Identification of Two Distinct Carcinoma-Associated Fibroblast Subtypes with Differential Tumor-Promoting Abilities in Oral Squamous Cell Carcinoma

Published OnlineFirst April 18, 2013; DOI: 10.1158/0008-5472.CAN-12-4150

Cancer Microenvironment and Immunology Research

Identiﬁcation of Two Distinct Carcinoma-Associated Fibroblast Subtypes with Differential Tumor-Promoting Abilities in Oral Squamous Cell Carcinoma

Daniela Elena Costea1, Allison Hills4, Amani H. Osman1, Johanna Thurlow5, Gabriela Kalna6, Xiaohong Huang4, Claudia Pena Murillo4, Himalaya Parajuli1, Salwa Suliman1,2, Keerthi K. Kulasekara7, Anne Chr. Johannessen1,3, and Max Partridge4

Abstract Heterogeneity of carcinoma-associated fibroblasts (CAF) has long been recognized, but the functional significance remains poorly understood. Here, we report the distinction of two CAF subtypes in oral squamous cell carcinoma (OSCC) that have differential tumor-promoting capability, one with a transcriptome and secretome closer to normal fibroblasts (CAF-N) and the other with a more divergent expression pattern (CAF-D). Both subtypes supported higher tumor incidence in nonobese diabetic/severe combined immunode- ficient (NOD/SCID) Ilg2(null) mice and deeper invasion of malignant keratinocytes than normal or dysplasia- associated fibroblasts, but CAF-N was more efficient than CAF-D in enhancing tumor incidence. CAF-N included more intrinsically motile fibroblasts maintained by high autocrine production of hyaluronan. Inhibiting CAF-N migration by blocking hyaluronan synthesis or chain elongation impaired invasion of adjacent OSCC cells, pinpointing fibroblast motility as an essential mechanism in this process. In contrast, CAF-D harbored fewer motile fibroblasts but synthesized higher TGF-b1 levels. TGF-b1 did not stimulate CAF-D migration but enhanced invasion and expression of epithelial–mesenchymal transition (EMT) markers in malignant keratinocytes. Inhibiting TGF-b1 in three-dimensional cultures containing CAF-D impaired keratinocyte invasion, suggesting TGF-b1–induced EMT mediates CAF-D–induced carcinoma cell invasion. TGF-b1–pretreated normal fibroblasts also induced invasive properties in transformed oral keratinocytes, indicating that TGF-b1–synthesizing fibroblasts, as well as hyaluronan-synthesizing fibroblasts, are critical for carcinoma invasion. Taken together, these results discern two subtypes of CAF that promote OSCC cell invasion via different mechanisms. Cancer Res; 73(13); 1–14. 2013 AACR.

Introduction and this was shown to be important for tumor progression (2, Induction of invasion is the key property differentiating 3). There is now convincing evidence that dysplasia-associated fi fi malignant from benign lesions. Similar to other carcinomas, broblasts (DAF) and carcinoma-associated broblasts (CAF) oral squamous cell carcinoma (OSCC) is a multistep process in differ from those associated with normal epithelium (4, 5), and which the altered epithelium undergoes malignant conversion that these adaptations have functional consequences for over a period of time (1). As carcinoma evolve, changes in the tumor progression and invasion (6, 7). Initially, these carcino- epithelia result also in concomitant adaptations of the adja- ma-promoting effects were attributed mainly to the subpop- fi cent normal stroma and one of its main cell type, the fibroblast, ulation of myo broblasts in the mixed CAF populations that had higher ability to secrete stimulative paracrine factors (3, 8, 9). Authors' Affiliations: 1Section for Pathology, The Gade Institute; 2Institute However, CAF were shown to show phenotypic and geno- 3 of Clinical Dentistry, University of Bergen; Department of Pathology, typic diversity (4, 10, 11), as well as complex changes in their Haukeland University Hospital, Bergen, Norway; 4Head and Neck Unit, Guy's and St. Thomas' Hospitals NHS Foundation Trust, London; 5Center secretory activity (12–14), but the functional significance of for Stem Cell Biology, Department of Biomedical Science, University of this heterogeneity has been poorly addressed so far. Only Sheffield, Sheffield; 6Beatson Institute, Glasgow, United Kingdom; and 7Rural Health School, La Trobe University, Bendigo, VIC 3552, Australia recently, the essential role of this heterogeneity and the cooperation between different subpopulations was shown for pros- Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/). tate tumorigenesis, with the differential TGF-b signaling in tumor stromal cells indicated to be pivotal (15). TGF-b1 Corresponding Author: Daniela Elena Costea, Section for Pathology, The Gade Institute, Haukeland University Hospital, 5021 Bergen, Norway. essential role in carcinogenesis, either as tumor suppressor at Phone: 47-5597-2564; Fax: 47-5597-3158; E-mail: early stages and/or as tumor promoter at late stages has been [email protected] long recognized (16). One of the processes through which TGF- doi: 10.1158/0008-5472.CAN-12-4150 b1 participates to carcinogenesis is the epithelial–mesenchy- 2013 American Association for Cancer Research. mal transition (EMT) phenomenon, via increased carcinoma

www.aacrjournals.org OF1

Costea et al.

cell motility, invasiveness, and ultimately metastasis, and this (RMAexpress; http://rmaexpress.bmbolstad.com/). CEL files phenomenon was shown to occur in OSCC as well (17). for samples and normalized data matrix are available from It was the aim of this study to investigate stromal hetero- http://www.ncbi.nlm.nih.gov/geo/ (GSE38517). Rank product geneity in OSCC at both molecular and functional levels, and analysis (20) was conducted to identify differentially expressed identify the mechanisms by which different CAF subsets genes between NOF (n ¼ 5), DAF (n ¼ 4), and CAF (n ¼ 7). Gene support oral carcinoma development and invasion. We show Ontology analysis was conducted for the differentially here a differential global gene expression profile for normal expressed genes using DAVID 6.7 (21). Data were analyzed oral fibroblasts (NOF), DAF, and CAF derived from OSCC, but using Spectral Clustering, an unsupervised algorithm (22), for the novelty is the observation that unsupervised clustering all strains maintained in 3D biomatrices and 2D monolayers, or could identify two subgroups of CAF. Detailed analysis of their for 3D alone only. migratory and secretory characteristics presented here indicates that the 2 subtypes of CAF are able to induce oral Tumor xenografting in NOD/SCID IL2rg(null) mice carcinoma cell invasion by different mechanisms. To assess tumor formation, 103 transformed, nontumorigenic oral dysplasic keratinocytes (DOK cell line; ref. 23) Materials and Methods obtained from European Collection of Cell Cultures were – Clinical samples, cell isolation, and characterization suspended in 50 mL of growth factor reduced matrigel (BD n ¼ 5 OSCCs, lesions with histologic evidence of dysplasia, and Biosciences), and inoculated alone ( 6) or together with 10 fi n ¼ n ¼ contra lateral normal oral biopsies were collected following broblasts of the strains NF5 ( 6), CAF1 ( 6), or CAF5 n ¼ ethical approval and written informed consent. CAF (n ¼ 18), ( 6) subcutaneously on the back of 12-week-old nonobese fi DAF (n ¼ 6), and NOF (n ¼ 17) were isolated and forwarded for diabetic/severe combined immunode cient (NOD/SCID) morphologic examination, 2-dimensional (2D) cultures, 3- IL2rg(null) mice (The Jackson Laboratory). Tumor incidence dimensional (3D) constructs in collagen type I biomatrices, and development (volume) was assessed at every 3 days. All fi gene microarray, quantitative reverse transcription PCR (qRT- mice were sacri ced 45 days after inoculation and tissues were PCR), immunohistochemistry (IHC), and flow cytometry anal- harvested for histologic assessment. The Norwegian Animal ysis (7, 18). For flow cytometry, subconfluent cells (105) were Research Authority approved all animal procedures. suspended in 100 mL PBS with 2% FBS and 1% HEPES (all Sigma), and incubated with 20 mLoffluorochrome-conjugated Tissue engineering and evaluation of carcinoma cell antibody fluorescein isothiocyanate (FITC) mouse anti-human invasion. epithelial-specific antigen (ESA; Biomeda), phycoerythrin (PE) Fibroblasts at passage 3 to 7 (split ratio 1:4) were embedded mouse anti-human CD31, CD45, CD146, CD140b (known as in collagen type I biomatrix (BD Biosciences), and seeded on platelet-derived growth factor receptor B—PDGFRB), FITC, or top with malignant or transformed oral keratinocytes in PE mouse immunoglobulin G 1 (IgG1)k—isotype control (all triplicate, as previously described (7). The majority of experi- BD Pharmingen) for 15 minutes and analyzed using a MoFlo ments were carried out using Ca1 malignant oral keratinocyte cell sorter (Beckman Coulter). For IHC, fixed cells and tissues cell line (24) that showed minimal invasion when seeded onto were stained with lineage-specific antibodies recognizing NOF-populated biomatrices. Nontumorigenic DOK cells and vimentin, aSMA (smooth muscle actin), CD31, pancytokeratin additional stains of malignant oral keratinocytes UK1, H357, (panCK), S100A4 (FSP; all DAKO; 1:50), and fibroblast-activat- 5PT, CaLH3 (25), and SCC25 (26) were used to validate the fi n ¼ ing protein (FAP; Affinity Bioreagents; 1:5), and vizualized with ndings. For some experiments, NOF ( 3) were pretreated EnVision kit (DAKO), as previously described (19). For scan- with 10 ng/mL TGF-b1 (BD Biosciences) for 10 days, then used ning electron microscopy (SEM), 2 104 fibroblasts were fixed to construct 3D biomatrices. 3D constructs were harvested, fi fi in 2% glutaraldehyde/0.1 mol/L phosphate buffer pH 7.2, for 2 formalin- xed, and paraf n-embedded. To measure depth of hours at 4C, mounted on grids and viewed using a Jeol JSM- invasion, 3-mm sections were stained for panCK (DAKO; 1:50). fi 7400 field emission-SEM. For transmission electron micros- Subsequently, each section was divided into fths. The central fi copy (TEM), cells were postfixed in 1% osmium tetroxide and the two outer fths were excluded from measurements, fi (Sigma) in PBS (30 minutes), dehydrated using graded etha- depth of invasion being assessed in the remaining two fths nols, embedded in epoxy resin, ultrathin sectioned, double only. For this, a horizontal line was drawn (using the software stained with uranyl acetate and lead citrate (Sigma). Speci- Olympus DP.Soft 5.0) through the uppermost remnants of the mens were examined using a TEM (JEOL 1230; Jeol Ltd.), and collagen gel to visualize the basement membrane zone; depth the micrographs processed using an Arcus II scanner (Agfa- of invasion was determined every 100 mm along this horizontal Gevaert N.V.). For detailed characterization, see Supplemen- line as the vertical distance from this line to the limit of tary Figs. S1 and S2. invading epithelial cells (19).

Gene expression profiling Secretory profile Cells (2 106) at passage 2 were seeded in 3D collagen type I Eighteen-hour serum-free conditioned media was collected (BD Biosciences) biomatrices in duplicates for 5 days. RNA was from cells maintained in 2D and 3D culture at similar passages extracted using RNA Stat (Biogenesis Ltd.) for analysis with and analyzed for levels of growth factors, cytokines, matrix U133 plus 2 arrays (Affymetrix Inc.). CEL file data were nor- metalloproteinases (MMP), and hyaluronan by ELISA with malized using Robust Multichip Average expression summary Luminex beads (R&D Systems, Inc.), the Widescreen Human

OF2 Cancer Res; 73(13) July 1, 2013 Cancer Research

Subtypes of Oral Carcinoma-Associated Fibroblasts

Cancer Panel 2 (Novagen, Millipore), the Flurokine MAP TGF-b TGF-b1 inhibition Multiplex Kit (R&D Systems, Inc.), and hyaluronan ELISA A potent and speciﬁc inhibitor of TGF-b1 superfamily type I (Corgenix). The results are presented with values normalized activin receptor-like kinase (ALK) receptors ALK4, ALK5, and for 106 cells; data represent the mean SD. Deposition in ALK7, the small molecule SB431542 (Sigma), was used at a patient tissue and 3D constructs of hyaluronan was visualized concentration of 10 mmol/L, and the antibody against TGF-b1 by histochemical staining in deparafﬁnized, rehydrated tissue (R&D Systems, Inc.) or the isotype control were used at a con- sections stained with a 5% acidic solution of Alcian blue centrationof 10 mg/mL. Inhibitorswere added 30 minutes before (Sigma) for 45 minutes and then counterstained with nuclear TGF-b1 treatment in monolayer. In 3D constructs, these reag- fast red (Sigma). ents were added each second day, with the change of medium.

Protein detection Motility assays Protein lysates were resolved by PAGE and the membranes Cell migration assays were conducted as previously were probed with antibodies recognizing Smad2, pMLC (both described (27) with and without prior exposure of cells to 10 Cell Signaling Technology Inc.; 1:1,000), Smad2/3 (R&D Sys- ng/mL TGF-b1 for 18 hours. For cell invasion assay, Transwell tems, Inc.; 1:1,000), Hyaluronan-mediated motility receptor were coated with 6 mg/mL Matrigel (Invitrogen). In some (RHAMM) (gift from V. Assmann, Center for Experimental experiments 0.3 mmol/L 4-MU (Sigma) was added to collagen Medicine, Institute of Tumor Biology, University Hospital matrix and media to block elongation of hyaluronan for 24 5 Hamburg, Germany; 1:400), and glyceraldehyde-3-phosphate hours prior the assessment. For time-lapse microscopy, 10 dehydrogenase (GAPDH; Abcam; 1:1,000) as previously cells were seeded on 1.7 mg/mL collagen (Nutacon) for 24 hours described (7). For validation of microarray data, additional and then treated with CellTracker Green CMFDA (Molecular Western blot analysis and IHC of monolayers and tissue Probes, Invitrogen). One hour before imaging, medium was sections was conducted with antibodies recognizing ITGA6, changed to 1% FBS/1% HEPES. Images were taken every 10 ITGA5 (all Cell Signaling Technology Inc.), GAPDH, DUSP4 minutes for 18 hours with a Zeiss LSM 510 META confocal (both Abcam), MFAP5, COL15A1 (all Sigma), PMEPA1, CADPS microscope to determine average track speed, straightness, (both Abnova), TAGLN (Leica), INTA11 (gift from Prof. D. length, and displacement using Open Source imaging, Fiji and Gullberg, Department of Biomedicine, University of Bergen, Imaris software (MeasurementPro and ImarisTrack; Bitplane). Norway), RAB27B (gift from M.C. Seabra, Imperial College London, UK), and CHI3L1 (gift from S. Werner, Institute of Statistical analysis – U Molecular Health Sciences, Zurich, Switzerland), all 1:1,000. All data are presented as mean SD. Mann Whitney test (SPSS version 18) was used for analysis of ELISA data, ANOVA post hoc Quantitative reverse transcription PCR with a Bonferroni test for RhoA/Rac G-LISA data, t Cells were lysed with RNA-Stat-60 (AMS Biotechnology paired Student test for the comparison of invasion scores, and t Europe Ltd.), total RNA was extracted following the manufac- independent Student test for analysis of population doublings ﬁ turer's instructions, and cDNA synthesis was conducted using data. To examine the biologic signi cance of the genes over- High-Capacity cDNA Archive Kit system (Applied Biosystems). expressed in CAFs, we examined whether they correlated with qRT-PCR was then conducted using inventoried TaqMan outcome using our independent microarray database of 71 assays with exon-spanning probes detecting MFAP5, CADPS, head and neck squamous cell carcinoma (22), with disease- ﬁ ASPN, TAGLN, TIAMI, MEST, EVI1, CDKNIC, EDIL3, SRGN, VIM, speci c death as the primary endpoint. Cox univariate and S100A4, TWST1, HMGA2, FAPa, PDGFRB, COL1A2, DDR, THY1, multivariate analysis was conducted using the R environment C (http://www.r-project.org) and Survival bioconductor package and ACTB. Comparative 2 DD t method was used to quantify (http://www.bioconductor.org). the relative mRNA expression.

Inhibition of hyaluronan Results For some experiments, 0.3 mmol/L 4-methylumbelliferone CAFs show increased expression of TGF-b target genes (4-MU; Sigma) was added to collagen matrix and media for 24 Rank product analysis showed that 335 genes were signif- hours to block elongation of hyaluronan chains and motility of icantly upregulated and 347 were downregulated [false dis- cells. Speciﬁc inhibition of hyaluronan synthesis was achieved covery rate (FDR) 0.01] when CAF and NOF were compared by hyaluronan synthase HAS2 short hairpin RNA (shRNA) (Supplementary Table S1). These genes were linked to sub- lentiviral particles transduction of CAFs 1 and 3. Fibroblasts strate adhesion, tissue remodeling, cell migration, secretion, were seeded at 105 cells per well in 6-well plates and after 24 growth regulation, and angiogenesis (Table 1). Gene Ontology hours were infected with HAS2 shRNA or control shRNA analysis conﬁrmed that the overexpressed genes clustered into lentiviral particles (Santa Cruz Biotechnology) at a multiplicity many functional groups (Supplementary Table S1). Notably, of of infection (MOI) of 100 in presence of polybrene (5 mg/mL; the top 100 genes (the unique gene symbols), 52 were TGF-b Santa Cruz Biotechnology), then centrifuged at 32C for 2 targets (Fig. 1A; ref. 28). In addition, many transcripts of factors hours at 2,300 rpm. After 48 hours, puromycin (2 mg/mL; Santa that bind to, or modulate the bioactivity and cellular response Cruz Biotechnology) was added for selection of transduced to TGF-b1, including ASPN (29), BGN (30), DCN (31), PMEPAI cells, and after additional 5 days, the cells were split 1:4 in (32), and CTHRC1 (33) were upregulated when CAF were puromycin-containing medium. compared with NOF (Table 1). Other upregulated genes were

www.aacrjournals.org Cancer Res; 73(13) July 1, 2013 OF3

Costea et al.

Table 1. Summary of the top overexpressed and underexpressed genes by ﬁbroblasts derived from normal human oral mucosa, dysplastic lesions, and OSCC

Top 30 genes overexpressed in CAF vs. NOF MFAP5, COL15A1, CH13L1, CADPS, SRGN1, IGFBP3, EDIL3, PSPH, RAB27B, WISP2, EVI2A2, LRG7, COL11A, ASPN, TAGLN, TMEM132B, SLC6A15, KCND2, MAB21L2, PPAPDC1A, NPTX2, THBD, FN1, SULF1, CSTA, KRTAP1-5, MGC34830, SERPINB7, BEX1, ODZ3

Top 30 TGFB targets overexpressed in CAF vs. NOF COL15A, CHI3L1, IGFPB3, EDIL3, WISP2, COL11A, ASPN, TAGLN, KCND2, THBD, FN1, TMEPAI, SCG2, DSP, BCAT1, TES, SPRP4, TIMP3, BGN, MGP, EV12A BCL2A1 GATA6, ODZ4, GLIPR1 HAS 2 ITGBL1, ITGA6, TGM2, BMPR1B

Top 30 genes overexpressed in DAF vs. NOF COL15A, MEST, PITX1, DBC1, OMD, ASPN, CFI, NFIB, EVI12A, HNCN1, CHL1, BGN, ADCY2, FOXL2, DPT, ALDH1A1, CPXM, DPT, IGFBP3, PLXDC1, TMEM132B, SLC6A15, KCND2, MAB21L2, PPAPDC1A

Function Overexpressed in CAF vs. NOF Under expressed in CAF vs. NOF

ECM-receptor intercation/adhesion MFAP5, COL15A, CHI3L1, COL11A1, MAP7, ATRNL1, GPC3, ROBO2, ASPN, FN1, SULF1, GPC6, HMCNI, EGFL6, SPON1, THBS4, SEMA4D, MEGF10, CDH13, MGP, ITGBL1, ITGA6, F11R, ITGA8, MATN2, PCDH10, ITGA11, COL10A1, PLOD2, HAS2, DPT, CILP, ITGA9, CDH18, DCN, TM4SF1, DSG2, PCDH7, CTHRC1, EMILIN2, FLRT3, CXCL12 ITGA10, LGALS1, SDC2 Tissue remodeling CSTA, SERPINB7, SERPINI1, TIMP3, MMP27, GPC3, CORIN, CTSC, CPXM2, TGM2, PLOD2, ADAM12, CTSH, SOD3, MMP12, MMP10 SERPINE1 Migratory machinery TAGLN, TIAM1, KBTBD10, CDH13, BGN, MYO3B, CXCL1, IL-8 MET, TM4SF1, TPM1, ACTC, CTHRC1, SDC2 RhoGTPase/actin cytoskeleton associated TAGLN, TIAM1, MTSS1, RIN2, TPM1, RAB38 DOCK2, ACTC, WASF2, DAAM1, CNN1, PP1CB, GARNL4, PDLIM5 Secretion/endocytosis CADPS, SRGN, RAB27B, SCG2, TACSTD2, SGNE1, INHBA, KCMNB4 Endothelial cell associated COL15A, EDIL3, THBD, DCBLD2, SPON1, EDG2, BMP4, PGF, PEAR1, BGN, NOTCH3, GDF5 KDR IL-8, RHOJ, PDGFD, PDPN Transcription TSHZ1, GATA6, NOTCH3, EBF, SPEN, INHBA, FOXL2, PITX1, SIX2, Growth regulation SRGN, IGFBP3, EREG, IGFBP5, IGFBP6, GDF10, F11R, FGF13, BMP4, INHBA, GDF5, GAS6, BMPR1B, SDC2 NOV, CXCL1, GMFG, CXCL12, WISP3, SPRY1, PDGFD, IL1RN Inﬂammatory/Immune response NPTX2, SCG2, ENPEP, ASB5, IL-24 LR8, FAM19A5, CXCL14, PTGS2, CXCL1, CD83, CCL11, IL-8, CXCL10, CCL 8, LY75, CXCL3, CXCL2, IL1RN Transcription factors TSHZ1, GATA6, NOTCH3, EBF, SPEN, PAX3 INHBA, FOXL2, PITX1, SIX2

Function Overexpressed in DAF vs. NOF Under expressed in DAF vs. NOF

ECM/adhesion COL15A1, OMD, HMCN1, DPT, SULF1, SRGN1, PCDH10, TNC, GPM6B, MGP, OGN, MFAP5, MFAP5, THBS4, CLDN1 Tissue remodeling ADAMTS5, SULF1, SULF2, ADAMTS15, CORIN, MASP1, MMP3 Migratory machinery BGN, KBTBD10, IL-8 Endothelial associated TEM7 (PLXDC1), JAG1, Transcription factors PITX1, NFIB, EVI1 FOXL2, MAF ID4, RUNX3, FOXL1 TSHZ1, PAX3, Inﬂammatory/immune responses CFI, IGSF10, CXCL14, CFD, IL-8, CCL11, IL-24, LY6K, FAM19A5

NOTE: The genes differentially expressed in CAF when compared with NOF (FDR 0.01) were linked to substrate adhesion, tissue remodeling, cell migration, secretion, growth regulation, and angiogenesis. TGF-b1 targets are shown in italics. Fewer changes were found when DAF and NOF were compared (127 genes signiﬁcantly upregulated and 75 downregulated, FD rate of 0.01). The most overexpressed functional group in DAF when compared with NOF were transcription factors.

OF4 Cancer Res; 73(13) July 1, 2013 Cancer Research

Subtypes of Oral Carcinoma-Associated Fibroblasts

ABCD PMEPA1 BGN CADPS

Color key 1.0

-2 1 Row Z-score 0.80.60.40.20.0

Disease-specific survival Disease-specific P = 0.025 P = 0.0058 P = 0.0000315

0102030 40 50 60 70 0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70 Follow-up (mo) Follow-up (mo) Follow-up (mo) F 160 dist(t(a), method = “euc”) hclust (*, “complete”) 1401201008060 CAF7 NF2 NF4 NF1 NF5 NF3 CAF6 CAF4 CAF2 CAF3 CAF5 CAF1 DAF4 DAF2 DAF1 DAF3 CAF7 Height NF1 DAF4 CAF5 CAF6 DAF3 NF5 DAF1

Divergent CAF1 DAF2 NF3 NF2 NF4 CAF4 CAF3 transcriptome CAF2 (CAF-D) Homogeneous, closer to normal, transcriptome ×10–4 (CAF-N) E G H Color key 5 4.5 1 Row Z-Score

216248_s_at / NR4A2 209505_at / NR2F1 0 Closer to normal, 204622_x_at / NR4A2 transcriptome 203666_at / CXCL12 204621_s_at / NR4A2 (CAF-N) 226281_at / DNER 209687_at / CXCL12 –5 222877_at / NRP2 V3 209506_s_at / NRF2F1 229225_at / NRP2 232136_s_at / CTTNBP2 205051_s_at / KIT 201860_s_at / PLAT –10 204273_at / EDNRB 236926_at / Tbx1 231666_at / PAX3

205884_at / ITGA4 1568791_s_at / EDNRB –15 CAF6 CAF7 CAF5 CAF1 CAF3 CAF2 CAF4 CAF7 CAF5 CAF6 CAF1 CAF3 CAF4 CAF2 –12 –10 –8 –6 –4 –2 0 2 4 6 CAF-D CAF-N CAF-D CAF-N V2 ×10–4

Figure 1. A, heatmap showing differentially expressed transcripts of TGF-b target genes between NOF and CAF. Expression values are log2, mean centered (red, higher expression; green, lower expression). B, Kaplan–Meier analysis of 71 head and neck SCC cases for PMEPA1 (probe set 222450_at). The black line shows the quartile with the lowest PMEPA1 expression (0–0.25; 4 events/18 cases); the red line the other quartiles (0.25–1; 23 events/53 cases). C, Kaplan–Meier analysis of 71 HNSCC cases for BGN (201261_x_at). The black line shows the quartile with lowest BGN expression (0–0.25; 1 events/18 cases); the red line the other quartiles (0.25–1; 26 events/53 cases). D, Kaplan–Meier analysis of 71 head and neck SCC cases for CADPS (1568603_at). The black line shows the quartiles with lowest CADPS expression (0–0.75; 15 events/54 cases); the red line shows the quartile with highest CADPS expression (0.75–1; 12 events/17 cases). E, spectral clustering showing distinct grouping of fibroblasts grown in 2D and 3D, and for those maintained in 3D separate grouping of NOF, DAF, and CAF. CAF clustered into 2 subgroups: one with a transcriptome closer to NOF (CAF 1-4, termed CAF-N), the other with a more divergent expression profile (CAF 5-7, termed CAF-D). F, hierarchical clustering showing the tight grouping of the more homogenous group of CAF-N (closest to DAF and NOF) when compared with the more heterogenous and divergent group of CAF-D. G, heatmap showing differentially expressed probe sets between CAF-N and CAF-D of genes enriching GO BP 5, GO:0016477cell migration. Expression values are log2, mean centered (red, higher expression; green, lower expression). H, heatmap showing differentially expressed probe sets between CAF-D and CAF-N of genes enriching GO:0014032neural crest cell development and GO BP 5, GO:0016477cell migration. Expression values are log2, mean centered (red, higher expression; green, lower expression). known modulators of cellular secretion, such as CADPS (34), regulated (FDR 0.01). Interestingly, many of the upregulated RAB27B (35), and SRGN (8-, 9-, and 6-fold, respectively; ref. 36). genes in DAF were transcription factors (Table 1). The prog- Fewer changes were found when DAF and NOF were com- nostic significance of the top-ranked 50 probe sets and TGF-b1 pared, with 127 genes significantly upregulated and 75 down- targets identified here by rank product analysis as being

www.aacrjournals.org Cancer Res; 73(13) July 1, 2013 OF5

Costea et al.

upregulated in CAF versus NOF was investigated on our support for carcinoma cell invasion using in vitro 3D con- independent microarray database of 71 head and neck SCC. structs showed that CAF-N were significantly more effective Univariate analysis showed that 20 of these probe sets were in supporting deeper invasion of carcinoma cells than CAF- significantly associated with reduced disease-free survival (P D(P ¼ 0.005; Fig. 2D), with a mix of small islands and single 0.05). These genes included PMEPA1 (P ¼ 0.025), BGN (P ¼ carcinoma cells invading the 3D biomatrix (Fig. 2E). Never- 0.0058), and CADPS (P < 0.001), Kaplan–Meier analysis showed theless, both subtypes of CAF supported significantly deeper the quartile of cases with the lowest expression of these genes invasion of oral carcinoma cells when compared with NOF had the best outcome (Fig. 1B–D). Significant changes in gene embedded in 3D biomatrix (Ca1 cell line; Fig. 2D and E). This expression were validated by qRT-PCR, IHC, and Western blot was confirmed using other cell lines (DOK, SCC25, CaLH3, analysis with additional strains of fibroblasts to those profiled and UK1; Supplementary Fig. S4). An intermediate pattern of on Affymetrix arrays maintained in 3D culture (Supplementary invasion was observed when DAF were incorporated into the Fig. S3). 3D biomatrix (Supplementary Fig. S4).

Gene expression profile identifies two subgroups of CAF The two CAF subgroups show a differential secretory derived from OSCC profile Spectral clustering showed a clear separation of fibro- To identify growth factors and multifunctional cytokines blasts grown in 2D cultures from those maintained in 3D that could be linked to the key differences in CAF behavior, biomatrices (Fig. 1E), and for the cells in 3D biomatrices, we investigated their secretory profile (Supplementary Fig. NOF and DAF grouped at different positions from CAF, S5). Multiplex ELISA assay of conditioned medium from reflecting their distinct gene expression signatures (Fig. fibroblasts maintained in 3D biomatrices showed that CAF 1E). Further analysis using hierarchical clustering (Fig. synthesized significantly more TGF-b1(P < 0.001), interleu- 1F), and various clustering arrangements of 3D cultures kin (IL)-1a (P < 0.001), acidic fibroblast growth factor (aFGF; alone, showed that CAF 1–4 tightly grouped together, indi- P ¼ 0.008), TNF-a (P ¼ 0.014), and keratinocyte growth factor cating that these strains were transcriptionally more similar (KGF) (P ¼ 0.04) than NOF, whereas more VEGF (P ¼ 0.003) to each other than the rest of the CAF strains (CAF 5–7). CAF and hepatocyte growth factor (HGF; P ¼ 0.007) was secreted 1–4 often clustered closely to NOF, and for these reasons by NOF (Supplementary Fig. S5B). CAF-N secreted signifi- were termed CAF-N; whereas the more transcriptionally cantly more KGF (P ¼ 0.019), HGF (P ¼ 0.002), aFGF (P ¼ divergent CAF 5–7 were termed CAF-D. Of note, the NOF 0.008), and MMP3 (P ¼ 0.016) than CAF-D (Fig. 4A), whereas strain (NF1) closest to CAF strains in some clustering CAF-D secreted 9 times more TGF-b1( P < 0.001;Fig.3A)and arrangements was from a juvenile, and the DAF strain more IL-1a (P ¼ 0.006) than CAF-N (Fig. 3A). (DAF3) that also clustered closest to CAF strains was isolated from a patient who later developed OSCC. Gene TGF-b1 increases invasion and expression of EMT Ontology analysis revealed that differentially expressed markers in OSCC cells, whereas TGF-b1 inhibition by genes between CAF-N and CAF-D clustered in the functional SB431542 impairs CAF-D–induced invasion of OSCC cells groups of cell migration and cell surface signal transduction in 3D constructs (Supplementary Table S1.11 and S1.12) Some of the genes Provided that CAF-D stimulated significantly deeper car- upregulated in CAF-N versus CAF-D belonged to cell migra- cinoma cell invasion when compared with NOF, the obser- tion (e.g., TBX1, KIT, ITGA4, CXCL12,andNR2F1)and vation that TGF-b1 was the most obvious change in their the angiogenesis functional group (Fig. 1G), whereas some secretory profile determined us to assess the TGF-b1effect of the genes upregulated in CAF-D versus CAF-N (e.g., PAX3, on carcinoma cell migration and invasion. Transwell assays NRP2,andEDNRB) were associated with mesenchymal/ showed significant increase of oral carcinoma cell migration neural crest development and amoeboidal cell movement and invasion after treatment with TGF-b1forallcelllines (Fig. 1H). tested (P ¼ 0.012 and 0.002, respectively), varying from 2- to 3-folds (Fig. 3B and C), Although the amplitude and kinetics The two CAF subgroups provide differential support for oftheresponsetoTGF-b1 treatment varied between the cell tumor formation and invasion lines investigated, qRT-PCR showed an increased expression DOK cells did not develop tumors when xenotransplanted of EMT-related markers after 5 hours of exogenous TGF-b1 subcutaneously in NOD/SCID IL2rg(null) mice. Tumor inci- exposure(Fig.3D).IHCshowedsignificant increase in dence increased from 0 (0 of 6), when DOK were inoculated vimentin expression as long as 5 days after TGF-b1treat- alone, to 66.66% (8 of 12) by coinoculating DOK with CAF. A ment (Fig. 3E–G). Significant decrease in cell invasion was significantly higher tumor incidence was observed for DOK observed when 3D biomatrices populated with CAF-D were coinoculated with CAF-N (83.33%, 5 of 6) when compared seeded with DOK and various other carcinoma cell lines on with DOK coinoculated with CAF-D (50%, 3 of 6; Fig. 2A). In top (n ¼ 3), and 24 hours later treated with SB431542 (Fig. addition, DOK/CAF-N tumors developed after a shorter lag 3F–I). Significant decrease in invasion was also observed for time (median of 7 days) than DOK/CAF-D tumors (median of Ca1 and 5PT cell lines occurring after treatment with the 14 days; Fig. 2A), and showed invasion into the musculature anti-TGF-b1-Ab, but this effect was not statistically signif- layer (Fig. 2B), in contrast to the well-circumscribed DOK/ icant (Fig. 3F–I). Inhibition of activation of TGF-b1down- CAF-D tumors (Fig. 2C). Quantification of the fibroblast stream pathways after treatment with inhibitors 30 minutes

OF6 Cancer Res; 73(13) July 1, 2013 Cancer Research

Subtypes of Oral Carcinoma-Associated Fibroblasts

6 A DOK/CAF1 DOK/CAF5 B

5 38 DOK ALONE 33 DOK/CAF1 4 28 DOK/CAF5 3 23 18

No. tumors 2 13 8 1 3 0 -2 7 d 14 d 21 d 28 d 35 d 45 d Figure 2. A, incidence and lag time 7 d 14 d 21 d 28 d 35 d 42 d of tumors formed by DOK cells C DOK/CAF1 D DOK/CAF5 coinoculated subcutaneously with CAF-N (CAF1) or CAF-D (CAF5) in NOD/SCID/Ilg2(null) mice. B, microphotograph showing the tumor front of invasion into the muscular layer of a tumor developed after coinoculation of DOK and CAF1; hematoxylin and eosin (H&E) staining. C, microphotograph showing a well demarcated tumor developed after coinjection of DOK and CAF5 (H&E staining; E magnification, 200; scale bar, 200 Ca1/CAF1 Ca1/CAF3 Ca1/CAF5 Ca1/CAF6 mm). D, the depth of invasion of malignant oral keratinocytes (Ca1) was significantly higher when CAF- Ns, as opposed to CAF-D and NOF, were embedded into the biomatrix. Significant differences (P ¼ 0.001) Ca1/CAF1 Ca1/CAF3 Ca1/CAF5 Ca1/CAF6 are marked by a star. E, top, representative H&E-stained sections of 3D constructs with malignant oral keratinocytes (Ca1) seeded onto CAF-N, CAF-D, or NOF. panCK H&E Bottom, panCK staining of serial sections. Magnification, CAF-N CAF-D Ca1/NF4 Ca1/NF7 200. Scale bar, 200 mm. F

200 180 160 140 120 Ca1/NF4 Ca1/NF7 100 80 60 40

Depth of invasion ( μ m) 20 0 panCK H&E CAF1 CAF3 CAF4 CAF5 CAF6 CAF7 NF1 NF4 NF6 NF7 CAF-N CAF-D NOF NOF before TGF-b1 stimulation could be shown with SB431542 4A). The presence of this subpopulation of migratory fibro- but not anti-TGF-b1-Ab (Fig. 3J). blasts predominantly in CAF-N was also indicated by increased expression of pSmad2 when CAF-N and NOF were The two CAF subgroups have distinct motility compared (Fig. 4B). The significantly higher levels of RhoA- phenotypes GTPase and Rac1/2/3-GTPase in CAF-N when compared To determine other critical differences between the 2 CAF with NOF (P ¼ 0.0001; Supplementary Fig. S6) reflected also subtypes that might be linked to their differential ability to the higher proportion of migratory fibroblasts in CAF-N support tumor development and invasion, we investigated when compared with NOF. Furthermore, time-lapse micros- fibroblast motility. Transwell migratory experiments showed copy showed that CAF-N moved faster and for longer dis- that CAF-N contained a significantly higher percentage of tances from the initial point (longer track displacement) intrinsically migratory fibroblasts than CAF-D or NOF (Fig. than NOF (Fig. 4C and D).

www.aacrjournals.org Cancer Res; 73(13) July 1, 2013 OF7

Costea et al.

100,000 A NOF CAF-N 10,000 CAF-D

Figure 3. A, signiﬁcant differences cells 1,000 6 in levels of secretion of various growth factors, cytokines, and 100 MMPs determined for NOF, CAF-N, and CAF-D maintained in

ng/mL per 10 10 3D biomatrices. CAF-N secreted higher levels of KGF and HGF than CAF-D, whereas more TGF-b1 1 was secreted by CAF-D, P < 0.05. KGF HGF MMP3 aFGF TGF-β1 IL-1α B, cell migration in a Transwell assay for OSCC-derived cell lines B 300 Ctrl. C and DOK cells. Cells treated with 1 TGF-β1 200 250 180 Ctrl. ng/mL TGF-b1 for 18 hours β 160 TGF- 1 migrated signiﬁcantly more than 200 140 the control group. C, cell invasion 150 120 in a Matrigel Transwell assay for 100 the same cell lines. TGF-b1– 80 100 treated cells invaded signiﬁcantly 60 50 40 more than the control group. D, the

Mean migrating cells/feild expression of mRNA for various

Mean invading cell/feild 20 0 0 EMT markers in Ca1 cells treated H357 Cal DOK H357 Cal DOK with TGF-b1 relative to control cells. E, expression of vimentin as β D EMT markers in TGF-β–treated Cal relativ to ctrl. E G Vimentin in Cal +/– TGF- detected by IHC in Ca1 cells. F, 2.5 80 Vimentin Ca1 ctrl. expression of vimentin as detected S100A4 70 – 2.0 Twist by IHC in TGF-b1 treated Ca1 HMGA2 60 cells. G, the number of vimentin- 1.5 50 positive cells after TGF-b1 40 treatment was significantly higher Ca1+TGF-β1 1.0 F 30 than in control cells. Statistical 20 significant differences between the 0.5 % of positive cells 10 groups P < 0.05 are highlighted by mRNA relative value 0 a star. H, representative 0.0 β (TGF- β treated vs. ctrl cells) Ctrl. TGF- 1 N.a.N. 1 h 5 h 24 h 72 h 120 h hematoxylin and eosin–stained sections of 3D constructs with H malignant oral keratinocytes CaLH3/CAF5 CaLH3/CAF5 CaLH3/CAF5 (CaLH3) seeded onto CAF-D and +SB 431542 +AB anti-TGFβ-1 24 hours later treated either with SB431542 or anti-TGF-b1 monoclonal antibody for 10 days (magnification, 200; scale bar, 200 mm). I, quantification of the depth of invasion of malignant keratinocytes in 3D constructs with and without addition of TGF- I 140.00 Ctrl. b1 inhibitors. J, Western blot SB 431542 120.00 analysis for pSMAD2 and SMAD2/ m) Anti TGFB1 Ab μ 3 showing inhibition of TGF-b1 100.00 J β Ctrl. TGF- 1 SB AB downstream pathway in Ca1 cells 80.00 treated with with SB431542 but pSMAD2 not when treated with anti-TGF- 1 60.00 b monoclonal antibody. 40.00 SMAD2/3 20.00 Depth of invasion ( invasion of Depth

0.00 Ca1 CaLH3 4PT DOK

CAF-N subpopulation of intrinsically motile fibroblasts fore hyaluronan was next investigated (37). Significantly is dependent on hyaluronan and is essential for higher levels of hyaluronan were found when comparing all supporting carcinoma cell invasion available CAF strains (n ¼ 9) with NOF (n ¼ 3) maintained in Differential intrinsic migratory activity has been linked to 3D biomatrices (P ¼ 0.008; Supplemental Fig. S4A), and the inherent differences in the production of hyaluronan; there- levels of secreted hyaluronan were significantly higher for

OF8 Cancer Res; 73(13) July 1, 2013 Cancer Research

Subtypes of Oral Carcinoma-Associated Fibroblasts

A B CAF-N NOF Figure 4. A, Transwell migration CAF1 CAF3 CAF4 NF1 NF2 NF4 assay showing various fibroblast pMLC strains with distinct migratory characteristics: CAF-N with the highest proportion of intrinsically Smad2 migratory cells and the relatively stationary CAF-D. B, Western blot pSmad2 analysis showing increased levels of % Migratory cells pMLC and pSmad2 in CAF-N when GAPDH compared with NOF, reflecting the presence of a higher proportion of intrinsically migratory fibroblast in CAF-N CAF-D NOF CAF-N. C and D, track speed (C) and C D track displacement (D) of CAF-N and NOF quantified by time-lapse microscopy over 18 hours: CAF-N migrated at higher speed and showed greater displacement (the

shortest path between the start and μ m μ m/s the endpoints of the total cell movement track) than NOF. Statistical signiﬁcant differences between groups, P < 0.05, are highlighted by a star.

CAF-N NOF CAF-N NOF

CAF-N than CAF-D (P ¼ 0.002; Fig. 5A). HAS2 was also significant inhibition of invasion of carcinoma cells in 3D upregulated 3-fold in CAF-N when compared with CAF-D constructs (P ¼ 0.0067; Fig. 5H and I). (Supplementary Table S1.4). Histochemical visualization of hyaluronan by Alcian blue staining showed abundant hya- TGF-b1 activation of NOFs is sufficient to promote luronaninthetumorstromafromwhichCAF-Nstrainswere invasion of transformed oral keratinocytes isolated (Supplementary Fig. S7D), and a fine, discontinuous Upregulation of TGF-b1 target genes in CAF prompted us to deposition of hyaluronan around cords of invasive cells in 3D establish whether this had any functional relevance for CAF constructs with Ca1 cells onto CAF3 (Supplementary Fig. motility and subsequent oral carcinoma cell invasion. Trans- S7E and S7F). IHC and Western blot analysis revealed that well assays revealed that after exposure to 10 ng/mL TGF-b1, a hyaluronan receptors RHAMM (Fig. 5B and C) and CD44 significantly higher proportion of migratory fibroblasts were (data not shown) were expressed by CAF. Blocking hyalur- present in CAF-N and NOF than in CAF-D (Fig. 6A), indicating onan chain elongation by addition of 4-MU resulted in a that CAF-D contained only low numbers of fibroblasts that significant decrease in the proportion of intrinsically migra- migrated in response to exogenous TGF-b1. This was also tory fibroblasts in CAF-N (Fig. 5D), identifying hyaluronan as indicated by the increased expression of Smad2 in TGF-b1– a key factor for the maintenance of this CAF-N migratory treated CAF-N and NOF (Fig. 6B), whereas pSmad2 was almost subpopulation. SEM of control and 4-MU–treated CAF-N undetectable in CAF-D either before or after exposure to revealed a reduction in the number of lamellipodia and exogenous TGF-b1 (data not shown). Broadly, similar percen- filopodia (Fig. 5E). The functional importance of hyaluro- tages of na€ve fibroblasts that migrated in response to exog- nan-dependent subpopulation of intrinsically migratory enous TGF-b1 were observed when CAF-N and NOF were CAF-N for oral carcinoma cell invasion was proven by a compared (Fig. 6A). These data show that CAF-N include, in significant reduction in the depth of invasion of oral carci- addition to the subpopulation of intrinsically motile fibro- noma cells when CAF-N pretreated with 4-MU were incor- blasts, another subset of na€ve fibroblasts that are able to porated into 3D constructs and seeded on top with Ca1 cells become motile after stimulation with exogenous TGF-b1, and (P ¼ 0.001; Fig. 5F and G). 4-MU is described as a selective that these subtypes of motile fibroblasts are largely absent in inhibitor of nonsulfated GlcUA-containing glycosaminogly- CAF-D (Fig. 6C). The functional importance for the TGF-b1– cans, and thus of hyaluronan production, but other effects migratory activation of fibroblasts for carcinoma cell invasion on related glycosaminoglycans cannot be excluded. For this was tested in 3D models by pretreating NOF with 10 ng/mL reason, we also specifically targeted hyaluronan synthesis TGF-b1 for 10 days and embedding afterward these activated with HAS2 shRNA lentiviral particles. qPCR showed a 84.50% fibroblasts into a 3D biomatrix seeded on top with DOK reduction in HAS2 mRNA in HAS2 shRNA-treated CAF3 cells. Nonactivated, matched fibroblasts supported only a compared with control shRNA-treated CAF3 (Supplementa- minimal DOK cell invasion (Fig. 6D), whereas TGF-b1–acti- ry Fig. S8), and this downregulation induced a statistically vated fibroblasts supported significantly deeper invasion in 3D

www.aacrjournals.org Cancer Res; 73(13) July 1, 2013 OF9

Costea et al.

A Secretion of hyaluronan B RHAMM RHAMM 50,000 45,000 40,000 Figure 5. A, CAF secreted

cells 35,000 significantly more hyaluronan 6 30,000 than NOF (P ¼ 0.008) and that 25,000 CAF-N synthesized significantly CAF-N NOF 20,000 C higher levels than CAF-D (P ¼ 15,000 CAF1 CAF3 CAF4 NF1 NF2 NF4 0.002). B, IHC and (C) Western ng/mL per 10 10,000 RHAMM blot analysis confirmed that CAF 5,000 GAPDH 0 and malignant oral keratinocytes NOF CAF-N CAF-D expressed hyaluronan-receptor RHAMM. D, significantly fewer D E CAF1-M CAF1+4MU CAF3-M CAF3+4MU CAF-N migrated through 50 Control Transwell after exposure to 45 4MU-treated 40 4-MU that prevents elongation of 35 hyaluronan chains, P < 0.001. 30 25 E, CAF-N morphology as 20 depicted by SEM showing larger, 15 10 flattened cells with fewer

% Migratory cells 5 filopodia after exposure to 4-MU 0 CAF1 CAF3 NF1 NF4 when compared with untreated CAF-N NOF controls. Top, magnification, F G 1,500; scale bar, 10 mm; 60 Control bottom, magnification, 5,000; 4MU-treated Ca1/CAF1 Ca1/4MU-treated CAF1 50 scale bar, 1 mm. F, panCK IHC 40 showing reduced invasion of Ca1 30 carcinoma cells after CAF-N were 20 exposed to 4-MU, confirmed

10 by measurement of the Depth of invasion ( μ m) depth of invasion (G; 0 CAF1 CAF3 NF1 magnification, 200; scale bar, CAF-N NOF 200 mm). H, representative HIhematoxylin and eosin–stained sections of 3D constructs with malignant oral keratinocytes (Ca1 200.00 shGFP-ctrl and CaLH3) seeded onto ctrl- 180.00 shHAS2 shRNA- and HAS2shRNA- 160.00 treated CAF3 (magnification, 140.00 200; scale bar, 200 mm). I, 120.00 quantification of the depth of 100.00 invasion of malignant 80.00 keratinocytes in 3D constructs 60.00 with ctrl-shRNA- and Depth of invasion ( μ m) 40.00 HAS2shRNA-treated CAF3. 20.00 CaLH3 Ca1 0.00 Ca1 CaLH3 4PT DOK

CAF3 ctrl-shRNA CAF3 HAS2 shRNA

biomatrices (Fig. 6E). The pattern of invasion induced by TGF- CAF-D subtype included few motile cells and the motility b1–activated NOF showed predominantly cords of cells and phenotype was largely unchanged in response to TGF-b1. single cells invading vertically in the biomatrix (Fig. 6F), similar Functional effects of the distinct CAF subtypes were eluci- to the pattern of carcinoma cell invasion induced by CAF-N dated by implanting transformed, nontumorigenic oral kera- (Fig. 6G). tinocytes together with CAF into NOD/SCID IL2rg(null) mice. CAF-N significantly increased tumor incidence and shortened Discussion the lag time for tumor formation as compared with CAF-D. Two distinct CAF subtypes were identified in this study by CAF-N also supported deeper oral carcinoma cell invasion transcriptome and secretome analysis: CAF-N with a gene than CAF-D in 3D models, with areas of noncohesive invasion expression pattern and secretory profile closer to NOF, and as well as vertical proliferation of cords of carcinoma cells, a CAF-D with a divergent transcriptome and secreting very high pattern of invasion previously correlated with poor prognosis levels of TGF-b1. The CAF-N subtype included a high percent- (37). The observation that CAF-N supported the best tumor age of intrinsically motile CAFs, and the number of migratory formation and invasion indicates the pivotal role of CAF cells could be increased in response to TGF-b1, whereas the migration for oral carcinoma cell invasion, in line with previous

OF10 Cancer Res; 73(13) July 1, 2013 Cancer Research

Subtypes of Oral Carcinoma-Associated Fibroblasts

A B 90 CAF-N NOF Ctrl. Control 80 TGF-β1–treated CAF3 CAF3 NF1 NF1 SF SF β β β 70 +TGF- 1 +TGF- 1 +TGF- 1 60 50 pSmad2 40 30 Smad2/3

% Migratory cells 20 10 GAPDH 0 CAF1 CAF3 CAF6 CAF7 NF1 NF4 CAF-N CAF-D NOF

C 100%

90%

80%

70% Stationary/ secretory 60%

50%

40% Migratory after TGF-β1 stimulation/ 30% naïve 20% Migratory/intrin 10% sically activated 0% CAF1 CAF3 CAF6 CAF7 NF1 NF4 CAF-N CAF-D NOF

D DOK/NF16 F TGF-β1-pretreated G DOK/CAF1 NF16

E DOK/TGF-β1-pretreated NF16

Figure 6. A, Transwell migration assay showing CAF distinct response to TGF-b1: CAF-N showed increase of the subpopulation of migratory cells after exposure to TGF-b1, whereas CAF-D were refractory to exposure to TGF-b1. B, Western blot analysis showing increased levels of pSmad2 after 1 hour TGF- b1 exposure of CAF-N and NOF. Skin fibroblasts (SF) were used as control for activation of TGF-b1 downstream pathway. C, the distribution of different cell phenotypes contributing to the heterogeneity of each fibroblast strain analyzed: (i) intrinsically motile cells; (ii) TGF-b1-responsive cells; and (iii) stationary/low-migratory and high-secretory cells. CAF-N and NOF contained significantly higher percentages of intrinsically and TGF-b1–induced motile cells when compared with CAF-D (P ¼ 0.05). CAF-D contained a significantly high percentage of low-motile, highly secretory cells (P < 0.01). D and E, hematoxylin and eosin staining of representative 3D constructs with DOK cells seeded onto control NOF (D) or matched NOF pretreated with TGF-b1(E), showing that TGF-b1–activated NOF supported a deeper invasion (magnification, 200; scale bar, 200 mm). F, closer view (magnification, 400; scale bar, 50 mm) showing invasion of DOK when seeded onto NOF pretreated with TGF-b1, with small islands, single cells, and cords of cells growing deep into the matrix (noncohesive pattern of invasion). G, closer view (magnification, 400; scale bar, 50 mm) showing invasion of DOK when seeded onto CAF-N with a noncohesive, vertical pattern of invasion, similar to the invasion pattern seen when DOK were seeded onto NOF pretreated with TGF-b1. publications (38). However, this was previously linked to the upregulated in the present study (Supplementary Table S1 and requirement for expression of the integrins A3 and A5. These Supplementary Fig. S3) and other reports (39, 40) may substi- integrins were not upregulated in the present study (Supple- tute for A3 and A5. mentary Table S1 and Supplementary Fig. S3), suggesting that Analysis of data from migration assays identified at least alternative integrins, including A6, A10, and A11 that were 3 different phenotypes of fibroblasts contributing to the

www.aacrjournals.org Cancer Res; 73(13) July 1, 2013 OF11

Costea et al.

heterogeneity of each fibroblast strain: (i) intrinsically motile progression, as indicated by our results here. The high secre- cells; (ii) TGF-b1–responsive cells; and (iii) stationary/low- tion of HGF might also maintain low the synthesis of TGF-b1 migratory cells (Fig. 6C). It has been previously shown in NOF; it lowers progressively in CAF-N, and switches that migratory cells are not active synthesizers (41), there- completely to low HGF and high TGF-b1 secretion in CAF- fore it is most likely that the third subpopulation of sta- D (Fig. 3A). The mechanism responsible for this shift is tionary/low migratory fibroblasts is responsible for synthesis currently unknown, as is the mechanism switching from of various growth factors, cytokines, MMPs, and extracellu- TGF-b1–responsive to TGF-b1–refractory CAF. A possible lar matrix (ECM) components including hyaluronan. As explanation might be loss of TGFBR2 by CAF as previously depicted in Fig. 6C, all fibroblast strains were heterogenous, suggested (44), or defects in Smad signaling as found for but presence of these subtypes in various proportions, and malignant keratinocytes (45). Increased expression of genes their cooperative effects was found to be essential for the that bind to, or ameliorate the response to TGF-b1, including functional differences observed between NOF, CAF-N, and BGN and ASPN (31, 46), or PMEPA1 and CTHRC1 that interfere CAF-D. CAF-N were most efficient at maximizing the poten- with phosphorylation of Smads 2/3 (32, 33, 47) may also tial contact with transformed keratinocytes due to the high maintain the distinct CAF subpopulations. proportion of intrinsically migratory fibroblasts that were Nevertheless, secretion of high levels of TGF-b1 not only able to create a network of tracks used by oral transformed induces NOF to evolve into CAF; it can also stimulate invasion cells for invasion, as previously suggested (38). The main- of malignant keratinocytes directly through EMT (48). In line tenance of this subpopulation and its effects on carcinoma with this, our results suggest that CAF-D, being high TGF-b1 cell invasion was dependent on hyaluronan secretion. The secretors, are able to stimulate EMT and oral cancer cell CAF-N strain that secreted the highest levels of hyaluronan invasion via TGF-b1 (Fig. 3). However, this high TGF-b1– (CAF3) supported the deepest invasion of malignant oral secreting CAF phenotype (CAF-D) did not support deep inva- keratinocytes, pinpointing the notion that cooperation sion in our experiments. Migration of keratinocytes deep into between different subpopulations of CAF (e.g., motile cells stroma probably requires additional mechanisms such as a and hyaluronan-synthesizers) results in increased ability to subpopulation of highly motile fibroblasts, and a subpopula- promote carcinoma invasion. The broad similarity between tion of cells that also secrete hyaluronan, MMPs, or growth CAF-N, which supported the highest rate of tumor formation factors that support proliferation (KGF and HGF). In vivo, these and the deepest invasion, and NOF was unexpected but is other subpopulations of cells might be continuously renewed consistent with the notion that na€ve fibroblasts undergo from bone marrow or adjacent normal connective tissue, but adaptations in response to signals in the milieu (2). whether or not the fibroblast subtypes identified in this study The microarray analysis conducted in this study indicated share a common lineage is presently unknown. Highly motile the pivotal role of TGF-b in the stromal adaptations that occur fibroblasts are also known to be present in wounds (49), thus it as oral carcinoma evolves, as we found that approximately 50% will be instructive to tease out differences between wound- (52 of 100) of the top genes upregulated in CAF compared with associated fibroblasts and CAF to establish whether they are NOF were TGF-b targets. The importance of TGF-b1 for CAF broadly similar, or whether CAF have unique characteristics. activation was confirmed by demonstration of significant Most likely, it is the persistence of the transcriptome changes in deeper invasion occurring when TGF-b1–pretreated NOF were fibroblasts in the developing tumor milieu that drives the embedded in 3D constructs and seeded on top with trans- altered balance in the subpopulations identified in the present formed oral keratinocytes, as compared with matched con- study. Our experiments indicate that such changes are of a trols. Most likely, inflammatory cells that are present in vivo in dynamic nature, and the reduced life span of CAF (Supple- the tumor stroma provide TGF-b1 that initiates this process mentary Fig. S1) may indicate that in vivo, the fibroblasts with a (42), but over time, the change in the transcriptome, with changed transcriptome are eliminated and replaced continu- increased expression of many TGF-b1 target genes, leads to ously. Nevertheless, these adaptations support tumor forma- emergence of the nonmotile (41), high TGF-b1–secreting tion and have prognostic significance, and we provide here for fibroblast subtype (14). By correlating Transwell motility assay the first time evidence that CAF subtypes have functional with secretome analysis, we found this subset of high TGF-b1 relevance via different mechanisms to support tumor forma- secretors to be present at significantly higher proportions in tion and invasion. CAF-D strains. This finding might indicate that CAF-N and CAF-D are actually 2 different stages of CAF in the OSCC Disclosure of Potential Conflicts of Interest progression, with CAF-N representing an earlier stromal No potential conflicts of interest were disclosed. change than CAF-D. This is also suggested by the findings that CAF-N are closer to NOF and DAF in many aspects, and Authors' Contributions Conception and design: D.E. Costea, A.H. Osman, J. Thurlow, H. Parajuli, their impact on carcinoma cell invasion is more dramatic, as it S. Suliman, A.C. Johannessen, M. Partridge might be necessary for inducing invasion at early, but not Development of methodology: D.E. Costea, A. Hills, A.H. Osman, J. Thurlow, late stages of carcinogenesis. It has been previously shown H. Parajuli, S. Suliman, K.K. Keerthi, M. Partridge Acquisition of data (provided animals, acquired and managed patients, that increased HGF synthesis suppresses TGF-b1 secretion, provided facilities, etc.): D.E. Costea, A.H. Osman, C.P. Murillo, H. Parajuli, and that changes in the balance between HGF and TGF-b1 S. Suliman, K.K. Keerthi, M. Partridge fi Analysis and interpretation of data (e.g., statistical analysis, biostatistics, synthesis by broblasts might be decisive in pathogenesis of computational analysis): D.E. Costea, A.H. Osman, J. Thurlow, G. Kalna, other diseases (43). Such switch might also occur with OSCC X. Huang, C.P. Murillo, H. Parajuli, S. Suliman, M. Partridge

OF12 Cancer Res; 73(13) July 1, 2013 Cancer Research

Subtypes of Oral Carcinoma-Associated Fibroblasts

Writing, review, and/or revision of the manuscript: D.E. Costea, A.H. for Clinical Odontology, University of Bergen) for assistance with the animal Osman, J. Thurlow, G. Kalna, H. Parajuli, S. Suliman, A.C. Johannessen, experiments. M. Partridge Administrative, technical, or material support (i.e., reporting or orga- nizing data, constructing databases): D.E. Costea, A. Hills, J. Thurlow, Grant Support X. Huang, K.K. Keerthi, A.C. Johannessen, M. Partridge This study was cofunded by King's Medical Research Trust (to M. Partridge), Study supervision: D.E. Costea, A.C. Johannessen, M. Partridge Bergen Medical Research Foundation (to D.E. Costea; grant no. 20/2009) and The Norwegian Cancer Research Association (to D.E. Costea; grant no. 515970/2011). The costs of publication of this article were defrayed in part by the payment of Acknowledgments page charges. This article must therefore be hereby marked advertisement in The authors thank Dr. Hege A. Dale [Molecular Imaging Center, Functional accordance with 18 U.S.C. Section 1734 solely to indicate this fact. Genomics (FUGE) Program, The Research Council of Norway, University of Bergen, Bergen, Norway] for assistance with TEM, SEM and confocal microscopy, and quantiﬁcation of time-lapse microscopy and to E.M. McCormack (Institute Received November 5, 2012; revised March 15, 2013; accepted March 31, 2013; for Internal Medicine, University of Bergen) and Prof. K. Mustafa (Institute published OnlineFirst April 18, 2013.

References

1. Lippman SM, Hong WK. Molecular markers of the risk of oral cancer. extended G2 cell cycle phase that is associated with apoptotic resis- N Engl J Med 2001;344:1323–6. tance. BMC Cancer 2010;10:166. 2. Kalluri R, Zeisberg M. Fibroblasts in cancer. Nat Rev Cancer 2006; 19. Costea DE, Kulasekara K, Neppelberg E, Johannessen AC, Vintermyr 6:392–401. OK. Species-specific fibroblasts required for triggering invasiveness 3. Marsh D, Suchak K, Moutasim KA, Vallath S, Hopper C, Jerjes W, et al. of partially transformed oral keratinocytes. Am J Pathol 2006;168: Stromal features are predictive of disease mortality in oral cancer 1889–97. patients. J Pathol 2011;223:470–81. 20. Breitling R, Armengaud P, Amtmann A, Herzyk P. Rank products: a 4. Tchou J, Kossenkov AV, Chang L, Satija C, Herlyn M, Showe LC, et al. simple, yet powerful, new method to detect differentially regulated Human breast cancer associated fibroblasts exhibit subtype specific genes in replicated microarray experiments. FEBS Lett 2004;573: gene expression profiles. BMC Med Genomics 2012;5:39. 83–92. 5. Lim KP, Cirillo N, Hassona Y, Wei W, Thurlow JK, Cheong SC, et al. 21. Dennis G Jr, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, et al. Fibroblast gene expression profile reflects the stage of tumour DAVID: database for annotation, visualization, and integrated discov- progression in oral squamous cell carcinoma. J Pathol 2011;223: ery. Genome Biol 2003;4:P3. 459–69. 22. Thurlow JK, Pena Murillo CL, Hunter KD, Buffa FM, Patiar S, Betts G, 6. Orimo A, Gupta PB, Sgroi DC, Arenzana-Seisdedos F, Delaunay T, et al. Spectral clustering of microarray data elucidates the roles of Naeem R, et al. Stromal fibroblasts present in invasive human breast microenvironment remodeling and immune responses in survival of carcinomas promote tumor growth and angiogenesis through elevat- head and neck squamous cell carcinoma. J Clin Oncol 2010;28: ed SDF-1/CXCL12 secretion. Cell 2005;121:335–48. 2881–8. 7. Gaballah K, Costea DE, Hills A, Gollin SM, Harrison P, Partridge M. 23. Chang SE, Foster S, Betts D, Marnock WE. DOK, a cell line established Tissue engineering of oral dysplasia. J Pathol 2008;215:280–9. from human dysplastic oral mucosa, shows a partially transformed 8. De Wever O, Mareel M. Role of tissue stroma in cancer cell invasion. non-malignant phenotype. Int J Cancer 1992;52:896–902. J Pathol 2003;200:429–47. 24. Mackenzie IC. Growth of malignant oral epithelial stem cells after 9. Sobral LM, Bufalino A, Lopes MA, Graner E, Salo T, Coletta RD. seeding into organotypical cultures of normal mucosa. J Oral Pathol Myofibroblasts in the stroma of oral cancer promote tumorigenesis Med 2004;33:71–8. via secretion of activin A. Oral Oncol 2011;47:840–6. 25. Harper LJ, Piper K, Common J, Fortune F, Mackenzie IC. Stem cell 10. Sugimoto H, Mundel TM, Kieran MW, Kalluri R. Identification of patterns in cell lines derived from head and neck squamous cell fibroblast heterogeneity in the tumor microenvironment. Cancer Biol carcinoma. J Oral Pathol Med 2007;36:594–603. Ther 2006;5:1640–6. 26. Rheinwald JG, Beckett MA. Tumorigenic keratinocyte lines requiring 11. Bauer M, Su G, Casper C, He R, Rehrauer W, Friedl A. Heterogeneity of anchorage and fibroblast support cultures from human squamous cell gene expression in stromal fibroblasts of human breast carcinomas carcinomas. Cancer Res 1981;41:1657–63. and normal breast. Oncogene 2010;29:1732–40. 27. Sapkota D, Costea DE, Blo M, Bruland O, Lorens JB, Vasstrand EN, 12. Lewis MP, Lygoe KA, Nystrom ML, Anderson WP, Speight PM, et al. S100A14 inhibits proliferation of oral carcinoma derived cells

Marshall JF, et al. Tumour-derived TGF-beta1 modulates myofibro- through G1-arrest. Oral Oncol 2012;48:219–25. blast differentiation and promotes HGF/SF-dependent invasion of 28. McBryan J, Martin FT. A database of TGF-B responsive genes. Dublin, squamous carcinoma cells. Br J Cancer 2004;90:822–32. Ireland: University College Dublin; 2007. 13. Davalos AR, Coppe JP, Campisi J, Desprez PY. Senescent cells as a 29. Nakajima M, Kizawa H, Saitoh M, Kou I, Miyazono K, Ikegawa S. source of inflammatory factors for tumor progression. Cancer Metas- Mechanisms for asporin function and regulation in articular cartilage. tasis Rev 2010;29:273–83. J Biol Chem 2007;282:32185–92. 14. Rosenthal E, McCrory A, Talbert M, Young G, Murphy-Ullrich J, 30. Lopez-Casillas F, Payne HM, Andres JL, Massague J. Betaglycan can Gladson C. Elevated expression of TGF-beta1 in head and act as a dual modulator of TGF-beta access to signaling receptors: neck cancer-associated fibroblasts. Mol Carcinog 2004;40: mapping of ligand binding and GAG attachment sites. J Cell Biol 116–21. 1994;124:557–68. 15. Kiskowski MA, Jackson RS II, Banerjee J, Li X, Kang M, Iturregui JM, 31. Hildebrand A, Romaris M, Rasmussen LM, Heinegard D, Twardzik DR, et al. Role for stromal heterogeneity in prostate tumorigenesis. Cancer Border WA, et al. Interaction of the small interstitial proteoglycans Res 2011;71:3459–70. biglycan, decorin and fibromodulin with transforming growth factor 16. Prime SS, Davies M, Pring M, Paterson IC. The role of TGF-beta in beta. Biochem J 1994;302(Pt 2):527–34. epithelial malignancy and its relevance to the pathogenesis of oral 32. Watanabe Y, Itoh S, Goto T, Ohnishi E, Inamitsu M, Itoh F, et al. cancer (part II). Crit Rev Oral Biol Med 2004;15:337–47. TMEPAI, a transmembrane TGF-beta-inducible protein, sequesters 17. Ramos DM, Dang D, Sadler S. The role of the integrin alpha v beta6 in Smad proteins from active participation in TGF-beta signaling. Mol Cell regulating the epithelial to mesenchymal transition in oral cancer. 2010;37:123–34. Anticancer Res 2009;29:125–30. 33. Li J, Cao J, Li M, Yu Y, Yang Y, Xiao X, et al. Collagen triple helix repeat 18. Harper LJ, Costea DE, Gammon L, Fazil B, Biddle A, Mackenzie IC. containing-1 inhibits transforming growth factor-b1-induced collagen Normal and malignant epithelial cells with stem-like properties have an type I expression in keloid. Br J Dermatol 2011;164:1030–6.

www.aacrjournals.org Cancer Res; 73(13) July 1, 2013 OF13

Costea et al.

34. Cisternas FA, Vincent JB, Scherer SW, Ray PN. Cloning and charac- 42. Bhowmick NA, Neilson EG, Moses HL. Stromal fibroblasts in cancer terization of human CADPS and CADPS2, new members of the Ca2þ- initiation and progression. Nature 2004;432:332–7. dependent activator for secretion protein family. Genomics 2003;81: 43. Mou S, Wang Q, Shi B, Gu L, Ni Z. Hepatocyte growth factor sup- 279–91. presses transforming growth factor-beta-1 and type III collagen in 35. Ostrowski M, Carmo NB, Krumeich S, Fanget I, Raposo G, Savina A, human primary renal fibroblasts. Kaohsiung J Med Sci 2009;25: et al. Rab27a and Rab27b control different steps of the exosome 577–87. secretion pathway. Nat Cell Biol 2010;12:19–30. 44. Meng W, Xia Q, Wu L, Chen S, He X, Zhang L, et al. Downregulation of 36. Kolset SO, Tveit H. Serglycin—structure and biology. Cell Mol Life Sci TGF-beta receptor types II and III in oral squamous cell carcinoma and 2008;65:1073–85. oral carcinoma-associated fibroblasts. BMC Cancer 2011;11:88. 37. Bryne M, Koppang HS, Lilleng R, Kjaerheim A. Malignancy grading of 45. Bornstein S, White R, Malkoski S, Oka M, Han G, Cleaver T, et al. the deep invasive margins of oral squamous cell carcinomas has high Smad4 loss in mice causes spontaneous head and neck cancer with prognostic value. J Pathol 1992;166:375–81. increased genomic instability and inflammation. J Clin Invest 38. Gaggioli C, Hooper S, Hidalgo-Carcedo C, Grosse R, Marshall JF, 2009;119:3408–19. Harrington K, et al. Fibroblast-led collective invasion of carcinoma cells 46. Kizawa H, Kou I, Iida A, Sudo A, Miyamoto Y, Fukuda A, et al. An with differing roles for RhoGTPases in leading and following cells. Nat aspartic acid repeat polymorphism in asporin inhibits chondrogenesis Cell Biol 2007;9:1392–400. and increases susceptibility to osteoarthritis. Nat Genet 2005;37: 39. Navab R, Strumpf D, Bandarchi B, Zhu CQ, Pintilie M, Ramnarine VR, 138–44. et al. Prognostic gene-expression signature of carcinoma-associated 47. LeClair RJ, Durmus T, Wang Q, Pyagay P, Terzic A, Lindner V. Cthrc1 is fibroblasts in non–small cell lung cancer. Proc Natl Acad Sci U S A a novel inhibitor of transforming growth factor-beta signaling and 2011;108:7160–5. neointimal lesion formation. Circ Res 2007;100:826–33. 40. Zhu CQ, Popova SN, Brown ER, Barsyte-Lovejoy D, Navab R, Shih W, 48. Yu C, Liu Y, Huang D, Dai Y, Cai G, Sun J, et al. TGF-beta1 mediates et al. Integrin alpha 11 regulates IGF2 expression in fibroblasts to epithelial to mesenchymal transition via the TGF-beta/Smad pathway enhance tumorigenicity of human non–small-cell lung cancer cells. in squamous cell carcinoma of the head and neck. Oncol Rep Proc Natl Acad Sci U S A 2007;104:11754–9. 2011;25:1581–7. 41. Masur SK, Dewal HS, Dinh TT, Erenburg I, Petridou S. Myofibroblasts 49. Wang C, Tammi M, Guo H, Tammi R. Hyaluronan distribution in the differentiate from fibroblasts when plated at low density. Proc Natl normal epithelium of esophagus, stomach, and colon and their can- Acad Sci U S A 1996;93:4219–23. cers. Am J Pathol 1996;148:1861–9.

OF14 Cancer Res; 73(13) July 1, 2013 Cancer Research

Identification of Two Distinct Carcinoma-Associated Fibroblast Subtypes with Differential Tumor-Promoting Abilities in Oral Squamous Cell Carcinoma

Daniela Elena Costea, Allison Hills, Amani H. Osman, et al.

Cancer Res Published OnlineFirst April 18, 2013.

Updated version Access the most recent version of this article at: doi:10.1158/0008-5472.CAN-12-4150

Supplementary Access the most recent supplemental material at: Material http://cancerres.aacrjournals.org/content/suppl/2013/04/19/0008-5472.CAN-12-4150.DC1

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

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerres.aacrjournals.org/content/early/2013/06/20/0008-5472.CAN-12-4150. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.