Oncogenic role of SFRP2 in p53-mutant osteosarcomadevelopment via autocrine and paracrine mechanismHuensuk Kima,b,c, Seungyeul Yood,e, Ruoji Zhouf,g, An Xuf, Jeffrey M. Bernitza,b,c, Ye Yuana,b,c, Andreia M. Gomesa,b,h,i,Michael G. Daniela,b,c, Jie Sua,b,c,j, Elizabeth G. Demiccok,l, Jun Zhud,e, Kateri A. Moorea,b,c, Dung-Fang Leea,b,f,g,m,n,1,2,Ihor R. Lemischkaa,b,c,o,1,3, and Christoph Schaniela,b,c,o,p,1,2

aDepartment of Cell, Developmental and Regenerative Biology, Icahn School of Medicine at Mount Sinai, New York, NY 10029; bThe Black Family Stem CellInstitute, Icahn School of Medicine at Mount Sinai, New York, NY 10029; cThe Graduate School of Biomedical Sciences, Icahn School of Medicine at MountSinai, New York, NY 10029; dDepartment of Genetics and Genomic Sciences, Icahn School of Medicine at Mount Sinai, New York, NY 10029; eIcahn Institutefor Genomics and Multiscale Biology, Icahn School of Medicine at Mount Sinai, New York, NY 10029; fDepartment of Integrative Biology and Pharmacology,McGovern Medical School, The University of Texas Health Science Center at Houston, Houston, TX 77030; gThe University of Texas MD Anderson CancerCenter, UTHealth Graduate School of Biomedical Sciences, Houston, TX 77030; hCentre for Neuroscience and Cell Biology, University of Coimbra, 3030-789Coimbra, Portugal; iInstitute for Interdisciplinary Research, University of Coimbra, 3030-789 Coimbra, Portugal; jCancer Biology and Genetics Program,Memorial Sloan Kettering Cancer Center, New York, NY 10065; kDepartment of Pathology, Icahn School of Medicine at Mount Sinai, New York, NY 10029;lDepartment of Pathology, Mount Sinai Hospital, Toronto, ON M5G 1X5, Canada; mCenter for Stem Cell and Regenerative Medicine, The Brown FoundationInstitute of Molecular Medicine for the Prevention of Human Diseases, The University of Texas Health Science Center at Houston, Houston, TX 77030;nCenter for Precision Health, School of Biomedical Informatics and School of Public Health, The University of Texas Health Science Center at Houston,Houston, TX 77030; oDepartment of Pharmacological Sciences, Icahn School of Medicine at Mount Sinai, New York, NY 10029; and pMount Sinai Institute forSystems Biomedicine, Icahn School of Medicine at Mount Sinai, New York, NY 10029

Edited by Randall T. Moon, University of Washington, Seattle, WA, and approved October 4, 2018 (received for review August 26, 2018)

Osteosarcoma (OS), the most common primary bone tumor, is highlymetastatic with high chemotherapeutic resistance and poor survivalrates. Using induced pluripotent stem cells (iPSCs) generated from Li–Fraumeni syndrome (LFS) patients, we investigate an oncogenic roleof secreted frizzled-related protein 2 (SFRP2) in p53 mutation-associated OS development. Interestingly, we find that highSFRP2 expression in OS patient samples correlates with poor survival.Systems-level analyses identified that expression of SFRP2 increasesduring LFS OS development and can induce angiogenesis. EctopicSFRP2 overexpression in normal osteoblast precursors is sufficientto suppress normal osteoblast differentiation and to promote OSphenotypes through induction of oncogenic molecules such asFOXM1 and CYR61 in a β-catenin–independent manner. Conversely,inhibition of SFRP2, FOXM1, or CYR61 represses the tumorigenic po-tential. In summary, these findings demonstrate the oncogenic roleof SFRP2 in the development of p53 mutation-associated OS and thatinhibition of SFRP2 is a potential therapeutic strategy.

SFRP2 | p53 | osteosarcoma | autocrine | paracrine

Osteosarcoma (OS) is the most common primary bone tumor.It accounts for about 60% of all primary bone tumors and

about 2% of all childhood cancers (1, 2). Despite significantadvances in OS treatment modalities, the 5-y overall survival ratehas remained stable over the last 20 y at 60–70% for patientswith primary OS and less than 30% for patients with metastasis(3, 4). This stagnation of clinical outcomes underlines the urgentnecessity for novel model systems to study the mechanism of OSin a patient-specific context and to identify molecular targets forthe development of new therapeutic strategies.The tumor suppressor p53 regulates cell cycle, apoptosis, se-

nescence, metabolism, and cell differentiation (5–7). Therefore, itis not surprising that aberrant p53 expression contributes signifi-cantly to cancer development (8, 9). Half of all human sporadicbone tumors have genetic lesions in TP53 (10, 11). Patients withLi–Fraumeni syndrome (LFS), which is caused by mutations inTP53, show a 500-fold higher incidence of OS relative to thegeneral population (12–14). Manipulation of p53 function con-firmed its significance in OS development (15, 16) and identifiedmesenchymal stem cells (MSCs) and preosteoblasts (pre-OBs) asthe cellular origin of OS (17, 18). Osteoblast (OB)-restricted de-letion of p53/Mdm2 or p53/Rb resulted in OS development at ahigh penetrance of about 60% and 100%, respectively (19, 20).The first secreted frizzled-related protein (SFRP) was identi-

fied as a WNT antagonist (21). As a known WNT antagonist,SFRP2 is considered a tumor suppressor. Indeed, several reports

showed that SFRP2 hypermethylation and its decreased ex-pression are associated with prostate, liver, colorectal, and gas-tric cancer (22–27). Originally, SFRP2 was reported as a secretedantiapoptosis-related protein (28, 29). Ectopic expression ofSFRP2 promotes cell growth and has antiapoptotic properties inrenal and breast cancer (30–32). The role of SFRP2 appears tobe cancer-type specific and remains controversial. Thus, in-vestigation and understanding of the role of SFRP2 in differenttypes of cancer, including OS, is warranted.Using induced pluripotent stem cells (iPSCs) derived from

LFS patients, we previously recapitulated the pathophysiological

Significance

Li–Fraumeni syndrome is a rare disorder caused by germlineTP53 mutations, predisposing patients to early-onset cancers,including osteosarcoma (OS). Here we demonstrate that strongexpression of SFRP2, a reported WNT antagonist, in OS patientsamples correlates with poor survival and that SFRP2 over-expression suppresses normal osteoblast differentiation, pro-motes OS features, and facilitates angiogenesis via autocrineand paracrine mechanisms in an induced pluripotent stem celldisease model. We show that these SFRP2-mediated phenotypesare canonical WNT/β-catenin independent and are mediatedthrough induction of oncogenes such as FOXM1 and CYR61. Wefurther demonstrate that inhibition of SFRP2, FOXM1, or CYR61represses tumorigenesis. Our data suggest that inhibition ofSFRP2 should be explored clinically as a strategy for treatmentpatients with p53 mutation-associated OS.

Author contributions: H.K., D.-F.L., I.R.L., and C.S. designed research; H.K., R.Z., A.X.,J.M.B., Y.Y., A.M.G., M.G.D., J.S., E.G.D., and K.A.M. performed research; K.A.M. contrib-uted new reagents/analytic tools; H.K., S.Y., E.G.D., J.Z., and D.-F.L. analyzed data; andH.K., S.Y., D.-F.L., I.R.L., and C.S. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Published under the PNAS license.

Data deposition: The RNA-sequencing data have been deposited in the Gene ExpressionOmnibus databank (accession nos. GSE102729 and GSE102732).1D.-F.L., I.R.L., and C.S. contributed equally to this work.2To whom correspondence may be addressed. Email: dung-fang.lee@uth.tmc.edu orchristoph.schaniel@mssm.edu.

3Deceased December 7, 2017.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1814044115/-/DCSupplemental.

Published online November 1, 2018.

E11128–E11137 | PNAS | vol. 115 | no. 47 www.pnas.org/cgi/doi/10.1073/pnas.1814044115

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

features of LFS-mediated OS development (33, 34). Taking ad-vantage of this platform, we observed increased expression ofSFRP2 during LFS iPSC-derived OB differentiation. As a result ofthese findings and because the exact function of SFPR2 in OS isnot clear, we investigated its role in LFS p53 mutation-mediatedabnormal OB differentiation, tumorigenesis, and OS develop-ment. Here, we report that SFRP2 overexpression (SFRP2OE)induces OS phenotypes, increases FOXM1 expression, and pro-motes angiogenesis and endothelial expression of the matricellularprotein CYR61. Conversely, targeting SFRP2OE in LFS and OShas therapeutic promise for OS subtypes with p53 mutations.

ResultsSFRP2OE Is Associated with p53 Mutation-Mediated Human OSDevelopment. To discover potential therapeutic targets for LFS-mediated OS, we compared the genome-wide transcripts of theLFS dataset (GSE58123) composed of MSCs differentiated to OBsin vitro from two LFS (P53p.G245D) patient iPSC lines, LFS1-Aand LFS2-B, and one control iPSC line, WT-1 (SI Appendix, TableS1) (33), with an OS tumor-initiating cell dataset (GSE33458) (Fig.1A) (35). We first identified differentially expressed genes (DEGs)between LFS and WT cells in the GSE58123 dataset. Because thedataset includes only one sample for each differentiation timepoint (D0, D7, D14, and D17), we performed a paired t test be-tween each of the two LFS patient iPSC-derived samples withWT cells and identified DEGs common to both LFS samples withrespect to WT cells (fold change >2, paired t test P < 0.01)(Dataset S1). This method enabled extraction of consistently up- ordown-regulated DEGs (Fig. 1A). We identified SFRP2, whichshowed the greatest fold change (>30), as an overexpressed genewith potential implication in OS development. Quantitative PCRand Western blot analyses confirmed that SFRP2 expression issignificantly overexpressed in OBs derived from different LFS iPSClines compared with WT controls (Fig. 1B). We confirmed thecorrelation between the P53p.G245D mutation and SFRP2 ex-pression by introducing the mutation into WT MSCs (SI Appendix,Fig. S1A). SFRP2 expression was significantly increased in WTMSCs edited to harbor the P53p.G245D mutation but not in WTMSCs with p53 depletion (shRNAs to p53) (SI Appendix, Fig.S1B). Fibroblasts from LFS patients with the G245D mutation alsohave higher SFRP2 expression (SI Appendix, Fig. S1C). IncreasedSFRP2 expression is not unique to P53p.G245D. OBs derived fromLFS iPSCs carrying a heterozygous P53p.Y205C mutation andfrom various p53-mutant embryonic stem cells (ESCs) (G245D,G245S, R248W, and R249S) overexpress SFRP2 compared withtheir WT counterparts (SI Appendix, Fig. S1 D and E). Addition-ally, SFRP2 expression was detected in several OS cell lines har-boring p53 mutations (HOS, MNNG, and 143B) or MDM2amplification (SJSA-1) but not in the p53 WT OS cell line U2OSor the p53-null line SaOS2 (Fig. 1C). SFRP2 was also detected ins.c. tumors obtained from mouse xenograft assays with LFS celllines (Fig. 1D) (33). These results demonstrate that SFRP2OE isinduced by LFS-associated p53 mutations.Next, we analyzed the expression of SFRP2 levels in human OS

tissue samples using a tissue microarray spotted with 151 OS pa-tient tissue samples. The OS patient samples were categorized bySFRP2 staining intensity: Groups 0 and 1were considered to benegative to low intensity, and groups 2 and 3 were high intensity(Fig. 1E and SI Appendix, Fig. S1F). Interestingly, high SFRP2expression (intensity scores 2 and 3, with 3 being the highest ex-pression) negatively correlated with patient survival (Fig. 1F andSI Appendix, Fig. S1G and Table S2). As the patients’ p53 statusassociated with this OS tissue array is not available, a correlationbetween SFRP2 expression and p53 mutation is not possible.There was no gender difference in tumor incidence (SI Appendix,Fig. S1H), and the area staining positive for SFRP2 is small in thegroup with low SFRP2 expression (SI Appendix, Fig. S1I). Thesedata indicate that SFPR2OE is associated with poor prognosis forOS patients.

Global Gene-Expression Analysis of SFRP2OE. To better understandthe molecular role of SFRP2OE in LFS cells, we generatedglobal gene-expression profiles of WT SFRP2OE cells, LFS cellswith SFRP2 knockdown (SFRP2KD), parental WT cells (WT2-C), and LFS cells (LFS2-B) at four time points (D0, D7, D14,and D17) during in vitro OB differentiation (SI Appendix, Fig.S2A). We used lentiviral vectors for doxycycline-inducible over-expression (36) and for shRNA depletion (37) to generate stableSFRP2OE or SFRP2KD cell lines, respectively, (SI Appendix,Fig. S2B). Principal component analysis (PCA) showed that WTand LFS cells are clearly distinct from each other (Fig. 2A),consistent with our previous LFS RNA-sequencing (RNA-seq)data (33). Expression patterns of SFRP2OE cells were also dis-tinct from those of WT cells and were more evident at laterstages of differentiation (Fig. 2A). Likewise, LFS datasets di-verged more strongly at late stages of differentiation, whileSFRP2KD data clustered together and were distinct from LFSgroups (Fig. 2A). This result shows that SFRP2OE significantlyalters gene expression in pre-OBs during differentiation. Linear

Fig. 1. SFRP2OE in LFS (G245D)-mediated OS. (A, Left) Global transcriptomeanalysis of LFS P53p.G245D- and WT iPSCs-derived MSCs differentiated to OBs.(Right) DEGs between LFS and WT cells were sorted by paired t test (P < 0.01)with a fold change >2. SFRP2 is an overexpressed gene that is also enriched inthe signature gene list of an OS gene set (GSE33458). (B) Quantitative PCRanalysis (Right) of SFRP2 expression (D4 of differentiation; data are shown asmean ± SD; n = 3 independent repeats in triplicate) in LFS P53p.G245D andWTMSCs (*P < 0.05; **P <0.0001; ANOVA). The Inset depicts Western blottingusing mouse monoclonal anti-SFRP2 antibody (catalog no. sc-365524; SantaCruz Biotechnology). (C) SFRP2 expression in human WT and LFS P53p.G245DiPSC-derived OBs (D4 of differentiation) and OS cell lines (data are shown asmean ± SD; n = 3 independent repeats in triplicate; **P < 0.0001, ANOVA). (D,Upper) Expression of SFRP2 protein in LFS P53p.G245D s.c. xenograft tumors.(Lower) Human prostate tissue was used as the positive staining control, andmouse muscle tissue was used as the negative staining control. (Scale bars:100 μm.) (E) Representative staining of SFRP2 protein expression levels in OStissue samples from 151 OS patients. In D and E, rabbit polyclonal anti-SFRP2(catalog no. PA5-29390; Thermo Fisher Scientific) was used. (F) Survival anal-ysis of OS patients linked to the tissue microarray. The survival curves ofSFRP2 high- and low-expression groups were calculated using the log-ranktest (χ2 = 11.13) (#P = 0.0009).

Kim et al. PNAS | vol. 115 | no. 47 | E11129

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

regression with controlling for time identified DEGs betweenSFRP2OE and WT, between LFS and WT, and betweenSFRP2KD and LFS groups [P < 0.001, corresponding permu-tation false-discovery rate (FDR) <0.01] (Fig. 2B and DatasetsS2–S4). Database for Annotation, Visualization and IntegratedDiscovery (DAVID) functional annotation was used to un-derstand the general biological effect of DEGs associated withSFRP2OE. WT signature genes (DEGs during OB differentia-tion) were generally involved in bone mineralization and bonemorphogenetic protein (BMP) and WNT signaling pathwaysrequired for normal OB differentiation (Fig. 2C). Interestingly,SFRP2OE signature genes were linked to cell proliferation, celladhesion, and cell migration, categories implicated in the tu-morigenic role of SFRP2 (31, 38, 39), whereas SFRP2KD sig-nature genes restored bone mineralization, apoptosis process,and negative regulation of cell proliferation (Fig. 2C). MouseGene Atlas analysis (40) of DEGs from WT, SFRP2OE, andSFRP2KD signatures confirmed that SFRP2OE significantlyaffects differentiation, while SFRP2KD recovered normal OBdifferentiation expression signatures (SI Appendix, Fig. S2C).To determine whether SFRP2OE is associated with oncogenic

molecular features of LFS, we compared SFRP2OE with LFS andSFRP2KD signature genes. There were 308 up-regulated and478 down-regulated genes common between SFRP2OE and LFSexpression signatures (Fisher’s exact test, P = 1.4 × 10−171 and2.27 × 10−160, respectively), while 277 down-regulated and 264 up-regulated genes were inversely overlapped between SFRP2KD andLFS expression signatures (Fisher’s exact test, P = 2.13 × 10−130

and P = 2.8 × 10−127, respectively) (SI Appendix, Table S3). Forgene functional analysis, we used the gene set enrichment analysis(GSEA) tool Enrichr (41, 42). Gene annotation analysis showedthat the up-regulated genes common to SFRP2OE and LFScompared with WT include cell-cycle and well-known oncogenicgenes frequently found in cancers such as AURKB, BUB1B, andFOXM1 (Fig. 2 D and E). Interestingly, Enrichr-embedded ChIPenrichment analysis (41–43) revealed that LFS/SFRP2OE up-regulated genes were enriched for FOXM1 targets (P = 1.52 ×10−16) (Fig. 2F). This suggests a correlation between SFRP2OE-mediated cell proliferation and the oncogenic function of FOXM1target genes in human OS, irrespective of a p53 mutation. Common

LFS/SFRP2OE down-regulated genes are associated with extra-cellular matrix and cell–cell adhesion (SI Appendix, Fig. S2D). In-terestingly, tumor-suppressor genes such as CADM1 andCADM4 [tumor-suppressor gene database (44)] were enrichedin the down-regulated gene set (Fisher’s exact test, P = 0.004) (SIAppendix, Fig. S2D).Finally, we compared SFRP2OE DEGs with the human OS Gene

Expression Omnibus (GEO) set (GSE36001) to determine the cor-relation with OS features through GSEA. SFRP2OE and LFS DEGswere highly correlated with human OS signature genes, whereasSFRP2KD andWTDEGs were highly concordant with OB signaturegenes (Fig. 2G and SI Appendix, Fig. S2E). Additionally, SFRP2OEsignature genes were significantly enriched in up-regulated genes ofthe NCI-60 panel of cell lines harboring TP53 mutations and in theRb1-knockout mouse, suggesting that SFRP2 strongly contributes tothe signature of well-known OS genetic lesions, RB1 deletion, andTP53 mutation (Fig. 2H and SI Appendix, Fig. S2F). Collectively, ourtranscriptome analyses revealed tumorigenic properties of SFRP2OEin LFS and OS.

SFRP2OE Dysregulates OB Differentiation Through BMP/WNTSuppression. We reported that LFS iPSC-derived MSCs andpre-OBs have defects in OB differentiation, supporting the ideathat compromised p53 dysregulates OB differentiation andpromotes the generation of abnormal MSCs and pre-OBs, thecells of origin of OS (33). Because our global transcriptomeanalyses provided evidence that SFRP2OE potentially perturbsOB differentiation (Fig. 2C), we wondered if SFRP2OE in LFSsuppresses normal OB differentiation as an oncogenic down-stream modulator of mutant p53. We performed alkaline phos-phatase (AP) and Alizarin red S staining on LFS, SFRP2OE, andWT cells at different OB differentiation time points (D0–D21).Normal OBs exert a high level of AP activity and secrete largeamounts of calcium (mineral deposition) detected by Alizarinred S staining (45–47). SFRP2OE in WT cells clearly suppressednormal OB differentiation, similar to its effect in LFS cells (Fig.3 A and B). In contrast, SFRP2KD in LFS cells rescued osteo-genic AP activity and mineral deposition (SI Appendix, Fig. S3 Aand B). In agreement with this observation, LFS and SFRP2OEOBs had lower levels of pre-OB markers (ALPL and COL1A1),

PC1

0

PC2

PC320

40

D7D7

D14

D17

D0D14

D17

D0

D7

D14

D7D14

D17

D0

D0D17

-40 -20 0 20 40 200 -20

LFS

SFRP2KD

SFRP2OE

WT

−2 2

Row Z−Score

WT SFRP2OE LFS SFRP2KD

755

517

542

584

Negative muscle proliferationBone mineralization

WNT pathway

BMP pathway

WT OBs

p-v

alu

e (-

log

10)

0 1

2

3

4 5

6

0 714 17 0 7 14 17 0 714 17 0 7 14 17

0

1

2

3

4

5

6

7

SFRP2OE

Cell adhesionPositive cell migration

Inflammatory response

Positive cell proliferation 0

1

2

3

4

5

6

7

Negative cell proliferation

Negative cell migration

Apoptosis

Postive bone migration

SFRP2KD

SFRP2OE_up

LFS_up

SFRP2KD_down

259 203

45263

2194

232 LFS_up & LFS_up & SFRP2KD_downSFRP2KD_down

Extracellular matrix organization: (p=8.0×10-6)Extracellular structure organization: (p=2.0×10-4)

LFS_up & LFS_up & SFRP2OE_upSFRP2OE_upMitotic cell cycle (p=1.8×10-10)Mitotic nuclear division (p=6.5×10-9)

WT

SFRP2OELFS

BUB1BNCAPGFOXM1PKMYT1AURKBNCAPHNUF2BUB1

CDC25CNDC80KIF2CSPC24FAM83DBRINP1ESPL1CHEK1NUSAP1GTSE1DLGAP5EZH2

1

2

3

4**

*

***

***

*

Log2(RPKM

)

FOXM1

ChIP Enrichment

Analysis

A B C

D E F G HSFRP2OE signatures

Up-regulated in mutated TP53

p-value<0.001NES=1.23

p-value<0.001NES=1.74

-0.2

0

.0

0

.3 0

.5E

nri

chm

ent

Sco

re (

ES

)

-0.2

0

.0

0.3

OS OB UpSFRP2OE

DownSFRP2OE

Fig. 2. Global gene-expression analysis reveals that SFRP2OE induces cell proliferation and perturbs the differentiation process. (A) PCA of RNA-seq datafrom SFRP2OE and SFRP2KD samples [proportion of variance; principal component 1 (PC1) = 0.4020; principal component 2 (PC2) = 0.1839; principal com-ponent 3 (PC3) = 0.0903]. (B) Heatmaps of DEGs in SFRP2OE and SFRP2KD samples (DEG cutoff P value < 0.001). (C) DAVID gene ontology (GO) term biologicalprocess analysis of DEGs from WT, SFRP2OE, and SFRP2KD cells. The top 20 GO terms of DEGs from each group are presented in the panels. The cutoff at P =0.05 is indicated by dashed lines. (D) Enrichr GO biological process analysis of common genes among SFRP2OE up-regulated genes, LFS up-regulated genes,and SFRP2KD down-regulated genes. (E) Relative expression of cell-cycle– and cell-proliferation–related DEGs up-regulated in both SFRP2OE and LFS cells.(Asterisks indicate cancer-associated genes.) (F) Common up-regulated DEGs are enriched in FOXM1 ChIP (in U2OS) data (Enrichr). (G) GSEA of the SFRP2OEsignature with the OS gene set (GSE36001). (H) GSEA of the SFRP2OE signature with the TP53 mutation gene set (MSigDB C6 analysis).

E11130 | www.pnas.org/cgi/doi/10.1073/pnas.1814044115 Kim et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

mature OB markers (BGLAP and PTH1R), osteocyte markers(ZNF521 and FGF23), and OB differentiation transcriptionalfactors (RUNX2 and OSX) (Fig. 3C). Interestingly, ATF4,whose expression is high in OS tissue and OS cell lines (48), wasup-regulated in SFRP2OE and LFS OBs (Fig. 3C). These re-sults, together with our finding that LFS4 (P53.pY205C)-derivedOBs also showed reduced mineral deposition (SI Appendix, Fig.S3C), indicate that SFRP2OE as a consequence of mutantp53 perturbs normal osteogenic processes, which is consistentwith the RNA-seq biological process analysis (Fig. 2C).The BMP/SMAD and canonical WNT pathways are required

for normal osteogenesis (49–51). In particular, BMP ligands fa-cilitate OB differentiation through the BMP/SMAD pathway(50, 52, 53). Using a luciferase-based phospho-SMAD reportersystem, we monitored BMP signaling activity throughout in vitrodifferentiation (D0–D14). While BMP signaling was relativelylow in MSCs and progenitors (D4), it increased in pre-OBs(D10–D14) (Fig. 3D). This result is consistent with publishedin vivo data (52, 54). As expected, SFRP2OE and LFS cellsshowed lower BMP signaling activity than WT OBs. It supportsthe idea that SFRP2OE perturbs normal OB differentiation inpart through inhibition of the BMP pathway.The effects of SFRP2 on canonical WNT signaling are context

dependent. Generally, SFRP2 suppresses WNT signaling bybinding WNT ligands, thus preventing their binding to WNTreceptors (30, 55, 56). However, SFRP2 has also been shown toactivate WNT signaling at low concentrations or in certain celltypes (35, 57–59). To understand the role of SFRP2 in canonicalWNT signaling in OS development, we performed Western blotand β-catenin immunostaining. Compared with WT cells, LFS,SFRP2OE, and HOS cells consistently had lower levels of cy-toplasmic β-catenin (SI Appendix, Fig. S3D), while the amount ofnuclear β-catenin increased only in WT groups with WNT3a

treatment (SI Appendix, Fig. S3E). Using the TOP/FOP flashluciferase WNT signaling reporter system, we observed low basalWNT activity in all groups (Fig. 3E, no treatment). We activatedWNT signaling using WNT3a (100 ng/mL) or LiCl (10 mM),which inhibits GSKβ. SFRP2OE showed a reduction in WNT3a-mediated canonical WNT activity (Fig. 3E). LFS and HOS cellsexhibited little or no response to either LiCl or WNT3a (Fig. 3Eand SI Appendix, Fig. S3F). Ectopic expression of constitutivelyactive β-catenin (DeltaN90) in LFS did not facilitate cell pro-liferation or promote cell transformation (SI Appendix, Fig.S3G). However, knockdown of β-catenin led to severe cell deathof LFS cells, confirming that β-catenin is necessary for homeo-stasis and viability of MSCs and OBs (SI Appendix, Fig. S3H) (55,56). Although it is widely accepted that hyper-WNT signaling isassociated with oncogenicity in many cancers, our data supportthe idea that hyper-WNT is not always involved in osteosarco-magenesis. This is consistent with the observation of inactivecanonical WNT activity in high-grade human OS (57).

SFRP2OE Transforms Pre-OBs into OS-Like Cells.We next investigatedthe oncogenic contribution of SFRP2OE by analyzing cell mor-phology and cell proliferation in ectopic SFRP2OE and SFRP2KDconditions. SFRP2OE in WT cells resulted in the development ofaggregated structures, which is reminiscent of tumor cell aggrega-tion, and was enhanced in ectopic SFRP2OE in LFS cells at D14–D17 of OB differentiation (Fig. 4A). Overexpression also increasedproliferation during MSC culture and OB differentiation (Fig. 4Band SI Appendix, Fig. S4A). In contrast, SFRP2KD cells did notform aggregated structures and showed decreased cell proliferationcompared with LFS cells (Fig. 4B and SI Appendix, Fig. S4B).It was reported that ectopic SFRP2 expression alters the G2

phase of the cell cycle and suppresses apoptosis (32). Our analysisof both cell cycle and apoptosis on D4 of OB differentiation

Fig. 3. SFRP2OE suppresses OB differentiation in vitro. (A) Osteogenic AP analysis in LFS P53p.G245D-, SFRP2OE-, and reverse tetracycline-controlled transactivator(rtTA)-expressingWT-derived pre-OB cells. (B) Alizarin red S staining in LFS P53p.G245D-, SFRP2OE-, and rtTA-expressing WT-derived pre-OB cells. (C) Quantitative PCRanalysis of the indicated osteogenic markers in LFS P53p.G245D, SFRP2OE, and WT cells (data are shown as mean ± SD; n = 3 independent repeats in triplicate; *P <0.05; ANOVA). (D) BMP signaling activity in LFS P53p.G245D, SFRP2OE, and WT cells as measured by a BMP reporter (pSBE3-Luc) (data are shown as mean ± SD; n =3 independent repeats in triplicate; **P < 0.0001; ANOVA). (E) WNT signaling activity as measured by TOP/FOP flash luciferase reporter. Activity levels were normalizedusing an internal control (dual luciferase assay system). Data are presented as mean ± SD (n = 3 independent repeats in triplicate; *P < 0.05; ANOVA).

Kim et al. PNAS | vol. 115 | no. 47 | E11131

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

revealed significantly more SFRP2OE cells in S phase (72.17 ±3.22%) compared with WT cells (29.82 ± 1.68%) (Fig. 4C and SIAppendix, Fig. S4C). This implies that SFRP2OE promotes S-phaseentry, cell-cycle progression, and, thus, cell proliferation. In contrast,SFRP2KD resulted in significantly less cell proliferation with moreSFRP2KD cells in G0/G1 (21.51 ± 2.68%) and fewer in S phase(77.36 ± 2.86%) compared with LFS cells (G0/G1: 8.60 ± 4.32%; Sphase: 90.51 ± 1.99%) (Fig. 4 B and C and SI Appendix, Fig. S4 Band D). No statistical difference in apoptosis was found among thevarious groups (Fig. 4C). Collectively, we observed that SFRP2OEpromotes S-phase entry at the pre-OB stage.Anchorage-independent growth is a hallmark of cancer cell

transformation. We used the soft agar colony-formation assay todetermine the transformation ability of SFRP2. SFRP2OE andLFS cells were able to form oncogenic colonies (Fig. 4D and SIAppendix, Fig. S4 E and F). Both the number and size of coloniesformed by SFRP2OE and LFS cells were significantly greaterthan those of WT cells. In contrast, SFRP2KD cells formedsignificantly fewer colonies than LFS cells (Fig. 4E). These dataclearly demonstrate the tumorigenic ability of SFRP2.To determine the tumorigenicity of SFRP2OE, we performed

in ovo chick chorioallantoic membrane (CAM) assays (SI Ap-pendix, Fig. S4G). This assay provides a convenient and commonway to check the oncogenicity of sarcoma cells (58, 59).SFRP2OE cells showed increased tumor formation and cellnumbers compared with WT cells (Fig. 4F). SFRP2KD cells hadsuppressed tumor formation compared with LFS cells (Fig. 4G).Since SFRP2 is a secreted molecule, we wondered if a blockingantibody would reduce OS features. We used the well-documented p53 mutant osteosarcoma cell line HOS. As hy-pothesized, adding a monoclonal anti-SFRP2 antibody to HOS

cells significantly reduced tumorigenesis (Fig. 4H). In serialtransplantation analysis, SFRP2OE cells maintained theirtumor-formation ability (SI Appendix, Fig. S4H), while the anti-body in HOS consistently suppressed it over three consecutivetransplantations (SI Appendix, Fig. S4I). Additionally, using theCAM assay, we checked whether FOXM1 inhibition could sup-press tumor formation. We used FOXM1 shRNA (FOXM1KD)and the small molecule FDI-6, a reported potent inhibitor ofFOXM1 (60). The result showed that both FOXM1KD andFDI-6 treatment effectively decreased SFRP2-mediated tumorformation (SI Appendix, Fig. S4 J and K). These in ovo datademonstrate that SFRP2OE contributes to LFS-mediated tu-morigenesis and OS development and that targeting SFRP2 withan antibody suppresses OS formation.

Secreted SFRP2 Facilitates Angiogenesis and Tumorigenesis ThroughCYR61 Induction in Endothelial Cells. Angiogenesis is a critical com-ponent of solid tumor growth. SFRP2 has been shown to functionas an angiogenic factor in human triple-negative breast cancer,angiosarcoma, and melanoma (38, 58, 61). However, whetherSFRP2 has any angiogenic properties in OS remains unknown.Because SFRP2 is a secreted protein, we hypothesized that MSCsand pre-OBs in LFS patients could secrete SFRP2 into the bonematrix where it initiates migration and sprouting of endothelial cells,thereby facilitating neoangiogenesis. To investigate this, we con-ducted an in vitro angiogenesis assay; we cultured human vascularendothelial cells (HUVECs) on Matrigel in WT-, LFS-, orSFRP2OE-conditioned medium (CM) for 16 h and then quantifiedthe tube formation rate. As expected, WT CM was not able tosignificantly induce tube formation (Fig. 5 A and B). In contrast,LFS and SFRP2OE CM dramatically increased tube formation, as

Fig. 4. SFRP2OE in pre-OBs induces cell transformation. (A) Morphological changes in WT/SFRP2OE and LFS (LFS2-B)/SFRP2OE cells (red arrow indicatesaggregated cells). (Scale bars: 100 μm.) (B) Cell-proliferation analysis using the CyQUANT cell fluorescent DNA content assay of WT cells expressing rtTA andWT SFRP2OE cells (Left) and LFS (LFS2-B expressing empty shRNA) and SFRP2KD cells (Right) [data are shown as mean ± SD (n = 4; *P < 0.05; **P < 0.0001, two-way ANOVA)]. (C) Cell-cycle analysis of rtTA-expressing WT and SFRP2OE cells (Left) and LFS (LFS2-B) and SFRP2KD (SFRP2 shRNA-1) cells (Right) using BrdU/7-AAD at D4 of OB differentiation (data are shown as mean ± SD; n = 3 independent repeats in duplicate; **P < 0.0001, two-way ANOVA). (D and E, Upper) Softagar colony formation of rtTA-expressing WT and SFRP2OE cells (Dox, doxycycline) (D) and LFS2-B (LFS), LFS2-B with empty shRNA (Empty), SFRP2 shRNA-1(KD-1) cells (E). Colonies over 50 μm in size were included in the quantification of colony numbers. (Scale bars: 100 μm.) (Lower) Quantification. Data areexpressed as mean ± SD; each symbol represents an individual colony >50 μm in size; **P < 0.0001, one-way ANOVA). (F–H) CAM assays (Upper) and analysis(Lower) of rtTA-expressing WT and SFRP2OE cells (F), LFS (LFS2-B) expressing empty shRNA (Empty) and SFRP2 shRNA-1 (SFRP2KD-1) cells (G), and HOS cellstreated with SFRP2 neutralizing antibody (catalog no. sc-365524; Santa Cruz Biotechnology) (H). White ring diameter, 10 mm. White lines outline tumors. In Fand G, representative images from two independent CAMs are shown for each condition. In G and H, D7 OBs were used. (F–H, Lower) Analysis (data areshown as mean ± SD; each symbol represents CAM of an individual egg; *P < 0.05; **P < 0.0001, one-way ANOVA in F and unpaired t test in G and H).

E11132 | www.pnas.org/cgi/doi/10.1073/pnas.1814044115 Kim et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

did the addition of recombinant SFRP2 protein to WT CM (Fig. 5A and B). Moreover, SFRP2KD CM dramatically decreased tubeformation to the levels of the negative control and WT CM. Thisresult is consistent with other reports showing the angiogenicfunction of SFRP2 in angiosarcoma and melanoma (38, 61).To investigate how SFRP2 increases angiogenesis, we performed

RNA-seq profiling of HUVECs after treatment with LFS,SFRP2OE, and WT CM. PCA with triplicate samples from eachcondition showed that the replicates clustered together but weredistinct among the conditions (Fig. 5C). To identify SFRP2-mediated effects on angiogenesis induction, we focused on iden-tifying and analyzing DEGs between SFRP2OE and WT condi-tions that are also differentially expressed between LFS and WTconditions. Using DESeq2 (Padj < 0.01), we identified 498 com-mon genes (258 up-regulated and 240 down-regulated) between1,826 LFS and 969 SFRP2OE DEGs (Fig. 5D and SI Appendix,Fig. S5 A and B). DAVID functional annotation of the 258 up-regulated common genes showed that secreted SFRP2 mainlyalters angiogenesis, antiapoptosis, and inflammatory responseprocesses of endothelial cells (Fig. 5E). Most up-regulated angio-genic genes are either well-known angiogenic markers or procancergenes such as VEGFC, CYR61, ANGPT2, VASP, EPAS1, JUNB,

and CDH13 (Fig. 5F). High expression levels of CYR61, known tofacilitate angiogenesis, were recently associated with poor prog-nosis in OS (62, 63). Interestingly, we also observed increased ex-pression of CYR61 in endothelial cells treated with SFRP2OE andLFS CM (Fig. 5 F and G). Thus, we investigated if SFRP2OEincreases angiogenesis and tumor formation through an action ofCYR61. Measuring the level of neovascularization and tumor sizein a CAM assay, we found that CYR61 antibody treatment de-creases angiogenesis and tumor formation in the presence ofSFRP2 (Fig. 5 H–J). This supports the idea that secreted SFRP2facilitates angiogenesis of endothelial cells and tumorigenesis, atleast in part, through CYR61.Our RNA-seq analysis revealed the up-regulation of antiapoptosis-

related genes in HUVECs by LFS and SFRP2OE CM (SI Appendix,Fig. S5D). Indeed, the number of viable HUVECs was higher inSFRP2OE CM and LFS CM than in WT CM (SI Appendix, Fig.S5E). These results suggest that SFRP2 facilitates angiogenesis andprotects endothelial cells from apoptosis by inducing the expressionof angiogenic/oncogenic factors and antiapoptosis regulators, re-spectively. Intriguingly, cell-cycle–related DEGs in HUVECs treatedwith LFS CM and SFRP2OE CM were mainly down-regulated(SI Appendix, Fig. S5 C and F and Datasets S5–S7). There was

Fig. 5. Secreted SFRP2 facilitates angiogenesis. (A) In vitro tube formation of human endothelial cells in CM as indicated. (B) Quantification of tube for-mation assay (data are shown as mean ± SD, n = 3 independent experiments; *P < 0.05; **P < 0.0001, ANOVA). (C) PCA of the transcriptome of endothelialcells treated with LFS, SFRP2OE, and WT CM (proportion of variance; PC1 = 0.483, PC2 = 0.167, PC3 = 0.133). (D) Cluster analysis of the 498 common LFS P53p.G245D and SFRP2OE DEGs in the transcriptome of HUVECs exposed to LFS, SFRP2OE, and WT CM. (E) DAVID GO functional analysis of up-regulated DEGscommon to LFS P53p.G245D and SFRP2OE. (F) Expression of angiogenic markers in HUVECs treated with WT, SFRP2OE, or LFS P53p.G245D CM. (G) qPCRanalysis of CYR61 expression in endothelial cells after WT, SFRP2OE, or LFS P53p.G245D CM treatment (data are shown as mean ± SD; n = 2 independentrepeats in triplicate; *P < 0.0001, ANOVA). (H) Representative images of a CAM inoculated with SFRP2OE pre-OBs (D7) in the presence of control rabbit IgG(Left) or CYR61 antibody (catalog no. PA1-16580; Thermo Fisher Scientific) (Right). White lines outline tumors. White ring diameter, 10 mm. (I) Tumor for-mation rate of LFS P53p.G245D and SFRP2OE pre-OBs in the CAM assay in the presence of control or CYR61 antibody (data are shown as mean ± SD; eachsymbol represents the CAM in an individual egg; *P < 0.005; **P < 0.0001, unpaired t test). (J) Quantification of neovascularization in the CAM assay afterCYR61 antibody treatment (data are shown as mean ± SD; each symbol represents the CAM in an individual egg; *P < 0.005, unpaired t test).

Kim et al. PNAS | vol. 115 | no. 47 | E11133

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

little overlap in cell-cycle–related DEGs between HUVECsactivated by LFS CM and SFRP2OE CM or in LFS andSFRP2OE OBs (SI Appendix, Fig. S5G). For example, positivecell-cycle regulators such as CDC25B, CKS1B, DLGAP5, GTSE1,JAG2, and NUSAP1 were down-regulated in endothelial cellsexposed to LFS and SFRP2OE CM but were up-regulated in OBsexposed to LFS and SFRP2OE (SI Appendix, Fig. S5G). Onlythree genes showed a common expression pattern in the two groups:JUNB and MYC, two tumorigenic factors, were up-regulated, whileKRT18, an intermediate filament gene, was down-regulated in bothOBs and HUVECs affected by SFRP2 signaling (SI Appendix, Fig.S5G). We speculated that HUVECs in LFS CM and SFRP2OE CMstopped proliferating and initiated differentiation to generate neo-vasculature. This result emphasizes that the function of SFRP2 iscontext dependent.

Targeting SFRP2OE Effectively Suppresses OS Development. To ex-amine if suppression of SFRP2 expression in LFS has any ther-apeutic effect on OS development in vivo, we s.c. injected LFSand SFRP2KD cells at D7 of OB differentiation into nu/nu mice.Tumors formed rapidly in mice injected with LFS3-A cells (at2 wk postinjection) or LFS2-B cells (at 3 wk postinjection) (SIAppendix, Fig. S6A). We attribute the difference in tumor-formation timing to lower SFRP2 expression in LFS2-B cellsthan in LFS3-A cells (Fig. 1B). In contrast, mice injected withSFRP2KD cells did not form tumors, or the tumors were of verysmall size (Fig. 6A and SI Appendix, Fig. S6A). Histologicalanalysis revealed that LFS tumors were composed of large,poorly differentiated malignant tumor cells with mitotic features

and large areas of necrosis reminiscent of the tissue morphologyseen in human osteoblastic or fibroblastic OS (Fig. 6B and SIAppendix, Fig. S6B) (64). A few of the small tumors in theSFRP2KD group showed some neoplastic features, while othersshowed induction of fibrous tissues (Fig. 6B and SI Appendix, Fig.S6B). Interestingly, LFS tumors showed lower levels of AP, amarker of mature OBs, than tumors formed from SFRP2KD cells(Fig. 6B). Despite higher β-catenin expression, β-catenin was lo-calized to the cytosol, granules, and cell membrane in theSFRP2KD group (Fig. 6B). This finding augments the idea thatSFRP2-mediated tumorigenesis is not directly linked to hyperactiveWNT signaling in LFS. Furthermore, we found more vaso-formation, as assessed by expression of the CD31 endothelialmarker, in LFS cell-induced tumors than in the SFRP2KD group(Fig. 6 C and D). These histological analyses support the idea thatSFRP2OE facilitates neovascularization and tumorigenesis. As inthe SFRP2KD group, targeting SFRP2 in malignant OS cells(MNNG) reduced OS tumor growth (SI Appendix, Fig. S6C). Ad-ditionally, we checked the expression of signature genes altered bySFRP2KD by quantitative PCR in these primary tumors (Fig. 6Eand SI Appendix, Fig. S6D). While CADM1 and CADM4 weresignificantly up-regulated in SFRP2KD OBs and in SFRP2KDprimary tumors compared with parental LFS and OS, oncogenicCYR61 and FOXM1 were only differentially expressed (i.e., de-creased) between SFRP2KD and empty-vector groups in primarytumors (Fig. 6E and SI Appendix, Fig. S6D).We next performed intratibial injection in mice (65, 66) to

investigate the orthotopic effect of SFRP2 depletion in OS de-velopment. In the LFS group, two mice generated tumors in the

Fig. 6. Targeting SFRP2 in LFS pre-OBs and the OS cell line suppresses OS development in vivo. (A) Tumor formation in nu/nu mice 4 wk after injection of LFSP53p.G245D or SFRP2KD (SFRP2 shRNA-1) cells (tumor volume = 3.14/6·L·W·H). Each symbol represents an individual mouse (data are shown as mean ± SD; n =5; *P < 0.001; **P < 0.0001, ANOVA). (B) Immunohistochemistry analysis of LFS P53p.G245D and SFRP2KD cell-induced tumors. (C) The CD31+ area in LFS P53p.G245D and SFRP2KD cell-induced tumors. (D) Quantification of CD31 expression in tumors. Each symbol represents an induced tumor in an individual mouse(data are shown as mean ± SD; *P < 0.05, unpaired t test). (E) Expression of CYR61 altered by SFRP2KD in LFS P53p.G245D OB- and MNNG-derived s.c. primarytumors (data are shown as mean ± SD; n = 4; *P <0.05, unpaired t test). (F) Intratibial tumor formation with MNNG cells with empty shRNA (Empty) or SFRP2-shRNA-1 (SFRP2KD). Red lines outline tumors. (G) Tumor size 4 wk after intratibial injection (D7 of differentiation) of MNNG with empty shRNA (Empty) orSFRP2 shRNA-1 (SFRP2KD) or WT OBs. Each symbol represents a tumor in an individual mouse (data are shown as mean ± SD; *P < 0.05, ANOVA). (H) Im-munohistochemistry analysis of representative MNNG and SFRP2KD tumors in G.

E11134 | www.pnas.org/cgi/doi/10.1073/pnas.1814044115 Kim et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

tibia 5 mo postinjection, and one mouse formed a tumor on theflank of the body but not intratibially (SI Appendix, Fig. S6E).This might have been caused by a failure of proper intratibialinjection. Three mice died within 3 mo of injection with no visibleexterior tumor formation but with clear signs of weight loss. Theactual cause of death was undetermined. In stark contrast, theSFRP2KD group did not form any osteosarcomatous lesions (SIAppendix, Fig. S6E). To corroborate this finding, we depletedSFRP2 in MNNG cells. Suppressing SFRP2 expression in MNNGcells significantly reduced tumor incidence and size (Fig. 6 F andG). Primary lesions showed characteristics of primary human OSwith osteolytic, mitotically active undifferentiated cells as well aswide bone destruction (Fig. 6H and SI Appendix, Fig. S6F). Mi-croscopically, MNNG tumors showed less AP activity, more pro-liferating (Ki67+) cells, and more neovascularized (CD31+) areasthan the SFRP2-depleted MNNG tumors (Fig. 6H). The bonestructure in the SFRP2KD group was relatively intact, althoughthe tumors still contained undifferentiated tumorigenic cells in theproximal tibia (Fig. 6G and SI Appendix, Fig. S6F). Conclusively,the in vivo xenograft study unequivocally demonstrates that de-pletion of SFRP2 in LFS-mediated tumorigenesis and OS devel-opment has potential therapeutic benefits.

DiscussionWe show that LFS patient-specific iPSCs and their MSCs/OBsprovide a platform for investigating the early development of OS,and we identify and evaluate a potential therapeutic target, SFRP2.Using molecular and functional experiments, traditional in vivostudies, and bioinformatic analyses, we found that SFRP2 expres-sion contributes significantly to LFS-mediated tumorigenesis andOS development by facilitating angiogenesis and cell proliferation.Most significantly, we demonstrated that targeting SFRP2 has apotential therapeutic application in LFS and OS.We found that SFRP2 is up-regulated in LFS patient cells and

OS cell lines harboring TP53 mutations. We showed thatSFRP2 exerts oncogenicity in MSCs/OBs both in vitro and in vivo.These findings conflict with the hyper-WNT signaling paradigm incancer biology. Generally, SFRP2 is known as a WNT pathwayantagonist due to its ability to sequester WNT ligand fromthe frizzled receptor and is considered to be a tumor suppressor.Indeed, promoter hypermethylation of SFRP2 is found with in-creased WNT signaling in many cancers (67, 68). One study foundthat enhanced WNT signaling as a consequence of SFRP2 meth-ylation is associated with OS cell invasion (69). However, in theLFS context, SFRP2 functions as an oncogenic factor. We con-sistently observed that SFRP2OE suppressed WNT signaling andthat the majority of β-catenin remained localized to the cytoplasm.Our data also indicate that lower β-catenin expression and inactiveβ-catenin are associated with the initiation of abnormal OB dif-ferentiation and are maintained to the terminal stage of osteo-sarcomagenesis. This finding is consistent with a report of inactiveβ-catenin and WNT signaling in high-grade human OS (57). Ourstudy adds evidence of the complexity of WNT pathway regulationand its diverse outcomes in different conditions.The mechanism by which SFRP2 exerts its role in OS develop-

ment is unknown and remains elusive, context dependent, and evencontroversial in many other cancers. A recent paper described anage-related increase in SFPR2 expression in fibroblasts that acti-vated a multistep signaling cascade in melanoma cells, resulting in areduction of β-catenin and MITF expression and consequently inthe loss of APE1. The absence of APE1 affected the DNA damageresponse of the cells, increased metastasis, and rendered the cellsmore resistant to chemotherapy (38). In our study, we observed thatSFRP2OE suppresses β-catenin, but MITF or APE1 expression wasnot altered. Instead, we found that SFRP2OE in OBs up-regulatesFOXM1 in OBs as an autocrine factor and CYR61 in endothelialcells as a paracrine factor in a β-catenin–independent way. There-fore, we concluded that SFRP2-associated osteosarcomagenesisdevelops through a mechanism different from that in melanoma.Oncogenesis is attributed to abnormal differentiation, aberrant

cell adhesion, a dysregulated cell cycle, and overproliferation. Our

transcriptomic analysis of SFRP2OE and SFRP2KD cells showedthat SFRP2 has multiple functions during abnormal differentiationand tumorigenesis in LFS OS development. Specifically, SFRP2OEchanges the cell-cycle profile and increases the expression of cell-proliferation–related genes such as AURKB and FOXM1. AURKBregulates chromosomal segregation during mitosis and meiosis, andaberrant expression of AURKB is observed in several cancers (70,71). FOXM1 is also overexpressed in many cancers (72, 73). Usingour LFS OB model in the in ovo CAM assay, we found that in-hibition of FOXM1 impedes OS development (SI Appendix, Fig.S4 J and K). FOXM1 regulates the transcription of cell-cycle andproliferation genes (74, 75), including the SKP2 and CKS1 sub-units of the SCF ubiquitin ligase complex. These two proteinsubiquitinate p21 (Cip1) and p27 (Kip1), regulators of cell-cycleprogression into S phase, for degradation (76). Therefore, theobserved increase in cell proliferation and in the number of cells inS phase of LFS and SFRP2OE cells may be explained by SFRP2OE-induced FOXM1. One may assume that SFRP2 depletion in LFS cellsshould reduce FOXM1 expression. However, this is not what we ob-served. We explain this by the autoregulation of FOXM1 expression(77, 78). Thus, once FOXM1 expression is induced, its expression ismaintained, and SFRP2 depletion has no apparent effect. Neverthe-less, FOXM1 inhibition by shRNA and FDI-6 small-molecule treat-ment effectively decreased tumor formation (SI Appendix, Fig. S4 Jand K). It is noteworthy that FOXM1 up-regulation was correlatedwith a poor prognosis in OS patients (79). This fits well with our ownobservation that high SFRP2 expression in OS tissues is linked todecreased patient survival (Fig. 1F and SI Appendix, Fig. S1G).Additionally, we found that down-regulated genes common to

LFS and SFRP2OE that are up-regulated in SFRP2KD cells in-clude cell-adhesion genes such as CADM1 and CADM4. They areknown tumor suppressors in various cancers (80–83). The signifi-cance and role of CAMD1 and CADM4 down-regulation in OSdevelopment is unknown. A possibility is that their down-regulationis linked to anchorage-independent cell growth and cell egress fromthe bone environment, resulting in metastasis. This warrants futureinvestigation. Taken together, the abnormal expression of thesespecific genes (AURKB, FOXM1, CADM1, and CADM4) clearlyindicates the oncogenic property of SFRP2OE in OS.Angiogenesis is a critical hallmark of cancer progression. In this

study we show that secreted SFRP2 facilitates angiogenesis in vitroand in vivo. Global transcriptome analysis revealed up-regulationof angiogenesis-related and antiapoptotic genes in endothelial cellsgrown in LFS, SFRP2 cell-conditioned, and SFRP2-supplementedmedia. Angiogenesis-related genes such as CDH13, EPAS1, andCYR61 are highly associated with focal adhesion and tumorigen-esis. Particularly, CYR61, a secreted extracellular matrix protein, isa proangiogenic/tumorigenic molecule validated in many cancers(84, 85). Interestingly, CYR61 mediates specific functions in dif-ferent types of cells through binding to distinct integrins duringangiogenesis (86, 87) and metastasis (88). CYR61 is overexpressedin OS compared with normal bone tissues, and its depletion causesinhibition of neoangiogenesis in the developing tumor (62). Wefound that CYR61 inhibition prevents SFRP2-mediated OS de-velopment in the in ovo CAM assay (Fig. 5 H and I). Collectively,the SFRP2–CYR61 axis in LFS may alter the extracellular matrixenvironment, thus promoting angiogenesis, tumorigenesis, and pos-sibly metastasis. Surprisingly, SFRP2 did not increase the expressionof cell-cycle–related genes in endothelial cells, in contrast to its ef-fects in LFS and SFRP2OE OBs.Overall, our data provide insights into the role of SFRP2 in OS

initiation and development. Using the powerful tool of iPSC-basedmodeling, we show that SFRP2 has a dual function. SFRP2 increasesthe proliferative capacity of pre-OBs and also can promote angio-genesis through cell-adhesion/cell-matrix factors; SFRP2-FOXM1 inLFS OBs/OS and SFRP2-CYR61 in endothelial cells contribute toOS development. We did not observe a significant effect of SFRP2in metastasis of LFS/OS. However, a recent report showed thatSFRP2 potentially plays a crucial role in metastatic OS, where itsectopic expression enhanced invasiveness and metastasis of humanand mouse OS cells without affecting cell proliferation (89). In

Kim et al. PNAS | vol. 115 | no. 47 | E11135

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

contrast, one study found that increased SFRP2 methylation result-ing in reduced expression is associated with OS invasiveness (69).These observed discrepancies in the role of SFRP2 in invasivenessand metastasis of OS might be due to differences in the progressivestage of the cells or in the specific OS subset studied. Detailed futureexploration of the SFRP2–FOXM1 and SFRP2–CYR61 axes duringOS development and metastasis will provide additional mechanisticinsight and help develop novel therapeutic strategies to treat LFSand OS.

Materials and MethodsWT and LFS iPSC and p53-Mutant ESC Culture. All iPSCs were generated usingthe Sendai virus method as previously reported (33, 90). ESCs and iPSCs (SIAppendix, Table S1) were cultured in DMEM/F12 supplemented with 20%knockout serum replacement, 10 ng/mL basic FGF (Invitrogen) on mousefeeder cells (GlobalStem), or in mTeSR (STEMCELL Technologies) on Matrigel-coated plates.

iPSC Differentiation to MSCs and OBs. iPSCs were differentiated to MSCs andOBs as previously described (33). For details see SI Appendix.

SFRP2 Gene Expression and Knockdown. SFRP2OE and SFRP2KD, respectively,were achieved by infecting iPSC-derived MSCs with lentiviral particlesencoding for inducible SFRP2 or shRNAs against SFRP2 (SI Appendix, TableS5). For details see SI Appendix.

OS Cell Lines. Human OS cell lines were obtained from Robert Maki (IcahnSchool of Medicine, New York) and Nino Rainusso (Texas Children’s Hospital,Houston). HOS, MNNG, and 143B cells were cultured in DMEM/10% FBS, andSJSA-1, U2OS, and SaOS-2 cells were cultured in RPMI/10% FBS at 37 °C and5% CO2 in a humidified incubator.

Genome Editing. The P53p.G245D mutation was introduced into WT iPSC-derived MSCs using CRISPR/Cas9, appropriate donor plasmid, and guideRNAs designed on the web tool at crispr.mit.edu (91). The various hetero-zygous p53 mutant ESC lines (SI Appendix, Table S1) were generated usingTALENs designed with the ZiFiT Targeter version 4.2 (zifit.partners.org/) (92)as described in ref. 93. For details see SI Appendix.

Angiogenesis/Tube-Formation Assay. Tube formation was assayed using the InVitro Angiogenesis Assay Kit (Abcam) as recommended by the manufacturer.Tube formation was captured 16 h post culture initiation with an EVOS FL CellImaging system (Thermo Fisher Scientific). Images were analyzed with theImageJ Angiogenesis Analyzer tool, which quantifies the tube-formationimages by extracting characteristic information about the formed tubemeshes and branches (94). For further details see SI Appendix.

CAM Xenograft Assay. The CAMassay with 1 × 106 differentiated OBs (D7 or D14)was conducted as described previously (95). For further details see SI Appendix.

Mouse Studies. Eight-week-old male nu/nu mice (Charles River) were used forintratibial and s.c. xenografts. Sample size was determined based on powercalculation (G*POWER software; with one-way ANOVA) and our estimationof surgical failure (96). All animal use was reviewed and approved by theIcahn School of Medicine’s Institutional Animal Care and Use Committee and

was conducted according to federal, state, and local regulations. For furtherdetails see SI Appendix.

Human OS Tissue Microarray. The OS tissue microarray set was obtained fromE.G.D. The use of the OS microarray set and data analysis were approved bythe Icahn School of Medicine’s Program for the Protection of Human Subjectsand its Institutional Review Board. For further details see SI Appendix.

RNA-Seq and Data Analyses. cDNA library preparation and sequencing wasoutsourced to Novogen. Sequence reads were aligned to human transcriptreference (RefSeq Genes, hg19) for expression analysis at gene levels usingTopHat (97) and HTSEq (98). The raw counts of reads aligned to each genewere further normalized by reads per kilobase of transcript per million readsmapped (RPKM) as the measurement of gene-expression abundance. For theoriginal dataset, we used linear regression to identify DEGs independent oftime (D0, D7, D14, and D17) among the different conditions (WT, SFRP2OE,LFS, or SFRP2KD cells) G∼time+condition. The P value cutoff (P < 0.001) wasvalidated by permutation test (FDR <0.01). For the HUVEC dataset, DESeq2(Bioconductor) was used to compare gene expression among WT, LFS, andSFRP2OE cells. An adjusted P value cutoff (Padj < 0.01) was applied to defineDEGs. We used DAVID (version 6.8) or Enrichr (41, 42) for functional ontol-ogy and pathway analyses of DEGs depending on which method wouldprovide the best visualization for the intended experimental purpose. Foradditional details and links to gene lists see SI Appendix.

Other Assays. Full details of quantitative real-time PCR, Western blotting, APand Alizarin red S staining, the luciferase reporter assay, β-catenin immunos-taining and quantification, the CyQUANT NF fluorescence proliferation assay(Thermo Fisher Scientific), and cell-cycle analysis can be found in SI Appendix.

Statistical Analysis. Student’s t test (two-sided) was applied, and changeswere considered statistically significant for P < 0.05. For comparison of morethan two groups, a one-way or two-way (if data included different timepoints) ANOVA was applied. Survival in human subjects and mouse experi-ments was represented with Kaplan–Meier curves, and significance was es-timated with the log-rank test. The statistical methods used for RNA-seqanalysis are described under RNA-seq data analysis above. The data areshown as mean ± SD of at least three biological replicates. Statistical analysiswas conducted using Microsoft Excel or GraphPad Prism software packages.

Additional Data. Histological data are available in SI Appendix, Table S6. Re-sources and reagents will be provided upon proper requests and use agree-ments. According to regulations, we do not share patient or clinicalinformation linked to the human OS tissue microarray set except as providedin this paper (SI Appendix, Table S2).

ACKNOWLEDGMENTS. We thank Drs. Robert Maki and Nino Rainusso forproviding OS cell lines, Dr. Marion D. Zwaka for sharing shRNAs to p53,Drs. Zichen Wang and Avi Ma’ayan for help in RNA-seq analysis of preliminarydata, and Dr. Avi Ma’ayan for reading and commenting on the manuscript.This work was supported by funds provided by the Graduate School of Bio-medical Sciences at the Icahn School of Medicine at Mount Sinai and byNational Cancer Institute Pre- to Postdoctoral Transition Award 1F99CA212489(to H.K.), NIH Pathway to Independence Award R00CA181496 and Cancer Pre-vention Research Institute of Texas Award RR160019 (to D.-F.L.), NIH Grant5R01GM078465 (to I.R.L.), and Empire State Stem Cell Fund through the NewYork State Department of Health Grant C024410 (to I.R.L.).

1. Kansara M, Teng MW, Smyth MJ, Thomas DM (2014) Translational biology of oste-osarcoma. Nat Rev Cancer 14:722–735.

2. Reed DR, et al. (2017) Treatment pathway of bone sarcoma in children, adolescents,and young adults. Cancer 123:2206–2218.

3. Jaffe N, Puri A, Gelderblom H (2013) Osteosarcoma: Evolution of treatment para-digms. Sarcoma 2013:203531.

4. Lin Y-H, et al. (2017) Osteosarcoma: Molecular pathogenesis and iPSC modeling.Trends Mol Med 23:737–755.

5. Bieging KT, Mello SS, Attardi LD (2014) Unravelling mechanisms of p53-mediatedtumour suppression. Nat Rev Cancer 14:359–370.

6. Lee D-F, et al. (2012) Regulation of embryonic and induced pluripotency by aurorakinase-p53 signaling. Cell Stem Cell 11:179–194.

7. Vogelstein B, Lane D, Levine AJ (2000) Surfing the p53 network. Nature 408:307–310.8. Brosh R, Rotter V (2009) When mutants gain new powers: News from the mutant

p53 field. Nat Rev Cancer 9:701–713.9. Muller PAJ, Vousden KH (2014) Mutant p53 in cancer: New functions and therapeutic

opportunities. Cancer Cell 25:304–317.10. Reed DE, Shokat KM (2014) Targeting osteosarcoma. Proc Natl Acad Sci USA 111:

18100–18101.

11. Toguchida J, et al. (1992) Mutation spectrum of the p53 gene in bone and soft tissuesarcomas. Cancer Res 52:6194–6199.

12. Birch JM, et al. (1994) Prevalence and diversity of constitutional mutations in thep53 gene among 21 Li-Fraumeni families. Cancer Res 54:1298–1304.

13. Hisada M, Garber JE, Fung CY, Fraumeni JF, Jr, Li FP (1998) Multiple primary cancers infamilies with Li-Fraumeni syndrome. J Natl Cancer Inst 90:606–611.

14. Zhou R, et al. (2017) Li-fraumeni syndrome disease model: A platform to developprecision cancer therapy targeting oncogenic p53. Trends Pharmacol Sci 38:908–927.

15. Jacks T, et al. (1994) Tumor spectrum analysis in p53-mutant mice. Curr Biol 4:1–7.16. Lozano G (2010) Mouse models of p53 functions. Cold Spring Harb Perspect Biol 2:

a001115.17. Rubio R, et al. (2014) Bone environment is essential for osteosarcoma development

from transformed mesenchymal stem cells. Stem Cells 32:1136–1148.18. Velletri T, et al. (2016) P53 functional abnormality in mesenchymal stem cells pro-

motes osteosarcoma development. Cell Death Dis 7:e2015.19. Lengner CJ, et al. (2006) Osteoblast differentiation and skeletal development are

regulated by Mdm2-p53 signaling. J Cell Biol 172:909–921.20. Walkley CR, et al. (2008) Conditional mouse osteosarcoma, dependent on p53 loss

and potentiated by loss of Rb, mimics the human disease. Genes Dev 22:1662–1676.

E11136 | www.pnas.org/cgi/doi/10.1073/pnas.1814044115 Kim et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0

21. Finch PW, et al. (1997) Purification and molecular cloning of a secreted, frizzled-related antagonist of Wnt action. Proc Natl Acad Sci USA 94:6770–6775.

22. Cheng YY, et al. (2007) Frequent epigenetic inactivation of secreted frizzled-related protein2 (SFRP2) by promoter methylation in human gastric cancer. Br J Cancer 97:895–901.

23. O’Hurley G, et al. (2011) The role of secreted frizzled-related protein 2 expression inprostate cancer. Histopathology 59:1240–1248.

24. Perry AS, et al. (2013) Gene expression and epigenetic discovery screen revealmethylation of SFRP2 in prostate cancer. Int J Cancer 132:1771–1780.

25. Takagi H, et al. (2008) Frequent epigenetic inactivation of SFRP genes in hepatocel-lular carcinoma. J Gastroenterol 43:378–389.

26. Suzuki H, et al. (2004) Epigenetic inactivation of SFRP genes allows constitutive WNTsignaling in colorectal cancer. Nat Genet 36:417–422.

27. Suzuki H, et al. (2002) A genomic screen for genes upregulated by demethylation andhistone deacetylase inhibition in human colorectal cancer. Nat Genet 31:141–149.

28. Melkonyan HS, et al. (1997) SARPs: A family of secreted apoptosis-related proteins.Proc Natl Acad Sci USA 94:13636–13641.

29. Polakis P (2000) Wnt signaling and cancer. Genes Dev 14:1837–1851.30. Lee J-L, Chang C-J, Wu S-Y, Sargan DR, Lin C-T (2004) Secreted frizzled-related protein

2 (SFRP2) is highly expressed in canine mammary gland tumors but not in normalmammary glands. Breast Cancer Res Treat 84:139–149.

31. Lee J-L, Lin C-T, Chueh L-L, Chang C-J (2004) Autocrine/paracrine secreted frizzled-related protein 2 induces cellular resistance to apoptosis: A possible mechanism ofmammary tumorigenesis. J Biol Chem 279:14602–14609.

32. Yamamura S, et al. (2010) Oncogenic functions of secreted frizzled-related protein2 in human renal cancer. Mol Cancer Ther 9:1680–1687.

33. Lee D-F, et al. (2015) Modeling familial cancer with induced pluripotent stem cells.Cell 161:240–254.

34. Gingold J, Zhou R, Lemischka IR, Lee D-F (2016) Modeling cancer with pluripotentstem cells. Trends Cancer 2:485–494.

35. Rainusso N, et al. (2011) Identification and gene expression profiling of tumor-initiating cells isolated from human osteosarcoma cell lines in an orthotopic mousemodel. Cancer Biol Ther 12:278–287.

36. Hockemeyer D, et al. (2008) A drug-inducible system for direct reprogramming ofhuman somatic cells to pluripotency. Cell Stem Cell 3:346–353.

37. Moffat J, et al. (2006) A lentiviral RNAi library for human and mouse genes applied toan arrayed viral high-content screen. Cell 124:1283–1298.

38. Kaur A, et al. (2016) sFRP2 in the aged microenvironment drives melanoma metastasisand therapy resistance. Nature 532:250–254.

39. Siamakpour-Reihani S, et al. (2011) The role of calcineurin/NFAT in SFRP2 inducedangiogenesis–A rationale for breast cancer treatment with the calcineurin inhibitortacrolimus. PLoS One 6:e20412.

40. Tan CM, Chen EY, Dannenfelser R, Clark NR, Ma’ayan A (2013) Network2Canvas: Net-work visualization on a canvas with enrichment analysis. Bioinformatics 29:1872–1878.

41. Chen EY, et al. (2013) Enrichr: Interactive and collaborative HTML5 gene list enrich-ment analysis tool. BMC Bioinformatics 14:128.

42. Kuleshov MV, et al. (2016) Enrichr: A comprehensive gene set enrichment analysisweb server 2016 update. Nucleic Acids Res 44:W90–W97.

43. Lachmann A, et al. (2010) ChEA: Transcription factor regulation inferred from in-tegrating genome-wide ChIP-X experiments. Bioinformatics 26:2438–2444.

44. Zhao M, Kim P, Mitra R, Zhao J, Zhao Z (2016) TSGene 2.0: An updated literature-basedknowledgebase for tumor suppressor genes. Nucleic Acids Res 44:D1023–D1031.

45. Savkovic V, et al. (2014) Mesenchymal stem cells in cartilage regeneration. Curr StemCell Res Ther 9:469–488.

46. Yang D, Okamura H, Nakashima Y, Haneji T (2013) Histone demethylase Jmjd3 regulatesosteoblast differentiation via transcription factors Runx2 and osterix. J Biol Chem 288:33530–33541.

47. Yeo H, Beck LH, McDonald JM, ZayzafoonM (2007) Cyclosporin A elicits dose-dependentbiphasic effects on osteoblast differentiation and bone formation. Bone 40:1502–1516.

48. Zeng H, et al. (2016) Crosstalk between ATF4 and MTA1/HDAC1 promotes osteosar-coma progression. Oncotarget 7:7329–7342.

49. Long F (2011) Building strong bones: Molecular regulation of the osteoblast lineage.Nat Rev Mol Cell Biol 13:27–38.

50. Rahman MS, Akhtar N, Jamil HM, Banik RS, Asaduzzaman SM (2015) TGF-β/BMP sig-naling and other molecular events: Regulation of osteoblastogenesis and bone forma-tion. Bone Res 3:15005.

51. Song L, et al. (2012) Loss of wnt/β-catenin signaling causes cell fate shift of pre-osteoblasts from osteoblasts to adipocytes. J Bone Miner Res 27:2344–2358.

52. Bandyopadhyay A, et al. (2006) Genetic analysis of the roles of BMP2, BMP4, andBMP7 in limb patterning and skeletogenesis. PLoS Genet 2:e216.

53. Kamiya N, et al. (2008) BMP signaling negatively regulates bone mass through scle-rostin by inhibiting the canonical Wnt pathway. Development 135:3801–3811.

54. Abzhanov A, Rodda SJ, McMahon AP, Tabin CJ (2007) Regulation of skeletogenicdifferentiation in cranial dermal bone. Development 134:3133–3144.

55. Georgiou KR, et al. (2012) Attenuated Wnt/β-catenin signalling mediates methotrexatechemotherapy-induced bone loss and marrow adiposity in rats. Bone 50:1223–1233.

56. Haÿ E, et al. (2014) N-cadherin/wnt interaction controls bone marrow mesenchymalcell fate and bone mass during aging. J Cell Physiol 229:1765–1775.

57. Cai Y, et al. (2010) Inactive Wnt/beta-catenin pathway in conventional high-gradeosteosarcoma. J Pathol 220:24–33.

58. Fontenot E, et al. (2013) A novel monoclonal antibody to secreted frizzled-relatedprotein 2 inhibits tumor growth. Mol Cancer Ther 12:685–695.

59. Sys G, et al. (2012) Tumor grafts derived from sarcoma patients retain tumor mor-phology, viability, and invasion potential and indicate disease outcomes in the chickchorioallantoic membrane model. Cancer Lett 326:69–78.

60. Gormally MV, et al. (2014) Suppression of the FOXM1 transcriptional programme vianovel small molecule inhibition. Nat Commun 5:5165.

61. Courtwright A, et al. (2009) Secreted frizzle-related protein 2 stimulates angiogenesisvia a calcineurin/NFAT signaling pathway. Cancer Res 69:4621–4628.

62. Habel N, et al. (2015) Cyr61 silencing reduces vascularization and dissemination ofosteosarcoma tumors. Oncogene 34:3207–3213.

63. Liu Y, et al. (2017) High expression levels of Cyr61 and VEGF are associated with poorprognosis in osteosarcoma. Pathol Res Pract 213:895–899.

64. Daft PG, et al. (2013) Alpha-CaMKII plays a critical role in determining the aggressivebehavior of human osteosarcoma. Mol Cancer Res 11:349–359.

65. Husmann K, et al. (2013) Matrix Metalloproteinase 1 promotes tumor formation andlung metastasis in an intratibial injection osteosarcoma mouse model. BiochimBiophys Acta 1832:347–354.

66. Yuan J, et al. (2009) Osteoblastic and osteolytic human osteosarcomas can be studiedwith a new xenograft mouse model producing spontaneous metastases. Cancer Invest27:435–442.

67. Marsit CJ, et al. (2005) Epigenetic inactivation of SFRP genes and TP53 alteration actjointly as markers of invasive bladder cancer. Cancer Res 65:7081–7085.

68. Nojima M, et al. (2007) Frequent epigenetic inactivation of SFRP genes and consti-tutive activation of Wnt signaling in gastric cancer. Oncogene 26:4699–4713.

69. Xiao Q, Yang Y, Zhang X, An Q (2016) Enhanced Wnt signaling by methylation-mediated loss of SFRP2 promotes osteosarcoma cell invasion. Tumour Biol 37:6315–6321.

70. González-Loyola A, et al. (2015) Aurora B overexpression causes aneuploidy andp21Cip1 repression during tumor development. Mol Cell Biol 35:3566–3578.

71. Smith SL, et al. (2005) Overexpression of aurora B kinase (AURKB) in primary non-small cell lung carcinoma is frequent, generally driven from one allele, and correlateswith the level of genetic instability. Br J Cancer 93:719–729.

72. Huang C, Du J, Xie K (2014) FOXM1 and its oncogenic signaling in pancreatic cancerpathogenesis. Biochim Biophys Acta 1845:104–116.

73. Radhakrishnan SK, et al. (2006) Identification of a chemical inhibitor of the oncogenictranscription factor forkhead box M1. Cancer Res 66:9731–9735.

74. Laoukili J, et al. (2005) FoxM1 is required for execution of the mitotic programme andchromosome stability. Nat Cell Biol 7:126–136.

75. Wang I-C, et al. (2008) FoxM1 regulates transcription of JNK1 to promote the G1/Stransition and tumor cell invasiveness. J Biol Chem 283:20770–20778.

76. Petrovic V, Costa RH, Lau LF, Raychaudhuri P, Tyner AL (2008) FoxM1 regulatesgrowth factor-induced expression of kinase-interacting stathmin (KIS) to promote cellcycle progression. J Biol Chem 283:453–460.

77. Cheng X-H, et al. (2014) SPDEF inhibits prostate carcinogenesis by disrupting a posi-tive feedback loop in regulation of the Foxm1 oncogene. PLoS Genet 10:e1004656.

78. Halasi M, Gartel AL (2009) A novel mode of FoxM1 regulation: Positive auto-regulatory loop. Cell Cycle 8:1966–1967.

79. Fan C-L, Jiang J, Liu H-C, Yang D (2015) Forkhead box protein M1 predicts outcome inhuman osteosarcoma. Int J Clin Exp Med 8:15563–15568.

80. Nagata M, et al. (2012) Aberrations of a cell adhesion molecule CADM4 in renal clearcell carcinoma. Int J Cancer 130:1329–1337.

81. Nowacki S, et al. (2008) Expression of the tumour suppressor gene CADM1 is associated withfavourable outcome and inhibits cell survival in neuroblastoma. Oncogene 27:3329–3338.

82. Sakurai-Yageta M, Masuda M, Tsuboi Y, Ito A, Murakami Y (2009) Tumor suppressorCADM1 is involved in epithelial cell structure. Biochem Biophys Res Commun 390:977–982.

83. Vallath S, et al. (2016) CADM1 inhibits squamous cell carcinoma progression by re-ducing STAT3 activity. Sci Rep 6:24006.

84. Chen C-C, Kim K-H, Lau LF (2016) The matricellular protein CCN1 suppresses hep-atocarcinogenesis by inhibiting compensatory proliferation.Oncogene 35:1314–1323.

85. Leask A, Abraham DJ (2006) All in the CCN family: Essential matricellular signalingmodulators emerge from the bunker. J Cell Sci 119:4803–4810.

86. Leu S-J, Lam SC-T, Lau LF (2002) Pro-angiogenic activities of CYR61 (CCN1) mediatedthrough integrins alphavbeta3 and alpha6beta1 in human umbilical vein endothelialcells. J Biol Chem 277:46248–46255.

87. Park YS, et al. (2015) CCN1 secreted by tonsil-derived mesenchymal stem cells promotesendothelial cell angiogenesis via integrin αv β3 and AMPK. J Cell Physiol 230:140–149.

88. Sun Z-J, et al. (2008) Involvement of Cyr61 in growth, migration, and metastasis ofprostate cancer cells. Br J Cancer 99:1656–1667.

89. Techavichit P, et al. (2016) Secreted frizzled-related protein 2 (sFRP2) promotes os-teosarcoma invasion and metastatic potential. BMC Cancer 16:869.

90. Zhou R, et al. (2018) Modeling osteosarcoma using Li-fraumeni syndrome patient-derived induced pluripotent stem cells. J Vis Exp, 10.3791/57664.

91. Hsu PD, et al. (2013) DNA targeting specificity of RNA-guided Cas9 nucleases. NatBiotechnol 31:827–832.

92. Reyon D, Khayter C, Regan MR, Joung JK, Sander JD (2012) Engineering designertranscription activator–Like effector nucleases (TALENs) by REAL or REAL-fast as-sembly. Current Protocols in Molecular Biology, eds Ausubel FM, et al. (John Wiley &Sons, Inc., Hoboken, NJ).

93. Xu A, et al. (2018) Establishment of a human embryonic stem cell line with homozygousTP53 R248Wmutant by TALENmediated gene editing. Stem Cell Res (Amst) 29:215–219.

94. Chevalier F, et al. (2014) Glycosaminoglycan mimetic improves enrichment and cell func-tions of human endothelial progenitor cell colonies. Stem Cell Res (Amst) 12:703–715.

95. Sys GML, et al. (2013) The in ovo CAM-assay as a xenograft model for sarcoma. J VisExp 17:e50522.

96. Dell RB, Holleran S, Ramakrishnan R (2002) Sample size determination. ILAR J 43:207–213.

97. Trapnell C, Pachter L, Salzberg SL (2009) TopHat: Discovering splice junctions withRNA-seq. Bioinformatics 25:1105–1111.

98. Anders S, Pyl PT, Huber W (2015) HTSeq–A Python framework to work with high-throughput sequencing data. Bioinformatics 31:166–169.

Kim et al. PNAS | vol. 115 | no. 47 | E11137

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

Nov

embe

r 14

, 202

0