posterior probabil-ities were 37% (95% CI, 22%–55%) and 10% (95% CI,4%–23%). Moreover, the signature score as a continuousvariable was also significant, increasing the risk of relapse(HR) 2.06 times per unit increase of the score (P ¼ 0.003;95% CI, 1.29–3.29).Thus, for both validation datasets, the classifier clearly

identified low-risk patients, particularly in the case of stage IIpatients, who would benefit from the omission of chemo-therapy, especially those who are T4N0 and thus clinicallyclassified as being at higher risk.Unfortunately, we only know the age, gender, recurrence

status, grade (only for GSE17538), stage, and survival timeof the patients in these cohorts, as the authors who uploadedthe data to the GEO did not provide information aboutother characteristics.

DiscussionMolecular biomarkers for predicting relapse of colorectal

cancer are needed to improve the selection of patients whowould benefit from an adjuvant treatment. Recently, thegreat importance of the stroma to the prognosis of varioustypes of cancer, especially those with a high degree of

desmoplasia, has become recognized. Our hypothesis is thatdifferential gene expression between NCF, CAF-PT, andCAF-LM would be useful for predicting relapse in primarytumors. We identified a short list of genes of stromalfibroblasts from anatomic sites involved in colorectal carci-nogenesis, whose transcriptional level was correlated withcancer progression. These genes display a gradual increase/decrease in expression during cancer progression, fromNCM to liver metastasis. The scalability observed for thisresponse may link the regenerative capacities of differentorgans after an injury (i.e., liver) and the greater ability ofCAF-LM to induce epithelial–mesenchymal transition(EMT) phenotypes in epithelial cells more efficiently thanin other types of myofibroblasts (36). Many of these genesare associated with wound healing and correlate with theactivation status of fibroblasts. As an example, SERPINE1,POSTN, HBEGF, PTX3 or TGFb2, IL1b, and TGFb1responsive genes, increase from NCF to CAF-LM, withIL1b and TGFb1 being the main inductors of fibroblastactivation. This transcriptomic program for wound responseseems to be more effective in the more aggressive liver andprimary tumors, and could explain the greater degree ofdesmoplasia that is associated with poor outcome (37).About downregulated genes, there is an enrichment of genesinvolved in inflammatory processes. This fact might explainthe better outcome of patients with an intense inflammation-related component infiltration. Probably the lack of suchCAF-derivedmolecules in themore aggressive tumors wouldavoid the recruitment of inflammatory cells to the tumor. Insummary, this result strongly suggests that changes incolorectal tumor stroma have a crucial role in disease pro-gression and outcome. Obviously, other fibroblast-specificgenes could have prognostic value, although their level ofexpression would not be associated with the transition fromnormal fibroblasts to highly activated fibroblasts, as thosegenes would be differentially expressed in a particular micro-environment (e.g., they could be expressed in hepatic metas-tases but be absent from primary tumors).Although fibroblasts from different demarcations display

particular transcriptomic programs (38), even consideringeach fibroblast type as a different cell, it is interesting toobserve that the transcriptome is dynamic and plastic when

Table 3. Multivariate Cox proportional hazard analysis for DFS and DSS in the validation set GSE17538

Multivariate analysis 19-gene signature GSE17538

DFS DSS

HR (95% CI) P HR (95% CI) P

Grade 1.59 (0.66–3.79) 0.3 1.63 (0.64–4.12) 0.3Stage 2.06 (1.1–3.84) 0.023 1.455 (1.07–5.51) 0.03319-gene classifier (H vs. L) 5.52 (2.23–13.65) <0.0001 6.029 (1.76–20.65) 0.004

NOTE:We included in the analysis only variables that were statistically significant in univariate analysis, then gender and age >55 yearswere excluded (n more clinical variables were available for the dataset at GEO).






confronted with a malignant cell, probably as a result of thecross-talk between the two compartments. This is particu-larly relevant if we consider that CAFs from the primarytumor are derived from local NCF when oncogenicallystimulated (same anatomic demarcation), but CAF-LM arecells with a different origin.Previous studies have not focused on the role of the three

microenvironments in which colorectal cancer cells interactduring tumor progression. In addition, different stromalclassifiers exist for breast (39), prostate (40), lung (41), andhepatocellular carcinoma (42), but this issue has not so farbeen addressed with respect to colorectal cancer; instead, thefocus has been on CAFs, the principal component of thestroma. We also identified a 19-gene signature that predictsdisease outcome with remarkable accuracy in whole-tumorsamples.The scalability of the stereotyped gene expression pat-

tern was obtained from isolated fibroblasts of colorectaltumor-associated demarcations, so the prognostic powerof the expression pattern is strong evidence of the impor-tance of the stroma in the oncogenesis of colorectaltumors. This was observed implicitly in a signature pre-viously obtained from whole-tumor samples, in whichmany of the 48 most relevant genes were of stromal origin(4). Because the method has been performed and validatedin whole-tumor samples, in which the stroma/epitheliumratio was not known (37), the expression signal detectedemphasizes the role of stroma. Most of the 19 genesidentified in the present study are strongly expressed byCAFs (with the exception of ULBP2 and CFTR) and to alesser extent by other cells of the tumor stroma. Inaddition, many of these genes have been associated pre-viously with tumorigenic processes. Stromal SERPINE1(codifying for PAI-1) has been involved in cell migrationand angiogenesis promotion in different desmoplastictumors (43, 44). INHBA (a TGFB family member) hasbeen recently observed to be expressed in the sequencefrom NCM–colonic adenoma–colorectal adenocarcinoma(45) and has been associated with prognosis (46). Elevatedexpression of POSTN is reported to be associated with theinvasion and anchorage-independent growth of the met-astatic process of head and neck squamous cell carcinoma(47) and the invasive neoplasm of the pancreas (48).Furthermore, secreted POSTN was reported to triggerEMT through coactivation of EGFR and alpha 5 integrin(49). Other genes such as NTM (neurotrimin) andDENND2A have not been yet involved in cancer processesneither in the stroma nor epithelium.

Figure 4. A, histogram for 19-gene classifier score values for the patientsof the validation dataset GSE33113 (90 patients stage II). The red linecorresponds to the cutoff (0.187) obtained from the Euclidian distancebetweensensitivity andspecificity for recurrenceprediction in the training

dataset. B, ROC curve for 19-gene signature performance in theGSE33113 (AUC, 0.71) validation dataset. C, waterfall plot for recurrencerisk of 90patients from validation datasetGSE33113 according to the 19-gene signature score (OR¼ 2.21; Sensitivity 68.4% and specificity 69%).D, DFS Kaplan–Meier curves for the performance of the 19-genesignature in the GSE33113 validation dataset (stage II) for signaturesclassified as high (red) or low (black) expression. HR, 3.84; likelihood ratiotest, P ¼ 0.0034. Five-year DFS rates were 60% and 87% for high- andlow-expression patients, respectively.


Berdiel-Acer et al.




According to previously reported gene signatures in colo-rectal cancer, we found overlap only between two genes(PDLIM3 and ULBP2) of our classifier and a recent molec-ular classification (50), contributing also these genes tocharacterize the group of worst outcome.We present different, complementary ways of interpreting

the results that could be of great relevance, depending on theclinical scenario. First, PAM analysis identifies the genes ofrelevance in terms of binary events (recurrence/nonrecur-rence) that are highly specific after cross-validation. Second,the signature score (equal weight for all genes) reveals thebiologic relevance of the genes selected by PAM, clearlydemonstrating that this signature is an independent predictorof outcome, even for stage III patients. This step is crucialbefore trying to develop a model that is more mathematicallyaccurate albeit without a biologic rationale. This reinforcesthe role of the stroma, and particularly fibroblasts, as impor-tant components of cancer progression, rather than as merecompanions of malignant cells.On the basis of our results, we provide accurate informa-

tion for clinical decision making, as our classifier performsvery well in identifying low-risk patients who would benefitfrom an omission of adjuvant chemotherapy if validated inindependent clinical samples. This could be very significantfor those stage II T4N0, perforated, obstructed, poorlydifferentiated, or microsatellite stable patients, who wouldotherwise be offered toxic adjuvant chemotherapy. In stageIII patients, our signature also differentiates low-riskpatients, but these results need to be corroborated in a largerseries and in trials to convince oncologists to withholdchemotherapy completely from these patients, and eventhose of the T2N1 subgroup, whose members have a betterclinical prognosis.In conclusion, using the transcriptomic signature of fibro-

blasts that we defined and specifically the 19-gene classifier

derived from it in a clinical setting, we can provide accurateinformation about the risk of recurrence, and may facilitatethe selection of patients at risk of recurrence, especially forhigh-risk stage II patients who would benefit from adjuvanttherapy.The findings from the stratification of patients based

exclusively on the CAF transcriptomic program indicate theneed to develop therapeutic strategies focused on these cells,as they are the main components of desmoplastic tumors.Prospective studies are needed to determine whether treat-ment decisions based on our stromal 19-gene classifier couldbenefit patients.

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

Authors' ContributionsConception and design: D.G. MollevíDevelopment of methodology: N.G. Díaz-Maroto, D.G. MollevíAcquisition of data (provided animals, acquired and managed patients, providedfacilities, etc.): M. Berdiel-Acer, X. Sanjuan, T. Serrano, V. Moreno, S. Goncalves-Ribeiro, A. Villanueva, D.G. MollevíAnalysis and interpretation of data (e.g., statistical analysis, biostatistics, compu-tational analysis): M. Berdiel-Acer, D. Cuadras, A. Berenguer, V. Moreno,A. Villanueva, D.G. MollevíWriting, review, and/or revision of the manuscript: M. Berdiel-Acer, X. Sanjuan,A. Berenguer, V. Moreno, S. Goncalves-Ribeiro, D.G. MollevíStudy supervision: R. Salazar, D.G. Molleví

Grant SupportThis work was supported by projects of the Instituto de Salud Carlos III (grants

PI07/0657 and PI10/1604) and the Ag�encia de Gesti�o d'Ajuts Universitaris i deRecerca (grant number AGAUR SGR-2009-681).

The costs of publication of this article were defrayed in part by the payment of pagecharges. This article must therefore be herebymarked advertisement in accordance with18 U.S.C. Section 1734 solely to indicate this fact.

Received March 5, 2014; revised April 25, 2014; accepted April 25, 2014;published OnlineFirst May 14, 2014.

References1. Lee AH, Dublin EA, Bobrow LG. Angiogenesis and expression of

thymidine phosphorylase by inflammatory and carcinoma cells inductal carcinoma in situ of the breast. J Pathol 1999;187:285–90.

2. Elenbaas B, Weinberg RA. Heterotypic signaling between epithelialtumor cells and fibroblasts in carcinoma formation. Exp Cell Res2001;264:169–84.

3. Salazar R, Roepman P, Capella G, Moreno V, Simon I, Dreezen C,et al. Gene expression signature to improve prognosis pre-diction of stage II and III colorectal cancer. J Clin Oncol 2011;29:17–24.

4. O'Connell MJ, Lavery I, Yothers G, Paik S, Clark-Langone KM, LopatinM, et al. Relationship between tumor gene expression and recurrencein four independent studies of patients with stage II/III colon cancertreated with surgery alone or surgery plus adjuvant fluorouracil plusleucovorin. J Clin Oncol 2010;28:3937–44.

5. Smith JJ, Deane NG, Wu F, Merchant NB, Zhang B, Jiang A, et al.Experimentally derived metastasis gene expression profile predictsrecurrence and death in patients with colon cancer. Gastroenterology2010;138:958–68.

6. Jorissen RN, Gibbs P, Christie M, Prakash S, Lipton L, Desai J, et al.Metastasis-associated gene expression changes predict poor out-comes in patients with dukes stage B and C colorectal cancer. ClinCancer Res 2009;15:7642–51.

7. Rinn JL, Bondre C, Gladstone HB, Brown PO, Chang HY. Anatomicdemarcation by positional variation in fibroblast gene expressionprograms. PLoS Genet 2006;2:e119.

8. Chang HY, Sneddon JB, Alizadeh AA, Sood R, West RB, MontgomeryK, et al. Gene expression signature of fibroblast serum responsepredicts human cancer progression: similarities between tumors andwounds. PLoS Biol 2004;2:E7.

9. Chang HY, Nuyten DS, Sneddon JB, Hastie T, Tibshirani R, Sorlie T,et al. Robustness, scalability, and integration of a wound-responsegene expression signature in predicting breast cancer survival. ProcNatl Acad Sci U S A 2005;102:3738–43.

10. Bissell MJ, HinesWC.Whydon't we getmore cancer? A proposed roleof the microenvironment in restraining cancer progression. Nat Med2011;17:320–9.

11. Wilson CL, Miller CJ. Simpleaffy: a BioConductor package forAffymetrix Quality Control and data analysis. Bioinformatics2005;21:3683–5.

12. Tusher VG, Tibshirani R, Chu G. Significance analysis of microarraysapplied to the ionizing radiation response. Proc Natl Acad Sci U S A2001;98:5116–21.

13. Tibshirani R, Hastie T, Narasimhan B, Chu G. Diagnosis of multiplecancer types by shrunken centroids of gene expression. Proc NatlAcad Sci U S A 2002;99:6567–72.






14. Abba M, Laufs S, Aghajany M, Korn B, Benner A, Allgayer H. Lookwho's talking: deregulated signaling in colorectal cancer. CancerGenomics Proteomics 2012;9:15–25.

15. Okazaki S, Ishikawa T, Iida S, IshiguroM, Kobayashi H, Higuchi T, et al.Clinical significance of UNC5B expression in colorectal cancer. Int JOncol 2012;40:209–16.

16. de Sousa EMF, Colak S, Buikhuisen J, Koster J, Cameron K, de JongJH, et al.Methylation of cancer-stem-cell-associatedWnt target genespredicts poor prognosis in colorectal cancer patients. Cell Stem Cell2011;9:476–85.

17. Calon A, Espinet E, Palomo-Ponce S, Tauriello DV, Iglesias M,Cespedes MV, et al. Dependency of colorectal cancer on a TGF-beta-driven program in stromal cells for metastasis initiation. CancerCell 2012;22:571–84.

18. Kim YH, Lee HC, Kim SY, Yeom YI, Ryu KJ, Min BH, et al. Epigenomicanalysis of aberrantly methylated genes in colorectal cancer identifiesgenes commonly affected by epigenetic alterations. Ann Surg Oncol2011;18:2338–47.

19. Xiao W, Wang J, Li H, Guan W, Xia D, Yu G, et al. Fibulin-1 is down-regulated through promoter hypermethylation and suppresses renalcell carcinoma progression. J Urol 2013;190:291–30

20. Cheng YY, Jin H, Liu X, Siu JM, Wong YP, Ng EK, et al. Fibulin 1 isdownregulated through promoter hypermethylation in gastric cancer.Br J Cancer 2008;99:2083–7.

21. Hibi K, Mizukami H, Saito M, Kigawa G, Nemoto H, Sanada Y. FBN2methylation is detected in the serum of colorectal cancer patients withhepatic metastasis. Anticancer Res 2012;32:4371–4.

22. Tsunoda S, Smith E, De Young NJ, Wang X, Tian ZQ, Liu JF, et al.Methylation of CLDN6, FBN2, RBP1, RBP4, TFPI2, and TMEFF2 inesophageal squamous cell carcinoma. Oncol Rep 2009;21:1067–73.

23. Kim TO, Park J, Kang MJ, Lee SH, Jee SR, Ryu DY, et al. DNAhypermethylation of a selective gene panel as a risk marker for coloncancer in patients with ulcerative colitis. Int J Mol Med 2013;31:1255–61.

24. Dannenberg LO, Chen HJ, Tian H, Edenberg HJ. Differential regulationof the alcohol dehydrogenase 1B (ADH1B) and ADH1C genes by DNAmethylation and histone deacetylation. Alcohol Clin Exp Res2006;30:928–37.

25. Huang YW, Jansen RA, Fabbri E, Potter D, Liyanarachchi S, ChanMW,et al. Identification of candidate epigenetic biomarkers for ovariancancer detection. Oncol Rep 2009;22:853–61.

26. Mithani SK, Smith IM, Califano JA. Use of integrative epigenetic andcytogenetic analyses to identify novel tumor-suppressor genes inmalignant melanoma. Melanoma Res 2011;21:298–307.

27. Richards KL, Zhang B, SunM, DongW, Churchill J, Bachinski LL, et al.Methylation of the candidate biomarker TCF21 is very frequent acrossa spectrum of early-stage nonsmall cell lung cancers. Cancer2011;117:606–17.

28. Weiss D, Stockmann C, Schrodter K, Rudack C. Protein expressionand promoter methylation of the candidate biomarker TCF21 in headand neck squamous cell carcinoma. Cell Oncol 2013;36:213–24

29. Louie E, Chen XF, Coomes A, Ji K, Tsirka S, Chen EI. Neurotrophin-3modulates breast cancer cells and the microenvironment to promotethe growth of breast cancer brain metastasis. Oncogene 2013;32:4064–77.

30. Moravcikova E, Krepela E, Prochazka J, Rousalova I, Cermak J,Benkova K. Down-regulated expression of apoptosis-associatedgenes APIP and UACA in non-small cell lung carcinoma. Int J Oncol2012;40:2111–21.

31. Ford CE, Jary E, Ma SS, Nixdorf S, Heinzelmann-Schwarz VA, WardRL. The Wnt gatekeeper SFRP4 modulates EMT, cell migration anddownstream Wnt signalling in serous ovarian cancer cells. PLoS ONE2013;8:e54362.

32. Chang YT, Wu CC, Shyr YM, Chen TC, Hwang TL, Yeh TS, et al.Secretome-based identification of ULBP2 as a novel serummarker forpancreatic cancer detection. PLoS ONE 2011;6:e20029.

33. Fujita A, Shida A, Fujioka S, Kurihara H, Okamoto T, Yanaga K. Clinicalsignificance of Rho GDP dissociation inhibitor 2 in colorectal carci-noma. Int J Clin Oncol 2012;17:137–42.

34. Li X, Wang J, Zhang X, Zeng Y, Liang L, Ding Y. Overexpression ofRhoGDI2 correlates with tumor progression and poor prognosis incolorectal carcinoma. Ann Surg Oncol 2012;19:145–53.

35. ChoHJ, BaekKE, Kim IK, Park SM,Choi YL, Nam IK, et al. Proteomics-based strategy to delineate the molecular mechanisms of RhoGDI2-induced metastasis and drug resistance in gastric cancer. J ProteomeRes 2012;11:2355–64.

36. Berdiel-Acer M, Bohem ME, Lopez-Doriga A, Vidal A, Salazar R,Martinez-Iniesta M, et al. Hepatic carcinoma-associated fibroblastspromote an adaptative response in colorectal cancer cells that inhibitproliferation and apoptosis: nonresistant cells die by nonapoptotic celldeath. Neoplasia 2011;13:931–46.

37. Huijbers A, Tollenaar RA, GWVP, Zeestraten EC, Dutton S,McConkeyCC, et al. The proportion of tumor-stroma as a strong prognosticatorfor stage II and III colon cancer patients: validation in the VICTOR trial.Ann Oncol 2013;24:179–85

38. Chang HY, Chi JT, Dudoit S, Bondre C, van de RijnM, Botstein D, et al.Diversity, topographic differentiation, and positional memory in humanfibroblasts. Proc Natl Acad Sci U S A 2002;99:12877–82.

39. Finak G, Bertos N, Pepin F, Sadekova S, Souleimanova M, Zhao H,et al. Stromal gene expression predicts clinical outcome in breastcancer. Nat Med 2008;14:518–27.

40. Jia Z, Rahmatpanah FB, Chen X, Lernhardt W, Wang Y, Xia XQ, et al.Expression changes in the stroma of prostate cancer predict subse-quent relapse. PLoS ONE 2012;7:e41371.

41. Navab R, Strumpf D, Bandarchi B, Zhu CQ, Pintilie M, Ramnarine VR,et al.Prognostic gene-expression signature of carcinoma-associatedfibroblasts in non-small cell lung cancer. Proc Natl Acad Sci U S A2011;108:7160–5.

42. Gao Q, Wang XY, Qiu SJ, Zhou J, Shi YH, Zhang BH, et al. Tumorstroma reaction-related gene signature predicts clinical outcome inhuman hepatocellular carcinoma. Cancer Sci 2011;102:1522–31.

43. Illemann M, Hansen U, Nielsen HJ, Andreasen PA, Hoyer-Hansen G,Lund LR, et al. Leading-edge myofibroblasts in human colon cancerexpress plasminogen activator inhibitor-1. Am J Clin Pathol2004;122:256–65.

44. Sieuwerts AM, Martens JW, Dorssers LC, Klijn JG, Foekens JA.Differential effects of fibroblast growth factors on expression ofgenes of the plasminogen activator and insulin-like growth factorsystems by human breast fibroblasts. Thromb Haemost 2002;87:674–83.

45. PessonM, Volant A, Uguen A, Trillet K, De La Grange P, Aubry M, et al.A gene expression and pre-mRNA splicing signature that marks theadenoma-adenocarcinoma progression in colorectal cancer. PLoSONE 2014;9:e87761.

46. OkanoM, Yamamoto H, OhkumaH, Kano Y, KimH, NishikawaS, et al.Significance of INHBA expression in human colorectal cancer. OncolRep 2013;30:2903–8.

47. Kudo Y, Ogawa I, Kitajima S, Kitagawa M, Kawai H, Gaffney PM, et al.Periostin promotes invasion and anchorage-independent growth inthe metastatic process of head and neck cancer. Cancer Res2006;66:6928–35.

48. Fukushima N, Kikuchi Y, Nishiyama T, Kudo A, FukayamaM. Periostindeposition in the stroma of invasive and intraductal neoplasms of thepancreas. Mod Pathol 2008;21:1044–53.

49. Yan W, Shao R. Transduction of a mesenchyme-specific geneperiostin into 293T cells induces cell invasive activity throughepithelial-mesenchymal transformation. J Biol Chem 2006;281:19700–8.

50. DeSousaEMF,WangX, JansenM,Fessler E, TrinhA, deRooij LP, et al.Poor-prognosis colon cancer is defined by a molecularly distinctsubtype and develops from serrated precursor lesions. Nat Med2013;19:614–8.


Berdiel-Acer et al.




2014;12:1254-1266. Published OnlineFirst May 14, 2014.Mol Cancer Res Mireia Berdiel-Acer, Daniel Cuadras, Natalia Guillén Díaz-Maroto, et al. for Colorectal Cancer Initiation, Progression, and MetastasisA Monotonic and Prognostic Genomic Signature from Fibroblasts

Updated version

10.1158/1541-7786.MCR-14-0121doi:

Access the most recent version of this article at:

Material

Supplementary

http://mcr.aacrjournals.org/content/suppl/2014/05/14/1541-7786.MCR-14-0121.DC1

Access the most recent supplemental material at:

Cited articles

http://mcr.aacrjournals.org/content/12/9/1254.full#ref-list-1

This article cites 50 articles, 12 of which you can access for free at:

Citing articles

http://mcr.aacrjournals.org/content/12/9/1254.full#related-urls

This article has been cited by 1 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://mcr.aacrjournals.org/content/12/9/1254To request permission to re-use all or part of this article, use this link



http://mcr.aacrjournals.org/lookup/doi/10.1158/1541-7786.MCR-14-0121

http://mcr.aacrjournals.org/content/suppl/2014/05/14/1541-7786.MCR-14-0121.DC1

http://mcr.aacrjournals.org/content/12/9/1254.full#ref-list-1

http://mcr.aacrjournals.org/content/12/9/1254.full#related-urls

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

mailto:[email protected]

http://mcr.aacrjournals.org/content/12/9/1254


A Monotonic and Prognostic Genomic Signature from Fibroblasts … · Genomics A Monotonic and...

Documents

Transcript of A Monotonic and Prognostic Genomic Signature from Fibroblasts … · Genomics A Monotonic and...