Stromal-Based Signatures for the Classi cation of Gastric Cancer - … · 2021. 2. 3. · Microenvironment and Immunology Stromal-Based Signatures for the Classiﬁcation of Gastric

Microenvironment and Immunology

Stromal-Based Signatures for the Classificationof Gastric CancerMark T. Uhlik1, Jiangang Liu1, Beverly L. Falcon1, Seema Iyer2, Julie Stewart1,Hilal Celikkaya2, Marguerita O'Mahony2, Christopher Sevinsky3,Christina Lowes3, Larry Douglass4, Cynthia Jeffries1, Diane Bodenmiller1,Sudhakar Chintharlapalli1, Anthony Fischl1, Damien Gerald2, Qi Xue2,Jee-yun Lee5, Alberto Santamaria-Pang3, Yousef Al-Kofahi3, Yunxia Sui3,Keyur Desai3, Thompson Doman1, Amit Aggarwal1, Julia H. Carter4,Bronislaw Pytowski2, Shou-ching Jaminet6, Fiona Ginty3, Aejaz Nasir1,Janice A. Nagy6, Harold F. Dvorak6, and Laura E. Benjamin2

Abstract

Treatment of metastatic gastric cancer typically involves che-motherapy and monoclonal antibodies targeting HER2 (ERBB2)and VEGFR2 (KDR). However, reliable methods to identifypatients who would benefit most from a combination of treat-ment modalities targeting the tumor stroma, including newimmunotherapy approaches, are still lacking. Therefore, we inte-grated a mouse model of stromal activation and gastric cancergenomic information to identify gene expression signatures thatmay inform treatment strategies. We generated a mouse model inwhich VEGF-A is expressed via adenovirus, enabling a stromalresponse marked by immune infiltration and angiogenesis at theinjection site, and identified distinct stromal gene expressionsignatures. With these data, we designed multiplexed IHC assays

that were applied to human primary gastric tumors and classifiedeach tumor to a dominant stromal phenotype representative ofthe vascular and immunediversity found in gastric cancer.We alsorefined the stromal gene signatures and explored their relation tothe dominant patient phenotypes identified by recent large-scalestudies of gastric cancer genomics (The Cancer Genome Atlas andAsian Cancer Research Group), revealing four distinct stromalphenotypes. Collectively, these findings suggest that a genomics-based systems approach focused on the tumor stroma can be usedto discover putative predictive biomarkers of treatment response,especially to antiangiogenesis agents and immunotherapy,thus offering an opportunity to improve patient stratification.Cancer Res; 76(9); 2573–86. �2016 AACR.

IntroductionGastric cancer is the second leading cause of cancer-related

death worldwide, with the highest prevalence in Asia (1, 2).Therapy for metastatic gastric cancer includes a combination ofchemotherapies and targeted therapies for Her2 (ERBB2) andVEGFR2 (KDR) as monoclonal antibodies. The prevalence ofHer2 overexpression is only �15% to 20%; thus, the majority

of patients are dependent on chemotherapy and antiangiogenesis.Although cancer immunotherapy is in the early phases ofdevelopment for gastric cancer, there are trials under way toexplore this modality both as monotherapy and in combinationwith the anti-VEGFR2 antibody ramucirumab (www.clinical-trials.gov and refs. 3 and 4). Both stromal therapies, antiangio-genesis and immunotherapy, are in need of reliable means toidentify those patients that will receive optimal therapeutic ben-efit, and to spare unnecessary side effects for those where thechance of clinical benefit is low.

One of the goals of the human genome project is to char-acterize the genetic and epigenetic variation of cancers,enabling their use to generate new therapeutic hypotheses. Ingastric cancer, there have been several large efforts publishedin recent years. The two largest come from The Cancer GenomeAtlas (TCGA; ref. 5) and Asian Cancer Research Group (ACRG;ref. 6). Both have published RNA profile analyses thatdivide patients into 4 subgroups. In the present study, wegenerated stromal-specific RNA and IHC-based signaturesthat represent different stages of stromal activation incancer. We then interrogated how this stromal approachcompares with the prior tumor-centric approaches to cometo a theoretical classification system to separate whichpatients might benefit most from antiangiogenesis and cancerimmunotherapy.

1Lilly Research Laboratories, Eli Lilly and Company, Indianapolis, Indi-ana. 2Lilly Research Laboratories, Eli Lilly and Company, New York,New York. 3General Electric Global Research Center, Niskayuna, NewYork. 4Department of Pathology,Wood Hudson Medical Center, Cov-ington, Kentucky. 5Department of Hematology-Oncology, SamsungMedical Center, Seoul, Seoul Korea. 6Department of Pathology andCenter for Vascular Biology Research, Beth Israel Deaconess MedicalCenter, Harvard Medical School, Boston, Massachusetts.

Note: Supplementary data for this article are available at Cancer ResearchOnline (http://cancerres.aacrjournals.org/).

M.T. Uhlik and J. Liu contributed equally to this article.

Corresponding Author: Laura E. Benjamin, Eli Lilly and Company, 450 East 29thStreet, 11th Floor, New York, NY 10016. Phone: 908-400-1544; E-mail:[email protected]

doi: 10.1158/0008-5472.CAN-16-0022

�2016 American Association for Cancer Research.

CancerResearch

www.aacrjournals.org 2573

on November 5, 2019. © 2016 American Association for Cancer Research. cancerres.aacrjournals.org Downloaded from

http://cancerres.aacrjournals.org/

Materials and MethodsGastric patient-derived xenograft models

Studies were conducted by Crown Bio, Shanghai, China. Sub-cutaneous primary human gastric models (HuPrime) were estab-lished and passaged in mice before use in efficacy studies asfollows: small 2 to 4 mm in diameter fresh tumor fragments wereimplanted subcutaneously into BALBC/nude mice (6–8 weeksof age). When tumors reached an average volume of �150 mm3

(132–193mm3), the animals were randomized by tumor volumeinto treatment groups (n¼ 7–8). For all statistical evaluations, thelevel of significance was set at P < 0.05.

Ad-VEGF-A flank gene array study design and bioinformaticsanalyses

The Ad-VEGF-A164flank model was performed as described in

refs. 7 and 8. Samples for messenger RNA (mRNA) profilingstudies were processed by Asuragen, Inc. using GeneChip MouseGenome 430 2.0 Array (Affymetrix) as described previously (9). Asummary of the image signal data, detection calls, and geneannotations for every gene interrogated on the arrays was gener-ated using the Affymetrix Statistical Algorithm MAS 5.0 (GCOSv1.3) algorithm (scaling factor ¼ 1,500) and quantile normali-zation across samples was applied. Log2 transformation andmean-centered standardization was performed. Accession num-ber in GEO is GSE76630.

Multiplexed IHCMultiplexed fluorescent IHC was performed as previously

described (10). Antibodies used were PECAM-1 (CD31) (AbCamab28364 at 1:50) or (Invitrogen A11006 at 1:100); aquaporin-1(AQP1) (Proteintech 20333-1-AP at 1:100 or 1:250); von Will-ebrand factor (vWF) (AbCam ab115771 at 1:50) or (MilliporeMAB3442 at 1:250); and CD34 (Biolegend 119302 at 1:100) or(Novocastra clone QBEnd/10 NCL-L-END at 1:250); Smoothmuscle actin (SMA; Sigma C6198 at 1:400); GLUT1 (Millipore07-1401 at 1:500).

GE multiomyx technologyMultiplexed tissue imaging analysis of gastric tissue microar-

rays (TMA) was performed using the GE Multiomyx platformessentially as described in 11 (see Supplementary Methods formore elaborate descriptions of these methods). Briefly, fourimmunostains contributed to blood vessel area detection: CD31,CD34, vWF, and AQP1. Five clusters of vessel phenotypes wereobtained: Early vessels were [CD34þ/AQP�/vWF] and [CD34þ/AQP1þ/vWF�], and Late vessels were [CD34þ/AQP1þ/vWFþ],[CD34�AQP1þvWFþ], and [CD34�, AQP1�/vWFþ]). Forimmune cell detection, CD3 (for T-cell detection) and CD163(formacrophage detection) were stained. Four phenotypic classesof gastric tumors from the TMAs were derived according tothreshold values for the percentage of early and late vessels (usinga value of 55% of late vessels as cutoff, where values above are"late" and values below are "early"), percentage of macrophagearea (using a value of 28% as a cutoff), and percentage of T-cellarea (using a value of 3.4% as a cutoff).

Human gastric cancer TMAArchived and de-identified FFPE tissues were provided by the

Wood Hudson Cancer Research under a HIPAA waiver. The TMAwas created from archived FFPE specimens. Clinical follow-up

informationwasobtained fromexistingmedical records inTumorRegistries. Archived surgical specimens were used to identify1-mm cores from Donor Blocks for embedding into a RecipientBlock of paraffin with receptacle holes prepared in a grid patternusing a Pathology Devices TMArrayer.

Imaging and image analysisTissues were imaged on a Marianas 575 system (Intelligent

Imaging Innovations). DAB-stained flanks were reviewed andclassified by pathologists using the classification established byDr. Harold Dvorak's group (12). Quantifications of the vascularmarker phenotypeswereperformedwith an iCys research imagingcytometer (Thor labs; ref. 10). Triplicate gastric tumor TMAs and acontrol TMA with normal and tumor tissues were stained andquantified using two panels. The first panel included CD31,CD34, and AQP1 and the second had CD31, vWF, and AQP1.The control TMA was used to standardize the gating and classi-fication from one stain run to the next. Any sample that had atleast 2 cores and a consistent diagnosis (intestinal versus diffuse)were used. To classify the cores, a multistep analysis was used.From the first panel, the percentage of CD31 phantoms that wereCD34þAQP�was classified as outcome 1. From the second panel,the percentage of CD31 that was AQPþvWF� was outcome 2,percentage of CD31 that was AQPþvWFþ was outcome 3, andfinally the percentage of CD31 that was AQP�vWFþwas outcome4. These four outcomes were normalized to 100% to reflect howmany of the total vessels fell into each outcome. The percentage ofvessels that fell into each of these outcomes was used for thevascular marker phenotyping to simplify the results and align thereadouts with the bioinformatics analysis. Greater than 15% ofoutcome 4 ¼ "mature." Less than 16% of outcome 1 or >50% ofoutcome 1 þ 2 ¼ "immature." Finally, all remaining cores wereclassified in "intermediate."

Molecular classification of stroma in gastric cancerThe distribution of the eight possible phenotypes within the

ACRG cohort was as follows (Signature 1/Signature 2/Signature3): �/�/�, 32.5%; �/�/þ, 6.02%; �/þ/�, 0.80%, �/þ/þ,13.65%;þ/�/�, 19.28%;þ/�/þ, 3.21%;þ/þ/�, 3.61%;þ/þ/þ,20.88%. Because four dominant phenotypes were recognized, thefour smaller subsets were distributed into the dominant classes asfollows: �/�/þ, �/þ/�, and þ/þ/� grouped with �/þ/þ;þ/�/þ grouped with þ/�/�. See Supplementary Materials andMethods for additional details.

Reverse-phase protein array dataReverse-phase protein array (RPPA) data from the TCGA cohort

were mined from Supplementary Data in ref. 5. TCGA patientswere classified into four stromal phenotypes as described, andmean protein expression values were reported for each group.Correlation analyses of analyte expression between stromal phe-notypes and molecular classifications were performed usingSpearman nonparametric analysis in JMP.

ResultsGastric cancer patient-derived xenograft tumor models exhibitstromal heterogeneity

As in humans, mouse models also display a variety ofresponses to antiangiogenesis. We tested 17 patient-derivedxenograft (PDX) gastric cancer models and found that they

Uhlik et al.

Cancer Res; 76(9) May 1, 2016 Cancer Research2574



varied from very responsive to unresponsive to anti-VEGFR2(DC101). A subset of these models is shown in Fig. 1A. Thedifferential response was not obviously due to differences intumor cell doubling time, histology or differentiation, but wedid observe that stromal patterns correlated to some degree withresponse. The two examples shown in Fig. 1B represent extremes:the top panel illustrates a responsive tumor with numerous, largeblood vessels with little hypoxia (as measured by Glut1 expres-sion); the bottompanel depicts a nonresponsive tumorwith smallcaliber vessels and pronounced hypoxia. We hypothesized thatstromal phenotypes might be used to differentiate gastric cancersfor susceptibility to stromal targeted therapies; i.e., blood vesselphenotypes for potential antiangiogenesis therapy and immuneinfiltrates for immunotherapy.

An adenoviral vector expressing VEGF can be used to modeltumor blood vessels and some other stromal elements

We started with a mouse model that drives a stromalresponse with an adenovirus expressing VEGF-A. This modelis not only valuable for its ability to generate time-dependentvascular differentiation, but also allows the study of stromalremodeling associated with immune cell infiltration, ECM, andfibrotic tissue remodeling reminiscent of wound healing, whichhas been so often compared to cancer (13). The absence oftumor cells enables an enriched stromal environment to gen-erate RNA signatures, which we then evaluated in humansamples of gastric cancer.

In this model, viral infection and VEGF-A production istransient and largely localized to the initial site infiltration ofimmune cells and initiation of an immature vascular network.The large fragile blood vessels in this stage have been termedmother vessels (MV). At intermediate stages (days 5–20), the

tissue is enriched by expansion of the vasculature and evolutionof MVs into glomeruloid microvascular proliferations (GMP)and capillaries. Also at this stage, lymphatic vessels begin toform, and the immune infiltrate is cleared. Later stages (beyondday 20) are marked by a resolution/maturation/stabilization ofvessels characterized by the heavy investment of pericytes,vascular malformations (VM), and remodeled extracellularmatrix (12, 14, 15). This model was previously used to dem-onstrate that early-stage and intermediate-stage vessel subsets(MVs, GMPs, and capillaries) were sensitive to aflibercept(VEGF/PLGF Ligand trap) therapy, whereas more mature sub-sets of vessels (VMs, feeding arteries, and draining veins) wereinsensitive (Supplementary Fig. S1; refs. 15, 16).

We performed a gene array analysis using flank samplesfrom days 0, 5, 20, and 60. Hundreds of differentially expressedgenes were identified between each of the time points (Fig. 2A),indicating the dynamic nature of this model throughout theentire time course. Hypergeometric analysis of the gene arraydataset indicates that there is an enrichment of genes associatedwith a variety of pathological and disease states, includingcancer and cardiovascular disease (Supplementary Fig. S2A;ref. 17) and a dynamic regulation of genes associated withVEGF-A (Supplementary Fig. S2B). In addition to endothelialgenes, there were differentially expressed genes involved inmultiple cell types of stroma. Temporally speaking, the genesthat were highly expressed early are involved in proliferation(Ki67), inflammation (Il-6, Pik3cg, granulocyte marker Tlr1,myeloid cell marker Cd68), and remodeling of matrix (Timp1and Tnc; Fig. 2B). Pan-endothelial expressed genes such aspecam1 (cd31), VE-Cadherin (Cdh5), and Vegfr2 (Flk1) showedpeak activation in the middle (days 5–20) time points. Pericytemarkers Pdgfrb, Rgs5, and Acta2 (Smooth Muscle Actin/SMA)

Figure 1.Gastric PDX response to anti-VEGFR2antibody (DC101). A, seven gastricPDX models treated with 40 mg/kgDC101 three times per week for 3weeks starting treatment at an initialtumor volume of �150 mm3 (n ¼ 7–8tumors per treatment per model).%T/C indicates a ratio of treated:control tumor volumeat last dayof thestudy. Doubling time was 5 to 7 daysfor the models shown. B, fluorescentimaging of cell nuclei (DAPI), bloodvessels (CD34) myofibroblasts/pericytes (smoothmuscle actin, SMA),hypoxic gene expression (GLUT1)reveals highly variable stromalarchitecture.

Stratification Hypothesis for Stromal-Targeted Therapies

www.aacrjournals.org Cancer Res; 76(9) May 1, 2016 2575



had peak later than the endothelial cell genes consistent withpericyte coverage following endothelial tube formation anddemarking the remodeling process that stabilizes vessels. TEK,a gene enriched in maturing blood vessels, is more highlyexpressed at later time points, while CD36, an immature bloodvessel marker, has higher expression at day 5. Genes involvedin lymphangiogenesis, such as VEGFC, have high expressionbetween days 5 and 20.

Bioinformatics analysis of differentially expressed genes fromtheAd-VEGF-A164model reveals a signature that definesdistinctstages of stromal remodeling

We used multiple sources of gene lists (see Materials andMethods) to select down to a smaller set of genes restricted tothose involved in tumor stroma: pathologic angiogenesis,wound-healing response, and tumor microenvironment bio-logy. This resulted in a pool of 1,549 genes that were used asseeds for additional bioinformatics analysis using ClaNC andPAMR algorithms to generate a 108 gene signature of the mostdifferentially expressed genes (Fig. 3A and Supplementary

Table S1), which segregated themselves into three stromalsignatures that reflected biological processes such as signature1 (immune cell activity), 2 (angiogenesis), and 3 (tissueremodeling) based on gene ontology (GO) classifications.

We applied this 108 gene set to a publicly available genearray data set from Avastin (anti-VEGF-A)-treated H1975NSCLC xenografts (Fig. 3B; ref. 18). Vehicle-treated sampleswere highly enriched in genes from signatures 1 and 2, whilesignature 3 genes were underrepresented. Conversely, Avastin-treated samples had a significant decrease of the signature 1 and2 genes with a concomitant enrichment of signature 3 genescompared with vehicle samples. This analysis reveals the selec-tive loss of less mature stroma (represented by signatures 1and 2) upon treatment with anti-VEGF, consistent with priorpreclinical and clinical studies (18–21). Avastin only blockshuman VEGF-A in this murine model, so we also explored ananti-mouse-VEGFA antibody (G6) in our Ad-VEGFA flankmodel. We observed similar results in that signatures 1 and2 were robustly diminished, while signature 3 remainedunchanged (Fig. 3C and Supplementary Fig. S3). An antibody

Figure 2.RNA expression profiles of genesillustrate the temporal shift in biologicprocesses in the model. A, the tableshows pairwise comparisons of thenumbers of differentially expressedgenes (DEG) between different timepoints. Criteria used for "total" DEGsare: probe signal called as "Present" or"Marginal" on one or more gene chips,FDR < 0.2, at least one probe havingmean raw signal greater than 250, andabsolute fold change greater than 1.5compared with day 0 control. B, rawsignal values are listed. Genes withhigh expression at day 5 includemarkers of proliferation, inflammation,and matrix remodeling. Angiogenesisgenes, such as blood vessel andpericytemarkers, were highest at days5 to 20.

Uhlik et al.




to VEGFR2 (DC101) in the same study was less robust atreducing signatures 1 and 2 but did significantly reduce (asopposed to increase) signature 3. These data provide evidencethat there are some biologic differences in targeting VEGFversus targeting VEGFR2, but whether these are significant

enough to translate to clinical differences remains unknown.Finally, we asked whether high signature 2 could be predictivefor DC101 efficacy, and in 7 gastric PDX models, the highestsignature 2 pulled out the only strong responder in that cohort(Fig. 3D).

Figure 3.Gene expression signatures revealdistinct phases of stromaldevelopment with differentialsensitivity to antiangiogenic agents.A, distinct classes of stromal geneexpression are visualized in the heatmap and boxplots of gene activationscores for study time points. Heatmap: red, high expression; blue, lowexpression. The identity of genesacross the x-axis is in the order listed inSupplementary Table S1. Highlyexpressed genes were defined asclusters representing phases ofstromal development (day 5,signature 1; day 20, signature 2; day60, signature 3; n¼ 5 samples per timepoint). Gene activation scores for eachtime point were calculated for eachgene signature and plotted with meanvalues shown. Refer to Materials andMethods for additional details. B, geneexpression from control (CON)- andbevacizumab (BEV)-treated H1975tumors (n ¼ 3 samples per treatmentgroup) was plotted as a heat map A.Gene activation scores for eachstromal signaturewere calculated andplotted for control and bevacizumab-treated tumors. C, gene activationscores for signatures 1–3 are shown asboxplots for nontreated (NT), DC101-,and G6-treated flank samples (n ¼ 5samples per treatment per group).Group means and P values ofcomparisons are shown in eachboxplot. Signature 1 scores are shownfor day 5 treatment time points,signature 2 scores are shown for day20 time points, and signature 3 scoresare shown for day 60 time points. D,mouse RNAseq of PDX models testedwith DC101 (Fig. 1).





Stage-specific vascular markers can distinguish betweenpathologic vessel subtypes

To develop IHC assays, we focused on four genes whose peakgene expression ranged from early (CD34), middle (AQP1),and late (vWF) time points. We optimized a multiplexedfluorescent two-panel angiophenotyping assay (10). CD31stained all vessel structures from each time point, consistentwith its role as a pan-endothelial marker (22). CD34 primarilystained early and intermediate vessels (seen at days 5 and 20),with very dim staining of later-stage vessels (represented by day60). AQP1 does not stain early-stage vessels, but strongly stainsintermediate vessels. vWF expression is primarily found in onlyintermediate- and late-stage vessels (Fig. 4A and B). Compar-ison of the multiplexed imaging analysis with previouslydescribed blood vessel subtypes (12) showed that these fourmarkers can be used to map to histologically defined subsets(Fig. 4B). The earliest vessel types that are formed in angio-genesis are MVs and GMPs, which only highly express CD34. Asvessels mature into VMs, they begin expressing AQP1. Withadditional maturation (and pericyte recruitment) the VMs andcapillaries also express vWF and start losing CD34 expression.The most mature form of vessels, the AVMs (including feedingarteries and draining veins) exclusively express vWF and do nothave CD34 or AQP1 expression (Fig. 4B, note arrows). There-fore, these endothelial markers can molecularly profile subsetsof vessels that have up to this point only been characterized bytheir morphology.

Multiplexed IHC assays are able to classify gastric tumors intodistinct stromal phenotypic groups

Eighty-five human primary gastric tumors were profiled inTMA format for vessel phenotypes using the previouslydescribed IHC assay (Fig. 5). Using this approach, we were ableto classify each of these tumors into their dominant stromalphenotype using vascular markers. Interestingly, we observeddifferences in the frequency of stromal/vascular phenotypesbetween the two major histologies in gastric cancer, Lauren's"diffuse" and "intestinal" (Fig. 5A) (21). About 50% of intes-tinal-type gastric tumors contained predominantly immaturestroma, as marked by high proportions of poorly lumenizedCD34-postive vessels with little to no vWF staining (Fig. 5B,left). Conversely, an overwhelming majority of diffuse-typetumors (�70%) contained predominantly intermediate stroma,indicated by the high proportions of their vessels stainingpositive for AQP-1 and with few CD34-single-positive (imma-ture) or vWF-single-positive (mature) vessels (Fig. 5B, right).

Because we had also observed a dynamic pattern of inflamma-tory cell infiltration from the bioinformatics analysis of theAd-VEGF-A164model and immune cells are likely to be importantfor emerging stromal therapies, we expanded our IHC method-ology to include an immune cell multiplexed panel also. UsingGE's Multiomyx platform (11) to analyze 20 vascular andimmune cell biomarkers on the same gastric TMA samples, wewere able to describe both the vascular and immune cell status ofthe samegastric cancer patient samples (Fig. 5C). The immune cellpanel included T-lymphocytes (CD3, CD8) andM2macrophages(CD163). Using this approach, we were able to classify fourprimary stromal phenotypes in gastric cancer, which we describeas vascular immature/noninflammatory (VINI), inflammatory(I), vascular mature/inflammatory (VMI), and vascular mature(VM). VINI tumors were characterized as having rudimentary

vessels found in immature stroma, but lacking the significantlymphocyte infiltration that also characterized the early infiltratethat we observed in the day 5 Ad-VEGF model. In contrast, tu-mors with immature angiogenic markers and with high levelsof lymphocytes were termed "Inflammatory." These vascularbeds could be characterized as having nonproductive earlyangiogenesis akin to what is seen following Notch inhibition orextreme hypoxia (23). Intermediate and Late vascular markerssignified the VM tumors, which could also be observed to haveeither immune infiltrate or not, so we called those withoutimmune infiltrate "VM" and those with a strong immune com-ponent "VMI." We hypothesized that these four stromal pheno-types represent fundamentally different biologic states thatmight have differential survival outcomes and/or implicationson therapeutic efficacy.

Gene signature patterns derived from the flank model definefour distinct stromal-based subsets of human gastric cancer

To further address our stromal phenotype hypothesis, weanalyzed a publicly available gene array dataset, TCGA (5), from249 patients with gastric cancer with our stromal signature (Sup-plementary Fig. S4). We observe that a majority of the diffusegastric tumors clustered together and show high expression ofsignature 2 and 3 genes, consistent with the IHC findings reportedin Fig. 5B.

The TCGA consortium devised a subclassification of gastriccancers into four distinct groups (C1–4) based upon a 40 genetumor-enriched RNA signature. A cross-comparison of ourstromal signature with these did not reveal any overlappinggenes. Using the "C1–4" designation, we clustered these patientsamples with our stromal signature (Supplementary Fig. S5A).Interestingly, as observed in the Lauren diffuse histology, theC1 tumors appear to have the highest expression of stromalsignatures 2 and 3. This group of patients was defined asgenomically stable with mRNA expression profiles characteris-tic of tumor cells undergoing epithelial-to-mesenchymal tran-sition (EMT).

Application of our signature to another independent gastriccancer dataset from the ACRG (6) revealed a high expression ofthe signature 2 and 3 genes in the group of patients characterizedas microsatellite stable MSS/EMT (Fig. 6A). 100% of TCGA C1subtype and 95%of ACRGMSS/EMT tumorswere predominantlyenriched for signature 2 (Fig. 6B). A third independent gastriccancer cohort (24) also showed a very high correlation of signa-ture 2 genes to their "Mesenchymal" molecular phenotype (Sup-plementary Fig. S5B and S5C). However, tumors from non-C1TCGA and non-MSS/EMT ACRG molecular classification groupsassociated strongly with stromal signatures 1 and 3, albeit withheterogeneous distributions. This was also apparent from theheatmaps (Fig. 6A and Supplementary Fig. S5A), where somepatients show high or low expression of genes from all clusterswithin the stromal signature or preferentially show gene expres-sion in two of the three clusters. Therefore, using this forcedclustering approach some patients will be placed into a particularcluster despite having an overall low gene activation score for thatcluster.

The heterogeneity of signatures in each patient suggested thatwe needed to move away from forced clustering into a gene-by-gene binary segmentation of gene activation scores approach todescribe a patient's stromal biology. We set a value of zero as themedian expression score, such that scores above zero would be

Uhlik et al.




scored "þ" and scores below zero would be scored "�"for sig-natures 1, 2, and 3, respectively. Whenwe did this, four dominantcombinatorial profiles were observed in both the TCGA and

ACRG datasets, �/�/�, þ/�/�, �/þ/þ, and þ/þ/þ, whichcompared favorably with the four phenotypes observed byIHC (Fig. 6C). These four profiles accounted for 88% of tumors

Figure 4.Vascular markers from distinct stromal states identify specific subsets of blood vessels in the Ad-VEGF-A164 model. A, vascular expression is marked by "þ"or "�" with a schematic of relative expression on the graph. Blood vessel phenotypes were histologically defined as in ref. 12. Cap, capillaries; AVM,arteriovenous malformations; FA, feeding arteries; DV, draining veins. B, multiplexed fluorescence staining with phase-specific vascular markers revealsdynamic expression patterns over time. Arrows, vessels stained with vWF that are negative for CD34 and AQP1. Scale bar, 100 mm.





Uhlik et al.




in the TCGA set and 86% of tumors in the ACRG set. The tumorswithin the other minor profiles were combined with the fourdominant profiles for future analyses (see Materials and Meth-ods). A comparison of the signature 1 genes confirms that thesegenes are highly correlated with inflammatory genes expressed inthe TCGA and ACRG cohorts (Supplementary Figs. S6 and S7).Conversely, both signatures 2 and 3 correlatedwith "Endothelial"cell gene expression. Therefore, tumors with signature 1 scoresabove zero were expected to have an inflammatory phenotype(Fig. 6C). Similarly, tumors with signature 2 scores above zerowere termed VM and are considered fundamentally associatedwith angiogenesis. Positive scores in both signatures 1 and 2 weregrouped as VMI and negative scores in bothwere grouped as VINI.In the end, signature 3 did not reveal robust differentiation in thegastric cancer samples.

Next, tumors from the ACRG and TCGA cohorts were clas-sified by these four stromal phenotypes, and the subsequentdistributions within each molecular subtype were calculated(Fig. 6D). Consistent with the earlier observation that TCGAC1 and ACRG MSS/EMT subtypes had enrichment for stromalsignature 2 genes, almost all tumors from these groups had aVM gene expression phenotype. However, a large percentage ofthese tumors (52% in TCGA C1 and 65% in ACRG MSS/EMT)also had expression of inflammatory genes, making them VMI.The TCGA C2 subtype was composed largely of tumors withinflammatory stromal phenotype (39%) with another signif-icant portion of tumors having a VMI phenotype (30%). TheACRG microsatellite instability (MSI) group had a similarproportion of inflammatory stromal phenotypes (38%) butfewer VMI tumors (13%). Both TCGA C2 and ACRG MSIclasses had very few purely VM stromal gene expression profiles(6% and 8%, respectively). Therefore, the TCGA C1 and ACRGMSS/EMT tumors can be classified as dominantly VM, whilethe TCGA C2 and ACRG MSI are dominantly inflammatory.The other two classes from the TCGA and ACRG cohortshad varied distributions among the four stromal phenotypes.Supplementary Table S2 describes relationships within ourstromal subtypes for important genetic features, such as MSI,EBV, Her2, and others.

Activation scores for multiple gene sets previously describedin the literature and associated with unique biology, such asstroma, inflammation, regulatory T cell (Treg) biology, werecalculated for each patient's tumor from the ACRG and TCGAcohorts and were subjected to pairwise correlation analysis(Fig. 6E and Supplementary Figs. S6 and S7). The gene activa-tion scores for various gene groups were consistent with theirstromal designations. For example, inflammatory tumors wereenriched in immune cell genes, including macrophages andT cells, whereas VM tumors were highly enriched in angiogen-esis genes. Additionally, tumors with a combined VMI pheno-type tended to have significantly higher expression of genes

associated with downmodulation of the immune response(Fig. 6E and F). These include genes such as HAVCR2 (TIM3ref. 25), IL10 (26, 27), and TGFb1 (26, 28) that are associatedwith Tregs and suppression of Th1 immune responses andVCAM1 (29), CD163 (30), and ITGAM (CD11b, ref. 31) thatare associated with M2 macrophage infiltration. Conversely,tumor with an inflammatory phenotype have significantlyhigher expression of immunostimulatory genes (TNF, IFNG,and ICAM1; refs. 26, 29, 31) and markers of M1 macrophageactivation (CD274, IDO1, and LAG3; refs. 26, 32). This suggeststhat the immune cell components within the Inflammatoryand VMI phenotypic groups represent distinct biologic activa-tion states and are likely to influence their respective tumorgrowth in disparate ways.

Stromal-based phenotypes differentiate gastric tumors byprotein expression profiles and patient survival in TCGA andACRG cohorts

We hypothesized that gastric tumors with similar stromalphenotypes would also share similarities in protein expressionand cellular signaling, and ultimately relate to overall patientsurvival. To explore these ideas, we first mined the RPPAdata from the TCGA cohort (5) to examine protein expressionprofiles within the different stromal phenotypic classes. Ofthe 191 analytes examined (5), 43 were highly differentiallyexpressed across the four stromal phenotypic classes (Fig. 7A).Plots of the mean relative expression value for these analytesshow two distinct patterns of expression, with the VM andVMI classes having very similar trend lines. The inflammatoryand VINI also share a common trend line with each other, butin an opposite direction to the VM and VMI expression profiles.A closer look at some of these analytes reveals striking differ-ences in expression of proteins involved in angiogenesis(VEGFR2, ACVRL1, Rictor; refs. 33, 34) and inflammatoryprocesses (Syk, CD49b, PAI1; refs. 35–37; Supplementary Fig.S8A). Analysis of the entire 191 RPPA set shows that the overallprotein expression profiles within the stromal phenotypicgroups correlate to the same molecular classes, as was demon-strated with the gene expression profiles (SupplementaryFig. S8B). This suggests that gastric tumors defined by differentgene expression stromal phenotypes also have differentialprotein expression patterns and, by extension, a different tumorbiology.

Previously, survival analysis of the ACRG cohort was per-formed with a patient follow-up time up to almost 9 years, witha minimum of 53 months of census data (6). A new Kaplan–Meier analysis was performed with updated data that include aminimum of 84 months of census data for each patient. Thesame trends that were previously reported still held true foreach of the four ACRG subclasses in this updated data (Sup-plementary Fig. S9). As observed with the RPPA data, two main

Figure 5.Human gastric cancers can be classified into distinct stromal phenotypic groups on the basis of phase-specificmarkers. A, tissue cores (n¼ 85 patient tumor cores intriplicate) were separated into the Lauren histological classes, diffuse and intestinal, and their dominant vascular expression profiles were determined. B,representative images of an intestinal tumorwith immature vascular phenotype and a diffuse tumorwithmature vessels. The intestinal tumor (left) has a vasculaturemarked by vessels expressing CD34 and AQP1, but not vWF. The diffuse gastric tumor has many vWF-positive vessels. Comp, composite image. Scale bar, 100 mm.C, Multiomyx staining of the human gastric tumor TMA was quantitated. Images shown are representative examples of tumors with the four defined stromalphenotypes. The table represents the averages across the 85 patient TMA. Immune cells are reported as a percentage of total cell area (macrophages) or total cellnumber (T cells). Scale bars, 100 mm.





Figure 6.Human gastric tumors classified into distinct stromal phenotypes by their gene expression profiles. A, the stromal signature was applied to treatment-na€�vegastric tumors from ACRG (N ¼ 249 tumor samples; ref. 6). (Continued on the following page.)

Uhlik et al.




stromal phenotypic classes appear to exist: VM and vascularimmature. Kaplan–Meier analysis of the ACRG cohort subdi-vided into angiogenic and nonangiogenic groups reveals arobust difference in survival (Fig. 7B). Therefore, the stromalstatus of gastric tumors has prognostic implications, such thatpatients with VM tumors fare more poorly than those withVascular Immature tumors. Although the differences betweeneach subgroup are smaller when using all four stromal phe-notypes (Fig. 7C), there is still statistical significance and someinteresting implications. Similar to the ACRG MSS/EMT andMSI classes, the VM group has the poorest overall survival while

the I group has the best survival. The VINI and VMI groups haveintermediate survivals, with the VMI trending closer to the VMgroup.

Given the available treatment options for gastric cancer andthe increasing interest in immunotherapy, differentiation of theinflammatory and VMI phenotypes in terms of Th1/2, dendriticcell activation and Tregs may help determine where combina-tions of antiangiogenesis and checkpoint inhibitor blockadecould be best tested with experimental agents. Because thesetumor samples were from treatment-na€�ve patients with mixedgrades and histology, there is the greatest opportunity for these

Figure 7.Stromal-based phenotypes differ in protein expression profiles and overall patient survival in TCGA and ACRG gastric cancer cohorts. A, RPPA datafrom TCGA cohort patients (5) for the top 43 differentially expressed proteins. In bold are trend lines for expression with analytes ordered lowest tohighest in the VM group. B, Kaplan–Meier plots of overall survival from the ACRG cohort are plotted for each of tumors segmented into VMI andVM and VINI and I. C, patients with VM phenotype tumors have the poorest overall survival, while those with I phenotypes have the best survival.Kaplan–Meier plots of overall survival from the ACRG cohort. D, statistics for B and C. E, integration of tumor and stromal phenotypes by potentiallyactionable subgroups using all 598 patients from both TCGA and ACRG.

(Continued.) ACRG classification of tumors is shown: MSS, microsatellite stable; MSI, microsatellite instable. B, schematic shows relationship of gastrictumors from the TCGA and ACRG cohorts with respect to the stromal signatures. C, gene activation scores from three stromal signatures were used toclassify tumors into four major stromal phenotypes. D, pie charts show the distribution of tumors from each of the ACRG and TCGA classes that have thestromal phenotypes shown. E, tumors with I and VMI phenotypes are highly enriched for immune cell gene expression. Boxplots of ACRG-based geneactivation scores for each of the stromal phenotypes are shown for gene categories as defined in ref. 43. P values for each group pairwise comparison areshown.. F, VMI gastric tumors have high expression of genes involved in immune cell suppression, while I gastric tumors are enriched gene expressionthat promote the immune response. Differential expression between groups is shown in the three right-most columns.





subgroups to most closely reflect a first-line setting, though itremains to be tested. In addition, there was a higher percentageof early-stage samples in the TCGA data set so in the end thedistribution of subsets from any analysis including this one willvary depending on the samples collected. Even when focusedon just the stage 3 and 4 samples, our stromal subtypes retainedtheir prognostic value. At this time, the only targeted therapy infirst-line gastric cancer is for IHC HER2 3þ and/or HER2-amplified tumors. Therefore, we assessed how these potentialstromal subgroups would divide when the HER2 high expres-sers were defined separately (Fig. 7D). Data with Her2 includedcan be found in Supplementary Table S2.

DiscussionExpression profiling has been used to explore molecular

classification of cancer across multiple indications. With theavailability of deep-sequencing data in recent years, there is anopportunity to take different views of subclassification andwhen appropriate, integrate data from the DNA, RNA, proteinprofiles with clinical outcome data. To date, the success inidentification of patient subsets for therapeutic matching hasprimarily been based on single gene mutations or high expres-sion in tumor cells. When this has been successful, such as foranti-Her2–targeted therapies in breast and gastric cancer, it canlargely be attributed to the targeting of dominant driver muta-tion. But for molecules with targets in the tumor stroma,whether antiangiogenesis or immunomodulatory, it wouldseem equally appropriate that our focus turn toward phenotyp-ing the stroma. Of course tumor genetics undoubtedly influ-ences the gene expression that contributes to a stromal phe-notype. Gastric cancer has few target-directed options and onlytrastuzumab is currently approved for HER2 high expressingcancers. Recently ramucirumab was approved in the second-line setting as a monotherapy or in combination with paclitaxel(38, 39), and there are several early-phase clinical trials ongo-ing with checkpoint inhibitors. This study was undertaken toinvestigate the heterogeneity of gastric cancer stroma to supportthe future study of available stromal-targeted therapies. Ulti-mately, whether stromal signatures, tumor signatures, geneticsor combinations of the above most potently predict therapyremains to be explored.

Technically, it is very difficult to develop a stromal signaturefrom tumor tissue, as there can be such variability in stromalcontent from sample to sample. For the TCGA profiling, aneffort was made to select sections enriched for tumor cells (5).Mouse tumor models pose an even greater challenge as we canreadily see the relative dearth of stroma in cell line xenografts aswell as PDX tumor models compared with human cancer(unpublished). Therefore, the stromal model used here hadclear advantages in being temporally and phenotypically repro-ducible. While this model uses a Nude mouse host, a robustmyeloid immune infiltrate found there was able to be extendedto lymphoid phenotypes by studying human samples. WhileVEGF-A is the dominant initiator of angiogenesis in this model,its viral expression is short lived, and additional angiogenicfactors arise to contribute to a complex vascular bed. There isalso some influence of the Adenovirus itself in the immune cellinflux.

Using molecular classification of cancer to supplement clas-sical pathology can reveal differences in prognosis or be pre-

dictive of treatment outcomes for patients that are otherwiseindistinguishable by histology. ACRG survival data paired withTCGA protein array data showed that the most striking differ-ence in our phenotypes and their prognosis is seen between thepatients with productive angiogenesis from those without it(Fig. 7). However, when more closely examined there weredifferences in prognosis also when considering the immunephenotypes in the VMI and inflammatory subgroups. Theprofile of the immune markers that dominate those groupssuggests that the inflammatory subgroup trends toward a moreproductive immune response with markers of cytotoxic T cellsand low incidence of immunosuppressive gene expression,similar to what has been reported (40). Notably, the MSIcohorts described by the TCGA most closely associated withthe nonangiogenic phenotype, but were split relatively equallybetween the I and VINI groups (Supplementary Table S2). MSIhigh tends to be enriched for activated T cells and checkpointregulators such as PD1 and PD-L1 (41). Interestingly, thus farclinical data suggest that only about half of the MSI highpatients respond to checkpoint inhibition.

On the other hand, the immunologic microenvironment inthe subgroup associated with angiogenesis was weightedtoward immunosuppression, perhaps not surprising given therole VEGF-A can play in promoting M2 myeloid phenotypesand inhibition of dendritic cell maturation (42). In this set-ting, we hypothesize that it might require more than justrelieving checkpoint inhibition to enable a robust immuneresponse and combination approach with antiangiogenicsadded to checkpoint inhibition may be worth exploring.Finally, it is worth mentioning that there was a subgroup withrelatively little productive angiogenesis or immune cell infil-tration, the VINI subgroup. This group is a major subset ofthe p53 mutant/microsatellite stable ACRG subgroup (seeFig. 6D), which was described by the ACRG consortium ashaving the most RTK amplification (6). Thus, more targetedtherapies may emerge that are well suited to this cohort in thefuture.

Disclosure of Potential Conflicts of InterestM.T. Uhlik is Vice President, Translational Oncology, at Biothera Phar-

maceutical, reports receiving commercial research grant from BiotheraPharmaceutical, and has ownership interest (including patents) in BiotheraPharmaceutical and Eli Lilly and Company. H.F. Dvorak reports receivingcommercial research grant from Lilly. No potential conflicts of interestwere disclosed by the other authors.

Authors' ContributionsConception and design: M.T. Uhlik, J. Liu, B.L. Falcon, S. Iyer, C. Lowes,S. Chintharlapalli, A. Fischl, D. Gerald, Q. Xue, S.-c. Jaminet, A. Nasir,H.F. Dvorak, L.E. BenjaminDevelopment ofmethodology:M.T. Uhlik, J. Liu, B.L. Falcon, S. Iyer, J. Stewart,M. O'Mahony, C. Sevinsky, D. Gerald, Q. Xue, Y. Al-Kofahi, F. Ginty, A. Nasir,H.F. Dvorak, L.E. BenjaminAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): M.T. Uhlik, J. Liu, B.L. Falcon, S. Iyer, J. Stewart,H. Celikkaya, M. O'Mahony, C. Sevinsky, C. Lowes, L. Douglass,C. Jeffries, D. Bodenmiller, Q. Xue, J.-y. Lee, J.H. Carter, A. Nasir, J.A.Nagy, H.F. DvorakAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): M.T. Uhlik, J. Liu, B.L. Falcon, S. Iyer, J. Stewart,H. Celikkaya, C. Sevinsky, L. Douglass, C. Jeffries, D. Bodenmiller, S. Chinthar-lapalli, A. Fischl, Q. Xue, A. Santamaria-Pang, Y. Al-Kofahi, Y. Sui, K. Desai,T. Doman, A. Aggarwal, J.H. Carter, B. Pytowski, S.-c. Jaminet, F. Ginty,A. Nasir, H.F. Dvorak, L.E. Benjamin

Uhlik et al.




Writing, review, and/or revision of the manuscript: M.T. Uhlik, J. Liu,B.L. Falcon, S. Iyer, C. Sevinsky, C. Lowes, S. Chintharlapalli, A. Fischl,D. Gerald, A. Santamaria-Pang, Y. Al-Kofahi, A. Aggarwal, F. Ginty, A. Nasir,J.A. Nagy, H.F. Dvorak, L.E. BenjaminAdministrative, technical, or material support (i.e., reporting or orga-nizing data, constructing databases): M.T. Uhlik, B.L. Falcon, S. Iyer,H. Celikkaya, L. Douglass, Q. Xue, A. Santamaria-Pang, J.H. Carter,B. Pytowski, F. GintyStudy supervision: M.T. Uhlik, S. Iyer, H.F. Dvorak, L.E. BenjaminOther (developed image and data analysis algorithms blood vessel segmen-tation and classification of immune cells): A. Santamaria-Pang

AcknowledgmentsThe authors thank G. Plowman and D. Ferry for critical reading and

editorial suggestions, C. Reinhard for helpful discussions, and ShengGuo at CrownBio for analysis of PDX expression profiles.

The costs of publication of this article were defrayed in part by the paymentof page charges. This article must therefore be hereby marked advertisementin accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Received January 6, 2016; revised February 18, 2016; accepted February 19,2016; published online May 2, 2016.

References1. Woo J, Cohen SA, Grim JE. Targeted therapy in gastroesophageal cancers:

past, present and future. Gastroenterol Rep (Oxf) 2015;3:316–29.2. Subhash VV, Yeo MS, Tan WL, Yong WP. Strategies and advancements in

harnessing the immune system for gastric cancer immunotherapy.J Immunol Res 2015;2015:308574.

3. Davidson M, Okines AF, Starling N. Current and future therapies foradvanced gastric cancer. Clin Colorectal Cancer 2015;14:239–50.

4. Abdel-Rahman O. Immune checkpoints aberrations and gastric cancer;assessment of prognostic value and evaluation of therapeutic potentials.Crit Rev Oncol Hematol 2016;97:65–71.

5. Cancer Genome Atlas Research Network. Comprehensive molecular char-acterization of gastric adenocarcinoma. Nature 2014;513:202–9.

6. Cristescu R, Lee J, NebozhynM, KimKM, Ting JC,Wong SS, et al. Molecularanalysis of gastric cancer identifies subtypes associatedwith distinct clinicaloutcomes. Nat Med 2015;21:449–56.

7. Nagy JA, Shih SC, Wong WH, Dvorak AM, Dvorak HF. Chapter 3.The adenoviral vector angiogenesis/lymphangiogenesis assay. MethodsEnzymol 2008;444:43–64.

8. Pettersson A, Nagy JA, Brown LF, Sundberg C, Morgan E, Jungles S, et al.Heterogeneity of the angiogenic response induced in different normaladult tissues by vascular permeability factor/vascular endothelial growthfactor. Lab Invest 2000;80:99–115.

9. Yadav V, Zhang X, Liu J, Estrem S, Li S, Gong XQ, et al. Reactivation ofmitogen-activated protein kinase (MAPK) pathway by FGF receptor 3(FGFR3)/Ras mediates resistance to vemurafenib in human B-RAFV600E mutant melanoma. J Biol Chem 2012;287:28087–98.

10. Falcon BL, Stewart J, Ezell S, Hanson J, Wijsman J, Ye X, et al. High-contentmultiplexed tissue imaging and quantification for cancer drug discovery.Drug Discov Today 2013;18:510–22.

11. GerdesMJ, Sevinsky CJ, SoodA, Adak S, BelloMO, Bordwell A, et al. Highlymultiplexed single-cell analysis of formalin-fixed, paraffin-embedded can-cer tissue. Proc Natl Acad Sci U S A 2013;110:11982–7.

12. Nagy JA, Chang SH, Shih SC, Dvorak AM, Dvorak HF. Heterogeneity of thetumor vasculature. Semin Thromb Hemost 2010;36:321–31.

13. Dvorak HF. Tumors: wounds that do not heal. Similarities between tumorstroma generation and wound healing. N Engl J Med 1986;315:1650–9.

14. Nagy JA, Dvorak HF. Heterogeneity of the tumor vasculature: the need fornew tumor blood vessel type-specific targets. Clin Exp Metastasis 2012;29:657–62.

15. Sitohy B, Nagy JA, Dvorak HF. Anti-VEGF/VEGFR therapy for cancer:reassessing the target. Cancer Res 2012;72:1909–14.

16. Sitohy B, Nagy JA, Jaminet SC, Dvorak HF. Tumor-surrogate blood vesselsubtypes exhibit differential susceptibility to anti-VEGF therapy. CancerRes 2011;71:7021–8.

17. Falcon S, Gentleman R. Using GOstats to test gene lists for GO termassociation. Bioinformatics 2007;23:257–8.

18. Cascone T, Herynk MH, Xu L, Du Z, Kadara H, Nilsson MB, et al. Upre-gulated stromal EGFR and vascular remodeling inmouse xenograftmodelsof angiogenesis inhibitor-resistant human lung adenocarcinoma. J ClinInvest 2011;121:1313–28.

19. Abramovitch R, Dafni H, Smouha E, Benjamin LE, Neeman M. In vivoprediction of vascular susceptibility to vascular susceptibility endothelialgrowth factor withdrawal: magnetic resonance imaging of C6 rat glioma innude mice. Cancer Res 1999;59:5012–6.

20. Benjamin LE, Keshet E. Conditional switching of vascular endothelialgrowth factor (VEGF) expression in tumors: induction of endothelial cell

shedding and regression of hemangioblastoma-like vessels by VEGF with-drawal. Proc Natl Acad Sci U S A 1997;94:8761–6.

21. Tolaney SM, Boucher Y, Duda DG, Martin JD, Seano G, Ancukiewicz M,et al. Role of vascular density and normalization in response to neoadju-vant bevacizumab and chemotherapy in breast cancer patients. Proc NatlAcad Sci U S A 2015;112:14325–30.

22. NewmanPJ, BerndtMC,Gorski J,WhiteGC2nd, LymanS, PaddockC, et al.PECAM-1 (CD31) cloning and relation to adhesion molecules of theimmunoglobulin gene superfamily. Science 1990;247:1219–22.

23. Noguera-Troise I,DalyC, PapadopoulosNJ, Coetzee S, BolandP,GaleNW,et al. Blockade of Dll4 inhibits tumour growth by promoting non-pro-ductive angiogenesis. Nature 2006;444:1032–7.

24. Li H, Yu B, Li J, Su L, YanM, Zhang J, et al. Characterization of differentiallyexpressed genes involved in pathways associated with gastric cancer. PLoSOne 2015;10:e0125013.

25. Monney L, Sabatos CA,Gaglia JL, RyuA,WaldnerH, Chernova T, et al. Th1-specific cell surface protein Tim-3 regulates macrophage activation andseverity of an autoimmune disease. Nature 2002;415:536–41.

26. Mia S, Warnecke A, Zhang XM, Malmstrom V, Harris RA. An optimizedprotocol for human M2 macrophages using M-CSF and IL-4/IL-10/TGF-beta yields a dominant immunosuppressive phenotype. Scand J Immunol2014;79:305–14.

27. Saraiva M,O'Garra A. The regulation of IL-10 production by immune cells.Nat Rev Immunol 2010;10:170–81.

28. Bhaskaran N, Quigley C, Weinberg A, Huang A, Popkin D, Pandiyan P.Transforming growth factor-beta1 sustains the survival of Foxp3 regulatorycells during late phase of oropharyngeal candidiasis infection. MucosalImmunol 2015 Nov 4, doi:10.1038/mi.2015.115.

29. Collins T, Read MA, Neish AS, Whitley MZ, Thanos D, Maniatis T.Transcriptional regulation of endothelial cell adhesion molecules: NF-kappa B and cytokine-inducible enhancers. FASEB J 1995;9:899–909.

30. Onofre G, Kolackova M, Jankovicova K, Krejsek J. Scavenger receptorCD163 and its biological functions. Acta Medica (Hradec Kralove)2009;52:57–61.

31. Tarique AA, Logan J, Thomas E, Holt PG, Sly PD, Fantino E. Phenotypic,functional, and plasticity features of classical and alternatively activatedhuman macrophages. Am J Respir Cell Mol Biol 2015;53:676–88.

32. WangH, Liu C,HanM, Cheng C, ZhangD. TRAM1 promotesmicrogliaM1polarization. J Mol Neurosci 2016;58:287–96.

33. Wang S, Amato KR, Song W, Youngblood V, Lee K, Boothby M, et al.Regulation of endothelial cell proliferation and vascular assembly throughdistinct mTORC2 signaling pathways. Mol Cell Biol 2015;35:1299–313.

34. JohnsonDW, Berg JN, BaldwinMA, Gallione CJ, Marondel I, Yoon SJ, et al.Mutations in the activin receptor-like kinase 1 gene in hereditary haemor-rhagic telangiectasia type 2. Nat Genet 1996;13:189–95.

35. Bohmer R, Neuhaus B, Buhren S, Zhang D, Stehling M, Bock B, et al.Regulation of developmental lymphangiogenesis by Syk(þ) leukocytes.Dev Cell 2010;18:437–49.

36. Arase H, Saito T, Phillips JH, Lanier LL. Cutting edge: the mouse NK cell-associated antigen recognized by DX5 monoclonal antibody is CD49b(alpha 2 integrin, very late antigen-2). J Immunol 2001;167:1141–4.

37. Dellas C, Loskutoff DJ. Historical analysis of PAI-1 from its discovery toits potential role in cell motility and disease. Thromb Haemost 2005;93:631–40.

38. Fuchs CS, Tomasek J, YongCJ, Dumitru F, Passalacqua R, Goswami C, et al.Ramucirumab monotherapy for previously treated advanced gastric or





gastro-oesophageal junction adenocarcinoma (REGARD): an internation-al, randomised, multicentre, placebo-controlled, phase 3 trial. Lancet2014;383:31–9.

39. Wilke H, Muro K, Van Cutsem E, Oh SC, Bodoky G, Shimada Y, et al.Ramucirumab plus paclitaxel versus placebo plus paclitaxel in patientswith previously treated advanced gastric or gastro-oesophageal junctionadenocarcinoma (RAINBOW): a double-blind, randomised phase 3 trial.Lancet Oncol 2014;15:1224–35.

40. Kim JW, Nam KH, Ahn SH, Park DJ, Kim HH, Kim SH, et al. Prognosticimplications of immunosuppressive protein expression in tumors as wellas immune cell infiltration within the tumor microenvironment in gastriccancer. Gastric Cancer 2016;19:42–52.

41. Llosa NJ, Cruise M, Tam A, Wicks EC, Hechenbleikner EM, Taube JM, et al.The vigorous immune microenvironment of microsatellite instable coloncancer is balanced by multiple counter-inhibitory checkpoints. CancerDiscov 2015;5:43–51.

42. Voron T, Marcheteau E, Pernot S, Colussi O, Tartour E, Taieb J, et al.Control of the immune response by pro-angiogenic factors. Front Oncol2014;4:70.

43. Angelova M, Charoentong P, Hackl H, Fischer ML, Snajder R, Krogs-dam AM, et al. Characterization of the immunophenotypes andantigenomes of colorectal cancers reveals distinct tumor escapemechanisms and novel targets for immunotherapy. Genome Biol2015;16:64.


Uhlik et al.



2016;76:2573-2586. Cancer Res Mark T. Uhlik, Jiangang Liu, Beverly L. Falcon, et al. Stromal-Based Signatures for the Classification of Gastric Cancer

Updated version

http://cancerres.aacrjournals.org/content/76/9/2573

Access the most recent version of this article at:

Material

Supplementary

http://cancerres.aacrjournals.org/content/suppl/2016/10/19/76.9.2573.DC1

Access the most recent supplemental material at:

Cited articles

http://cancerres.aacrjournals.org/content/76/9/2573.full#ref-list-1

This article cites 42 articles, 11 of which you can access for free 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://cancerres.aacrjournals.org/content/76/9/2573To request permission to re-use all or part of this article, use this link



http://cancerres.aacrjournals.org/content/suppl/2016/10/19/76.9.2573.DC1

http://cancerres.aacrjournals.org/content/76/9/2573.full#ref-list-1

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

mailto:[email protected]



Documents

Stromal-Based Signatures for the Classi cation of Gastric Cancer - … · 2021. 2. 3. · Microenvironment and Immunology Stromal-Based Signatures for the Classiﬁcation of Gastric