Transposon mutagenesis identifies genes thatcooperate with mutant Pten in breastcancer progressionRoberto Rangela, Song-Choon Leeb, Kenneth Hon-Kim Banb,c, Liliana Guzman-Rojasa, Michael B. Manna,Justin Y. Newberga, Takahiro Kodamaa, Leslie A. McNoed, Luxmanan Selvanesand, Jerrold M. Wardb,1, Alistair G. Ruste,2,Kuan-Yew Chinb, Michael A. Blackd, Nancy A. Jenkinsa,b,3, and Neal G. Copelanda,b,3,4

aCancer Research Program, Houston Methodist Research Institute, Houston, TX 77030; bDivision of Genomics and Genetics, Institute of Molecular and CellBiology, Agency for Science, Technology and Research, Biopolis, Singapore 138673; cDeparment of Biochemistry, Yong Loo Lin School of Medicine, NationalUniversity of Singapore, Singapore 138673; dDepartment of Biochemistry, University of Otago, Dunedin 9016, New Zealand; and eExperimental CancerGenetics, Wellcome Trust Sanger Institute, Hinxton, Cambridge CB10 1HH, United Kingdom

Contributed by Neal G. Copeland, October 4, 2016 (sent for review July 14, 2016; reviewed by Kent W. Hunter and Branden Moriarty)

Triple-negative breast cancer (TNBC) has the worst prognosis ofany breast cancer subtype. To better understand the genetic forcesdriving TNBC, we performed a transposon mutagenesis screen in aphosphatase and tensin homolog (Pten) mutant mice and identi-fied 12 candidate trunk drivers and a much larger number ofprogression genes. Validation studies identified eight TNBC tumorsuppressor genes, including the GATA-like transcriptional repres-sor TRPS1. Down-regulation of TRPS1 in TNBC cells promotedepithelial-to-mesenchymal transition (EMT) by deregulating multi-ple EMT pathway genes, in addition to increasing the expression ofSERPINE1 and SERPINB2 and the subsequent migration, invasion,and metastasis of tumor cells. Transposon mutagenesis has thusprovided a better understanding of the genetic forces driving TNBCand discovered genes with potential clinical importance in TNBC.

Sleeping Beauty | breast cancer | TRPS1 | metastasis | tumor suppressors

Breast cancer is the second leading cause of cancer-relateddeaths in the United States. The Cancer Genome Atlas (TCGA)

network has classified breast cancer into four main subtypes: luminalA, luminal B, HER2+, and basal-like (1–5). Basal-like or triple-negative breast cancer (TNBC) constitutes 10–20% of all breastcancers and has a higher rate of distal recurrence and a poorerprognosis than other breast cancer subtypes. Less than 30% ofwomen with metastatic TNBC survive 5 y and almost all die fromtheir disease despite adjuvant chemotherapy (1, 3–5). Mutations,rearrangements, or deletions in highly penetrant genes such asBRCA1, BRCA2, TP53, CDH1, STK11, and PTEN are importantdrivers of TNBC (6–8). PTEN is a dual-specificity phosphatasethat antagonizes the PI3K/AKT pathway through its lipid phos-phatase activity and negatively regulates the MAPK pathwaythrough its protein phosphatase activity (9, 10). Mutations inPTEN drive epithelial–mesenchymal transition (EMT) and pro-mote metastasis in TNBC (11–13). Similarly, in mice, heterozy-gous deletion of Pten induces mammary tumors with basal-likecharacteristics (14–17).Despite all of the cancer genome-sequencing efforts, there is

still an incomplete understanding of the genes and genetic net-works driving TNBC. New technologies that would provide amore complete understanding of the genetics of TNBC are stillneeded to deconvolute the complexity of this deadly cancer. Ourlaboratory and others have pioneered the use of transposonmutagenesis in mice as a tool for cancer gene discovery (18–26).Transposons induce cancer by randomly inserting into the mousegenome, mutating, and disrupting potential cancer genes. Trans-poson insertions in tumors thus serve as molecular tags for thehigh-throughput cloning and identification of cancer genes. Inaddition, because transposon insertions are PCR-amplified be-fore they are sequenced, insertional mutations in cancer genesthat are present in only a small fraction of tumor cells can be

identified. Transposon mutagenesis can thus identify genes thatare functioning at the tips of the cancer evolutionary tree andhelp deconvolute tumor evolution on a scale that is not yetpossible through the sequencing of human tumors.To identify genetic drivers of TNBC, we induced Sleeping

Beauty (SB) transposition in breast epithelial cells of mice thatwere heterozygous for a Pten-null allele. Breast tumors weresubsequently classified using a PAM50 subtyping approach andfound to represent a collection of different breast cancer sub-types including basal-like (45%), luminal A (39%), HER2 (11%),and normal-like (5%). Cloning and sequencing of the transposoninsertion sites in tumors identified 12 candidate trunk drivers anda much larger number of progression genes. Subsequent validationstudies identified eight TNBC tumor suppressor genes (TSGs),including a tumor and metastasis suppressor with clinical rele-vance in TNBC.

Significance

Triple-negative breast cancer (TNBC) is the most aggressivebreast cancer subtype. Despite extensive cancer genome-sequencing efforts, there is still an incomplete understandingof the genetic networks driving TNBC. Here, we used SleepingBeauty transposon mutagenesis to identify genes that co-operate with mutant Pten in the induction of TNBC. We iden-tified 12 candidate TNBC trunk drivers and a larger number ofprogression genes. Subsequent functional validation studiesidentified eight human TNBC tumor suppressor genes, includingthe GATA-like transcriptional repressor TRPS1, which was shownto inhibit lung metastasis by deregulating the expression ofmultiple serpin and epithelial-to-mesenchymal transition (EMT)pathway genes. Our study provides a better understanding ofthe genetic forces driving TNBC and discovered genes withclinical importance in TNBC.

Author contributions: R.R., S.-C.L., K.H.-K.B., N.A.J., and N.G.C. designed research; R.R., S.-C.L.,L.G.-R., T.K., L.A.M., and L.S. performed research; R.R., S.-C.L., K.H.-K.B., M.B.M., J.Y.N.,J.M.W., A.G.R., K.-Y.C., M.A.B., N.A.J., and N.G.C. analyzed data; and R.R., N.A.J., and N.G.C.wrote the paper.

Reviewers: K.W.H., National Cancer Institute; and B.M., University of Minnesota.

The authors declare no conflict of interest.

Freely available online through the PNAS open access option.1Present address: Global Vet Pathology, Montgomery Village, MD 20886.2Present address: Tumor Profiling Unit, The Institute of Cancer Research, Chester BeattyLaboratories, London SW3 6JB, United Kingdom.

3N.A.J. and N.G.C. contributed equally to this work.4To whom correspondence should be addressed. Email: ncopeland@houstonmethodist.org.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1613859113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1613859113 PNAS | Published online November 14, 2016 | E7749–E7758

GEN

ETICS

PNASPL

US

Dow

nloa

ded

by g

uest

on

Nov

embe

r 12

, 202

1

ResultsSBMutagenesis Promotes the Development of Multiple Breast CancerSubtypes in Ptenfl/+ Mice. Loss of the TSG PTEN is implicated inbreast cancer progression, clonally selected in TNBC, and favorsthe activation of the EMT pathway to promote metastasis (11–13).To identify genes that cooperate with PTEN in the progression ofbreast cancer, we crossed Ptenfl/fl mice with K5-Cre transgenic miceto generate K5-CreTg/+;Ptenfl/+ mice. The mice were then crossedto mice carrying one of two conditional SB transposition systems(SB11fl/fl;T2/Onc2Tg/Tg) (18) or (SB11fl/fl;T2/Onc3Tg/Tg) (27) togenerate K5-CreTg/+;Ptenfl/+;SB11fl/+;T2/Onc2Tg/+ (SB/Pten–Onc2) orK5-CreTg/+;Ptenfl/+;SB11fl/+;T2/Onc3Tg/+ (SB/Pten–Onc3) mice. SB/Pten–Onc2 mice carry 350 copies of T2/Onc2, all linked togetherat a single site on chromosome 1, whereas SB/Pten–Onc3 micecarry a 30-copy T2/Onc3 transposon concatamer located onchromosome 9 (20, 27). By using two different transposon con-catamers located on different donor chromosomes, we were ableto eliminate problems caused by local hopping (28) and achievegenome-wide coverage of SB mutagenesis. K5-CreTg/+ is active inearly mammary progenitors (29). Therefore, K5-driven Cre ex-pression should lead to excision of the conditional floxed allele

from the entire mammary epithelium, which is consistent withour LacZ reporter assays (Fig. S1 A and B) and those of others(30) who showed that LacZ is expressed in all basal cells andmost luminal cells in the mammary epithelium. K5-CreTg/+

should therefore induce mammary tumors with both luminaland basal cell origins.SB mutagenesis accelerated mammary tumor formation in

Ptenfl/+ mice with a median survival of 250 d for SB/Pten–Onc2mice and 313 d for SB/Pten–Onc3 mice (Fig. 1A). Tumor latencyfor SB/Pten–Onc2 mice was significantly earlier than SB/Pten–Onc3 mice (P = 0.003), which may reflect the higher number oftransposons carried by SB/Pten–Onc2 mice. Fifty-nine percent ofthe tumors were classified as adenocarcinomas, whereas 29%were classified as adenosquamous carcinomas and 12% as ade-nomyoepitheliomas (Dataset S1, Table S1).Adenocarcinomas were more frequent in SB/Pten–Onc2 mice

(74%), whereas adenomyoepitheliomas were only identified inSB/Pten–Onc3 mice. Hematoxylin and eosin (H&E) staining ofindividual tumors revealed a mixed histology, whereas immu-nohistochemical staining showed that both basal (cytokeratin 14)and luminal (cytokeratin 18) markers were often expressed in the

Fig. 1. SB mutagenesis promotes the developmentof multiple mammary tumor subtypes. (A) Kaplan–Meier survival curves of five different genotypiccombinations of mice. Pten/SB–Onc2 and Pten/SB–Onc3 mice showed significant tumor accelerationcompared with various control mice (Pten/SB–Onc2,P = 0.0003; Pten/SB–Onc3, P = 0.0001). (B and C) H&Eor immunohistochemical staining of mammary ad-enocarcinoma (B) or adenosquamous carcinoma (C).The adenocarcinoma shows areas of less differenti-ation (Upper H&E). The adenosquamous carcinomaalso has areas of no squamous differentiation in-vading muscle (Lower H&E). Both tumors showed ahigh degree of heterogeneity, expressing both basal(CK14) and luminal (CK18) cytokeratins. Both tumorshave low or focal high-proliferation rate tumors,based upon their Ki67 staining (NOTE-B shows afocally high rate), and express high levels of nuclearSBT protein. (Scale bar, 100 μm.) (D) Mammary tumorsubtype classification based upon its PAM50 expres-sion signature (31). The heat map displays gene ex-pression data (log scale, right legend) for the PAM50breast cancer subtype classifier for each mouse tumor(columns). The left side indicates the centroids foreach breast cancer subtype. The rows in the heat maprepresent genes in the PAM50 panel, and columnsrepresent each mammary tumor. Top panels showproliferation scores (blue, low; red, high) and PAM50subtypes: basal-like (black), Her2 (purple), luminal A(light blue), and normal-like (light green).

E7750 | www.pnas.org/cgi/doi/10.1073/pnas.1613859113 Rangel et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 12

, 202

1

same tumor (Fig. 1 B and C), which is suggestive of extensiveintratumor heterogeneity (Dataset S1, Table S1). Immunohis-tochemical analysis also showed that SB transposase (SBT) wasexpressed at high levels in tumors, consistent with their SB-induced origins (Fig. 1 B and C).To further define the mammary tumor subtypes derived from

SB mutagenesis, gene expression arrays were performed on 21mammary tumors. For each tumor, an “intrinsic subtype” wasassigned based on the previously described PAM50 subtypingapproach (31). Mouse orthologs for the PAM50 genes wereidentified, and the microarray data were used to determine theclosest intrinsic subtype centroid for each sample, based onSpearman correlation using logged mean-centered expressiondata. A gene proliferation signature was also used to generate aproliferation score for each sample (32). Basal-like (45%) andluminal A (39%) were the most abundant tumor subtypes, al-though HER2 (11%) and normal-like (5%) were detected atlower frequencies (Fig. 1D). As expected, basal-like tumors weregenerally more proliferative than the other tumor subtypes. TheSB–Pten mouse model thus provides a resource for functionalgenomic studies across multiple human breast cancer subtypes.

Identification of Candidate Cancer Genes. To identify genes mu-tated by SB that drive tumor development, we PCR-amplifiedand sequenced the transposon insertions from 18 SB/Pten–Onc2and 16 SB/Pten–Onc3 tumors using 454 next-generation se-quencing. Using the Wellcome Trust Sanger Institute’s trans-poson common insertion site calling pipeline (33), we identified105,540 mapped transposon reads corresponding to 23,137 uniquetransposon insertion sites. Using the locus-centric Gaussian kernelconvolution (GKC) (33) and gene-centric common integration site(gCIS) (34) calling methods, we identified 448 statistically signif-icant candidate cancer genes (CCGs) by combining the two lists ofCIS genes (Dataset S1, Table S2). CISs are genomic regions thatcontain more transposon insertions than predicted by chance andare thus likely to mark the location of CCGs. Pten was the mosthighly mutated CIS gene (Table 1 and Dataset S1, Table S2),

which likely reflects the strong selective pressure to inactivate thewild-type Pten allele present in tumor cells.T2/Onc2 and T2/Onc3 contain transcriptional stop cassettes in

both orientations and can inactivate the expression of a TSG ifinserted within one. They also contain promoters and down-stream splice donor sites and can deregulate the expression of aproto-oncogene if inserted upstream or in the 5′ end, in the sametranscriptional orientation. The pattern of transposon insertionsin CCGs can therefore be used to infer whether an insertionallymutated gene is an oncogene or TSG. Visual analysis of insertionpatterns of the 448 CCGs identified in mammary tumors sug-gested that most (97%) are functioning as TSGs. In addition, inmany cases, only a single transposon insertion was present at aCCG in tumor cells, suggesting that many CCGs are functioningas haploinsufficient TSGs. This is similar to what has been ob-served in other SB mutagenesis screens performed in solid tu-mors (18–26).

Comparative Oncogenomic Filtering. To assess the biological rele-vance of the 446 SB-identified CCGs with human orthologs tohuman breast cancer, we performed a number of cross-speciescomparisons. We found that 51 CCGs are listed in the CancerGene Census database (35), a catalog of known cancer genes,which is a highly significant overlap (P = 3.61E–18, two-sidedFisher’s exact test; Fig. 2A and Dataset S1, Table S3). We alsocompared our CCGs to the 127 significantly mutated genesfound in 12 major human cancer types (36) and found that 16 areSB-identified CCGs (P = 1.23E–08, two-sided Fisher’s exact test;Fig. 2B and Dataset S1, Table S4), indicating that our screen isselecting for genes that are relevant to many types of cancer.Finally, we used the TCGA breast cancer database (37, 38) toidentify genes mutated in human breast cancer (i.e., missense,nonsense, sense, and in-frame mutations). Among the 1,375 mu-tated genes listed in TCGA, 70 are SB-identified CCGs (P = 9.21E–12, two-sided Fisher’s exact test; Fig. 2C and Dataset S1, Table S5).SB-identified CCGs in mice therefore appear to be highly relevantto human breast cancer.

Table 1. Most highly mutated CIS genes identified in SB–Pten mammary tumors

Gene ChromosomeNo. of

tumors in CISUnique insertion

sitesTotal no.

of insertionsGKC P value adjusted

across genome

Pten 19 25 41 46 <1.00E–16Axin1 17 10 22 22 <1.00E–16Fat1 8 13 12 13 <1.00E–16Nf1 11 18 35 37 <1.00E–16Rasa1 13 11 19 20 <1.00E–16Trps1 15 10 20 20 <1.00E–16Tbl1xr1 3 10 11 11 <1.00E–16Gnaq 19 10 13 13 <1.00E–16Erbb2ip 13 4 12 12 <1.00E–16Pum2 12 7 8 9 <1.00E–16Cox7b2 5 7 1 7 <1.00E–16Man1a 10 7 1 7 <1.00E–16Dcun1d3 7 8 10 10 3.43E–12Smek1 12 5 7 7 1.66E–05Fbxw7 3 7 9 9 3.04E–03Jup 11 7 7 7 6.33E+03Ipo7 7 6 7 7 0.0127Mkl1 15 10 11 11 0.0224Sos2 12 3 7 7 0.0242Ccny 18 11 11 12 0.0271Spag9 11 6 8 8 0.0321Nfib 4 7 8 9 0.0364Zfp326 5 6 6 6 0.0431Arih1 9 5 9 9 0.0496

Rangel et al. PNAS | Published online November 14, 2016 | E7751

GEN

ETICS

PNASPL

US

Dow

nloa

ded

by g

uest

on

Nov

embe

r 12

, 202

1

Identification of Trunk Driver Genes. Most CCGs identified by SBmutagenesis are thought to function during late stages of tumorprogression (18–26). To identify the CCGs that function early inmammary tumor development, we selected transposon insertion sitesrepresented by the highest number of sequencing reads (SB insertionsin the CCG in ≥3 tumors with ≥9 reads per tumor), arguing thatthese insertions would be present in the largest number of tumorcells. This analysis identified 12 CCGs that we subsequently refer toas trunk drivers (Fig. 2D). Strikingly, 50% are known TSGs, includingPten, Arhgap35, Nf1, Rasa1, Axin1, and Apc. Transposon insertions in

these genes were mostly located throughout the coding regions,consistent with their TSG function (Fig. 2E and Fig. S2). The oneexception was Jup (also called gamma catenin), which contains acluster of insertions on the sense strand near exon 2, suggesting thatJup might function as an oncogene (Fig. S2). Recent reports haveshown that Jup overexpression is crucial for maintaining circulatingtumor cells as clusters and for facilitating homing to the lungs (39).

Altered Signaling Pathways and Cellular Processes. Using Enrichrfunctional annotation (40), we identified 22 pathways and cellular

Fig. 2. Comparative oncogenomic and pathway analysis of SB-identified CCGs. Cross-species comparison of SB-identified CCGs against genes listed in (A) theCancer Gene Census, (B) Pan-cancer 12, and (C) TCGA breast cancer databases. Significant enrichment of SB-identified CCGs was obtained using the Fisherexact test. (D) SB-identified trunk drivers along with mean and median sequence read counts for insertions containing ≥9 sequence reads and mutations in ≥3tumors. (E) List of the top 12 trunk drivers, along with their mean read counts, predicted effect of transposon insertions on gene expression and genefunction. (F) Biological pathway (yellow nodes) interaction network of CCG connections between 14 human orthologs that show a clinical association be-tween RNA abundance and breast cancer patient survival (blue nodes) and reported cancer gene drivers (red nodes), including known breast cancer driversPTEN, MAP2K4, TBL1XR1, AXIN1, and NF1. Chromatin, chromatin remodeling; EGFR, epidermal growth factor receptor signaling; MAPK, mitogen-activatedprotein kinase signaling; NTR, neurotrophin signaling; TNF/NF–κB, tumor necrosis factor and nuclear factor kappa B signaling; and WNT, Wnt signaling.

E7752 | www.pnas.org/cgi/doi/10.1073/pnas.1613859113 Rangel et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 12

, 202

1

processes (FDR P < 0.05) (Dataset S1, Table S6) that areenriched in SB-identified CCGs. Notably, the top six pathways andcellular processes contained 51% of the SB-identified CCGs withknown function. The Wnt pathway contained the most CCGs, withloss-of-function mutations detected in Apc, Axin1, and Gsk3b,which are involved in beta-catenin degradation, in addition toTcf7l2, Amer1, and Smad4 (41, 42). We also identified numerousinsertions in Pten, Rasa1, and Nf1, which are important regulatorsof EGFR, TNF-alpha, and MAPK signaling and are commonlyaltered in patients with breast cancer.

SB-Identified CCGs Have Clinical Relevance. To determine whetherSB-identified CCGs belonging to the top six signaling pathwayshave clinical relevance, we interrogated a breast cancer expressionmicroarray dataset that also reports recurrence-free patient sur-vival. All CCGs showing a significant association were also ana-lyzed within each breast cancer subtype. This analysis identified 14CCGs that were significantly associated with patient survival inone or more breast cancer subtypes (Fig. 2F and Dataset S1, TableS7). SB-identified CCGs thus appear to have clinical relevance.

SB-Identified CCGs Act as Tumor Suppressors in Human Breast Cancer.To provide additional evidence that SB-identified CCGs areimportant in human breast cancer, we selected 20 CCGs that aremutated in human breast cancer, possess tumor suppressor ac-tivity in loss-of-function studies, and have been identified asCCGs in other SB mutagenesis screens. We then used smallhairpin RNAs (shRNAs) delivered by lentivirus to silence theexpression of these CCGs in two TNBC cell lines, HCC70 andMDA-MB-468, which were selected because they carry mutationsin PTEN (Figs. S3 and S4). The cells were then orthotopicallyinjected into the mammary fat pad of athymic nude mice, andafter 45 d, the animals were necropsied and tumor volumesmeasured. The silencing of eight CCGs, including Man1a1, Pkp4,Rab10, Rasa1, Trps1, Vps26a, Xpnpep3, and Znf326, acceleratedtumor growth, whereas the silencing of R3hcc1l reduced tumorgrowth (Table 2). The silencing of Ppp1r12a and Pum2 inducedcell death during puromycin selection, suggesting that these genesare essential for cellular growth. These results show that theseCCGs are enriched for mammary cancer TSGs.

TRPS1 Is a Tumor Suppressor in TNBC. Down-regulation of Trps1resulted in the largest acceleration of tumor growth in ourfunctional validation studies. Trps1 encodes a GATA-like tran-scription factor that regulates gene expression by acting as atranscriptional repressor (43, 44). Mutations in TRPS1 lead toTricho-rhino-phalangeal syndrome, an autosomal-dominant dis-order characterized by craniofacial and skeletal malformations(45). Mutations in TRPS1 have been reported in several humancancers, including leukemia, prostate, colon, endometrial, andbreast cancer (46–49). Quantitative immunohistochemistry hasalso shown that TRPS1 is a prognostic marker in early-stage breastcancer and I–II ER+ patients receiving antihormone therapy (50).Stage II/III breast cancer patients with high expression levels ofTRPS1 also have a better survival outcome compared with lowTRPS1 expression patients (51), suggesting that tumors arisingfrom TRPS1-negative/low cells are intrinsically more aggressive.This is in agreement with our TRPS1 expression analysis acrossdifferent human breast cancer subtypes, which showed that TNBChas the lowest expression levels of TRPS1 (Fig. S5).Down-regulation of TRPS1 in TNBC results from activation of

the RAS–RAF–MEK pathway, which is common in basal-likebreast cancers (52). This in turn leads to the activation of the Fosfamily transcription factor FOSL1 and its downstream targetmiR-221/miR-222, followed by the down-regulation of TRPS1,which is a downstream target of miR-221/miR-222 (52). Theseresults predict a negative correlation between the expression ofTRPS1 and FOSL1/miRNA-221/222 in breast cancer, which weconfirmed by quantitative real time (qRT)-PCR (Fig. S6). To de-termine whether down-regulation of TRPS1 expression acceleratestumor growth in vivo, we used two different TRPS1 shRNAs todown-regulate TRPS1 expression in HCC70 and HCC1569 TNBCcells (Fig. S7 A and B). Down-regulation of TRPS1 significantlyaccelerated tumor growth in nude mice following orthotopic in-jection (Fig. 3 A and B). Finally, to determine whether TRPS1overexpression inhibits tumor growth, we overexpressed TRPS1 inMDA-MB-231 and HCC1954 TNBC cells (Fig. S7 C and D), whichhave reduced TRPS1 expression. We used a lentivirus con-struct expressing the TRPS1 ORF minus the 3′UTR to avoidmiRNA-221/222 targeting degradation. Tumor cells were injectedinto the fat pad of nude mice, and after several weeks, we observed

Table 2. Functional validation screening of CCGs

CIS orthologPredicted effecton gene function Function or pathway

Tumor growth, gene knockdownvs. control

AFTPH Disruption Membrane trafficking No significant differenceASH1L Disruption Histone methyltransferase No significant differenceEP400 Disruption Transcriptional activation No significant differenceERBB2IP Disruption EGFR signaling No significant differenceLPP Disruption Cell adhesion No significant differenceMAN1A1 Disruption Metabolism AccelerationNIPBL Disruption Cell cycle No significant differencePKP4 Disruption Cell adhesion AccelerationPPP1R12A Disruption Regulation of actin cytoskeleton Cell deathPUM2 Disruption Translational control Cell deathR3HCC1L Activation Unknown DecreaseRAB10 Disruption Cytoskeletal signaling AccelerationRASA1 Disruption EGFR and FGFR signaling AccelerationSOS2 Disruption PI3K-Akt signaling No significant differenceTRPS1 Disruption Transcriptional repressor AccelerationVPS26A Disruption WNT signaling AccelerationXPNPEP3 Disruption Aminopeptidase AccelerationYTHDF3 Disruption mRNA processing No significant differenceZNF143 Activation Transcriptional activation No significant differenceZNF326 Disruption Transcriptional activation Acceleration

Genes in bold indicate a significant tumor growth difference.

Rangel et al. PNAS | Published online November 14, 2016 | E7753

GEN

ETICS

PNASPL

US

Dow

nloa

ded

by g

uest

on

Nov

embe

r 12

, 202

1

a statistically significant reduction in tumor growth (Fig. 3 C andD). These results show that TRPS1 functions as a TSG in TNBC.

TRPS1 Is a Breast Cancer Metastasis TSG. miR-221/222 promotesEMT in breast cancer cells in part by down-regulating TRPS1expression, which is a direct transcriptional repressor of ZEB2and a driver of EMT (52). miR-221/222 expression also has beenshown to enhance the migration and invasion of nontransformedhuman mammary epithelial MCF10A cells, whereas syntheticoligo inhibitors of miR-221/222 attenuate the migration and in-vasion of MDA-MD-231 cells through the basement membranematrix (52). To determine whether TRPS1 is a breast cancermetastasis TSG, we examined the ability of MDA-MB-231 TNBCcells overexpressing the TRPS1 ORF, minus the 3′UTR, to colo-nize the lung. TRPS1-overexpressing cells were injected into themammary fat pad of athymic nude mice, and lung metastaticprogression was monitored by bioluminescence. Five weeks afterinjection, we observed a significant decrease in luciferase activityin the lungs of injected mice compared with control mice (Fig. 3E and F). These results indicate that TRPS1 is a TSG that in-hibits lung metastasis.

TRPS1 Regulates the Expression of Multiple Genes in the EMTPathway. To determine whether other EMT pathway genes inaddition to ZEB2 are transcriptionally regulated by TRPS1, we

knocked down TRPS1 expression in HCC70 TNBC cells usingtwo independent TRPS1 shRNAs and then used an EMT PCRgene expression array to identify genes that might be regulatedby TRPS1. The EMT PCR array profiles the expression of 84 keygenes that either change their expression during the process ofEMT or regulate the expression of these genes. We observed asignificant up-regulation of BMP2, MMP2, MMP9, SERPINE1,SNAI2, TFPI2, TGFB2, and ZEB1 and a significant down-regu-lation of COL5A, FN1, KRT14, SNAI1, SNAI3, and SOX10 usingthis array (Fig. 4A). These results indicate that TRPS1 regulatesmultiple genes in the EMT pathway. To further explore the bi-ological effects of these expression changes, we performed migrationand invasion assays using TNBC cells that over- or underexpressTRPS1. We observed increased migration and invasion of HCC70cells that had reduced TRPS1 expression (Fig. 4 B and C) anddecreased migration and invasion of MDA-MB-231 andHCC1954 cells that had increased TRPS1 expression (Fig. 4 Band C). Collectively, these results show that TRPS1 regulatesmultiple genes in the EMT pathway that have effects on cellmigration and invasion and are consistent with previous studiesthat measured the effects of miR-221/222 expression on themigration and invasion of tumor cells (52).

SERPINE1 and SERPINB2 Expression Is Negatively Regulated by TRPS1.SERPINE1 was the gene whose expression was most increased inTRPS1 knockdown cells (Fig. 4A). SERPINE1 encodes endo-thelial plasminogen activator inhibitor-1 (PAI1), a member ofthe serine protease inhibitor family that inhibits tissue-typeplasminogen activator (PLAT) and urokinase-type plasminogenactivator (PLAU), which breaks down fibrin clots. Interestingly,high SERPINE1 expression has been shown to enhance cellmigration and apoptosis resistance in head and neck carcinomapatients (53), whereas its down-regulation in nasopharyngealcarcinoma has been associated with reduced metastasis (54).This led us to explore the effects of TRPS1 knockdown on theexpression of other serpin family members, which showed that

Fig. 3. TRPS1 reduces tumor growth and suppresses metastasis in breastcancer. (A and B) Inactivation of TRPS1 in HCC70 and HCC1569 TNBC cell linesaccelerates tumor growth in orthotopic xenografts. (C and D) Ectopic over-expression of TRPS1 reduces tumor growth in HCC1954 and MDA-MB-231orthotopic xenografts. n =10 mice per group. One-sided unpaired t test wasrun for all experiments. (E) TRPS1 expression in MDA-MB-231 cells reducesmetastasis. Tumor lung colonization was measured by bioluminescent im-aging. The panels show mouse images at different time points. (F) Linegraphs represent three time-point measurements of the radiance averagefrom five injected animals. All data were evaluated using two-sided t test.Error bars represent SEM.

Fig. 4. TRPS1 regulates the expression of genes in the EMT pathway.(A) EMT gene expression in HCC70 TRPS1 knockdown cells. Two independentTRPS1 shRNA knockdown clones showed consistent mRNA expression levelsover control. (B) Migration and (C) invasion assays of TRPS1 shRNA knockdownin HCC70 cells and TRPS1 overexpression in MDA-MB-231 and HCC1954 cells.Data represent means ± SEM of three independent experiments.

E7754 | www.pnas.org/cgi/doi/10.1073/pnas.1613859113 Rangel et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 12

, 202

1

SERPINB2 is also highly overexpressed in TRPS1 knockdowncells (Fig. 5A). Plasmin from reactive brain stroma provides adefense against metastatic invasion of lung and breast cancercells (55). Plasmin does this by converting membrane-boundastrocytic FasL into a paracrine death signal for cancer cells andby inactivating the axon guidance molecule L1CAM, whichmetastatic cells express for spreading along brain capillaries andfor metastatic outgrowth. Importantly, brain metastatic lungand breast cancer cells that express high levels of neuroserpinand serpin B2 inhibit plasmin generation and its metastasis-suppressive effects (55).To confirm that SERPINE1 and SERPINB2 are secreted from

TRPS1 knockdown cells, we grew the cells in serum free-media for

24 h and then collected the supernatants. Sandwich ELISAshowed that SERPINE1 and SERPINB2 proteins are highlysecreted from TRPS1 knockdown TNBC cells (Fig. 5B). Finally,we used qRT-PCR to confirm that SERPINE1 and SERPINB2expression is negatively regulated by TRPS1 in orthotopic tumorxenografts (Fig. 5 C and D). These results indicate that the sup-pressive effects of TRPS1 on tumor metastases are mediated inpart through the suppressive effects of TRPS1 on SERPINE1 andSERPINB2 expression.

TRPS1 Is a Direct Transcriptional Repressor of SERPINE1 andSERPINB2. To determine whether TRPS1 is a direct transcriptionalrepressor of SERPINE1 and SERPINB2, we asked whether TRPS1

Fig. 5. TRPS1 directly regulates the expression of SERPINE1 and SERPINB2. (A) Gene expression analysis of different serpin family members. SERPINE1 andSERPINB2 were up-regulated after TRPS1 depletion in HCC70 cells. qRT-PCR validation was performed in triplicates. *P < 0.001. (B) Increased serpin proteinlevels in the supernatants of TRPS1 knockdown cells were detected by ELISA, *P < 0.001. (C and D) Serpin expression levels are regulated by TRPS1 in tumorxerographs. TRPS1 and SERPINE1 and SERPINB2 expression levels are negatively correlated in (C) HCC70 (n = 10 per group) and (D) MDA-MB-231 (n = 7) tumorxenografts. mRNA expression levels were detected by qRT-PCR. (E) ChIP assays confirmed that TRPS1 binds to SERPINB2 and SERPINE1 promoters. (Top)Possible TRPS1 binding sites in SERPINB2 and SERPINE1 promoters and control regions. (Bottom) Fold enrichment of TRPS1 binding to SERPINE1 and SERPINB2promoters. To validate anti-TRPS1 specificity, we used TRPS1 knockdown (shTRPS1) and nontargeted control (NTC) cells. The ZEB2 promoter was included as apositive control as previously described (52). (F and G) Luciferase reporter expression assays were performed using SERPINE1 and SERPINB2 promoters in (F)HCC70 control and TRSP1 shRNA cells and (G) MDA-MB-231 control and TRPS1-overexpressing cells. ZEB2 and GAPDH promoters were used as positivecontrols. Random control promoter (RCP) was used as a negative control. All promoters contained 1,000 base pairs of DNA. In all experiments, the results arereported as means ± SEM of three independent experiments. P values were calculated using two-sided t test.

Rangel et al. PNAS | Published online November 14, 2016 | E7755

GEN

ETICS

PNASPL

US

Dow

nloa

ded

by g

uest

on

Nov

embe

r 12

, 202

1

could bind to the SERPINE1 and SERPINB2 promoters usingchromatin immunoprecipitation (ChIP) assays. Because TRPS1represses transcription from GATA-containing binding sites (43,56–59), we searched for conserved GATA sites in the promoterregions of SERPINE1 and SERPINB2. We then designed oligo-nucleotides specific for these binding sites. As a negative control,we targeted a serpin promoter region that lacked a GATA bindingsite. ChIP assays showed that TRPS1 could bind to the SERPINE1and SERPINB2 promoters but only in parental HCC70 cells andnot in cells expressing TRPS1 shRNA (Fig. 5E). To further examineTRPS1 repressor activity, we performed luciferase reporter assays.SERPINE1 and SERPINB2 promoter-luciferase DNA constructswere transfected into TRPS1 knockdown HCC70 or TRPS1-over-expressing MDA-MB-231 cells. Subsequent quantification of lu-ciferase activity showed that SERPINE1 and SERPINB2 luciferaseactivity was significantly increased in TRPS1-deficient HCC70 cellscompared with control cells. Conversely, SERPINE1 and SERPINB2luciferase activity was significantly decreased in MDA-MB-231cells overexpressing TRPS1. These results are in agreementwith our in vitro and in vivo studies showing that TRPS1 isa transcriptional repressor of SERPINE1 and SERPINB2(Fig. 5 F and G).To determine whether SERPINE1 and SERPINB2 are im-

portant for the tumor acceleration observed in TRPS1 knock-down cells, we performed a rescue experiment by disruptingSERPINE1 and SERPINB2 expression in TRPS1 knockdownHCC70 cells and then measuring the effect of this disruption ontumor growth in transplanted mice (Fig. 6A). As shown in Fig.6B, knockdown of SERPINE1 or SERPINB2 in TRPS1 knock-down cells inhibited tumor growth, confirming that SERPINE1or SERPINB2 is important for the tumor acceleration seen inTRPS1 knockdown cells.

TRPS1 Expression Levels Are Clinically Associated with PatientSurvival. Finally, to determine whether TRPS1 expression levelsare clinically associated with patient survival, we queried apublically available breast cancer database (60). We found thathigh TRPS1 expression is associated with increased relapse-freesurvival but only in luminal A (ER+) breast cancer patients (Fig.S8A). This is in agreement with previous reports showing thatTRPS1 is a positive prognostic marker in ER+ breast cancer(51). Luminal B and basal-like breast cancer patients with re-duced expression of TRPS1 showed a negative trend in survival,but the results do not reach statistical significance (Fig. S8A).Interestingly, SERPINE1, SERPINB2, and FOSL1 expressionwas inversely correlated with TRPS1 expression in ER– breastcancer (Fig. S8B) (61), in agreement with our findings. Collec-tively, these results suggest that high TRPS1 expression in ER+breast cancer represses EMT and serpin gene expression, leadingto reduced tumor growth and metastases. However, in patientswith ER– breast cancer, where FOSL1 levels are high andmir221/222 expression is increased, reduced TRPS1 expressionleads to an increase in EMT and serpin gene expression, withconcomitant increased tumor growth and tumor metastasis(Fig. S8C).

DiscussionHere we have used SB mutagenesis to identify genes that co-operate with mutant Pten in the induction of breast cancer.Previous studies have shown that Pten mutant mice developwell-differentiated adenocarcinomas with prominent stromalproliferation (16) in addition to well-differentiated fibroade-nomas and pleomorphic adenocarcinomas (15). By using geneexpression microarrays and molecular signatures that are as-sociated with different breast cancer subtypes, we were able toidentify multiple mammary tumor subtypes in our SB–Ptenmodel. The two major subtypes we identified were luminal Aand basal-like tumors, suggesting that transposon mutagenesisis occurring in all mammary epithelium cell populations. Con-sistent with this, studies in K5Cre transgenic mice have identi-fied Cre activity in early mammary cell progenitors (30). K5Creshould therefore inactivate Pten and activate SB transpositionin both luminal and basal cell progenitors and might explainthe development of different breast cancer subtypes in our SBmouse model.Cloning and sequencing of SB insertion sites in tumors led

to the identification of 12 candidate trunk drivers and a muchlarger number of tumor progression genes. Comparativeoncogenomic filtering showed that these genes are enriched forgenes causally associated with human cancer, including breastcancer and cancers of many other cell types. Strikingly, 6 ofthe 12 trunk drivers identified by SB are known TSGs, includingPten, Arhgap35, Nf1, Rasa1, Axin1, and Apc. Pathway analysisshowed that these genes function in multiple signaling pathwaysand cellular processes important in cancer. Notably, the top sixpathways and cellular processes contained 51% of the SB-identified cancer genes, with the Wnt pathway showing a veryhigh enrichment for mutations in negative regulators of β-catenin.We also identified a high frequency of mutation in Map2k4 andMapk8. These genes are important components in the activa-tion of apoptosis in response to stress (62), which might ex-plain why these genes are inactivated in SB–Pten tumors.Finally, we also identified a high frequency of transposon in-sertions in chromatin-remodeling and -modifying enzymes.Collectively, these data suggest that SB is targeting cooperativesignaling networks that promote tumor growth without induc-ing apoptosis.An interesting finding in our studies was the identification of

tumor suppressors that have already been described in humanbreast cancer. These results indicate that humans and mice havesimilar selection pressures and provide evidence that these

Fig. 6. Serpins mediate tumor growth in TRPS1-deficient tumor cells.(A) qRT-PCR analysis of TRPS1, SERPINB2, and SERPINE1 expression in stablelentivirus-transfected HCC70 cells. (B) Tumor growth is significantly reducedin serpin-deficient TRPS1 shRNA tumor cells. n = 8 mice per group. Two-sidedt test was run for all experiments. Error bars represent SEM.

E7756 | www.pnas.org/cgi/doi/10.1073/pnas.1613859113 Rangel et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 12

, 202

1

mutations are important driving events in tumorigenesis. Al-though SB mutagenesis can uncover hundreds of genes drivingtumor progression and enable system-level studies, this tech-nology also possesses a challenge for functional validation. Here,we show that we can use high-throughput shRNA screens inhuman breast cancer cells in conjunction with SB mutagenesis tovalidate genes important for human breast cancer progression.Among 20 candidate TSGs we assayed using this high-through-put approach, the silencing of eight genes was found to accel-erate tumor growth, whereas the silencing of one gene reducedtumor growth. The silencing of two genes induced cell deathduring puromycin selection, suggesting that they are essentialgenes. Gene inactivation in orthotopic tumor xenografts there-fore appears to be a highly efficient platform for functionalvalidation studies of candidate TSGs.A major finding of our screen was the discovery and func-

tional validation of TRPS1 as a metastasis tumor suppressor inhuman TNBC. Consistent with these results, in SB–Pten tu-mors, Trsp1 was insertionally mutated only in basal-like tumors.This is in contrast to what has been reported in ER+ breastcancer, where increased TRPS1 expression correlates withimproved survival and a favorable response to antihormonetherapy. Remarkably, tumor cells from ER+ breast cancer pa-tients after antihormone therapy have decreased TRPS1 expres-sion and increased expression of mesenchymal markers (63),suggesting that breast tumors with low TRPS1 expression mightbe more resistant to chemotherapy and have a higher probabilityto metastasize.TRPS1 is a GATA-like transcription factor, which functions as

a transcriptional repressor or activator, depending on cell type,stage of development, or pathological conditions. In early hairfollicle progenitors, TRPS1 is a transcriptional activator of genesthat inhibit Wnt signaling (56), whereas it is a repressor ofRUNX2, a gene important for osteoclast differentiation andmaturation (57). Moreover, in odontoblast development, Trsp1 isan inhibitor of Dspp expression (58), whereas in chondrocytes,Trps1 controls proliferation and survival by repressing Stat3 ex-pression (59). In pathological conditions, TRPS1 represses theexpression of ZEB2, a component of the EMT pathway (52).Studies by Stinson et al. (52) have demonstrated that mir221/222targets the 3′UTR of the TRPS1 mRNA transcript. This degra-dation leads to increased ZEB2 expression as well as increasedmigration and invasion of tumor cells. In our studies, we showthat TRPS1 regulates many other genes in the EMT pathway inaddition to ZEB2. We also show that TRPS1 directly repressesthe expression of SERPINE1 and SERPINB2 in TNBC. Recentstudies using a breast metastasis model have shown that SERPINB2protects metastatic cells from plasminogen activator plasmin inthe brain, providing a metastatic advantage (55). Another serpinfamily member, Serpine2, was initially discovered in a model ofbreast cancer metastasis to bone and later reported to facilitatetumor cells to form vascular-like networks enabling perfusionand lung metastasis (64, 65). Likewise, SERPINE1 has been usedas a prognostic marker of breast cancer and evaluated duringchemotherapy administration (66, 67). In rescue experiments, wedemonstrated that tumor growth in TRPS1 knockdown cells ismediated by SERPINE1 and SERPINB2. This is in agreementwith mouse serpine1 knockout studies, which have also revealed adefect in tumor growth (68). Collectively, our studies suggest thatTRPS1 is a master regulator of several gene networks that reducetumor growth and metastasis.

In summary, our studies have provided insights into the ge-netic and evolutionary forces driving breast cancer and identifiedeight TNBC TSGs. Our SB–Pten mouse model can be used tostudy many subtypes of breast cancer, including basal-like, lu-minal A, HER2, and normal-like. Finally, our studies havehelped to elucidate an important signaling pathway in TNBCwith potential clinical importance to a disease that currentlyhas limited treatment options.

Materials and MethodsMice. SB–Pten mice were generated by crossing the following mice: K5-Cretransgenic mice (K5-CreTg/+) (69), Pten conditional knockout mice (C;129S4-Ptentm1Hwu/J) (15), Rosa26-loxP-STOP-LoxP-SB11 transposase conditionalknock-in mice [Gt(ROSA)26Sortm2(sb11)Njen] (70), T2/Onc2 (SB-6113) trans-poson transgenic mice [TgTn(sb-T2/Onc2)6113Njen] (20), and T2/Onc3 (SB-12740) transposon transgenic mice [TgTn(sb-T2/Onc3)12740Njen] (27). Toexamine the expression of Cre within the mammary epithelium, we used thefollowing mice: K5-Cre transgenic mice (K5-CreTg/+) (69) and LacZ transgenicreporter mice [B6;129S4-Gt(ROSA)26Sortm1Sor/J; Jackson Laboratory, StockNo. 003309]. Mice were housed in a specific pathogen-free facility with a12-h light/12-h dark cycle. Mice with the selected genotypes were aged andmonitored twice a week for mammary tumor development. Kaplan–Meieranalysis was performed, and survival curves were generated using Prism6software (GraphPad). Tumors >5 mm in diameter were dissected at nec-ropsy; half of the tumor was frozen for DNA sequence analysis, and theother half was fixed with 4% (vol/vol) paraformaldehyde. Fixed tissues werethen processed to paraffin blocks. H&E staining was used to characterizetumor histopathology. All mouse procedures were approved by the In-stitutional Animal Care and Use Committee, Institute of Molecular and CellBiology, Singapore, and the Animal Care and Use Committee, HoustonMethodist Research Institute, Houston.

Cell Culture, Functional Validation Screening, and Generation of Cell Clones.Human breast cancer cell lines HCC70 (CRL-2315), HCC1569 (CRL-2330),HCC1954 (CRL-2338), MDA-MB-231 (HTB-26), MDA-MB-468 (HTB-132), andBT-549 (HTB-122) were obtained from the American Type Culture Collection.MDA-MB-231 Luciferase (AKR-231) was obtained from Cell Biolabs. All celllines were cultured according to the vendor’s instructions. All cell lines usagewas approved by the IRB at the Houston Methodist Research Institute.Lentiviral shRNA particles were originally obtained from Open Biosystems.For the functional validation screening, we plated MDA-MB-468 and HCC70cells at a density of 5 × 105 cells per well in a six-well plate the night beforelentiviral transduction. The next day, the cells were infected overnight inserum-free medium containing polybrene (8 μg/mL) mixed with lentiviralparticles at a multiplicity of infection (MOI) of 6, consisting of individualshRNA NTC or pools of three CCG shRNAs. After overnight incubation, com-plete medium was added. At 48 h postinfection, puromycin selection mediumwas added (1.5 μg/mL), and the medium was change every 2 d until >95% ofthe cells expressed GFP fluorescence. All shRNA lentiviral clones were obtainedfrom Open Biosystems (Dataset S1, Table S8). For the generation of tumor cellsexpressing single shRNA clones, we used shRNA-NTC (RHS4348), shRNA-TRPS1(V3LHS_366300, V3LHS_366303), and TRPS1-ORF (OHS5900-224626817).Moreover, we used lentivirus expressing shRNA-SERPINE1, shRNA-SERPINB2,and shRNA-TFPI2 with a blasticidin resistant marker. The virus particles wereobtained from GenTarget Inc. For stable shRNA knockdown studies, HCC70and HCC1569 cells were selected with 1.5 μg/mL of puromycin for 1 wk. Forknockdown or overexpression studies, all cell lines were selected with blas-ticidin at 5 μg/mL, except for MDA-MB-231 (20 μg/mL), for 2 wk.

ACKNOWLEDGMENTS. We thank K. Rogers, S. Rogers and the Institute ofMolecular and Cell Biology histopathology core for performing histologicalanalysis, and P. Cheok, N. Lim, D. Chen, C. Wee, E. Freiter, and H. Lee formonitoring mice and animal technical assistance. The Biomedical ResearchCouncil, Agency for Science, Technology and Research (A-STAR), Singapore,and The Cancer Research Institute of Texas (CIPRIT), supported this research.A.G.R. was supported by the Cancer Research UK and the Wellcome Trust.Both N.G.C. and N.A.J. are CPRIT Scholars in Cancer Research.

1. Banerji S, et al. (2012) Sequence analysis of mutations and translocations across breast

cancer subtypes. Nature 486(7403):405–409.2. Curtis C, et al.; METABRIC Group (2012) The genomic and transcriptomic architecture

of 2,000 breast tumours reveals novel subgroups. Nature 486(7403):346–352.3. Ellis MJ, et al. (2012) Whole-genome analysis informs breast cancer response to ar-

omatase inhibition. Nature 486(7403):353–360.

4. Stephens PJ, et al.; Oslo Breast Cancer Consortium (OSBREAC) (2012) The landscape of

cancer genes and mutational processes in breast cancer. Nature 486(7403):400–404.5. Cancer Genome Atlas Network (2012) Comprehensive molecular portraits of human

breast tumours. Nature 490(7418):61–70.6. Arnedos M, et al. (2015) Precision medicine for metastatic breast cancer–Limitations

and solutions. Nat Rev Clin Oncol 12(12):693–704.

Rangel et al. PNAS | Published online November 14, 2016 | E7757

GEN

ETICS

PNASPL

US

Dow

nloa

ded

by g

uest

on

Nov

embe

r 12

, 202

1

7. Couch FJ, Nathanson KL, Offit K (2014) Two decades after BRCA: Setting paradigms inpersonalized cancer care and prevention. Science 343(6178):1466–1470.

8. Shiovitz S, Korde LA (2015) Genetics of breast cancer: A topic in evolution. Ann Oncol26(7):1291–1299.

9. Chalhoub N, Baker SJ (2009) PTEN and the PI3-kinase pathway in cancer. Annu RevPathol 4:127–150.

10. Song MS, Salmena L, Pandolfi PP (2012) The functions and regulation of the PTENtumour suppressor. Nat Rev Mol Cell Biol 13(5):283–296.

11. Engebraaten O, Vollan HK, Børresen-Dale AL (2013) Triple-negative breast cancer andthe need for new therapeutic targets. Am J Pathol 183(4):1064–1074.

12. Foulkes WD, Smith IE, Reis-Filho JS (2010) Triple-negative breast cancer. N Engl J Med363(20):1938–1948.

13. Shah SP, et al. (2012) The clonal and mutational evolution spectrum of primary triple-negative breast cancers. Nature 486(7403):395–399.

14. Li Y, et al. (2001) Deficiency of Pten accelerates mammary oncogenesis in MMTV-Wnt-1transgenic mice. BMC Mol Biol 2:2.

15. Li G, et al. (2002) Conditional loss of PTEN leads to precocious development andneoplasia in the mammary gland. Development 129(17):4159–4170.

16. Stambolic V, et al. (2000) High incidence of breast and endometrial neoplasia re-sembling human Cowden syndrome in pten+/- mice. Cancer Res 60(13):3605–3611.

17. Saal LH, et al. (2008) Recurrent gross mutations of the PTEN tumor suppressor gene inbreast cancers with deficient DSB repair. Nat Genet 40(1):102–107.

18. Bard-Chapeau EA, et al. (2014) Transposon mutagenesis identifies genes driving he-patocellular carcinoma in a chronic hepatitis B mouse model. Nat Genet 46(1):24–32.

19. Dorr C, et al. (2015) Transposon mutagenesis screen identifies potential lung cancerdrivers and CUL3 as a tumor suppressor. Mol Cancer Res 13(8):1238–1247.

20. Dupuy AJ, Akagi K, Largaespada DA, Copeland NG, Jenkins NA (2005) Mammalianmutagenesis using a highly mobile somatic Sleeping Beauty transposon system.Nature 436(7048):221–226.

21. Mann KM, et al.; Australian Pancreatic Cancer Genome Initiative (2012) SleepingBeauty mutagenesis reveals cooperating mutations and pathways in pancreatic ad-enocarcinoma. Proc Natl Acad Sci USA 109(16):5934–5941.

22. Mann MB, et al. (2015) Transposon mutagenesis identifies genetic drivers of Braf(V600E)melanoma. Nat Genet 47(5):486–495.

23. Moriarity BS, et al. (2015) A Sleeping Beauty forward genetic screen identifies newgenes and pathways driving osteosarcoma development and metastasis. Nat Genet47(6):615–624.

24. Rahrmann EP, et al. (2013) Forward genetic screen for malignant peripheral nervesheath tumor formation identifies new genes and pathways driving tumorigenesis.Nat Genet 45(7):756–766.

25. Riordan JD, et al. (2013) Identification of rtl1, a retrotransposon-derived imprintedgene, as a novel driver of hepatocarcinogenesis. PLoS Genet 9(4):e1003441.

26. Takeda H, et al. (2015) Transposon mutagenesis identifies genes and evolutionaryforces driving gastrointestinal tract tumor progression. Nat Genet 47(2):142–150.

27. Dupuy AJ, et al. (2009) A modified sleeping beauty transposon system that can beused to model a wide variety of human cancers in mice. Cancer Res 69(20):8150–8156.

28. Keng VW, et al. (2005) Region-specific saturation germline mutagenesis in mice usingthe Sleeping Beauty transposon system. Nat Methods 2(10):763–769.

29. Moumen M, et al. (2012) The proto-oncogene Myc is essential for mammary stem cellfunction. Stem Cells 30(6):1246–1254.

30. Chiche A, et al. (2013) Somatic loss of p53 leads to stem/progenitor cell amplificationin both mammary epithelial compartments, basal and luminal. Stem Cells 31(9):1857–1867.

31. Parker JS, et al. (2009) Supervised risk predictor of breast cancer based on intrinsicsubtypes. J Clin Oncol 27(8):1160–1167.

32. Anjomshoaa A, et al. (2008) Reduced expression of a gene proliferation signature isassociated with enhanced malignancy in colon cancer. Br J Cancer 99(6):966–973.

33. March HN, et al. (2011) Insertional mutagenesis identifies multiple networks of co-operating genes driving intestinal tumorigenesis. Nat Genet 43(12):1202–1209.

34. Brett BT, et al. (2011) Novel molecular and computational methods improve the ac-curacy of insertion site analysis in Sleeping Beauty-induced tumors. PLoS One 6(9):e24668.

35. Futreal PA, et al. (2004) A census of human cancer genes. Nat Rev Cancer 4(3):177–183.

36. Kandoth C, et al. (2013) Mutational landscape and significance across 12 major cancertypes. Nature 502(7471):333–339.

37. Ciriello G, et al.; TCGA Research Network (2015) Comprehensive molecular portraitsof invasive lobular breast cancer. Cell 163(2):506–519.

38. Gao J, et al. (2013) Integrative analysis of complex cancer genomics and clinicalprofiles using the cBioPortal. Sci Signal 6(269):pl1.

39. Aceto N, et al. (2014) Circulating tumor cell clusters are oligoclonal precursors ofbreast cancer metastasis. Cell 158(5):1110–1122.

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

41. Anastas JN, Moon RT (2013) WNT signalling pathways as therapeutic targets in can-cer. Nat Rev Cancer 13(1):11–26.

42. Wang X, et al. (2008) Association of genetic variation in genes implicated in theβ-catenin destruction complex with risk of breast cancer. Cancer Epidemiol BiomarkersPrev 17(8):2101–2108.

43. Malik TH, et al. (2001) Transcriptional repression and developmental functions of theatypical vertebrate GATA protein TRPS1. EMBO J 20(7):1715–1725.

44. van den Bemd GJ, Jhamai M, Brinkmann AO, Chang GT (2003) The atypical GATAprotein TRPS1 represses androgen-induced prostate-specific antigen expression inLNCaP prostate cancer cells. Biochem Biophys Res Commun 312(3):578–584.

45. Momeni P, et al. (2000) Mutations in a new gene, encoding a zinc-finger protein,cause tricho-rhino-phalangeal syndrome type I. Nat Genet 24(1):71–74.

46. Asou N, et al. (2007) Concurrent transcriptional deregulation of AML1/RUNX1 andGATA factors by the AML1-TRPS1 chimeric gene in t(8;21)(q24;q22) acute myeloidleukemia. Blood 109(9):4023–4027.

47. Chang GT, Jhamai M, van Weerden WM, Jenster G, Brinkmann AO (2004) The TRPS1transcription factor: Androgenic regulation in prostate cancer and high expression inbreast cancer. Endocr Relat Cancer 11(4):815–822.

48. Hong J, Sun J, Huang T (2013) Increased expression of TRPS1 affects tumor progres-sion and correlates with patient prognosis of colon cancer. BioMed Res Int 213:1–6.

49. Liang H, et al. (2012) Whole-exome sequencing combined with functional genomicsreveals novel candidate driver cancer genes in endometrial cancer. Genome Res22(11):2120–2129.

50. Radvanyi L, et al. (2005) The gene associated with trichorhinophalangeal syndromein humans is overexpressed in breast cancer. Proc Natl Acad Sci USA 102(31):11005–11010.

51. Chen JQ, et al. (2013) Prognostic value of the trichorhinophalangeal syndrome-1(TRPS-1), a GATA family transcription factor, in early-stage breast cancer. Ann Oncol24(10):2534–2542.

52. Stinson S, et al. (2011) TRPS1 targeting by miR-221/222 promotes the epithelial-to-mesenchymal transition in breast cancer. Sci Signal 4(177):ra41.

53. Pavón MA, et al. (2015) Enhanced cell migration and apoptosis resistance may un-derlie the association between high SERPINE1 expression and poor outcome in headand neck carcinoma patients. Oncotarget 6(30):29016–29033.

54. Sang Y, et al. (2015) TEL2 suppresses metastasis by down-regulating SERPINE1 innasopharyngeal carcinoma. Oncotarget 6(30):29240–29253.

55. Valiente M, et al. (2014) Serpins promote cancer cell survival and vascular co-option inbrain metastasis. Cell 156(5):1002–1016.

56. Fantauzzo KA, Christiano AM (2012) Trps1 activates a network of secreted Wnt in-hibitors and transcription factors crucial to vibrissa follicle morphogenesis. Development139(1):203–214.

57. Napierala D, et al. (2005) Mutations and promoter SNPs in RUNX2, a transcriptionalregulator of bone formation. Mol Genet Metab 86(1-2):257–268.

58. Napierala D, et al. (2012) Transcriptional repression of the Dspp gene leads to den-tinogenesis imperfecta phenotype in Col1a1-Trps1 transgenic mice. J Bone Miner Res27(8):1735–1745.

59. Suemoto H, et al. (2007) Trps1 regulates proliferation and apoptosis of chondrocytesthrough Stat3 signaling. Dev Biol 312(2):572–581.

60. Gy}orffy B, Surowiak P, Budczies J, Lánczky A (2013) Online survival analysis softwareto assess the prognostic value of biomarkers using transcriptomic data in non-small-cell lung cancer. PLoS One 8(12):e82241.

61. Ringnér M, Fredlund E, Häkkinen J, Borg Å, Staaf J (2011) GOBO: Gene expression-based outcome for breast cancer online. PLoS One 6(3):e17911.

62. Widmann C, Johnson NL, Gardner AM, Smith RJ, Johnson GL (1997) Potentiation ofapoptosis by low dose stress stimuli in cells expressing activated MEK kinase 1.Oncogene 15(20):2439–2447.

63. Creighton CJ, et al. (2009) Residual breast cancers after conventional therapy displaymesenchymal as well as tumor-initiating features. Proc Natl Acad Sci USA 106(33):13820–13825.

64. Eckhardt BL, et al. (2005) Genomic analysis of a spontaneous model of breast cancermetastasis to bone reveals a role for the extracellular matrix. Mol Cancer Res 3(1):1–13.

65. Wagenblast E, et al. (2015) A model of breast cancer heterogeneity reveals vascularmimicry as a driver of metastasis. Nature 520(7547):358–362.

66. Duffy MJ (2004) The urokinase plasminogen activator system: Role in malignancy.Curr Pharm Des 10(1):39–49.

67. Harbeck N, et al. (2004) Urokinase-type plasminogen activator (uPA) and its inhibitorPAI-I: Novel tumor-derived factors with a high prognostic and predictive impact inbreast cancer. Thromb Haemost 91(3):450–456.

68. Bajou K, et al. (1998) Absence of host plasminogen activator inhibitor 1 preventscancer invasion and vascularization. Nat Med 4(8):923–928.

69. Ramirez A, et al. (2004) A keratin K5Cre transgenic line appropriate for tissue-specificor generalized Cre-mediated recombination. Genesis 39(1):52–57.

70. Starr TK, et al. (2009) A transposon-based genetic screen in mice identifies genesaltered in colorectal cancer. Science 323(5922):1747–1750.

71. R Core Team (2015) R: A Language and Environment for Statistical Computing (RFoundation for Statistical Computing, Vienna, Austria). Available at www.R-project.org/.

72. Gautier L, Cope L, Bolstad BM, Irizarry RA (2004) affy–Analysis of Affymetrix Gen-eChip data at the probe level. Bioinformatics 20(3):307–315.

73. Parman C, Halling C, Gentleman R (2016) affyQCReport: QC Report Generation foraffyBatch Objects, R package version 1.44.0. Available at bioconductor.org/packages/affyQCReport/. Accessed October 14, 2014.

74. Soon WW, et al. (2011) Combined genomic and phenotype screening reveals secre-tory factor SPINK1 as an invasion and survival factor associated with patient prognosisin breast cancer. EMBO Mol Med 3(8):451–464.

E7758 | www.pnas.org/cgi/doi/10.1073/pnas.1613859113 Rangel et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 12

, 202

1