Transition of Mesenchymal and Epithelial Cancer Cells Depends … · Tumor Biology and Immunology Transition of Mesenchymal and Epithelial Cancer Cells Depends on a1-4 Galactosyltransferase-Mediated

Tumor Biology and Immunology

Transition of Mesenchymal and Epithelial CancerCells Depends on a1-4 Galactosyltransferase-Mediated GlycosphingolipidsFrancis Jacob1, Shahidul Alam1,2, Martina Konantz3, Ching-Yeu Liang2,Reto S. Kohler2, Arun V. Everest-Dass4, Yen-Lin Huang2, Natalie Rimmer1,Andre Fedier2, Andreas Sch€otzau2, Monica Nunez Lopez2, Nicolle H. Packer4,5,Claudia Lengerke3,6, and Viola Heinzelmann-Schwarz2,7

Abstract

The reversible transitions of cancer cells between epithelial andmesenchymal states comprise cellular and molecular processesessential for local tumor growth and respective dissemination.Wereport here that globoside glycosphingolipid (GSL) glycosyltrans-ferase-encoding genes are elevated in epithelial cells and correlatewith characteristic EMT signatures predictive of disease outcome.Depletion of globosides through CRISPR-Cas9–mediated dele-tion of the key enzyme A4GALT induces EMT, enhances chemore-sistance, and increased CD24low/CD44high cells. The choleratoxin–induced mesenchymal-to-epithelial transition occurred

only in cells with functional A4GALT. Cells undergoing EMT lostE-cadherin expression through epigenetic silencing at the pro-moter region of CDH1. However, in DA4GALT cells, demethyla-tion was able to rescue E-cadherin–mediated cell–cell adhesiononly in the presence of exogenous A4GALT. Overall, our datasuggest another class of biomolecules vital for epithelial cancercells and for maintaining cell integrity and function.

Significance: This study highlights the essential role ofglycosphingolipids in the maintenance of epithelial cancer cellproperties. Cancer Res; 78(11); 2952–65. �2018 AACR.

IntroductionThe requirement for epithelial-to-mesenchymal transition

(EMT) is particularly well reported in ovarian carcinogenesis.Here, metastasis is established by the EMT-driven delaminationof tumor cells from the in situ tumor, followedby their penetrationinto the surrounding peritoneal cavity (1). The adaptation ofspreading ovarian cancer cells requires EMT for disseminationwhile its reversed process, mesenchymal-to-epithelial transition(MET), then fosters local tumor growth at themetastatic site. Thus,

EMT and MET frequently alternate and are actively involved indifferent phases of tumor progression (2).

Although the transition of cancer cells between epithelial andmesenchymal states is a central mechanism of tumorigenesis, itsmolecular regulation is not fully understood (3). Migrating cellsare exposed to a different microenvironment, which might con-versely again inducemorphologic andmolecular changes. DuringEMT, epithelial cells lose their cell–cell junctions and apical–basalcell polarity to undergo morphologic changes toward prolifera-tion, migration, and invasion as well as resistance to anoikis (4).The presence of a specific surface architecture of glycosphingoli-pids (GSL) in epithelial andmesenchymal cells suggests that thesestructuresmight be importantmediating changes in cell behavior.However, the precise functional role of GSLs in this context ispoorly understood.

Multiple oncogenic events and signaling pathways have beenimplicated in the induction of EMT, especially mediated by theTGFb (5, 6). Besides TGFb, the glucosylceramide synthase inhib-itor EtDO-P4 was also capable of inducing partial EMT accom-panied with decreased E-cadherin and increased vimentin expres-sion (7). Later, Guan and colleagues pointed out that the GSLrepertoire changes during treatment with EtDO-P4 and TGFb inNMuMG, HCV29, and MCF7 cell lines (8). Furthermore, anassociation of the mesenchymal transcription factor Zeb1 withelevated St3Gal5 and gangliosides was reported in epithelialNMuMG cells (9).

We have previously demonstrated that individual glycans areprognostic markers in ovarian carcinoma (10, 11). Hereby, GSLsare heterogeneously expressed in various cancer cells derived fromcell lines as well as primary patient samples (12, 13). In general,GSLs are divided into galactosylated or glucosylated ceramides.

1Glyco-Oncology, Ovarian Cancer Research, Department of Biomedicine, Uni-versity Hospital Basel, University of Basel, Basel, Switzerland. 2Ovarian CancerResearch Program, Department of Biomedicine, University Hospital Basel,University of Basel, Basel, Switzerland. 3Stem Cells and Hematopoiesis, Depart-ment of Biomedicine, University Hospital Basel, University of Basel, Basel,Switzerland. 4Institute for Glycomics, Griffith University, Gold Coast, Queens-land, Australia. 5Department of Chemistry & Biomolecular Sciences, Biomolec-ular Discovery & Design Research Centre, Faculty of Science and Engineering,Macquarie University, North Ryde, New South Wales, Australia. 6Division ofHematology, University Hospital Basel, University of Basel, Basel, Switzerland.7Hospital for Women, Department of Gynecology and Gynecological Oncology,University Hospital Basel, University of Basel, Basel, Switzerland

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

F. Jacob, S. Alam, and M. Konantz are the co-first authors of this article.

Corresponding Author: Francis Jacob, Glyco-Oncology, Ovarian CancerResearch, Department of Biomedicine, University Hospital Basel, Basel CH-4031,Switzerland. Phone: 41-61-265-92-48; E-mail: [email protected]

doi: 10.1158/0008-5472.CAN-17-2223

�2018 American Association for Cancer Research.

CancerResearch

Cancer Res; 78(11) June 1, 20182952

on April 14, 2021. © 2018 American Association for Cancer Research. cancerres.aacrjournals.org Downloaded from

Published OnlineFirst March 23, 2018; DOI: 10.1158/0008-5472.CAN-17-2223

http://crossmark.crossref.org/dialog/?doi=10.1158/0008-5472.CAN-17-2223&domain=pdf&date_stamp=2018-5-21

http://cancerres.aacrjournals.org/

Glucosylceramides are further classified into three major classesbased on the action of specific glycosyltransferases; globo-(A4GALT), ganglio- (B4GALNT1 and ST3GAL5), and neo-lacto(B3GNT5) –series (Fig. 1A). Besides different glycosyltransferases,further elongationof the carbohydrate chains is determinedby theintracellular localization within the endomembrane system andby the regulation of specific glycosyltransferases within the GSLsynthesis (14). The functional role of GSLs in EMT/MET processesis still unexplained, particularly the role of globo, (neo-) lacto,and ganglio series GSLs (Fig. 1A). Here, we address this questionand dissect the role of globosides during EMT by showing that (1)globosides are required for E-cadherin–mediated cell–cell adhe-sion and (2) that loss of globosides triggers EMT through amechanism involving DNA methylation at the CDH1 promoter.

Materials and MethodsCell culture

A total number of thirty-four different cell lineswere acquired viadifferent sources and maintained in house in appropriate growthmedia (Supplementary Table S1). All cell lines were cultured at37�C in a 95%humidified atmosphere containing 5%CO2. All celllines were authenticated using short tandem repeat (STR) profilingand regularly tested for the absence of Mycoplasma.

CRISPR-Cas9–mediated depletion of globosidesA detailed description of the CRISPR-Cas9 design, molecular

cloning, cell sorting strategy, and characterization of homo-zygously deleted cancer cells is provided in SupplementaryInformation.

Flow cytometryImmunostaining and flow cytometry was performed as

described previously (10, 12).

Analysis of glycans released from GSLs using PGC LC-ESI-MS/MS

A detailed description of analysis of enzymatically releasedmembrane glycans is provided in Supplementary Information.

In vitro assays for characterization of globoside-depleted cancercells

A detailed description of in vitro assays for studying cell pro-liferation, motility, colony formation, and anoikis is provided inSupplementary Information.

Molecular cloning and lentiviral transductionofA4GALT rescueand E-cadherin constructs

A detailed description is provided in SupplementaryInformation.

RT-qPCRTo extract RNA, 105 cells were seeded in 6-well plates, grown to

70%–80% confluence, and then washed twice with sterile PBSprior to total RNA extraction using the ReliaPrep RNA CellMiniprep System (Promega). RNA was eluted in 60 mL RNase-free water and RNA concentration measured using a NanoDropND-1000 spectrophotometer (Thermo Fisher Scientific). Anamount of 1 mg of RNA in a total volume of 20 mL was reversetranscribed using the iScript Reverse Transcription Supermix forRT-qPCR (Bio-Rad Laboratories). RT-qPCR was performed on

CDH1, VIM, A4GALT, and reference genes HSPCB, SDHA, andYWHAZ in 10 mL reactions containing 10 ng cDNA (initial totalRNA), 400 nmol/L forward and reverse primer (SupplementaryTable S2), nuclease-free water, and 1� GoTaq qPCR Master Mixwith low ROX as reference dye (Promega) on a ViiA 7 Real-TimePCR System (Applied Biosystems, Thermo Fisher Scientific).Quantitative PCR was performed in triplicates and analyzed asdescribed previously (15).

Western blot analysisWhole-cell lysates were obtained from subconfluent cultures.

Cells were lysed for Western blot analysis according to standardlaboratory protocols. The protein concentration of cell lysates wasdetermined by BCA Protein Assay Kit (Pierce, Perbio Science).Equal amounts of protein (20 mg) were loaded and separatedusing SDS-PAGE, followed by blotting onto a polyvinylidenedifluoride membrane (Amersham Biosciences). Later, the mem-brane was blocked with 5% (w/v) BSA (Sigma) in TBST andincubated with primary mAbs E-cadherin (1:1,000), vimentin(1:1,000), and tubulin (1:1,000) diluted in 5% (w/v) BSA in TBSTat 4�C overnight. Afterwards, the membranes were washed inTBST and incubatedwith corresponding horse radish peroxidase–conjugated secondary antibodies (anti-rabbit and -mouse) in 3%BSA in TBST for 3 hours at room temperature. Finally, afterwashing in TBST, detection was carried out with the Super SignalWest Dura Extended Duration Substrate (Life Technologies).

Confocal fluorescence microscopyCells were grown on polylysine glass slides attached to a 8-well

chamber, fixed with 4% paraformaldehyde (Polysciences) for 15minutes, permeabilized with 0.3% Triton X-100 (Sigma), andincubated with blocking buffer [5% (w/v) BSA fraction V (Sigma)in PBS] for 1 hour. Cells were then stained with E-cadherinantibody and mounted in ProLong Gold Antifade Reagent withDAPI (Cell Signaling Technology #8961). Fluorescence imageswere taken on a LSM 780 confocal microscope (Zeiss).

Zebrafish xenograft modelThe "Kantonales Veterinaeramt Basel- Stadt" approved animal

experiments and zebrafish husbandry. A4GALT and correspond-ing control cells were labeled with the fluorescent CellTrackerCM-DiI (Life Technologies) as previously described. In brief,nacre/transparent zebrafish were maintained, collected, grownand staged in E3 medium at 28.5�C according to standard pro-tocols (16). For xenotransplantation experiments, zebrafishembryos were anesthetized in 0.4% tricaine (Sigma-Aldrich) at48hours postfertilization and 75–100humanovarian cancer cellswere microinjected into the vessel-free area of the yolk or thezebrafish common cardinal vein (Duct of Cuvier) of a transgenicTg(kdrl:eGFP) line, respectively. Embryos were incubated for 1hour at 28.5–29�C for recovery and then screened for the presenceof fluorescent human cancer cells in the yolk. Fish harboring redcells were incubated at 35�C as described before (17). Five daysafter transplantation, embryos were screened microscopically fortumor formation, extravasation, and secondary tumor mass for-mation using a Zeiss LSM 710 or a Leica TCS SP5 confocalmicroscope. Fish were furthermore dissociated into single cellsas described previously (18) and cells analyzed on a BDAccuri C6for CM-DiI–positive cells. For each experiment, 75–100 cells perfish were transplanted and at least 5 fish for each condition wereanalyzed in multiple biological replicates.

Globosides Are Vital for MET

www.aacrjournals.org Cancer Res; 78(11) June 1, 2018 2953




Figure 1.

Expression of globoside-encoding genes is elevated in epithelial cancer cells. A, GSL biosynthesis pathway. Glucose is attached to ceramides and furtherelongated into three major series, all with the precursor lactosylceramide (LacCer). GSL-series are highlighted by different colors. Carbohydratestructures provided along with the name were investigated by antibodies in this study. B, Linear discriminant analysis in three transcriptomic ovariancancer datasets separating epithelial (E, dark green) from mesenchymal (M, red) tissue samples. Wilks' lambda and corresponding P value is providedalong with each dot plot. C, Kaplan–Meier curves for overall survival and relapse-free survival in four EMT states (E, IE, IM, and M) in the Tothill dataset. Themedian overall and progression-free survival is provided in months (mths) along with the figure legend. D, Table summarizing globo and ganglio-seriessynthesizing glycosyltransferase-encoding genes predicting overall and relapse-free survival determined for by Cox proportional hazard model.E, Boxplots showing gene expression of CDH1, A4GALT, VIM, and ST3GAL5 among four different EMT states in three independent ovarian cancerdatasets (Tothill, Bonome, and The Cancer Genome Atlas). Color indicates typical expression of genes up in epithelial (green) or mesenchymal (red),epithelial (E), intermediate epithelial (IE), intermediate mesenchymal (IM), and mesenchymal (M). F, Heatmap of the abundance of GSLs across normaland cancer cell lines. The heatmap illustrates the unclustered distribution of cell lines from the mean GSL expression out of three independent experimentsusing flow cytometry. GSLs markers arranged by average linkage clustering. Cell line origin is provided along with the heatmap.

Jacob et al.

Cancer Res; 78(11) June 1, 2018 Cancer Research2954




Bisulfite sequencingFour different web-based bioinformatic engines were applied

as describedpreviously (15) to predict the genomic locationof theCpG island in the CDH1 gene (UCSC Genome Browser,CpGPLOT, Methprimer, and CpG island searcher). For analysisof the methylation status, genomic DNA was isolated fromexponentially growing cells as described previously (15). Anamount of 1 mg DNA was bisulfite converted using the EZMethylation-Gold Kit (Zymo Research Inc., Lucerna-Chem AG)and was further subjected to PCR reaction using MyTaq HS DNApolymerase (Bioline, Abc biopply). PCR reactions contained 1 �MyTaq reaction buffer, MyTaq HS polymerase, 300 nmol/LCDH1_Bis_1F and CDH1_Bis_1R (DNA sequence is listed inSupplementary Table S2), and nuclease-free water to a finalvolume of 10 mL performed under following conditions: 95�Cfor 10 minutes, 35 cycles of 95�C for 15 seconds, 50�C for 15seconds, and 72�C for 20 seconds, and finished by elongation at72�C for 3minutes. PCR products were subjected to a second PCRusing CDH1_Bis_2F and CDH1_Bis_2R following same condi-tions as described previously (Supplementary Table S2). The PCRproduct were visualized on a 2%agarose gel and purified from thegel using the Wizard SV Gel and PCR clean up system (Promega)and cloned into pGEM-T Easy (Promega). Plasmid minipreps(PureYield Plasmid MiniPrep System, Promega) from individu-ally grown colonies were sequenced using SP6 primer. Bisulfitesequencing data were analyzed and visualized using the quanti-fication tool for methylation analysis web-based application(http://quma.cdb.riken.jp/top/index.html).

Cholera toxin and 5-Aza treatmentsCells were treated with 100 ng/mL of cholera toxin (CTx;

Sigma), which was replenished every 3–4 days over a period of14 to 16 days. Cells were split to a ratio of 1:6 every 3–4 daysduring the treatments. In regards to global DNA demethylation,cells were treated with 2.5 mmol/L 5-aza-20-deoxycytidine (5-Aza)for up to 96 hours with replenishing media every 24 hours.

Native chromatin immunoprecipitationChromatin immunoprecipitation (ChIP) was performed as

previously described with modifications (19). In brief, SKOV3and SKOV3ip cells were lysed and treated bymicrococcal nuclease(Nuclease S7, Roche, #10107921001). ChIPwas carried out againwith antibody H3K4me3 (Cell Signaling Technology, #9727). Arabbit IgG (Santa Cruz Biotechnology Inc., #sc-2027) was used asa negative control. After incubationwith proteinGMag Sepharose(GE Healthcare Life Sciences, #28-9440-08) and washing, theprecipitated chromatin was eluted with 1% SDS in TE bufferfollowing digestion with proteinase K (Promega, #MC5005).Finally, precipitated DNA was purified by phenol–chloroformextraction/ethanol precipitation andwas analyzedbyquantitativePCR (Promega GoTaq qPCR Master Mix #A6001). Primers usedwere CDH1 and VIM (Supplementary Table S2).

Data acquisition and statistical analysisPublicly available transcriptomic datasets (GSE26712,

GSE68661, and GSE9899) were downloaded from Gene Expres-sion Omnibus (http://www.ncbi.nlm.nih.gov/geo/). Statisticalanalysis and figures were obtained through the use of the softwareR version 3.1.3 (www.R-project.org). Gene expression data forthe "Cancer Cell Line Encyclopedia" were accessed through thecBioPortal using R (www.cbioportal.org) using cgdsr, a R-based

application programming interface (API) that provides a basic setof R functions for querying the Cancer Genomics Data Serverhosted by the Memorial Sloan-Kettering Cancer Center (NewYork, NY).

Overall survival and relapse-free survival was investigated usingKaplan–Meier curves. Gene expression was dichotomized usingtree-based model partitioning ("ctree" in R package "party").Computations of the predicted survivor curves were used for Coxproportional hazards regression model (R package "survival").Results were presented as HRs with 95% confidence interval (CI)and corresponding P values.

All experiments were performed in triplicates and statisticalevaluation was done using unpaired Student t tests. P valuesof <0.05 were considered statistically significant (��, P < 0.001;��, P < 0.01; �, P < 0.05; �, P < 0.1).

ResultsGloboside-encoding genes are associated with EMT/MET

To gain insight into the involvement ofGSLglycosyltransferase-encoding genes (glyco-genes) in reversible transition of cancercells between epithelial and mesenchymal phenotypes, we inves-tigated the Tothill (20), Bonome (21), and The Cancer GenomeAtlas (22) transcriptome datasets representing epithelial (E),intermediate epithelial (IE), intermediate mesenchymal (IM),and mesenchymal (M) states. We applied a previously developedEMT scoreon transcriptomic data touniversally predict EMTstatesacross various cancer types (23).

First, we tested whether the expression of glycosidase- andglycosyltransferase-encoding genes discriminates epithelial frommesenchymal cancer cells in those transcriptomic datasets. Weselected all available Kyoto Encyclopedia of Genes and Genomes(KEGG) annotated glyco-genes involved in the synthesis ofglobo-, neo-lacto, and ganglio-series GSLs and demonstrated thatglyco-genes significantly discriminate E from M (Fig. 1B). Theseresults suggest that glyco-genes are differentially expressed duringtransition of ovarian cancer cells. EMT was previously shown topredict disease outcome in ovarian cancer (24–26), thus wefurther tested the EMT score in regards to overall survival andprogression-free survival. Patientswithmesenchymal features (IMand M) experienced a shorter overall survival (likelihood ratiotest ¼ 14.4, P ¼ 0.00239) and earlier disease recurrence(likelihood ratio test ¼ 13.1, P ¼ 0.00446) compared with IEand E subtype (Fig. 1C). With regard to GSL-encoding genes, theCox proportional hazardmodel revealed A4GALT as predictor forlonger overall survival and disease recurrence with a HR of 0.119.In contrast, the glyco-gene associated with ganglioside ST8SIA1predicted earlier disease recurrence with a HR of 2.715(P ¼ 0.006; Fig. 1D).

Next, we investigated the individual glyco-gene expression andfound that all three datasets displayed a classical EMT markerspectrum including the E-cadherin–encoding gene CDH1 andVIM revealing differential gene expression among the four dif-ferent EMT states (E, IE, IM, and M; Fig. 1E). We observedsignificantly decreased A4GALT expression particularly in themesenchymal signature (Fig. 1A and E). The A4GALT geneencodes the a1-4 galactosyltransferase, a key enzyme responsiblefor synthesis of globoside GSLs. In contrast, IM and M showedelevated expressionof ST3GAL5 encodinga2-3 sialyltransferase 5,which is involved in ganglioside GSL synthesis (Fig. 1E). A moredetailed analysis of the Tothill dataset with regard to all





http://quma.cdb.riken.jp/top/index.html

http://www.ncbi.nlm.nih.gov/geo/

www.R-project.org

www.cbioportal.org


KEGG-annotated glycosyltransferases involved in GSL synthesisrevealed generally decreased globoside and elevated gangliosidegene expression profiles among the four EMT states (Supplemen-tary Fig. S1A and S1B).

These data demonstrate that globo series–related genes, and inparticular, A4GALT, are elevated in epithelial cancer cells andassociated with improved patient's outcome. To set up an exper-imental model for studying the relevance of A4GALT and globo-sides during EMT/MET, we next determined the abundance of sixGSLs (lactosylceramide- LacCer; Gb3, SSEA3, paragloboside-nLc4, P1, and GM1) in 34 cell lines using flow cytometry (Fig.1A and F). In general, all normal and cancer cell lines displayed aheterogeneous expression of GSLs, with a peak of GM1 (82.0 �26.7% FITCþ, Fig. 1F). In contrast, P1 displayed the lowestexpression among all cell lines (4.6� 15.9% FITCþ). The remain-ing testedGSLs, including globosides Gb3 and SSEA3, displayed amore heterogeneous expression ranging from 0% to 97.7%FITCþ. Interestingly, the ovarian cancer cell line IGROV1 waspositive for all GSLs. Moreover, IGROV1 displayed an interme-diated expression of epithelial and mesenchymal markers in theCell Line Encyclopedia (27) and has previously been consideredas IE (Fig. 1; Supplementary Fig. S2). We therefore selectedIGROV1 as a model to study globoside depletion by genomeediting and the subsequent effects on EMT/MET.

Establishment of globoside-depleted cancer cells usingCRISPR-Cas9

As IGROV1 cells express a comprehensive GSL repertoireincluding globo, (neo-) lacto, and ganglio series, we genome-edited A4GALT in this cell line to deplete the glycosidic productsof this glycosyltransferase. Our paired-sgRNA–based strategy tar-gets the entire open reading frame (ORF) of A4GALT, a 1,335 bpgenomic region at chromosome 22q13 (Fig. 2A). Transient trans-fection of PX458 containing two specific sgRNAs following GFP-based clonal selection revealed various genotypes and finallyidentified three homozygous knockout clones out of 86 screened(DA4GALT; 3.5% efficacy; Fig. 2B; Supplementary Fig. S3). Inser-tions and deletions at the Cas9-active sites (indels) were detectedby DNA sequencing. However, due to the use of paired sgRNAstargeting the entire ORF, the presence of indels did not influencethe knockout ofA4GALT (Fig. 2C). The off-target analysis revealedno sequence variation (insertion/deletion) at highest rankedpredicted off-target sites (Fig. 2D). Examination of GSLs by flowcytometry revealed a depletion of globosides (Gb3 1%, SSEA31%) and P1 (1%), while LacCer (3–20%), nLc4 (50%–85%), andGM1 (70%–100%) remained largely unchanged in all threeDA4GALT cell clones (Fig. 2E).

According to the transcriptomic data, the gene expression ofspecific ganglioside synthesizing glycosyltransferases (e.g.,ST8SIA1) increases among the EMT spectrum. Thus, we per-formed quantitative analysis of the glycosylation of the GSLsexpressed by these cells by enzymatically releasing the glycansfrom the isolated GSLs for analysis using LC-ESI-MS/MS, to studythe structure of the glycan component of the GSL product afterdeletion of A4GALT. In line with our flow cytometry data, theglycanmass ofm/z 505.31� corresponding to themonosaccharidecomposition of globoside Gb3 [(Hex)3], which is the directproduct of A4GALT, was not detectable in DA4GALT cells(Supplementary Table S3). Interestingly, the ganglioside GM1[glycan mass m/z 999.5 ((Hex)3(HexNAc)1(NeuAc)1)] wasincreased significantly in DA4GALT cells (29.4%) compared with

wild-type (13.4%), indicating that certain gangliosides increasedas a consequence of globoside depletion (Supplementary Fig. S4).

Globoside-depleted cells acquire mesenchymal propertiesin vitro and in vivo

We next sought to ascertain the phenotypic changes of cellsdisrupted forA4GALT. When comparedwithwild-type,DA4GALTIGROV1 cells (IE) displayed a mesenchymal morphology. Wealso observed a loss of cell-clumps in cell cultures upondisruptionof A4GALT with DA4GALT cells growing in monolayer instead ofon top of each other. With regards to cell proliferation, weobserved no significant differences (P ¼ 0.05 at 120 hours; Fig.3A). Next, we examined the influence of A4GALT on anchorage-dependent and -independent growth in standard tissue cultureplates and soft-agar, respectively. Both assays revealed significant-ly reduced numbers of colonies in DA4GALT cells as comparedwith parental wild-type (anchorage-dependent P ¼ 0.0001 and-independent P ¼ 0.0002; Fig. 3B and C). This striking differencein anchorage-independent growth was further investigated inregards to cell detachment–induced apoptosis (anoikis), whichis a crucial step for tumor cells undergoing malignant transfor-mation or adapting to a new microenvironment. Resistance toanoikis is typical for ovarian cancer cells that can survive in asciticfluid before forming metastatic foci at distant sites (28). Weobserved prolonged cell viability in DA4GALT cells after 7 and10 days of cultivation (Fig. 3D). This was in line with an increasedpercentage of propidium iodide–positive cells in wild-type com-pared with DA4GALT IGROV1 cells (Fig. 3D). The significantdecrease of apoptotic DA4GALT cells coincides with reducedcleaved PARP, another marker of cellular apoptosis (Fig. 3E).

We next examined the migratory and invasive phenotype ofDA4GALT cells using in vitro directed cell motility assayswith FCS as chemoattractant. Cells harboring DA4GALT indeeddisplayed enhanced cell migration (Fig. 3F, P ¼ 0.0015) andinvasion (Fig. 3G, P ¼ 0.024). Intrigued by these results, weinjected both cell types into zebrafish embryos to study cellmotility and proliferation in vivo. Zebrafish embryos wereexamined after 3 days postinjection (dpi). In line with ourin vitro data, enhanced in vivo dissemination and outgrowthwas observed in DA4GALT compared with wild-type cells (Sup-plementary Fig. S5). Similarly, cell proliferation was not affectedin the in vivo model, as revealed by flow cytometry quantifica-tion of CM-DiI–positive cells in corresponding transplantedzebrafish embryos (Fig. 3H). We also studied early events ofmetastasis by transplanting cancer cells into the vasculature oftransgenic Tg(kdrl:eGFP) zebrafish. Here, deletion of A4GALTsignificantly increased the number of extravasated cells andmore clusters are formed as compared with parental cells (Fig.3I–J). These data indicate that the deletion of A4GALT leads toincreased cell motility and invasiveness in vitro as well as in vivo,features commonly accompanied with EMT.

E-cadherin–mediated cell–cell adhesion requires A4GALT-dependent GSLs

Because gene disruption of A4GALT revealed both in vitro andin vivo phenotypic changes associated with EMT, we next charac-terized globoside-depleted cells in more detail. Specifically, weinvestigated the major player in EMT- E-cadherin, which has apivotal role in cell–cell adhesion.

To confirm whether the A4GALT deletion mediates itsfunctional effect toward E-cadherin, we restored A4GALT in

Jacob et al.





DA4GALT cells by ectopically expressing the ORF of A4GALT in adoxycycline-inducible manner (rescue DXD). The aspartate-anyresidue-aspartate (DXD) motif, which is present in A4GALT, isimportant for enzymatic activity of several glycosyltransferases(29–31). Thus, a mutant (D194A) of A4GALT was generated asbeing enzymatically inactive (rescue DXA) and tested for expres-sion of various GSLs. The rescue DXD revealed similar levels forLacCer, Gb3, nLc4, P1, and GM1. However, SSEA3 was notreexpressed in rescue DXD cells. We also demonstrate for the firsttime that the DXD-motif in A4GALT is required for enzymaticactivity, as none of the A4GALT-mediated GSLs were present inrescued DXA cells (Supplementary Fig. S6A and S6B). Next,we determined the expression of E-cadherin in all four cell lines,wild-type, DA4GALT, rescue DXD, and DXA. In addition, wetransduced DA4GALT cells with human full-length E-cadherin aswell as the cytoplasmic chimera IL2R/E-cadherin (not capable of

engaginghomophilic adhesive activity; ref. 32). Thewild-type andDA4GALT cells (re-) expressing E-cadherin and IL2R/E-cadherinrevealed high expression of E-cadherin with an expected shifttoward lower molecular weight for IL2R/E-cadherin (Fig. 4A). Wealso observed a faint band for E-cadherin in the rescue DXD cells,namely DA4GALT cells reexpressing A4GALT and globosides. Incontrast, both DA4GALT and rescue DXA cells were not capable ofexpressing E-cadherin (Fig. 4A).

Intact cell–cell adhesion via E-cadherin usually appears asmembranous staining in epithelial cells andbecomes intermittentand jagged during EMT, indicating a breakdown of cell–celljunctions (33). This dispersed membrane-associated E-cadherinstaining was reported in various epithelial cells undergoingEMT (34–36). To determine whether A4GALT-mediated GSLsaffect cell-cell junctions, we analyzed genetically engineered cellsfor the distribution of E-cadherin, the integral component of the

Figure 2.

Generation of a stable and site-specificA4GALTmutant ovarian cancer cell line using CRISPR-Cas9 technology.A, Strategy to delete theA4GALT gene (DA4GALT) -two different sgRNAs targeting the entire open reading frame (ORF, light blue; Cas9 active sites, yellow). DA4GALT cells were identified using three differentprimers (CRISPR_F, CRISPR_R andORF_R; blue), resulting in a 1,335-bp deletion.B, Identification of homozygousDA4GALT cells. Single-cell cloneswere assayed bythree genotyping PCRs (1, Deletion PCR; 1,863 bp and 500 bp; 2, wild-type–specific PCR; 1,378 bp; and 3, inversion PCR; 937 bp) showing various genotypes(wild type, wt; deletion, del; inversion, inv; wt/del, inv/del, inv/inv, and del/del). C, DNA sequence variation (indels) at the CRISPR-Cas9 active sites. HomozygousDA4GALT (1335bp) was detected in DA4GALT _1 and DA4GALT _2, while DA4GALT _3 was heterozygously deleted (1,335 bp and 1,208 bp). Protospaceradjacent motif (PAM, red); red arrows, predicted deletion sites. D, Off-target data displayed no sequence variation compared with wild-type. PredictedCas9-active off-target sites are displayed in a representative DA4GALT clone (red arrows). E, Depletion of globosides and P1 in DA4GALT clones. Counterplotrepresents the data obtained by flow cytometry for wild-type and DA4GALT, showing negative control (unstained cells, red) and GSL-positive cells (green). Thevalue provided in each counterplot refers to the percentage of GSL-positive cells. Corresponding bar chart displays the mean � SD of three independentexperiments for wild-type (light gray) and three DA4GALT clones (dark gray). Student t test (�� , P < 0.01; �� , P < 0.001).






Figure 3.

Appearance of mesenchymal features upon deletion of globosides. A, MTT cell viability and proliferation assay for wild-type (A4GALT, gray line) and DA4GALT cells(black line). B, Representative wells with cell colonies and bar chart for quantification of anchorage-dependent growth. C, Anchorage-independent growth withreduced colonies in DA4GALT cells compared with wild-type. D, Representative histograms for cell detachment–induced apoptosis (anoikis) in wild-type andDA4GALT cells, unstainedor negative control (white), andpercentage of propidium iodide–stained cells (dead cells, gray). The percentage in eachhistogram refers to thepropidium iodide positivity. Corresponding line chart showing viable cells (y-axis) measured byMTT absorbance ratio (normalized to 100% at day 0) between day 0 andday 10 in both wild-type (A4GALT) and DA4GALT cells. E, Western blot data with increased cleaved PARP during day 7 and day 10 in wild-type cells comparedwith DA4GALT cells. F, Cell migration assay. G, Cell invasion (Matrigel) assay. H, For flow cytometry analyses, transplanted embryos were enzymatically dissociatedinto single cells. Depicted here are corresponding counterplots and a bar chart showing summarized data from n ¼ 3 experiments with at least n ¼ 5 fish groupand experiment (right side, percentage of CM-Dil–positive cancer cells). I and J, Extravasation and cluster formation are more pronounced in fish transplantedwithDA4GALT (J, bottom) comparedwith control (J, top) cells, when cells are transplanted into the zebrafish vasculature (Duct of Cuvier) of a transgenic Tg(kdrl:eGFP)line. Arrowheads, extravasated cancer cells that are not in contact with the endothelium and invaded tissues after extravasation. Red arrows, secondary tumormass formation (cluster formation) in proximity of the circulatory loop between the dorsal aorta and the caudal vein. Shown are representative pictures andsummarized data from n ¼ 2 biological replicate experiments with n ¼ 5 or more fish per group and experiment. For the graph, numbers of extravasated cancercells and cluster formation were counted; P values were derived using a Mann–Whitney U test. Bar chart with mean � SD (� , P < 0.05; �� , P < 0.01).

Jacob et al.





adherens junctions. Despite the low E-cadherin expression in therescue DXD (Fig. 4A), only these cells revealed areas of typicalmembranous E-cadherin staining as observed in IGROV1 byconfocal fluorescence microscopy (Fig. 4B). Neither full-lengthnor IL2R-fused E-cadherin–transduced DA4GALT cells displayedmembranous staining as observed in IGROV1.

The A4GALT DXD expression did not substantially rescue theexpression of E-cadherin (Fig. 4A). Thus, we cloned the full-lengthE-cadherinC-terminal taggedwith EGFP into pUltra togetherwitheither DXD or DXA (Fig. 4C). The bicistronic expression of both

proteins was confirmed by flow cytometry and Western blotanalysis (Fig. 4D and E). Investigating anchorage-independentgrowth in those cells by performing soft-agar assay revealedspheroids after 12 days only in cells with functional A4GALT(DXD, Fig. 4F and G). Tumor xenograft experiments in zebrafishembryos were also in agreement showing that the loss of A4GALTenhances cell motility as observed by a significant increase in thenumber of extravasated cells and larger cluster formation in DXAcompared with DXD (Supplementary Fig. S7). Taken together,these results indicate that A4GALT-mediatedGSLs are required for

E-cadherin EGFP A4GALT

T2A

hUbc promoter

pUltra-E-cadherinEGFP-A4GALTDXD

E-cadherin EGFP A4GALT

T2A

hUbc promoter

pUltra-E-cadherinEGFP-A4GALTDXA D194A

C D IGROV1 DXD DXA

PE

EGFP

Unstained control and Gb3 expression

E1) IGROV1

2) IGROV1 ΔA4GALT

3) IGROV1 ΔA4GALTEcadherinGFP DXD

4) IGROV1 ΔA4GALTEcadherinGFP DXA

1 2 3 4

α E-cadherin

α GFP

α Vimentin

α Tubulin

F IGROV1

ΔA4GALT

DXD

DXA

0

100

200

300

400

500

600

IGR

OV1

DXD

DXA

% o

f Sph

eroi

ds

ΔA4G

ALT

DXD

DXA

Brightfield E-cadherinEGFPG

A B A4GALT

A4GALTA4GALTA

4GA

LT

A4G

ALT

A4G

ALT

A4G

ALT

E-ca

dher

in

SKO

V3

DXD

DXA

SKO

V3

E-cadherin

Tubulin

E-cadherin

Tubulin

DXD DXA E-cadherinIL2R/E-cadherin

cytoplasmic

IL2R

/E-c

adhe

rincy

topl

asm

ic

Rescue A4GALT Overexpression

Figure 4.

E-cadherin–mediated cell–cell adhesion depends on the upstream glycosphingolipids. A, Western blot showing loss of E-cadherin in DA4GALT cells andreexpressionof E-cadherin (full length and cytosolic). SecondWestern blot analysis showspartial rescue of E-cadherin expression inDA4GALT reexpressingA4GALT(DXD) in contrast to mutant A4GALT (DXA). B, Confocal fluorescence images showing loss of membranous E-cadherin staining in DA4GALT compared withwild-type A4GALT and in DXD rescued cells displaying partial membranous E-cadherin staining, whereas overexpression of E-cadherin (full length andcytosolic) in DA4GALT cells was diminished E-cadherin membranous staining. White bar, 5 mm. C, Graphical representation of constructs for coexpressionof E-cadherin C-terminal tagged with EGFP and A4GALT in a pUltra backbone. Bicistronic expression was achieved by the self-cleaving peptide T2A. D,Representative flow cytometry data showing expression of Gb3 (PE) and E-cadherin (EGFP). E, Western blot analysis showing expression of E-cadherin, GFP,vimentin, and tubulin in four different cell lines (legend next to the image). F, Representative well of a spheroid assay (anchorage-independent growth) withquantification as bar chart (mean� SD). G, Brightfield and corresponding fluorescence image (EGFP) of a representative spheroid in IGROV1 DA4GALT E-cadherinEGFP DXD and a single cell in IGROV1 DA4GALT E-cadherin EGFP DXA. White bar, 50 mm.






E-cadherin–mediated cell–cell adhesion and are thus importantfor MET.

The presence of globosides is crucial for intermediate EMT cellsCDH1 and A4GALT expression are positively correlated in

epithelial ovarian cancer tissue samples (Tothill, Pearson corre-lation 0.26, P ¼ 6.25e�06). A4GALT and its enzymatic productsseem to be preferentially expressed in cells with epithelial char-acteristics and are required for cell–cell adhesion via E-cadherin.To better understand the correlation between A4GALT andE-cadherin in epithelial cells, we asked whether deletion ofA4GALT affects globosides and the function of E-cadherin in otherovarian cancer cells representing the full spectrum of EMT. Thus,we tested the transition properties of ovarian cancer cells that haveundergone genetic disruption of A4GALT in a panel of cell lines. Atotal of ten ovarian cancer cell lines were selected showingdifferential expression of CDH1 and VIM. A4GALT expressioncoincideswithCDH1 expression in those cell lines (Fig. 5A). Next,we tested for expression of E-cadherin and vimentin together withglobosides in matching cell lines. Gb3 and SSEA3 were generallyelevated in cell lines being primarily positive for E-cadherinand almost negative for vimentin, indicating that globosides arepreferentially expressed on epithelial cancer cells (Fig. 5B).

In addition to IGROV1, cells representing an IE state, weselected BG1, SKOV3, and A2780 cell lines displaying the remain-ing threedifferent EMT states (E, IM, andM, respectively) andusedthem for depletion (via CRISPR-Cas9–mediated A4GALT editing;BG1, SKOV3) and, respectively, overexpression of globosides vialentiviral transduction with pUltraA4GALT; A2780). As expected,genomic deletion (Supplementary Fig. S8A and S8B) or over-expression of A4GALT resulted in the absence or presence of Gb3and SSEA3, respectively (Fig. 5C; Supplementary Fig. S8C–S8G).However, E-cadherin expression levels were only reduced inSKOV3 DA4GALT cells, whereas BG1 (E) and A2780 (M) did notalter in E-cadherin expression (Fig. 5C). Interestingly, BG1DA4GALT cells showed partly intermittent and jagged stainingat the cell–cell borders and significantly reduced colony forma-tion (Supplementary Fig. S8D and S8E) supporting previousfindings in IGROV1.

To further demonstrate the involvement of globosides in tran-sition of cancer cells, we investigated SKOV3 inmore detail, whichusually exhibits an IM phenotype (1). Studies on SKOV3ip, aderivative of SKOV3 obtained after passaging in amouse, showedsignificantly increased growth rate, enhanced colony formation insoft agar and tumor formation in vivo compared with parentalSKOV3 (37). Indeed, we observed elevated E-cadherin expressioncoinciding with increased globosides in SKOV3ip compared withSKOV3 cells (Fig. 5C). We also observed significantly increasedanchorage-dependent growth and increased membranous stain-ing, which correlated with stronger E-cadherin expression inSKOV3ip cells (Fig. 5DandE).Moreover, zebrafish embryo tumorxenografts revealed significantly enhanced extravasation and clus-ter formation upon A4GALT deletion in SKOV3 (Fig. 5F; Supple-mentary Fig. S9). These are characteristics ofMET shifting parentalIM SKOV3 into IE SKOV3ip cells and our results indicate thatespecially for IE/IM cancer cells, globosides are essential forepithelial properties involved in E-cadherin expression andcell–cell adhesion.

As cancer cells transitioning towards a mesenchymal state areassociated with elevated resistance to a variety of conventionalchemotherapeutics (38), we tested how loss of A4GALT affects

resistance to doxorubicin in ovarian cancer cells (Fig. 5G). First,we observed increased resistance to doxorubicin in cancer cellsreflecting the various EMT states represented by BG1 (E), IGROV1(IE), and SKOV3 (IM). In contrast to BG1, IGROV1, and SKOV3displayed 4.25- and 1.80-fold higher resistance to doxorubicin forDA4GALT cells, respectively. The SKOV3ip cell line was 2.12-foldmore sensitive to doxorubicin compared with parental SKOV3,thus confirming for the first time that intermediate EMT cell linesacquire doxorubicin resistance correlating with expression ofglobosides (Fig. 5G).

As EMT enables cancer cells to acquire stem cell–like propertiesentering into amore mesenchymal state accompanied by the lackof E-cadherin (39), we next studied the impact of A4GALT on thewidely used cancer stem cell markers CD24 and CD44. We foundthat DA4GALT cell populations undergoing EMT displayed anincrease in the proportion ofCD24low/CD44high cells (Fig. 5H). Inregards to MET, CD24low/CD44high remained unchanged inDA4GALT cells treated with CTx, an MET-inducing agent (40).In contrast, wild-type A2780 (M) and SKOV3 (IM) switched fromCD24low/CD44high to CD24high/CD44low phenotypes after CTxtreatment (Fig. 5H). Changes in the level of both markers inSKOV3 were accompanied by an increase of the globoside Gb3(4 to 19% FITCþ) and enhanced E-cadherin expression (Fig. 5Iand J). Neither globoside nor E-cadherin expressionwas observedin A2780 and DA4GALT cells (IGROV1, BG1, and SKOV3) treatedwith CTx (Fig. 5H–J). Taken together, these data highlight notonly the importance of globosides inMET, but also their potentialimpact on the expression of CD24 and CD44, thus furtherindicating that GSLs are directly involved in the reversible tran-sition of cancer cells from an epithelial to a mesenchymalphenotype.

A4GALT deletion–induced EMT is accompanied by DNAhypermethylation at the promoter region of CDH1

Epigenetic programs, including DNA methylation and chro-matin modifications, play a key role in EMT (3). Recent work byPattabiraman and colleagues suggests that the protein kinase Apathway triggers epigenetic reprogramming by the histonedemethylase PHF2. However, in our case, CTx treatment onlymarginally affected expression of E-cadherin, CD24 and CD44markers, especially in DA4GALT cells (IGROV1, BG1, andSKOV3). Thus, we tested the impact of DNA methylation at thepromoter region of CDH1. Independent of the program applied,the analysis of the promoter region of CDH1 predicted a CpGisland surrounding the transcription start site (TSS) of CDH1(Fig. 6A). As most changes in DNAmethylation occur in a regionof �250/þ250 bp relative to the TSS (41), we analyzed the DNAmethylation status within the genomic region �116/þ124 of theTSS CDH1 comprising 18 CpGs (Fig. 6A). Ovarian cancer celllines in the E state (OVCAR3 and BG1) revealed DNA hypo-methylation at the investigated locus in contrast to A2780, whichshowed 84% � 2.65% DNA methylation (Fig. 6B). Interestingly,we observed increased DNA methylation at the promoterregion of CDH1 in both DA4GALT cell lines compared withparental IGROV1 and SKOV3 (Fig. 6B). Next, we hypothesizedthat E-cadherin expression inDA4GALT cells canbe reexpressed bythe hypomethylating agent 5-aza-20-deoxycytidine (5-Aza).Parental IGROV1, DA4GALT, and rescue DXD cells showedE-cadherin expression after 96 hours of treatment with 5-Aza(Fig. 6C). Finally, we investigated membranous staining of allthree cell lines and observed increased membranous staining in

Jacob et al.





TOV112D

TYK-nu

A2780

KURAMOCHI

SKOV3

IGROV1

OAW42

OVCAR4

OVCAR3

BG1

Relative mRNA levels (log10-fold)

A

E-cadherin

Vimentin

Tubulin

Gb3SSEA3

BG

1

OVC

AR

3

OVC

AR

4

OAW

42

IGR

OV1

SKO

V3

KU

RA

MO

CH

IA

2780

TYK

-nu

T OV1

12D

B

E-cadherin

Vimentin

Tubulin

Gb3SSEA3

BG1 SKOV3

SKO

V3ip

A2780

DA4GALT

DA4GALT

OE

A4GALT

Par

enta

l

Par

enta

l

Par

enta

l

C

E-cadherin/ D

AP

I

SKOV3ipSKOV3

ΔA4GALTSKOV3D

E

SKOV3

SKOV3 ΔA4GALTTg(kdrl-eGFP)la

116

SKOV3ip

F

SK

OV

3

SK

OV

3 ΔA4GALT

SK

OV

3ip

0

10

20

30

40

Num

ber o

f ext

rava

sate

d ce

lls *** *

SKO

V3ip

SKO

V3ΔA

4GA

LTSK

OV3

G

Rel

ativ

e ce

ll vi

abili

ty

BG1

Doxorubicin (nmol/L)

IGROV1

Doxorubicin (nmol/L) Doxorubicin (nmol/L)

SKOV3

H I

E-cadherin

Vimentin

Tubulin

A4GALT

CTx

J

Parental IC50 257 ± 64 nmol/LΔA4GALT IC50 463 ± 100 nmol/LSKOV3ip IC50 121 ± 40 nmol/L

Parental IC50 36 ± 10 nmol/LΔA4GALT IC50 153 ± 46 nmol/L

Parental IC50 4.6 ± 1.5 nmol/LΔA4GALT IC50 5.1 ± 2.5 nmol/L

SK

OV

3

SK

OV

3 T

SK

OV

3ip

ΔA4GA

0

2

4

6

8

10

Num

ber o

f clu

ster

s

* n.s.

1.0

P = 0.0001

P = 0.0001

P = 0.002

0.8

0.6

0.4

0.2

0.00 10 20 30

Color key

0 20 40 60 80 0 20 40 60 80

Color key

40 50 0 100

1.0

0.8

0.6

0.4

0.2

0.0

1.0

0.8

0.6

0.4

0.2

0.0200 300 400 500 0 200 400 600 800 1,000

Figure 5.

Thepresence of globosides is vital in intermediate EMTcells.A,CDH1, VIM, andA4GALT expression amongovarian cancer cell lines representing the full EMT spectrum(determined by RT-qPCR).B, CorrespondingWestern blot analysis showing expression of E-cadherin, vimentin, and tubulin. Matching heatmap shows themeans ofGb3 andSSEA3 expression of three independentflowcytometry experiments for each cell line.C,Deletion or overexpression ofA4GALT leads to absence or presenceof globosides, respectively. Wild-type and DA4GALT cells are shown for BG1 and SKOV3; overexpression of A4GALT (OE A4GALT) is shown next to parental A2780.D, Representative immunofluorescence images for E-cadherin (green) and DAPI (blue) in SKOV3, SKOV3 DA4GALT, and SKOV3ip cells. White bar, 20 mm.E,Corresponding anchorage-dependent growth (colony formation assay) andbar chart summarizing three independent experiments.F,Decrease of globosideGSLs(A4GALT-active cells) results in enhanced cancer cell motility in vivo. Representative confocal imaging (z-stack images) with an increased number of metastaticDA4GALT SKOV3 cells in the zebrafish embryo compared with SKOV3 and SKOV3ip. Corresponding bar charts with number of extravasated cells and clusterformation. Statistical evaluation was performed by nonparametric one-way ANOVA for 12–16 zebrafish embryos analyzed in three independent biologicalexperiments.G, Increasing resistance to doxorubicin in intermediate EMT cells deleted for A4GALT (line charts on three independent MTT assays); IC50s for parental(wild-type, black), DA4GALT (red), and SKOV3ip (green) are provided below each line chart. H, Representative dot plot for un- and CTx-treated cancer cell linesunstained (red) anddouble-stained for CD24 andCD44 (light blue). I,Abundance ofGSLs (LacCer, Gb3, and SSEA3) uponCTx treatment in corresponding cancer celllines determined by flow cytometry. J, Expression of E-cadherin, vimentin, and tubulin in parental and DA4GALT cell lines treated with CTx.






5-Aza–treated IGROV1 and rescue DXD cells, whereas DA4GALTcells did not reveal such a pattern (Fig. 6D).Here, we show thatwesubstantially rescued E-cadherin–mediated cell–cell adhesiononly in cells with a functional a1-4 galactosyltransferase and itsdirect glycolipid products (globosides), indicating that DNAmethylation coincides with the expression of E-cadherin uponloss of globosides.

In this study, we identified that the protein level of E-cadherinin SKOV3ip is higher than in SKOV3, but both cell lines showedDNA hypomethylation at the CDH1 promoter region, indicatingthat DNA methylation is not the only regulatory mechanism

silencing E-cadherin expression in SKOV3. The downregulationofCDH1/E-cadherin canbemediated by its transcriptional repres-sion through the binding of transcription factors implicated inEMT (42, 43). These transcription factors (e.g., SNAIL, SLUG, orTWIST) are known to bind to E-boxmotifs in theCDH1 promoterlocus. Initially, we investigated the Tothill dataset for inverselyexpressed EMT-associated genes and identified several transcrip-tion factors (Supplementary Fig. S10A). However, a reducedexpression of SLUG was only found in SKOV3ip (SupplementaryFig. S10B). It is well known that DNA methylation and histonemodifications on promoter regions are involved in the regulation

Figure 6.

Degree of DNA methylation at the promoter region of CDH1 correlates with presence of A4GALT and its glycosidic products. A, CpG island [green, predicted(depending on the algorithm used)] at the transcription start site of human CDH1; exon 1 and 2 (black bars); genomic region analyzed by bisulfite sequencing inmoredetail (red bar) and further magnified for CpGs analyzed throughput this study. B, Lollipop plots showing degree of DNA methylation at the promoter regionof CDH1, representing EMT spectrum based on E-cadherin expression. Total of methylated CpGs provided along with SE between sequences (n ¼ 6). C, Westernblot analysis showing reexpression of E-cadherin in mock (DMSO) and 5-Aza–treated cancer cells IGROV1, IGROV1 DA4GALT, and IGROV1 DA4GALT rescueDXD cells. D, Membranous staining of E-cadherin only present in cells positive for A4GALT and globosides. E-cadherin expression in 5-Aza–treated IGROV1DA4GALT rescue DXD with membranous staining for E-cadherin; IGROV1 DA4GALT negative for membranous E-cadherin staining. White bar, 10 mm.

Jacob et al.





of gene transcription, and DNA methylation is considered ahallmark for long-term gene silencing (44). Furthermore, weobserved that the level of histone H3 lysine 4 trimethylation(H3K4me3), a mark associated with active transcription andalso preventing DNA methylation, was more abundant onCDH1 promoter in SKOV3ip compared with SKOV3 (Supple-mentary Fig. S10C). These results suggest that the transcriptionfactor binding at the CDH1 promoter region or the dynamics ofhistone modifications might play a more important role inregulating E-cadherin expression in SKOV3 and SKOV3ip, notDNA methylation.

DiscussionGSLs have been studied for decades and linked to the embry-

onic development and cell differentiation in normal and diseasedcells. In this study, we identified novel roles for GSLs as EMTinducers and mediators of cell–cell adhesion in ovarian cancercells. We showed that cancer cells undergoing EMT upon genomicdeletion of the key glycosyltransferase A4GALT display silencedE-cadherin expression, resulting in loss of cell–cell adhesionproperties. Remarkably, reexpression of globosides by A4GALTrescue partially restored homophilic cell–cell adhesion, whereasectopic E-cadherin was unable to. The pivotal role of globosideswas further demonstrated by membranous E-cadherin stainingonly in cells with functional A4GALT and concomitant DNAdemethylation.

Epithelial and intermediate EMT cells were shown to generallyexpress higher levels of globo series GSLs and underwent EMTupon A4GALT deletion. Moreover, quantitative analysis of gly-cans released from the GSLs displayed increased gangliosides inDA4GALT cells. On the other hand, SKOV3ip showed increasedlevels of Gb3 and E-cadherin, suggesting that these cells under-wentMET in themouse.Our observed changes in the level ofGSLsduring EMT is in line with work by Hakomori and colleaguesdescribing a ganglioside enrichment and globoside (e.g., SSEA3)reduction in a hydroxytamoxifen inducible TWIST-EMT modelsystem of human mammary epithelial cells (45, 46). Our tran-scriptomic data analyses support these findings as all significantlydifferentially expressed genes encoding for glycosyltransferasesinvolved in the ganglio series GSL synthesis are upregulated inmesenchymal phenotypes. In this context, specifically targetingkey glycosyltransferases (e.g., ST3GAL5 and ST8SIA1) or theirglycosidic products may be a therapeutic option to induceMET inmesenchymal cancer cells for suppression of metastatic spread inovarian cancer patients.

Epithelial ovarian cancer seems to be an optimal model forstudying EMT/MET as it has a high rate of local intraabdominalspread and 60.5% of ovarian cancer cell lines can be categorizedinto IE and IM states (1). Moreover, this is also similarly reflectedin ovarian cancer tissue samples and in the cell line encyclopaedia(23). During EMT, cells undergo a series of sequential events todisintegrate cell–cell contacts and the regulatory machinery thatcontrols cell polarity (3). Epithelial cells become progressivelyseparated from mesenchymal-like cells by gradually looseningcell–cell contacts to transit from polarized epithelium to postEMT-mesenchymal state to metastasize (47). More recent studiesassume an even greater flexibility exists in the EMT process, withcells rather moving through a spectrum of intermediate phasesinstead of being clearly epithelial or mesenchymal (3). In EMT,cells harbor the greatest plasticity because they are able to either

reverse to an epithelial state or progress towards a mesenchymalphenotype (48). Our observations on EMT upon A4GALT dele-tion support findings in the literature and further point outthe importance of GSLs in IGROV1 and SKOV3 cell lines(both intermediate EMT and E-cadherinþ/vimentinþ). In con-trast, epithelial (BG1, E-cadherinþ/vimentin�) and stable mes-enchymal (A2780, E-cadherin�/vimentinþ) cells did not fullyundergo transition upon deletion or ectopic expression ofA4GALT, respectively, but in these cells, we observed changesin E-cadherin cell–cell adhesion and differential expression ofCD24 and CD44, both markers applied in solid tumors dis-tinguishing cancer stem cells from bulk tumor populations(49) and with regard to GSLs (40, 45).

Interestingly, DA4GALT IGROV1 also exhibited anoikis resis-tance as well as an increased migratory and invasive phenotypecompared with wild-type cells, all of which have typical char-acteristics of EMT. DA4GALT cells acquired traits associated withmetastatic characteristics, which lead to enhanced cell dissemi-nation and extravasation in vivo (Fig. 3; ref. 50). In light of ourresults, it becomes more evident that ovarian cancer cells inintermediate EMT/MET states are characterized by a specificglycan composition on lipids such as ceramides, which mightfunctionally act upstream of E-cadherin affecting homophiliccell–cell adhesion. Apart from that, the important role of cera-mides has been shown in tumor growth andmetastatic diseases bylipidomic studies, preclinical models as well as clinical trials(reviewed in ref. 51). Thus, the impact of ceramides should notbe neglected, as they are part of lipid raft domains and involved inthe crosstalk between stromal and tumor cells.

Our findings suggest a role of A4GALT-mediated GSLs inhuman cancer cells undergoing "type 3" EMT. This seems to bein contrast to A4galt�/� mice, which follow "type 1 EMT" and donot show differences in embryonic formation and organ devel-opment (52). However, to study A4GALT-mediated GSLs in thebackground of A4galt�/�mice and neoplastic transformation, thedevelopment of a de novo tumormousemodel harbouring a Pax8-Cre–driven inactivation of a combination of Brca, Tp53, and Ptengenes might be the first choice in this model (53).

E-cadherin–mediated cell–cell adhesion usually provides aspatial cue for cells to distinguish an unbound (apical) from abound (basolateral) cell surface and to accumulate actingmembrane proteins at the required surface (54) Althoughreexpression of A4GALT resulted only in partial rescue ofE-cadherin and cell–cell adhesion, interestingly, no membra-nous staining was observed in either full-length E-cadherin orIL2R/E-cadherin cytotail expression in DA4GALT IGROV1cells. Treatment of cells with an agent inducing global DNAdemethylation (5-Aza) increased E-cadherin expression inDA4GALT cells; however, only coexpression of A4GALT in thesecells re-stored the wild-type E-cadherin–mediated cell–celladhesion. Thus, we propose that the composition of globo-sides, the "lipid glyco-fingerprint," intrinsically defines cell–celladhesion between cancer cells. There is controversy in thediscussion how the recognition between two cells, beforeforming adhesion via E-cadherin occurs: either by carbohy-drate–carbohydrate or carbohydrate–protein recognition(55, 56). Interestingly, differentially expressed glyco-genesinvolved in glycosylation of ceramides during EMT/MET havebeen described. One study in breast cancer found 44 genesinvolved in mesenchymal (CD24low/CD44high) metabolicpathways including genes encoding for enzymes involved in






GSL synthesis, like UGCG and ST6GAL1 (57). Our proposedrole of GSLs in EMT is also supported by recent work demon-strating that the CTx induces MET in several cell lines via thePKA pathway, inducing epigenetic changes that can modulateMET together with loss of mesenchymal characteristics (40).CTx is known to bind the ganglioside GM1 (Galb1-3GalNacb1-4(Neu5Ac(a2-3)Galb1-4Glc-ceramide) and associated structures(58). Considering that gangliosides seem to be preferentiallyexpressed in mesenchymal cells, CTx might indeed induce METby selectively binding to ganglioside-expressing cancer cells withmesenchymal features. In conclusion, our findings identify GSLsas major players in the transition of ovarian carcinoma cellsbetween epithelial and mesenchymal states. To translate our datato ovarian cancer tissue samples, the abundance of various GSLsmight be investigated using MALDI imaging, as mass spectralimaging of enzymatically released glycans from N-glycoproteinshas been recently demonstrated in tissue (59).

Disclosure of Potential Conflicts of InterestN. Packer is a professor at Institute for Glycomics, Griffith University. No

potential conflicts of interest were disclosed by the other authors.

Authors' ContributionsConception and design: F. Jacob, S. Alam, M. Konantz, R.S. Kohler, V.Heinzelmann-SchwarzDevelopment of methodology: F. Jacob, S. Alam, M. Konantz, C.-Y. Liang,R.S. Kohler, A.V. Everest-Dass, Y.-L. Huang, N. Packer, V. Heinzelmann-SchwarzAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): F. Jacob, M. Konantz, A.V. Everest-Dass, C. Lengerke,V. Heinzelmann-SchwarzAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): F. Jacob, S. Alam, M. Konantz, A.V. Everest-Dass,A. Sch€otzau, C. Lengerke, V. Heinzelmann-Schwarz

Writing, review, and/or revision of the manuscript: F. Jacob, S. Alam,M. Konantz, R.S. Kohler, A.V. Everest-Dass, A. Fedier, N. Packer, C. Lengerke,V. Heinzelmann-SchwarzAdministrative, technical, or material support (i.e., reporting or organizingdata, constructing databases): F. Jacob, S. Alam, A.V. Everest-Dass, C. Lengerke,V. Heinzelmann-SchwarzStudy supervision: F. Jacob, N. Packer, V. Heinzelmann-SchwarzOther (GSL staining, flow cytometry): Y.-L. HuangOther (cell culture, Western blot, supporting zebrafish experiments):N. RimmerOther (cell line experiments): A. FedierOther (cell culture and Western blot): M.N. Lopez

AcknowledgmentsThis work was supported by Swiss National Foundation (310030-156982,

310030_143619, -32, and Sinergia 171037; to V. Heinzelmann-Schwarz),Krebsliga Schweiz (KFS_3013-08-2012 to V. Heinzelmann-Schwarz andKFS-3841-02-2016 to F. Jacob), Krebsliga beider Basel (06-2013 to F. Jacob),Novartis Foundation for Biomedical Research (13B093 to F. Jacob),FreeNovation 2016 provided by Novartis (to F. Jacob), the Department ofBiomedicine, University Hospital Basel and University of Basel. We alsoacknowledge funding from the Australian Research Council Centre ofExcellence in Nanoscale Biophotonics CE140100003 (to N.H. Packer andA.V. Everest-Dass).

We are grateful to the Flow Cytometry Facility (Danny Labes, EmmanuelTraunecker, and Lorenzo Raeli) and Microscopy Facility (Michael Abanto andBeat Erne) for providing all necessary support. We would also like to acknowl-edge the following people for their various contributions to this publication:Jo€elle S. M€uller, Tatiana Pochechueva, and Yasmin Grether. Finally, we wouldlike to thank Gerhard Christofori for carefully reading our manuscript and hissupport on this project.

The costs of publication of this articlewere defrayed inpart by the payment ofpage charges. This article must therefore be hereby marked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Received July 27, 2017; revisedNovember 13, 2017; acceptedMarch20, 2018;published first March 23, 2018.

References1. Huang RY, Wong MK, Tan TZ, Kuay KT, Ng AH, Chung VY, et al. An

EMT spectrum defines an anoikis-resistant and spheroidogenic inter-mediate mesenchymal state that is sensitive to E-cadherin restorationby a src-kinase inhibitor, saracatinib (AZD0530). Cell Death Dis2013;4:e915.

2. Ahmed N, Abubaker K, Findlay J, Quinn M. Epithelial mesenchymaltransition and cancer stem cell-like phenotypes facilitate chemoresistancein recurrent ovarian cancer. Curr Cancer Drug Targets 2010;10:268–78.

3. Nieto MA, Huang RY, Jackson RA, Thiery JP. Emt: 2016. Cell 2016;166:21–45.

4. Diepenbruck M, Christofori G. Epithelial-mesenchymal transition (EMT)and metastasis: yes, no, maybe? Curr Opin Cell Biol 2016;43:7–13.

5. Teicher BA.Malignant cells, directors of the malignant process: role oftransforming growth factor-beta. Cancer Metastasis Rev 2001;20:133–43.

6. Massague J. TGFbeta in Cancer. Cell 2008;134:215–30.7. Guan F, Handa K, Hakomori SI. Specific glycosphingolipids mediate

epithelial-to-mesenchymal transition of human and mouse epithelial celllines. Proc Natl Acad Sci U S A 2009;106:7461–6.

8. Guan F, Schaffer L, Handa K, Hakomori SI. Functional role ofgangliotetraosylceramide in epithelial-to-mesenchymal transitionprocess induced by hypoxia and by TGF-{beta}. FASEB J 2010;24:4889–903.

9. MathowD, Chessa F, RabionetM, Kaden S, JennemannR, Sandhoff R, et al.Zeb1 affects epithelial cell adhesion by diverting glycosphingolipidmetab-olism. EMBO Rep 2015;16:321–31.

10. Jacob F, Anugraham M, Pochechueva T, Tse BW, Alam S, Guertler R, et al.The glycosphingolipid P(1) is an ovarian cancer-associated carbohydrateantigen involved in migration. Br J Cancer 2014;111:1634–45.

11. Jacob F, Goldstein DR, Bovin NV, Pochechueva T, Spengler M, Caduff R,et al. Serum antiglycan antibody detection of nonmucinous ovariancancers by using a printed glycan array. Int J Cancer 2012;130:138–46.

12. Alam S, Fedier A, Kohler RS, Jacob F. Glucosylceramide synthase inhibitorsdifferentially affect expression of glycosphingolipids. Glycobiology 2015;25:351–6.

13. Anugraham M, Everest-Dass AV, Jacob F, Packer NH. A platform for thestructural characterization of glycans enzymatically released from glyco-sphingolipids extracted from tissue and cells. Rapid Commun Mass Spec-trom 2015;29:545–61.

14. D'Angelo G, Capasso S, Sticco L, Russo D. Glycosphingolipids: synthesisand functions. FEBS J 2013;280:6338–53.

15. Kohler RS, Anugraham M, Lopez MN, Xiao C, Schoetzau A, Hettich T,et al. Epigenetic activation of MGAT3 and corresponding bisectingGlcNAc shortens the survival of cancer patients. Oncotarget 2016;7:51674–86.

16. Krauss J, Astrinidis P, Frohnhofer HG, Walderich B, Nusslein-Volhard C.transparent, a gene affecting stripe formation in Zebrafish, encodes themitochondrial protein Mpv17 that is required for iridophore survival.Biol Open 2013;2:703–10.

17. HaldiM, TonC, SengWL,McGrathP.Humanmelanoma cells transplantedinto zebrafish proliferate, migrate, produce melanin, form masses andstimulate angiogenesis in zebrafish. Angiogenesis 2006;9:139–51.

18. Svoboda O, Stachura DL, Machonova O, Zon LI, Traver D, Bartunek P. Exvivo tools for the clonal analysis of zebrafish hematopoiesis. Nat Protoc2016;11:1007–20.

19. Umlauf D, Goto Y, Feil R. Site-specific analysis of histone methylation andacetylation. Methods Mol Biol 2004;287:99–120.

Jacob et al.





20. Tothill RW, Tinker AV, George J, Brown R, Fox SB, Lade S, et al. Novelmolecular subtypes of serous and endometrioid ovarian cancer linked toclinical outcome. Clin Cancer Res 2008;14:5198–208.

21. Bonome T, Lee JY, Park DC, Radonovich M, Pise-Masison C, Brady J, et al.Expression profiling of serous low malignant potential, low-grade, andhigh-grade tumors of the ovary. Cancer Res 2005;65:10602–12.

22. Cancer Genome Atlas Research Network. Integrated genomic analyses ofovarian carcinoma. Nature 2011;474:609–15.

23. Tan TZ, Miow QH, Miki Y, Noda T, Mori S, Huang RY, et al. Epithelial-mesenchymal transition spectrum quantification and its efficacy in deci-phering survival and drug responses of cancer patients. EMBO Mol Med2014;6:1279–93.

24. Davidson B, Berner A, Nesland JM, Risberg B, Berner HS, Trope CG, et al. E-cadherin and alpha-, beta-, and gamma-catenin protein expression is up-regulated in ovarian carcinoma cells in serous effusions. J Pathol 2000;192:460–9.

25. Blechschmidt K, Sassen S, Schmalfeldt B, Schuster T, Hofler H, Becker KF.The E-cadherin repressor Snail is associated with lower overall survival ofovarian cancer patients. Br J Cancer 2008;98:489–95.

26. Hosono S, Kajiyama H, Terauchi M, Shibata K, Ino K, Nawa A, et al.Expression of Twist increases the risk for recurrence and for poor survival inepithelial ovarian carcinoma patients. Br J Cancer 2007;96:314–20.

27. Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S,et al. The Cancer Cell Line Encyclopedia enables predictive modelling ofanticancer drug sensitivity. Nature 2012;483:603–7.

28. He X, Chien J, Shridhar V. Assessment of resistance to anoikis in ovariancancer. Methods Mol Biol 2013;1049:347–54.

29. Li J, Rancour DM, Allende ML, Worth CA, Darling DS, Gilbert JB, et al. TheDXD motif is required for GM2 synthase activity but is not critical fornucleotide binding. Glycobiology 2001;11:217–29.

30. Persson M, Letts JA, Hosseini-Maaf B, Borisova SN, Palcic MM, Evans SV,et al. Structural effects of naturally occurring human blood group Bgalactosyltransferase mutations adjacent to the DXD motif. J Biol Chem2007;282:9564–70.

31. Ihara H, Ikeda Y, Koyota S, Endo T, Honke K, Taniguchi N. A catalyticallyinactive beta 1,4-N-acetylglucosaminyltransferase III (GnT-III) behaves asa dominant negative GnT-III inhibitor. Eur J Biochem 2002;269:193–201.

32. Gottardi CJ, Wong E, Gumbiner BM. E-cadherin suppresses cellular trans-formation by inhibiting beta-catenin signaling in an adhesion-indepen-dent manner. J Cell Biol 2001;153:1049–60.

33. Maeda M, Johnson KR, Wheelock MJ. Cadherin switching: essential forbehavioral but not morphological changes during an epithelium-to-mes-enchyme transition. J Cell Sci 2005;118:873–87.

34. Hoot KE, Lighthall J, Han G, Lu SL, Li A, Ju W, et al. Keratinocyte-specificSmad2 ablation results in increased epithelial-mesenchymal transitionduring skin cancer formation and progression. J Clin Invest 2008;118:2722–32.

35. Bassaganas S, Carvalho S, Dias AM, Perez-Garay M, Ortiz MR, Figueras J,et al. Pancreatic cancer cell glycosylation regulates cell adhesion andinvasion through the modulation of alpha2beta1 integrin and E-cadherinfunction. PLoS One 2014;9:e98595.

36. Busche S, Kremmer E, Posern G. E-cadherin regulates MAL-SRF-mediatedtranscription in epithelial cells. J Cell Sci 2010;123:2803–9.

37. Yu D, Wolf JK, Scanlon M, Price JE, Hung MC. Enhanced c-erbB-2/neuexpression in human ovarian cancer cells correlates with more severemalignancy that can be suppressed by E1A. Cancer Res 1993;53:891–8.

38. Mani SA, Guo W, Liao MJ, Eaton EN, Ayyanan A, Zhou AY, et al. Theepithelial-mesenchymal transition generates cells with properties of stemcells. Cell 2008;133:704–15.

39. Kalluri R, Weinberg RA. The basics of epithelial-mesenchymal transition. JClin Invest 2009;119:1420–8.

40. Pattabiraman DR, Bierie B, Kober KI, Thiru P, Krall JA, Zill C, et al.Activation of PKA leads to mesenchymal-to-epithelial transition and lossof tumor-initiating ability. Science 2016;351:aad3680.

41. Kelly TK, Liu Y, Lay FD, Liang G, Berman BP, Jones PA. Genome-widemapping of nucleosome positioning and DNA methylation within indi-vidual DNA molecules. Genome Res 2012;22:2497–506.

42. Hajra KM, Chen DY, Fearon ER. The SLUG zinc-finger protein repressesE-cadherin in breast cancer. Cancer Res 2002;62:1613–8.

43. Bolos V, Peinado H, Perez-Moreno MA, Fraga MF, Esteller M, Cano A. Thetranscription factor Slug represses E-cadherin expression and inducesepithelial to mesenchymal transitions: a comparison with Snail and E47repressors. J Cell Sci 2003;116:499–511.

44. Rose NR, Klose RJ. Understanding the relationship between DNA meth-ylation and histone lysine methylation. Biochim Biophys Acta 2014;1839:1362–72.

45. Liang YJ,Ding Y, Levery SB, LobatonM,HandaK,Hakomori SI. Differentialexpression profiles of glycosphingolipids in humanbreast cancer stem cellsvs. cancer non-stem cells. Proc Natl Acad Sci U S A 2013;110:4968–73.

46. Liang YJ, KuoHH, Lin CH, Chen YY, Yang BC, Cheng YY, et al. Switching ofthe core structures of glycosphingolipids from globo- and lacto- to ganglio-series upon human embryonic stem cell differentiation. Proc Natl Acad SciU S A 2010;107:22564–9.

47. Huang RY, Guilford P, Thiery JP. Early events in cell adhesion andpolarity during epithelial-mesenchymal transition. J Cell Sci 2012;125:4417–22.

48. Huang RY, Chung VY, Thiery JP. Targeting pathways contributing toepithelial-mesenchymal transition (EMT) in epithelial ovarian cancer.Curr Drug Targets 2012;13:1649–53.

49. Clevers H. The cancer stem cell: premises, promises and challenges.Nat Med 2011;17:313–9.

50. Tam WL, Weinberg RA. The epigenetics of epithelial-mesenchymal plas-ticity in cancer. Nat Med 2013;19:1438–49.

51. Beloribi-Djefaflia S, Vasseur S, GuillaumondF. Lipidmetabolic reprogram-ming in cancer cells. Oncogenesis 2016;5:e189.

52. Okuda T, TokudaN, Numata S, ItoM,OhtaM, Kawamura K, et al. Targeteddisruption of Gb3/CD77 synthase gene resulted in the complete deletionof globo-series glycosphingolipids and loss of sensitivity to verotoxins.J Biol Chem 2006;281:10230–5.

53. Perets R, Wyant GA, Muto KW, Bijron JG, Poole BB, Chin KT, et al.Transformation of the fallopian tube secretory epithelium leads to high-grade serous ovarian cancer in Brca;Tp53;Pten models. Cancer Cell 2013;24:751–65.

54. Wang AZ, Ojakian GK, Nelson WJ. Steps in the morphogenesis of apolarized epithelium. I. Uncoupling the roles of cell-cell and cell-substra-tum contact in establishing plasma membrane polarity in multicellularepithelial (MDCK) cysts. J Cell Sci 1990;95:137–51.

55. Bucior I, Scheuring S, Engel A, Burger MM. Carbohydrate-carbohydrateinteraction provides adhesion force and specificity for cellular recognition.J Cell Biol 2004;165:529–37.

56. Hakomori S. Carbohydrate-to-carbohydrate interaction in basic cell biol-ogy: a brief overview. Arch Biochem Biophys 2004;426:173–81.

57. Shaul YD, Freinkman E, Comb WC, Cantor JR, Tam WL, Thiru P, et al.Dihydropyrimidine accumulation is required for the epithelial-mesenchy-mal transition. Cell 2014;158:1094–109.

58. Merritt EA, Sarfaty S, van den Akker F, L'Hoir C,Martial JA, HolWG. Crystalstructure of cholera toxin B-pentamer bound to receptor GM1 pentasac-charide. Protein Sci 1994;3:166–75.

59. Everest-Dass AV, Briggs MT, Kaur G, Oehler MK, Hoffmann P, Packer NH.N-glycan MALDI imaging mass spectrometry on formalin-fixed paraffin-embedded tissue enables the delineation of ovarian cancer tissues.MolCellProteom 2016;15:3003–16.






2018;78:2952-2965. Published OnlineFirst March 23, 2018.Cancer Res Francis Jacob, Shahidul Alam, Martina Konantz, et al.

1-4 Galactosyltransferase-Mediated GlycosphingolipidsαTransition of Mesenchymal and Epithelial Cancer Cells Depends on

Updated version

10.1158/0008-5472.CAN-17-2223doi:

Access the most recent version of this article at:

Material

Supplementary

http://cancerres.aacrjournals.org/content/suppl/2018/03/23/0008-5472.CAN-17-2223.DC1

Access the most recent supplemental material at:

Cited articles

http://cancerres.aacrjournals.org/content/78/11/2952.full#ref-list-1

This article cites 59 articles, 21 of which you can access for free at:

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

Subscriptions

Reprints and

[email protected]

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

Permissions

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

.http://cancerres.aacrjournals.org/content/78/11/2952To request permission to re-use all or part of this article, use this link



http://cancerres.aacrjournals.org/lookup/doi/10.1158/0008-5472.CAN-17-2223

http://cancerres.aacrjournals.org/content/suppl/2018/03/23/0008-5472.CAN-17-2223.DC1

http://cancerres.aacrjournals.org/content/78/11/2952.full#ref-list-1

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

mailto:[email protected]

http://cancerres.aacrjournals.org/content/78/11/2952


Documents

Transition of Mesenchymal and Epithelial Cancer Cells Depends … · Tumor Biology and Immunology Transition of Mesenchymal and Epithelial Cancer Cells Depends on a1-4 Galactosyltransferase-Mediated