Altered Expression and Splicing of ESRP1 in Malignant...

Research Article

Altered Expression and Splicing of ESRP1 inMalignant Melanoma Correlates with Epithelial–Mesenchymal Status and Tumor-AssociatedImmune Cytolytic ActivityJun Yao1, Otavia L. Caballero2,3, Ying Huang4, Calvin Lin4, Donata Rimoldi5,Andreas Behren6, Jonathan S. Cebon6, Mien-Chie Hung1, John N.Weinstein7,Robert L. Strausberg8, and Qi Zhao2,4

Abstract

Melanoma is one of the major cancer types for which newimmune-based cancer treatments have achieved promisingresults. However, anti–PD-1 and anti–CTLA-4 therapies are effec-tive only in some patients. Hence, predictive molecular markersfor the development of clinical strategies targeting immunecheckpoints are needed. Using The Cancer Genome Atlas (TCGA)RNAseq data, we found that expression of ESRP1, encoding amaster splicing regulator in the epithelial–mesenchymal tran-sition (EMT), was inversely correlated with tumor-associatedimmune cytolytic activity. That association holds up across mul-tiple TCGA tumor types, suggesting a link between tumor EMTstatus and infiltrating lymphocyte activity. In melanoma, ESRP1mainly exists in a melanocyte-specific truncated form transcribedfrom exon 13. This was validated by analyzing CCLE cell line data,

public CAGE data, and RT-PCR in primary cultured melanomacell lines. Based on ESRP1 expression, we divided TCGA mela-noma cases into ESRP1-low, -truncated, and –full-length groups.ESRP1-truncated tumors comprise approximately two thirds ofmelanoma samples and reside in an apparent transitional statebetween epithelial and mesenchymal phenotypes. ESRP1 full-length tumors express epithelialmarkers and constitute about 5%of melanoma samples. In contrast, ESRP1-low tumors expressmesenchymal markers and are high in immune cytolytic activ-ity as well as PD-L2 and CTLA-4 expression. Those tumorsare associated with better patient survival. Results from ourstudy suggest a path toward the use of ESRP1 and otherEMT markers as informative biomarkers for immunotherapy.Cancer Immunol Res; 4(6); 552–61. �2016 AACR.

IntroductionMelanoma is one of the most invasive malignancies, and

patients with melanoma mainly die from tumor metastasis. Thereversible phenotype switch between proliferative and invasivecells that drives themetastasis of tumor cells has been proposed asa model for melanoma progression (1). Many transcriptionfactors and signaling pathways induced by changes in the micro-

environment have been implicated in tumorigenesis and thisphenotypic switch (2, 3). MITF, a master transcriptional regulatorinvolved in melanocyte differentiation and pigment production,has been shown to play a critical role in melanoma development(4, 5). High expression of MITF is associated with transformationof melanocytes and hyperproliferation of the transformed cells.Downregulation of MITF expression leads to greater invasivepotential of melanoma cells (3, 6, 7). Changes in the Wntsignaling pathway have also been implicated in the phenotypicswitch (8–11). The inversely expressed genes ROR1 and ROR2 arereported to be associated with proliferative and invasive pheno-types in melanoma, respectively (12). However, many of theseobservations are based on in vitro gain-of-function or loss-of-function experiments and might not reflect the complexity ofmicroenvironment changes and cohesive signaling networksin vivo.

ESRP1 encodes a cell type–specific splicing regulator proteinexclusively expressed in epithelial cells such as those comprised ofectoderm-derived tissues and endothelial organ lining. Expres-sion of ESRP1 controls cell type–specific alternative splicing ofmany genes, such as FGFR2, CD44, and CTNND1, that areimportant signaling molecules mediating cell division, growth,and differentiation under specific microenvironment settings(13). For example, two alternative forms of FGFR2, IIIb and IIIc,which exhibit different affinity with their FGF ligands, are pre-dominantly expressed in epithelial and mesenchymal cells,respectively. The selective expression of the FGFR2 IIIb or IIIc

1Department of Molecular and Cellular Oncology, The University ofTexas MD Anderson Cancer Center, Houston, Texas. 2Ludwig Collab-orative Laboratory, Johns Hopkins University School of Medicine,Baltimore,Maryland. 3OrygenBiotecnologia, SA., S~aoPaulo, SP,Brazil.4Regeneron Pharmaceuticals Inc., Tarrytown, New York. 5ClinicalTumor Biology and Immunotherapy Unit, Ludwig Center, Universityof Lausanne, Switzerland, Lausanne, Switzerland. 6Cancer Immuno-biology Laboratory, Olivia Newton-John Cancer Research Institute,Victoria, Australia. 7Department of Bioinformatics andComputationalBiology,TheUniversityof TexasMDAndersonCancerCenter, Houston,Texas. 8Ludwig Cancer Research, New York, New York.

Note: Supplementary data for this article are available at Cancer ImmunologyResearch Online (http://cancerimmunolres.aacrjournals.org/).

CorrespondingAuthors:Qi Zhao, Regeneron Pharmaceuticals Inc., 777 Old SawMill River Road, Tarrytown, NY 10514; Phone: 914-847-5961; Fax: 914-847-7414;E-mail: [email protected]; and Robert L. Strausberg, Ludwig CancerResearch, 666 3rd Avenue, New York, NY 10017, E-mail: [email protected]

doi: 10.1158/2326-6066.CIR-15-0255

�2016 American Association for Cancer Research.

CancerImmunologyResearch

Cancer Immunol Res; 4(6) June 2016552

on June 16, 2018. © 2016 American Association for Cancer Research. cancerimmunolres.aacrjournals.org Downloaded from

Published OnlineFirst April 4, 2016; DOI: 10.1158/2326-6066.CIR-15-0255

http://cancerimmunolres.aacrjournals.org/

isoforms is under the control of ESRP1 (14). We previouslyreported that dominance of epithelial or mesenchymal celltypes exists in renal cell carcinoma (RCC; ref. 15). Our resultssupport the notion that clear cell renal cell carcinoma (ccRCC)originates from mesenchymal-like stem cells embedded in adultkidney, unlike other subtypes of RCC, such as chromophobe,that likely have an epithelial-like stem cell origin. Thus, ESRP1and status of its target genes might serve as molecular markers oftumor cells with epithelial or mesenchymal properties.

Cutaneous melanoma originates from melanocytes derivedfrom multipotent neural crest cells (NCC). NCCs undergo epi-thelial–mesenchymal transition (EMT) before delaminatingfrom the dorsal neural tube and can differentiate into many celltypes, such as neurons of the peripheral nervous system, glial andSchwann cells, melanocytes, and craniofacial cartilage. Althoughthe EMT process enables NCC migration, formation of NCCs'derivative structures requires re-aggregationof cells, often coupledwith the reverse mesenchymal–epithelial transition (MET) fol-lowing migration (16, 17). Behavior of neural crest stem cellsduring embryogenesis and their existence in adult NCC-derivedtissues raise questions about the origin and ultimate fate ofmelanoma cells (18).

Transformation ofmelanocytes is driven by somatic changes incancer genomes. However, these somatic changes are not suffi-cient to fully explain the highly heterogeneous and metastaticnature of the disease. A comprehensive genomic landscape ofalterations in cutaneous melanoma was developed based onmultidimensional high-throughput data (19). The substantialheterogeneity of tumor cells and their microenvironments isreflected in the complex patterns of genomic, epigenomic, andtranscriptomic alternations.With rapid advances of immunother-apy in melanoma and other types of cancers, we urgently need todiscover diagnostic and prognostic markers that could be appliedin clinical settings. In this study, we further utilized datasetsgenerated for melanoma tumors by The Cancer Genome Atlas(TCGA) to investigate ESRP1-based cell-type plasticity in mela-noma and its implication in melanoma biology and treatment.

Materials and MethodsSamples and datasets

TCGA RNA-Seq expression and methylation (HumanMethyla-tion450 probes) data and corresponding clinical metadata wereaccessed from TCGA public access data portal released beforeJanuary 27, 2015 (https://tcga-data.nci.nih.gov/tcga/dataAccess-Matrix.htm). Exon level mapping files generated for Cancer CellLine Encyclopedia (CCLE) RNA-Seq gene expressionwere accessedthrough the ArrayStudio OncoLand portal (http://www.omicsoft.com). Cap analysis gene expression (CAGE) data were queriedfromFANTOM5 (http://fantom.gsc.riken.jp). Other RNA-Seq dataincludingmelanocyte cell lines (GSM1138580 and GSM819489),shMITF knockdown experiments (SRP048577), mouse melano-cyte and melanoma samples (SRX478034, SRX478033,ERX386690, and SRX475297) were downloaded from NCBI'sShort Reads Archive. Somatic point mutation data were accessedfrom GDAC Firehose (http://gdac.broadinstitute.org/).

The melanocyte cell line used in real-time PCR was pur-chased from SicenCell Research Laboratory, Inc., in 2004.Other melanoma cell lines, either commercially available orestablished from low-passage culture of cutaneous melanomalesions and authenticated by branches of Ludwig Cancer

Research, were acquired between 1998 and 2013. Specifically,Hs294T was acquired from ATCC. T618A, Me246, Me260, andMe275 were established and authenticated at the Ludwig Insti-tute for Cancer Research, Lausanne Branch. The SK-MEL seriesof melanoma cell lines were established and authenticated atthe Ludwig Institute for Cancer Research, New York branch, atthe Memorial Sloan Kettering Cancer Center. LM-MEL-62 wasestablished and authenticated at the Ludwig Institute for Can-cer Research Melbourne, Austin Branch, at Melbourne, Austra-lia. Cell lines were usually cultured for a maximum of threepassages.

Defining ESRP1 groups based on exon expressionCalculation of exon expression is based on exon quantification

file; gene expression is based on normalized gene expressionvalues in RSEM (RNA-Seq by Expectation-Maximization). TheESRP1-low cutoff is defined by ESRP1 gene expression less than125 RSEMafter examination of the ESRP1 expression distributionin all samples. Among the rest that have ESRP1 gene expression>125 RSEM, samples with an exon truncation ratio �0.65 aredefined as ESRP1 truncated; samples with a truncation ratiosmaller than 0.65 are grouped as ESRP1 full length. Consideringthat exons 14 and 15 are alternatively expressed exons, the exonratio used in assigning ESRP1-based subgrouping is defined asfollows: sum (exon 13 þ exon 16) RSEM/sum (exon (1–12) þexon 13þ exon 16) RSEM. The truncation ratio cutoff of 0.65 wasdefined after examination of its distribution. The heatmap in Fig.1 was generated without median removal. Median removal wasused in all other expression heatmaps.

Gene expression signaturesCytolytic activity is calculated as the geometric mean of GZMA

and PRF1 using log2 expression values. We used normal tissuegene expression data obtained from theGTEx portal (http://www.gtexportal.org/home/) to identify two skin-specific genes (KRT2and LOR).With the highest KRT2/LOR expressing non-skin tissue,itsmedian expression of KRT2/LOR plus 3� standard deviation islower than RPKM (reads per kilobase of transcript per millionmapped reads) 16. We therefore call melanoma samples withKRT2 or LOR expression above RPKM 16 as keratinocytecontaminated.

Quantitative real-time reverse transcription PCRAnalysis of cDNA samples was performed in duplicate for

ESRP1 and for the reference gene within each experiment usingthe CFX96 Real-Time PCR system (Biorad) and Taqman platform(Applied Biosystems). TFRC was amplified as an internal refer-ence gene. The PCR primers and probes for ESRP1(Hs00214472_m1, flanking exons 11–2; Hs00936411_m1,flanking exons 13–14) and TFRC internal control gene(4326323E) were purchased from Applied Biosystems. Primersused for PCR amplification were chosen to encompass intronbetween exon sequences to avoid amplification of genomic DNA.

Cloning of full-length ESRP1 and transfection into SKMEL28PCR was undertaken with High Fidelity Taq polymerase (Invi-

trogen) using cDNA from the BT20 cell line: Forward 50-ATGACGGCCTCTCCGGATTACTTGGTG-30 and reverse 50-AATA-CAAACCCATTCTTTGGGT-30. The PCR product was recoveredand cloned in pCDNA pcDNA3.1/V5-His using the pcDNA3.1/V5-His TOPO TA Expression Kit (Invitrogen). The correct full-

ESRP1 Expression Marks Melanoma Cellular State

www.aacrjournals.org Cancer Immunol Res; 4(6) June 2016 553




length clone was determined by Sanger sequencing using the T7promoter and V5 reverse primer.

Transfection was performed using Effectene (Qiagen) accord-ing to the manufacturer's instructions. Western blotting wasperformed as previously described (15). The antibodies usedincluded: mouse monoclonal anti-V5 (clone 2F11F7; Life Tech-nologies) and anti–ESRP-1/2 (clone 23A7.C9; Rockland Immu-nochemicals) and rabbit polyclonal anti-actin (20-33; Sigma-Aldrich).

Statistical testsThe Krustal–Wallis test (nonparametric ANOVA) was used for

multigroup significance tests. Breslow thickness was categorizedin four levels as I, < 1mm; II, < 2mm; III, < 4mm; and IV, 4þmm.The Fisher exact test is used for testing sample distribution. TheKaplan–Meier log-ranked test was performed for survival bene-fits. All tests were done in R, Graphpad, SSPSv12 or ArrayStudio.

ResultsInverse correlation of ESRP1 expression to tumor-associatedimmune cytolytic activity

The use of antibodies to CTLA-4 or PD-1 has induced remark-able and durable antitumor responses in patients withmelanomaand other types of cancers (20). However, only a minority of

patients treated achieve complete tumor regression (21, 22).Therefore, identifying predictive immunotherapy biomarkers iscritical for both selecting appropriate patients for immunotherapyand for our complete understanding of the mechanism of actionof these agents. It has been reported that PD-L1 expression isregulated bymicroRNA-200/ZEB1 in lung cancer (23), suggestingthat tumor EMT status may be associated with immunotherapyresponse. To examine the relationship between EMT and tumorimmune cytolytic activity, we interrogated multiple cancer typesin the TCGA database.

We used ESRP1 as a biomarker of EMT status because itsexpression is highly restricted to epithelial cells and its level istightly linked to the selective expression of epithelial or mesen-chymal-specific alternative splice forms of multiple genes such asFGFR2 (15, 24). We also applied a two-gene expression signature(PRF1 and GZMA) of immune cytolytic activity (25). These genesencode perforin and granzyme A proteins, which are specificallyexpressed by CD8þ cytotoxic T cells upon activation. The two-gene signature significantly correlates with T-cell markers, includ-ing cytotoxic T lymphocyte (CTL) markers, as well as with pan-cancer survival benefits. As shown in Fig. 1, ESRP1 expression wasinversely correlated to this signature of cytolytic activity in manycancer types, including lung, colon, prostate, breast, bladder, andthyroid, with significant P values (< 1 � 10�10; SupplementaryTable S1). Although a trend of inverse correlation exists in other

Colon

BladderBLAD PAAD THCA

BRCALUSCCOADC

ytol

ytic

act

ivity

Cyt

olyt

ic a

ctiv

ity

Cyt

olyt

ic a

ctiv

ityC

ytol

ytic

act

ivity

Cyt

olyt

ic a

ctiv

ityC

ytol

ytic

act

ivity

Pancreas Thyroid

108

610

50

86

4

42

108

64

20

10

105

0

128

64

2

9

r = −0.40 r = −0.31 r = −0.25

r = −0.32r = −0.45r = −0.37

1010 811 12 12 14

10ESRP1ESRP1ESRP1

9 11 12108 1210864 12

13 9 10 11 12 13 14

Lung squamous Breast non-TNBC

Figure 1.ESRP1 expression is inversely correlatedwith immune cytolytic activity in TCGAcarcinomas. Cytolytic activitymeasurementwasundertakenas described inMaterialsand Methods. x-axis, ESRP1 expression in log2 scale; y-axis, cytolytic activity in log2 scale. Primary and metastatic tumors have been included but samplesthat do not express ESRP1 (RSEM < 1) have been removed. BLCA, bladder urothelial carcinoma; BRCA, breast invasive carcinoma; COAD, colon adenocarcinoma;LUSC, lung squamous carcinoma; PAAD, pancreatic adenocarcinoma; THCA, thyroid carcinoma; TNBC, triple-negative breast cancer.

Yao et al.

Cancer Immunol Res; 4(6) June 2016 Cancer Immunology Research554




cancer types, the Pearson correlation coefficient has shown lessstatistical significance (Supplementary Table S1).

Tissue-specific expression of truncated ESRP1 transcript inmelanoma and normal melanocyte

Intriguingly, melanoma, a tumor of neural crest origin (26),had high ESRP1 expression compared with other neural crest–derived tumors, such as glioblastoma and low-grade glioma(Supplementary Fig. S1A). This was also observed in cancer celllines [Cancer Cell Line Encyclopedia (CCLE) dataset], whereESRP1 was highly expressed in cells of epithelial origin, as is thecase in most carcinoma cell lines, but not expressed in cells ofmesenchymal origin, such as sarcomas and lymphomas, norin other neuronal tumor cell lines (Supplementary Fig. S1B).ESRP1 expression was high in cell lines of "skin" melanomaorigin (61/62 CCLE skin cell lines are from melanoma).

A close examination of ESRP1 expression at the exon level inmelanoma revealed that a majority of tumors do not express full-length ESRP1 transcript. Instead, only the last four exons (exons13–16) were highly expressed and presumably would not gen-erate ESRP1 protein that carries normal function. Based on theabsolute ESRP1 expression level and the relative exon 13–16expression ratio (see Materials and Methods), we classified mel-anomas in three subgroups (Fig. 2A). In approximately 20% ofmelanomas, designated ESRP1-low, ESRP1 was expressed at lowlevels (RSEM value of less than 125). In approximately two thirdsof melanomas, the ESRP1 truncated form was prominent (exonratio above 0.65), whereas about 10% of all tumors express full-length ESRP1 transcripts in substantial amounts (Table 1). Nota-bly, the one normal control sample in the TCGA melanomadataset fell into the ESRP1-truncated group, suggesting that thetruncated transcript form was not tumor specific, but ratherproduced during normal melanocyte differentiation. We did notobserve any aberrant change in the ESRP1 homolog, ESRP2,which also has epithelial-specific expression, although higherESRP2 expression was associated with the ESRP1 full-lengthgroup, as expected (Fig. 2A).

When tumor sites were aligned to ESRP1 groups, we observedenrichment of primary tumors inside the ESRP1 full-length group(Fig. 2A and Table 1). This raised a concern about whether full-length ESRP1 expression was from contaminating keratinocytes,which are often found in primary melanomas. Using gene expres-sion data from normal tissue, obtained from the GTEx portal(http://www.gtexportal.org/home/), we identified two skin-spe-cific genes (KRT2 and LOR) with strong expression (Supplemen-tary Fig. S2A), and used these two genes to estimate keratinocytecontamination in TCGA samples. Indeed, about half of ESRP1full-length tumors were KRT2 and/or LOR positive. After remov-ing these contaminated samples, a portion of melanomas stillexpressed full-length ESRP1. Thesemelanomas had been collectedfrom different tumor sites, including primary, regional subcuta-neous, lymph node, and distant metastasis sites, with no enrich-ment of any particular tumor site (Supplementary Fig. S2B).Similar findings were observed for ESRP1 expression in the CCLEcell line database, in which 4 out of 53 melanoma cell lines hadfull-length ESRP1 expression and 10 cell lines had low ESRP1 ex-pression (Fig. 2B). Therefore, although ESRP1 full-length expres-sion in some samples may be introduced by keratinocyte contam-ination, some melanoma tumors did express full-length ESRP1.To avoid signals from contaminating keratinocytes, we discardedKRT2/LOR-positive samples from further analysis (Table 1).

We further validated our findings by real-time PCR with aseparate collection of 42 melanoma cell lines derived fromcultures of primary tumors and a melanocyte cell line (seeMaterials and Methods). To detect the truncated transcripts, weused two sets of PCR primers to amplify cDNA flanking ESRP1exons 11 and 12 or exons 13 and 14, respectively, and determinethe relative abundance of exon 13–14 versus exon 11–12 pro-ducts. In comparison with normal kidney and colon samples, allmelanoma cell lines except one (SKMEL128), as well as thenormal melanocyte cell line (HEM), exhibited an increased usageof exons 13–14, confirming that the truncated ESRP1 transcriptpredominates in themajority of melanomas (Fig. 2C). Because inthe TCGA dataset we only detected expression of this truncatedform inmelanoma, and thepresence of this particular transcript inother types of cancers is rare (Supplementary Fig. S1B), expressionof the truncated ESRP1 was apparently melanocyte specific.

Mechanism of truncated ESRP1 expression in melanomaTo further investigate the truncated form of ESRP1, we per-

formed additional analyses. First, we tested its origin by cloningfull-length cDNA of ESRP1 into an expression vector and tran-siently transfecting into SKMEL28, a melanoma cell line thatpredominantly expresses the truncated ESRP1 transcripts (Fig.2C). Only full-length ESRP1 protein was detected in the trans-fected cell lysate by Western blotting after 48 and 72 hours (datanot shown). This suggests that SKMEL28 cells could express andsustain expression of full-length ESRP1 and the truncated tran-script was not the post-transcriptional processed product from thefull-length ESRP1. Because the region of ESRP1 recognized by themonoclonal antibody raised against the full-length mouse Esrp1is unknown, we could not exclude the possibility that the trun-cated ESRP1 protein was also produced in these cells.

We further hypothesized that an alternative transcriptional startsite (TSS) could direct expression of the truncated ESRP1 tran-script. To test this assumption, we queried high-throughputsequencing mapping outputs of CAGE libraries, which capturethe 50 end sequence of mRNAs in the FANTOM project (http://fantom.gsc.riken.jp). Indeed, in two skin melanoma cell lines(COLO-679 and G-361), most CAGE reads for ESRP1 weremapped to a region inside exon 13 (Fig. 3A). As a control, CAGEreads were seen to be aligned at the 50 end of MLANA (themelanoma-specific melan-A gene) promoter region, as expected.In other epithelial cell lines, CAGE reads were almost exclusivelymapped to ESRP1 50 end promoter region. Thus, the truncatedESRP1 resulted from alternative TSS usage at an approximatelocation of chr8:95690440 (hg19) at the start of exon 13 inmelanoma. The two known functional domains (RNaseH andRRM domains) are missing from the truncated ESRP1, so theproduct from this new isoform is unlikely to retain its normalfunction (Fig. 3B).

Upstream sequence analysis identified a consensus E-boxmotif(CACGTG) located�57-bp upstream of ESRP1 exon 13 (Fig. 3B),which is a known binding site for MITF, a master regulator ofmelanocyte development and a known melanoma oncogene (4,5). Several lines of evidence support the idea that MITF regulatedthe expression of truncated ESRP1 through binding to the E-boxmotifs inside intron 12 and initiated new transcription fromwithin exon 13. First, correlation of global gene expression totruncated ESRP1 expression resulted in MITF as the most corre-lated gene, with a Pearson coefficient of 0.77 (Fig. 3C), and withmany of the remaining top correlated genes being MITF target






A Low Truncated Full ESRP1exon_1exon_2exon_3exon_4exon_5exon_6exon_7exon_8exon_9exon_10exon_11exon_12exon_13exon_14exon_15exon_16

exon_1ESRP1

log2 (exon RSEM)Normal−9 4

log2 (exon RSEM)−4 82

8exon 11−12exon 13−14, normalexon 13−14, tumor

Normal Melanoma cell lines

1.0

0.5

0.0exon

-14

pct

6

4

2

0

Nor

m e

xon

expr

Kid

ney

Col

on

HE

M

Hs2

94T

T618

A

ME

260

LMM

EL6

2

SK

ME

L23

SK

ME

L28

SK

ME

L128

ME

275

ME

246

exon_2exon_3exon_4exon_5exon_6exon_7exon_8exon_9exon_10exon_11exon_12exon_13exon_14exon_15exon_16

ESRP1RSEM

Trunc.ratio

1

keratinprimarycu/subculymphdistant

ESRP2RSEM

B C

Kid

ney

Col

on

HE

M

Hs2

94T

T618

A

ME

260

LMM

EL6

2

SK

ME

L23

SK

ME

L28

SK

ME

L128

ME

275

ME

246

Figure 2.Detection of truncated ESRP1 transcripts inmelanoma. A, ESRP1 expression at gene and exon levels inmelanoma (tumor samples, n¼471; normalmelanocyte, n¼ 1).Melanoma samples are organized into three groups (see Materials and Methods) and ordered by ESRP1 expression. ESRP1 exon expression heatmap is shownat top, followed by ESRP1 expression values in a barplot. A truncation ratio of 0.65 was used to call full-length ESRP1 tumors (see Materials and Methods). Positionof the normal sample is marked by a vertical yellow line. Tumor sites for primary, regional subcutaneous, lymph node, and distant metastasis tumors aremarked by dark-blue lines. The "keratin" lane marks potential contamination of keratinocytes (see Materials and Methods). B, expression of ESRP1 at exon levelin CCLE cancer cell lines. C, detection of low, truncated, and full-length ESRP1 expression in melanoma cell lines by real-time PCR. Two probe sets amplifyregions flanking exon 11–12 and exon 13–14, respectively. Top, relative signal intensity of the two probe sets normalized by normal kidney. Bottom, percentageof exon 13–14 amplicon out of the total signal. HEM, Hermes cell line.

Yao et al.





genes. Second, the consensus E-box found in humans is conservedin monkeys and apes but not in mice and rats (Fig. 3B andSupplementary Fig. S3). Consistently, we detected little mouseEsrp1 expression using mouse nevi and melanoma RNA-Seqsamples (see Materials andMethods). Third, knockdown ofMITFby shRNA in 501MEL and Hermes cell lines leads to reducedexpression of MITF target genes (e.g. CDK2) and of truncatedESRP1 (Fig. 3D). Together, these results suggest that MITF reg-ulates truncated ESRP1 expression in melanomas.

Epithelial and mesenchymal cell phenotypes are associatedwith ESRP1 status in melanoma

We examined the expression of a set of epithelial and mesen-chymal markers in the three ESRP1 subgroups after removingkeratinocyte-contaminated samples (Fig. 4A). ESRP1-low tumorsexhibited a strong expression signature of mesenchymal markerslike N-cadherin (CDH2) and TWIST1. About half of melanomaswith full-length ESRP1 expression displayed higher expression ofan epithelialmarker panel including genes such as for keratin. This

Table 1. Distribution of melanoma tumors on ESRP1 expression status and tumor site before and after (before/after) removing potentially keratinocytecontaminated samples

ESRP1-low ESRP1-trun ESRP1 full-length Total

Primary tumor 4a/3a 69/56 32a/5 105/64Regional lymph node 56/52 154/153 12/12 222/217Cutaneous and subcutaneous 19/19 49/47 6/3 74/69Distant metastasis 13/13 49/47 6/5 68/65Total 92 321/303 56/25 469/415a, P value < 0.0001 by the Fisher exact test.

Figure 3.Melanomas express a noveltranscriptional isoform of ESRP1. A,alternative TSS of ESRP1. Top, CAGElibraries capture reads mapped to TSSsites at MLANA (top) and ESRP1(bottom) gene loci. Depth of readscoverage is labeled on the right. Exonsin alternative transcripts are drawn infilledboxes. B, ESRP1 coding structure.The E-box location is marked out by ared box. Exon 13 is labeled to show therelative exon positions. AlternativeTSS for the truncated ESRP1 withinexon 13 is marked by a red arrow.Functional domains encoded byESRP1, including RNaseH-like domainand RRM (RNA recognition motif) areshown in colored boxes in the genestructure. Sequence surrounding thejunction of intron 12 and exon 13 isshown belowwith comparison amongother species. C, correlation betweenMITF and ESRP1 gene expression.Samples are categorized into ESRP1-low and ESRP1-trun groups. D, shMITFdownregulates expression of thetruncated ESRP1. CDK2, a known MITFtarget, andACTB (Actin beta) are usedas positive and negative controls ofthe experiment, respectively.






was not restricted to primary tumors, but was also found intumors from regional cutaneous/subcutaneous locations, lymphnodes, and distant metastatic sites. Approximately two thirds oftumors expressing truncated ESRP1 seemed to reside in a statebetween epithelial and mesenchymal, even though ESRP1 wasfunctionally compromised in these tumors. This suggested thatloss of ESRP1 function per se is not sufficient to render a mesen-chymal phenotype, at least inmelanoma. Surprisingly, E-cadherin(CDH1) had higher expression in both ESRP1 full-length– andESRP1-truncated groups. We also examined the isoform switch inthe FGFR2 gene, which showed a clear transition from the epi-thelial isoform (FGFR2-b) in ESRP1 full-length tumors to themesenchymal isoform (FGFR2-c) in tumors with truncated and

low ESRP1 (Fig. 4B). This finding not only validated our compu-tational prediction of ESRP1 full-length expression, but alsoconfirmed that truncated ESRP1 was unable to exert its normalfunction to induce isoform switch in the FGFR2 gene.

Previously, several studies defined molecular subtypes of mel-anoma using gene expression profiling (3, 27–29). Among those,Hoek and colleagues classifiedmelanoma into "proliferative" and"invasive" subtypes based on the microarray profiling of 86cultured melanomas (3). When we compared the 105-gene sig-nature from that study with our ESRP1 subgroups, we observed agood correlation of the ESRP1-low group to the invasive subtype,as well as the truncated-ESRP1 group to the proliferative subtype(Fig. 4C). The ESRP1 full-length group, however, comprised

LowA

B

C

Truncated Full ESRP1

Mesenchym

alE

pithelial

TWIST1CDH2ZEB1TGFB3PDGFALAMB1LAMC1TGFB1CRB3DSPGRHL2KRT5KRT14PROM2CLDN4CDH1

ESRP1

FGFR2b

FGFR2c

Exon_1

ESRP1

MITFRSEM

Proliferative

Invasive

Exon_21

Low Truncated Full

Low Truncated Full

Figure 4.ESRP1 subgroups are associated withepithelial–mesenchymal cellularproperties. A, expression heatmap ofmolecular markers for epithelial andmesenchymal cells. B, FGFR2expression at exon level shows that themesenchymal and epithelial forms aredominantly expressed in non–ESRP1full-length and ESRP1 full-lengthsubgroups, respectively. C, heatmap ofthe 105-gene signature and MITFexpression levels (RSEM) are alignedto ESRP1 subgroups. Samples in allheatmaps are in the same order asin Fig. 2.

Yao et al.





samples with both proliferative and invasive signatures. MITF isapparently a dominant factor in the proliferative subtype becausemany MITF target genes were in the 105-gene list and highlyexpressed in the proliferative subtype, whereas in the ESRP1-low/invasive group, MITF expression was low (Fig. 4C). This impliesthat increased MITF level may actively block a mesenchymalphenotype, in agreement with previous findings (3).

ESRP1-low melanomas are associated with increased immunecytotoxicity and favorable patient survival

We examined immune cytolytic activity using cytotoxic T-cellmarker genes and the two-gene signature in three ESRP1-basedmelanoma subgroups after removing all primary melanomasamples, because these are not particularly relevant to immunecheckpoint therapy. This showed a statistically significant increaseof cytolytic activity in the ESRP1-lowmelanomas (Fig. 5A), whichwas consistent with our findings in other cancer types (Fig. 1).Predicted cytolytic activity between the ESRP1 full-length and theESRP1-truncated groups was not significantly different, suggest-ing the correlation of cytolytic activity was more related to themesenchymal phenotype rather than to the epithelial phenotype.

We next examined the expression of the PD-1, PD-L1, PD-L2, andCTLA-4 genes in ESRP1 subgroups (Fig. 5A and B). All thesemolecules, which are targeted by immune checkpoint blockageantibodies, had increased expression in the ESRP1-low tumors.We also examined the possible relationship of ESRP1 grouping tomutational status and clinical parameters. No particular associ-ation was found between BRAF and NRAS mutation and ESRP1groups, although KIT mutations seem to be absent in ESRP1-lowtumors (Fig. 5A). The ESRP1 groups were also not associated withthe total number of somatic mutations. In the TCGA melanomadataset, the ESRP1 grouping had a moderate association toBreslow depth, with ESRP1-low tumors having the thinnestBreslow depth (Fig. 5B). Kaplan–Meier survival analysis on TCGAmelanoma patients stratified by ESRP1 status showed a trend ofmore favorable survival of ESRP1-low tumors (P ¼ 0.057, Fig.5C), which is consistent with the lower Breslow depth values andhigher active immune signatures found within this group.

In summary, our data show an inverse correlation betweenESPR1 expression and tumor-associated immune cytolytic activityin multiple tumor types. Through the study of altered ESRP1transcription in melanoma, we gained further insights on the

A Low Truncated Full ESRP1PDCD1LG2/PD-L2CD274/PD-L1CTLA4CD8ACD3EGZMAPRF1PDCD1/PD-1NRASBRAFNF1KITBreslow

Low

Trun Fu

ll

Low

Trun Fu

ll

Low

Trun Fu

ll

Low

Trun Fu

ll

Low

Trun Fu

ll

Stage

B PD-L2 CTLA-4 Cytolytic Breslow Tumor size

C

15 P = 4.2e-6 P = 0.0014 P = 0.0014 P = 0.0028 P = 0.14

10

Gen

e ex

pres

sion

Thic

knes

s (m

m)

Pat

holo

gic

T

5

0

15

10

5

0

LowTrun/Full

1.0

0.8

Pro

babi

lity

0.6

0.4

0.2

0 2,000 6,000

P = 0.057

10,000Survival days

0.0

15

10

5

0

15 65

43210

10

5

0

Figure 5.Association of ESRP1 expressionstatus with tumor-associatedcytolytic activity and clinicalparameters. A, comparison of immunecytolytic activity and related genes inESRP1 subgroups. Top, heatmap ofexpression of immune checkpointtarget genes and active T-cellmarkers. Bottom, alignment ofmutations and clinical parameters toESRP1 subgroups. B, boxplotpresentation of expression levels ofPD-L2, CTLA-4, cytolytic activity,Breslow depth, and tumor T stages inESRP1 subgroups. C, Kaplan–Meiersurvival analysis of TCGA melanomacases stratified by ESRP1 status.






complex heterogeneity ofmelanoma cell properties during tumorprogression, and strengthened a link between tumor-intrinsicEMT status and tumor microenvironment in melanoma. Resultsfrom our study highlight the potential utility of ESRP1 status inpredicting response to checkpoint blockade immunotherapy. Theconceptual link between tumor EMT status and activated infil-trating lymphocytes may have further implications in the immu-notherapy field and will need further elucidation.

DiscussionCancer immunotherapy by immune checkpoint inhibitors,

adoptive T-cell therapy, and therapeutic vaccines has advancedto the first line of treatments for many metastatic cancers, espe-cially melanoma (20, 30, 31). Contributing to this opportunity isthe rich stromal cell–orchestrated tumor microenvironment, par-ticularly those tumors with high numbers of tumor-infiltratinglymphocytes (TIL), as is the case inmelanoma.Here, we show thatmelanoma mesenchymal-like tumors were associated withenhanced immune cytolytic activity.Moreover, this phenomenonwas corroborated by a negative correlation between immunecytolytic activity and epithelial phenotypes, and especially withthe fact thatESRP1 expressionwas inversely associatedwithhighercytolytic activity in many other cancer types (Fig. 1). RCC isanother cancer type that has shown relatively high sensitivity toimmunotherapy (32). Previously, we have demonstrated thatccRCChas amoremesenchymal-like phenotype (15).Our currentresults suggest that tumor cells of mesenchymal nature are able tobetter acquire TILs and are well-suited to immunotherapy. Withthe success of immune checkpoint blockage therapy (anti–PD1/PDL-1 and/or anti–CTLA-4) in some patients, an important issueis why some patients are either not responsive or later developresistance. Among the many factors that contribute to tumorimmunity, such asHLAabundance, neoantigen load, and intrinsicproperties, our findings suggest that the cellular state also has animpact on immune response susceptibility.

In this study, we have described the identification of a melano-cyte-specific ESRP1 mRNA transcript. This highly expressed iso-form is a result of de novo transcription starting from exon 13,leading to a shorter mRNA (truncated ESRP1) that comprises onlythe last four exons and loses the region coding for the twofunctional domains of the protein. Although the TCGAmelanomadataset contains only one normal control sample, which showedtruncatedESRP1expression,wedid confirmthisfindingbyRT-PCRin a normal melanocyte cell line (HEM; Fig. 2C). Additionally,publicly accessible RNA-Seq data of independently derived mela-nocyte specimens (NCBIGEOGSE46805 andGSE33092) all showtruncated ESRP1 expression in melanocytes (33, 34).

The TCGA melanoma dataset is a heterogeneous collection ofsamples fromprimary tumors, regional, lymphnodes, anddistantmetastases. Our initial findings that primary tumors were signif-icantly enriched in ESRP1 full-length tumors (Fig. 2A) raised theconcern of keratinocyte contamination as the cause of ESRP1 full-length expression. Using a two-gene signature (KRT2/LOR), wewere able to predict and remove a majority if not all keratinocytecontaminated samples from our study. KRT2 encodes keratin 2E,an epidermal keratin expressed largely in the upper spinous layerof epidermal keratinocytes. LOR encodes loricrin,which is amajorprotein component of the cornified cell envelope found in ter-minally differentiated epidermal cells.We show that expression ofKRT2 and LOR is correlated and highly skin specific (Supplemen-

tary Fig. S2A), and the function of these genes suggests they areauthentic keratinocyte genes. Even after removal of contaminatedsamples, a small portion of melanomas still expressed full-lengthESRP1, which suggests that expression of full-length ESRP1 couldbe an intrinsic tumor property.

Based on ESRP1 status, we divided melanomas into threesubgroups. Approximately two thirds of melanomas expressedthe truncated ESRP1 form that was also observed in normalcontrols, so this likely represents the default ESRP1 status. Basedon the expression of EMT markers, the state of these samples isneither distinctly mesenchymal nor epithelial. Compared withloss of expression of other epithelial genes, the E-cadherin(CDH1) gene is still highly expressed in this subgroup. We haveobserved that some tumors lose ESRP1 expression and are mes-enchymal, andwe show that thismay be linked to the loss ofMITFexpression. For other tumors, which have epithelial characteris-tics, full-length ESRP1 is expressed. The underlying basis andmechanisms for these alterations will need further investigation.However, our preliminary results revealed loss ofESRP1promotermethylation in the ESRP1 full-length group, indicating theinvolvement of epigenetic regulation. Also, a few samples thatexpressed full-length ESRP1 did not highly express epithelialmarkers. A closer examination of these individual samples iden-tified at least one sample that expressed another novel truncatedform, starting from exon 3 of ESRP1, and therefore had escapedour initial computational screen. Consequently, other factorsmay influence the functional status of ESRP1. The ESRP1-lowsubgroup was associated with better TILs and overall survivalcompared with the other two subgroups.

The identification of effective markers for diagnosis and pre-diction of treatment response in cancer immunotherapy is impor-tant to help guide effective intervention strategies for individualpatients. Our study indicates that ESRP1-low melanomas withgreater immune cytolytic activity have a higher probability ofresponding to immune checkpoint blockade therapy and suggestsa path toward the use of ESRP1 and other EMT markers asinformative biomarkers for immunotherapy.

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

Authors' ContributionsConception and design: J. Yao, R.L. Strausberg, Q. ZhaoDevelopment of methodology: J. Yao, O.L. Caballero, J.S. Cebon, Q. ZhaoAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): O.L. Caballero, D. Rimoldi, A. Behren, J.S. Cebon,Q. ZhaoAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): J. Yao, Y. Huang, J.S. Cebon, Q. ZhaoWriting, review, and/or revision of the manuscript: J. Yao, O.L. Caballero,C. Lin, A. Behren, J.S. Cebon, R.L. Strausberg, Q. ZhaoAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): A. Behren, M.-C. Hung, J.N. WeinsteinStudy supervision: J.S. Cebon, Q. Zhao

Grant SupportThis work is funded by Ludwig Cancer Research and Regeneron Pharma-

ceuticals Inc.The costs of publication of this articlewere defrayed inpart by the payment of

page charges. This article must therefore be hereby marked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Received October 14, 2015; revised February 29, 2016; accepted March 3,2016; published OnlineFirst April 4, 2016.


Yao et al.




References1. Hoek KS, Goding CR. Cancer stem cells versus phenotype-switching in

melanoma. Pigment Cell Melanoma Res 2010;23:746–59.2. Ghislin S, Deshayes F, Middendorp S, Boggetto N, Alcaide-Loridan C.

PHF19 and Akt control the switch between proliferative and invasive statesin melanoma. Cell Cycle 2012;11:1634–45.

3. Hoek KS, Schlegel NC, Brafford P, Sucker A, Ugurel S, Kumar R, et al.Metastatic potential of melanomas defined by specific gene expres-sion profiles with no BRAF signature. Pigment Cell Res 2006;19:290–302.

4. Levy C, Khaled M, Fisher DE. MITF: master regulator of melanocytedevelopment and melanoma oncogene. Trends Mol Med 2006;12:406–14.

5. Strub T, Giuliano S, Ye T, Bonet C, Keime C, Kobi D, et al. Essential role ofmicrophthalmia transcription factor for DNA replication, mitosis andgenomic stability in melanoma. Oncogene 2011;30:2319–32.

6. Carreira S,Goodall J, Denat L, RodriguezM,Nuciforo P,HoekKS, et al.Mitfregulation ofDia 1 controlsmelanomaproliferation and invasiveness. GenDev 2006;20:3426–39.

7. Koludrovic D, Davidson I. MITF, the janus transcription factor of mela-noma. Fut Oncol 2013;9:235–44.

8. Grossmann AH, Yoo JH, Clancy J, Sorensen LK, Sedgwick A, Tong Z, et al.The small GTPase ARF6 stimulates beta-catenin transcriptional activityduring WNT5A-mediated melanoma invasion and metastasis. Sci Sig2013;6:ra14.

9. Bachmann IM, Straume O, Puntervoll HE, Kalvenes MB, Akslen LA.Importance of P-cadherin, beta-catenin, and Wnt5a/frizzled for progres-sion of melanocytic tumors and prognosis in cutaneous melanoma. ClinCancer Res 2005;11:8606–14.

10. ChienAJ, ConradWH,MoonRT. AWnt survival guide: fromflies to humandisease. J Invest Dermatol 2009;129:1614–27.

11. ChienAJ,Moore EC, Lonsdorf AS, Kulikauskas RM, Rothberg BG, Berger AJ,et al. ActivatedWnt/beta-catenin signaling inmelanoma is associated withdecreased proliferation in patient tumors and a murine melanomamodel.Proc Natl Acad Sci U S A 2009;106:1193–8.

12. O'Connell MP, Marchbank K, Webster MR, Valiga AA, Kaur A, Vultur A,et al. Hypoxia induces phenotypic plasticity and therapy resistance inmelanoma via the tyrosine kinase receptors ROR1 and ROR2. CancerDiscov 2013;3:1378–93.

13. Warzecha CC, Shen S, Xing Y, Carstens RP. The epithelial splicing factorsESRP1 and ESRP2 positively and negatively regulate diverse types ofalternative splicing events. RNA Biol 2009;6:546–62.

14. Warzecha CC, Sato TK, Nabet B, Hogenesch JB, Carstens RP. ESRP1 andESRP2 are epithelial cell-type-specific regulators of FGFR2 splicing. MolCell 2009;33:591–601.

15. Zhao Q, Caballero OL, Davis ID, Jonasch E, Tamboli P, Yung WK, et al.Tumor-specific isoform switch of the fibroblast growth factor receptor 2underlies the mesenchymal and malignant phenotypes of clear cell renalcell carcinomas. Clin Cancer Res 2013;19:2460–72.

16. Theveneau E, Mayor R. Neural crest delamination and migration: fromepithelium-to-mesenchyme transition to collective cell migration. DevBiol 2012;366:34–54.

17. AcloqueH,AdamsMS, FishwickK, Bronner-FraserM,NietoMA.Epithelial-mesenchymal transitions: the importance of changing cell state in devel-opment and disease. J Clin Invest 2009;119:1438–49.

18. ShakhovaO.Neural crest stem cells inmelanoma development. CurrOpinOncol 2014;26:215–21.

19. Cancer Genome Atlas Network. Genomic classification of cutaneous mel-anoma. Cell 2015;161:1681–96.

20. Pico de Coana Y, Choudhury A, Kiessling R. Checkpoint blockade forcancer therapy: revitalizing a suppressed immune system. Trends Mol Med2015;21:482–91.

21. TopalianSL,Hodi FS, Brahmer JR,Gettinger SN, SmithDC,McDermottDF,et al. Safety, activity, and immune correlates of anti-PD-1 antibody incancer. N Engl J Med 2012;366:2443–54.

22. Wolchok JD, Kluger H, CallahanMK, PostowMA, Rizvi NA, Lesokhin AM,et al. Nivolumab plus ipilimumab in advanced melanoma. N Engl J Med2013;369:122–33.

23. Chen L, Gibbons DL, Goswami S, Cortez MA, Ahn YH, Byers LA, et al.Metastasis is regulated viamicroRNA-200/ZEB1 axis control of tumour cellPD-L1 expression and intratumoral immunosuppression. Nat Commun2014;5:5241.

24. Warzecha CC, Jiang P, Amirikian K, Dittmar KA, Lu H, Shen S, et al. AnESRP-regulated splicing programme is abrogated during the epithelial-mesenchymal transition. Embo J 2010;29:3286–300.

25. Rooney MS, Shukla SA, Wu CJ, Getz G, Hacohen N. Molecular and geneticproperties of tumors associated with local immune cytolytic activity. Cell2015;160:48–61.

26. Sommer L. Generation ofmelanocytes fromneural crest cells. Pigment CellMelanoma Res 2011;24:411–21.

27. Bittner M, Meltzer P, Chen Y, Jiang Y, Seftor E, Hendrix M, et al. Molecularclassification of cutaneous malignant melanoma by gene expression pro-filing. Nature 2000;406:536–40.

28. John T, Black MA, Toro TT, Leader D, Gedye CA, Davis ID, et al. Predictingclinical outcome through molecular profiling in stage III melanoma. ClinCancer Res 2008;14:5173–80.

29. Winnepenninckx V, Lazar V, Michiels S, Dessen P, Stas M, Alonso SR, et al.Gene expression profiling of primary cutaneous melanoma and clinicaloutcome. J Natl Cancer Inst 2006;98:472–82.

30. Rosenberg SA, Restifo NP. Adoptive cell transfer as personalized immu-notherapy for human cancer. Science 2015;348:62–8.

31. Sharma P, Allison JP. The future of immune checkpoint therapy. Science2015;348:56–61.

32. Yang JC, Hughes M, Kammula U, Royal R, Sherry RM, Topalian SL, et al.Ipilimumab (anti-CTLA4 antibody) causes regression of metastatic renalcell cancer associated with enteritis and hypophysitis. J Immunother2007;30:825–30.

33. Flockhart RJ, Webster DE, Qu K, Mascarenhas N, Kovalski J, Kretz M, et al.BRAFV600E remodels the melanocyte transcriptome and induces BANCRto regulate melanoma cell migration. Genome Res 2012;22:1006–14.

34. Pawlikowski JS, McBryan T, van Tuyn J, Drotar ME, Hewitt RN, Maier AB,et al. Wnt signaling potentiates nevogenesis. Proc Natl Acad Sci U S A2013;110:16009–14.






2016;4:552-561. Published OnlineFirst April 4, 2016.Cancer Immunol Res Jun Yao, Otavia L. Caballero, Ying Huang, et al. Tumor-Associated Immune Cytolytic Activity

Mesenchymal Status and−Correlates with Epithelial in Malignant Melanoma ESRP1Altered Expression and Splicing of

Updated version

10.1158/2326-6066.CIR-15-0255doi:

Access the most recent version of this article at:

Material

Supplementary

http://cancerimmunolres.aacrjournals.org/content/suppl/2016/04/02/2326-6066.CIR-15-0255.DC1

Access the most recent supplemental material at:

Cited articles

http://cancerimmunolres.aacrjournals.org/content/4/6/552.full#ref-list-1

This article cites 34 articles, 11 of which you can access for free at:

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

Subscriptions

Reprints and

[email protected]

To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department

Permissions

Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://cancerimmunolres.aacrjournals.org/content/4/6/552To request permission to re-use all or part of this article, use this link



http://cancerimmunolres.aacrjournals.org/lookup/doi/10.1158/2326-6066.CIR-15-0255

http://cancerimmunolres.aacrjournals.org/content/suppl/2016/04/02/2326-6066.CIR-15-0255.DC1

http://cancerimmunolres.aacrjournals.org/content/4/6/552.full#ref-list-1

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

mailto:[email protected]

http://cancerimmunolres.aacrjournals.org/content/4/6/552


Altered Expression and Splicing of ESRP1 in Malignant...

Documents

Transcript of Altered Expression and Splicing of ESRP1 in Malignant...