stem cell differentiation and lumen formation in ... · on lumen formation and stem cell...

Tumor and Stem Cell Biology

Stem Cell Differentiation and Lumen Formation in ColorectalCancer Cell Lines and Primary Tumors

Neil Ashley, Trevor M. Yeung, and Walter F. Bodmer

AbstractSingle cancer stem–like cells (CSC) from colorectal cancers can be functionally identified by their ability to

form large lumen-containing colonies in three-dimensional Matrigel cultures. These colonies contain thethree types of differentiated colorectal epithelial cells, and single cells obtained from them can reproducethemselves and form tumors efficiently in immunodeficient mice. In this study, we show how hypoxia affectsthese CSC-derived lumens to control differentiation of stem-like cells and enterocytes via the homeobox geneCDX1. Lumens were identified by F-actin staining and they expressed many characteristics associated withnormal differentiated intestinal epithelium, including brush border enzymes, polarization, and tight junc-tions. RNA interference–mediated silencing of CDX1 reduced lumen formation. Inhibitory effects of hypoxiaon lumen formation and stem cell differentiation, including suppression of CDX1 expression, could bemimicked by inhibiting prolyl-hydroxylases that activate HIF1, suggesting that HIF1 is a critical mediator ofthe effects of hypoxia in this setting. Cell line–derived lumens were phenotypically indistinguishable fromcolorectal tumor glandular structures used by pathologists to grade tumor differentiation. Parallel results tothose obtained with established cell lines were seen with primary cultures from fresh tumors. This in vitroapproach to functional characterization of CSCs and their differentiation offers a valid model to studycolorectal tumor differentiation and differentiation of colorectal CSCs, with additional uses to enable high-throughput screening for novel anticancer compounds. Cancer Res; 73(18); 5798–809. �2013 AACR.

IntroductionColorectal tumors originate in the epithelial layer of the large

intestine, which is lined with microinvaginations called crypts.A stemcell population resides at the crypt base and gives rise tothe 3 main differentiated lineages in colon: enterocytes, gobletcells, and entero-endocrine cells (1). Enterocytes lining thecrypts of the intestine express a layer of microvilli on theirapical membranes known as the brush border, in which manydifferentiation markers such as villin are segregated in apolarized manner (2–4). Full development of the brush borderand associated cytoskeletal and enzyme activities correspondswith intestinal differentiation (5, 6). Although cellular differ-entiation is ultimately deranged in colonic adenocarcinomas,colorectal tumors usually retain some degree of differentiation,such as glandular structure and expression of differentiationmarkers, and also frequently express brush border–enrichedenzymes such as DPPIV/CD26 and alkaline phosphatase (7).

The glandular form of tumors and general mucinous contentare used by pathologists to help grade tumors (8). Tumors areclassed as either well/moderately differentiated or "low grade"if they contain many "neoplastic glands" that have a crypt-likestructure or as poorly differentiated or "high grade" if they lacksuch glandular structures (9). Low-grade tumors generallyexhibit a better prognosis for the patient (10).

Much evidence indicates that adenocarcinomas are drivenby a subset of cells with characteristics of stem cells, includingthe ability to self-renew and to differentiate into all 3 coloniclineages (11). We have previously shown that a subpopulationof single cells derived from certain colorectal cell lines can,when grown for several weeks under 3-dimensional (3D)conditions, form either large crypt-like structures consistingof polarized cells surrounding a cell-free lumen or instead formsmall non-lumen colonies (12–15). The large lumen coloniesexpress differentiation markers for all 3 colon cell lineages,indicating the multipotent nature of the original cell. Singlecells derived from large lumen colonies were capable of form-ing new large lumen colonies, as well as small non-lumencolonies, whereas cells derived from small non-lumen coloniescould only give rise to further small colonies (14). Lumencolony–derived cells were also more tumorigenic in mousexenografts than cells derived from small non-lumen colonies.The proportion of cells that gives rise to lumen colonies couldbe increased by enrichment of cells with high expression ofboth the stem cell markers CD44 and CD24 (14). Collectively,these characteristics indicate that lumen formation is drivenby high clonogenic cancer stem cells that can be derived from

Authors' Affiliation: Cancer and Immunogenetics Laboratory, WeatherallInstitute of Molecular Medicine, Department of Oncology, University ofOxford, John Radcliffe Hospital, Oxford, United Kingdom

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

Corresponding Author: Walter F. Bodmer, Oxford University WeatherallInstitute of Molecular Medicine, John Radcliffe Hospital, Headington,Oxford OX3 9DS, United Kingdom. Phone: 44-0186-5222-416; Fax: 44-0186-5222-737; E-mail: [email protected]

doi: 10.1158/0008-5472.CAN-13-0454

�2013 American Association for Cancer Research.

CancerResearch

Cancer Res; 73(18) September 15, 20135798

Cancer Research. by guest on August 29, 2020. Copyright 2013 American Association forhttps://bloodcancerdiscov.aacrjournals.orgDownloaded from

https://bloodcancerdiscov.aacrjournals.org

lumen-forming cell lines. Low clonogenic cells are unable togive rise to large lumen colonies and form small non-lumencolonies. Therefore, formation of lumens from single cells canbe used to characterize cancer stem cell differentiation.Interestingly, some colorectal cell lines such as DLD1 do not

differentiate under 3D conditions and instead form disorga-nized colonies (14, 15). These cell lines retain self-renewalcapacity of stem cells but lack the ability to fully differentiateand are correspondingly more aggressive in xenografts andmore clonogenic than lumen-forming cell lines (14). Onepossibility is that differentiation is inhibited in these lines byreduced expression of key transcription factors such as CDX1.CDX1 is a gut transcription factor crucial to colorectal differ-entiation (16, 17), and it controls the transcription of a numberof intestinal differentiation markers including villin (18), cyto-keratin 20 (16), and FABP1 (19). In colorectal cancers, CDX1 isfrequently transcriptionally silenced by promoter methylation(16, 17).In this study, we have characterized the nature of differen-

tiation of 3D colonies formed by different cell lines and primarytumor-derived cancer stem cells.

Materials and MethodsCell cultureDetails of cell line origins and validation can be found in the

Supplementary Text. All cell lines were cultured in completeDulbecco's Modified Eagle Media (DMEM) supplemented with10% FBS and 1% penicillin/streptomycin (Invitrogen). Hypoxiccells were grown in humidified 1% oxygen and 10% CO2

environment using a MiniGalaxy incubator (RS Biotech Ltd).Mediumwas replaced 2 times per week after being prewarmedand equilibrated overnight in 1% oxygen. For dimethyloxaloyl-glycine (DMOG) treatment, culture medium containing 2mmol/L DMOG or dimethyl sulfoxide (DMSO; vehicle control)was overlaid over the gels and this was changed every 24 hours.For primary culture, 6 moderate/well-differentiated tumor

specimens, obtained from fully consenting donors underNational Research Ethics study, 07/H0606/120, were mechan-ically disrupted into small pieces and cultured in serum0freemedium (Excell-620, Sigma-Aldrich Ltd.) supplemented with2% StemPro hESC supplement (Life Technologies Inc.) onnonadherent plastic, until spheroids were formed. The cul-tures were checked by EpCam immunofluorescence andreverse transcriptase PCR to consist of CDX1-expressingepithelial cells (not shown). Established colonies were trans-ferred to Matrigel and grown for 7 days, before fixation andimmunostaining, as detailed below.

F-actin assay of colonies grown in MatrigelSingle-cell suspensions were achieved by fluorescence-

activated cell sorting (FACS) or filtration through 30- and20-mm filters (Celltrics, Partec GmbH). Five hundred cellswere suspended in 40 mL of an ice-cold mixture of Matrigeland diluted 1:1 with ice-cold DMEM. Cell suspensions wereseeded in triplicate into 96-well plates precoated with asolidified 1:1 Matrigel/DMEM layer (20 mL per well). OverlaidMatrigel was allowed to set for 20 minutes at 37�C, andculture medium was then added to the wells. Colonies were

grown for 2 weeks, unless otherwise specified, with mediumchanges every 3 days. The medium was removed and 100 mLof 4% (v/v) paraformaldehyde in PBS was added for 20minutes. Fluid was then removed and 100 mL of PBS contain-ing 1:200 dilution of Triton-x was added for 10 minutes. Thefluid was removed and the wells were washed 4 times with50 mmol/L glycine in PBS. About 100 mL TRITC-phalloidindiluted at 1:1,000 in PBS was added and incubated at 4�Covernight. The colonies were then rinsed 3 times with PBS asabove. A total of 100 mL of 10 mg/mL 406-diamidino-2-phenylindole (DAPI; Sigma) in PBS was added to each welland the plate was imaged using a Zeiss LSM510 laserscanning confocal microscope (Carl Zeiss Ltd.).

FACS analysisCells were harvested and 1� 106 cells and labeled with anti-

CD44 and anti-CD24. Cell viability was assessed with SytoxBlue. Cells were run on a DakoMoFLow cytometer. Gates wereset to exclude dead cells, debris, and doublets and the top orbottom 5% of gated CD44/CD24 (see Fig. 3). Labeled cells werecollected into separate tubes containing 200 mL ice-coldDMEM. Cells were then embedded in Matrigel as describedabove and assayed after 2-week growth.

Western blottingWestern blotting was conducted on cell lysates as described

previously (20).

In vivo tumorigenic assaysSingle cellswere injected subcutaneously into theflanks of 6-

to 10-week-old female NOD/SCID immunodeficient mice,obtained from the John Radcliffe Hospital Biomedical Services(Oxford, UK) as described previously (14). Animal studies wereconducted according to the University of Oxford institutionalguidelines andwithin the limits of the Project License issued bythe Home Office, United Kingdom.

Further details ofmaterials andmethods can be found in theSupplementary Data.

ResultsLumen characterization

F-actin (filamentous actin) was identified as a useful markerof lumens by staining normal human colon cryosections withfluorescent phalloidin, which is specific for F-actin (Fig. 1A). Innormal colon, F-actin is intensely enriched at the apical surfaceof enterocytes lining the crypt, corresponding to the brushborder, presumably due to the presence ofmicrovilli on colonicenterocytes (21, 22). We next examined F-actin labeling insingle-cell–derived colonies from a panel of 6 colon cancer celllines, grown in a 3DMatrigel matrix for 2 weeks. In an SW1222colony, confocal Z-sectioning showed intense F-actin labelingon the apical cell membranes facing the lumen, which wassurrounded by polarized cells expressing cytokeratin 20 ontheir basolateralmembranes (Fig. 1B). LS180, SW1222, and C80formed colonies consisting of polarized cells surroundingcentral lumens visible by light microscopy (Fig. 1C). All phasevisible lumens were clearly labeled by intense F-actin staining.HT29, HCT116, and DLD1 did not form observable lumen

Lumen Formation in Colorectal Cancers

www.aacrjournals.org Cancer Res; 73(18) September 15, 2013 5799



Figure 1. Polarized F-actin is a marker ofcolonic brush borders in vivo and labelslumens in vitro. A, confocal images of normalhuman colon cryosections labeled withTRITC-phalloidin (red) and DAPI (blue) tovisualize F-actin and nuclei, respectively.Left image shows longitudinal section; rightimage shows transverse section. Scale bar,50 mm. B, z-stack gallery of a SW1222colony labeledwith F-actin (red), DAPI (blue),and cytokeratin 20 (green). Numbersrepresent distance in z-axis from top. Scalebar, 10 mm. C, single cells from various celllines were grown in Matrigel for 6 days.Colonieswere labeledwith TRITC-phalloidin(red) to visualize F-actin and DAPI tovisualize nuclei. Bar, 10 mm.

Ashley et al.

Cancer Res; 73(18) September 15, 2013 Cancer Research5800



colonies using phase contrast and did not exhibit organizedbright foci of F-actin staining, although F-actin was visible inplasmamembranes. A gallery of F-actin–labeled colonies fromvarious cell lines is shown in Supplementary Fig. S1.We next examined whether other intestinal brush border

markers were also present in lumens. Villin, an F-actin–inter-acting protein associated with intestinal differentiation, wasstrongly enriched in lumens and its staining overlapped thatfor F-actin (Fig. 2A). HCT116 andDLD1did not express villin. InHT29, villin was generally expressed but it was not polarizedand did not overlap F-actin. Ezrin, another F-actin–interactingprotein enriched in the brush border, was also stronglyenriched in lumens where it was clearly located at the apicalcell membrane (Fig. 2B). Ezrin was only weakly present at non-apical plasma membranes. Ezrin was strongly expressed inHCT116 in a nonpolarized manner.The brush border enzyme CD26/dipeptidyl peptidase-4

(DPPIV; ref. 7) and the brush border–enriched tight junctionmarker Zo-1(23) were also polarized and enriched in lumenapical membranes (Supplementary Fig. S2A and S2B). Incontrast, non-lumen–forming cell lines failed to show orga-nized expression of either marker.We next examined whether formation of F-actin–enriched

lumens in colonies derived from a single-cell precluded thepresence of secretory lineages. Co-staining of SW1222 colonieswith phalloidin and anti-MUC2, a goblet cell–specific mucin

(Fig. 2C), showed that goblet cells were present in F-actincolonies grown from single cells. We also examined expressionof carcinoembryonic antigen (CEA); a membrane-associatedand secreted glycoprotein commonly used as a cancer marker.CEA was present almost exclusively in the lumens of C80 andSW1222 colonies (Fig. 2D).

The above evidence indicates that lumen formation incolonies derived from single colorectal cells represents brushborder differentiation but does not preclude differentiationalong secretory lineages.

Cancer stem cell enrichmentTo determine whether stem cell–enriched cell fractions

gave rise to increased frequencies of lumens with F-actin/brush border differentiation, we FACS-sorted SW1222 andLS180 cells according to expression of cancer stem cellmarkers CD24 and CD44, as described previously (14). Singlesorted cells were then plated in Matrigel and grown intocolonies for 2 weeks. In Fig. 3A and B, the top 5% of CD24/44(CD24 and CD44 high)-expressing SW1222 cells gave rise tosignificantly more large colonies with F-actin–labeled lumensthan unsorted/bulk cell populations and significantly fewersmall colonies with no lumens. The bottom 5% CD24/44-expressing cells (CD24 and CD44 low) cells showed very lowclonogenicity, and resulting colonies were virtually all-smallwith absent or poorly developed F-actin foci. Similar results

Figure 2. Lumens represententerocyte brush borderdifferentiation but also containsecretory lineages. Single cellsfrom specified cell lines wereembedded in Matrigel and grownfor 10 days. Colonies were fixedand labeled with phalloidin (red),DAPI (blue), and immunolabeledwith anti-villin (A) or anti-ezrin (B). Cand D, single SW1222 or C80 cellswere grown in Matrigel for 10 daysbefore immunolabeling with eitheranti-Muc2 antibodies (C) or anti-CEA (D). F-actin was colabeledwith TRITC-phalloidin. Scale bar,20 mm.





were obtained with LS180 cells (Fig. 3C and D). Thus, stemcell–enriched populations give rise to a larger proportion oflarger colonies with increased brush border/enterocyte dif-ferentiation and a much smaller proportion of small non-lumen–forming colonies.

Effect of hypoxia and Hif1a stabilization ondifferentiation and CDX1 expression

Lumen formation can be inhibited by long-term hypoxia,accompanied by a downregulation of CDX1 in an Hif1-depen-dent manner (15). To determine whether F-actin was similarly

downregulated by hypoxia, we compared LS180 and SW1222cells grown in Matrigel for 14 days in either normoxia orhypoxia (1% O2). Figure 4A shows representative images ofF-actin labeling of the colonies under each condition. Hypoxiccolonies were smaller and F-actin focus size and intensity werereduced, compared with normoxia. The number of F-actin fociappeared to be similar. Quantification of colonies indicated asignificant drop in the proportion of colonies with F-actin–labeled lumens and an increase of colonies without well-defined lumens (Fig. 4B). These results show that hypoxiainhibits brush border development in colon cancer stem cells.

Figure 3. F-actin can be used to identify lumen colonies in stem cell–enriched populations. A and C, scattergram plots showing the gating of the extreme5% of CD44/CD24 high- and low-expressing cells populations of SW1222 or LS180 cells sorted by FACS. The gates were set so that 5% of the totalpositive populations were selected in both the top right quadrant (CD44/CD24 high) and the bottom left quadrant (CD44/CD24 low). B and D, isolatedCD24/44-high and -low SW1222 or LS180 cells were grown for 2 weeks inMatrigel and stained with TRITC-phalloidin. Colonies weremanually characterizedinto the following categories: large colony, diameter larger than 40 mm; lumen colony, F-actin foci larger than 10 mm. Bar graphs show totalcounts of different colony classifications from 3 replicate wells for each gating strategy. B, for SW1222 colonies, Fisher exact test showed that single cellswith high CD24/44 expression gave rise to a significantly higher proportion of large lumen colonies at the expense of small non-lumen colonies whencomparedwith unsorted cells (P¼ 0.0007).P values represent Fisher exact test of significance of the change in the proportion of lumen to non-lumen coloniesand indicate a significant shift towards lumen colony formation for cells enriched for CD24/44 expression. c2 2 � 4 table also showed that single cells withhigh CD24/44 expression gave rise to a significantly higher proportion of lumen colonies to non-lumen colonies irrespective of colony size (P ¼ 0.0003),compared with unsorted cells. The CD24/44 low-expressing cells showed a much lower clonogenicity than either unsorted or CD24/44-high cells.Although there was a clear trend towards an increase in the ratio of small non-lumen colonies to large lumen colonies, the very small number of coloniesprecluded statistical analysis. D, for LS180, the Fisher exact test showed a similar significant increase in the proportion of large lumen to small non-lumencolonies derived from cells sorted for high CD24/44 expression, compared with CD24/44 low-expressing cells (P ¼ 0.0002), and a c2 2 � 4 table showeda significant difference between CD24/44-high and -low when comparing all lumen colonies to non-lumen colonies irrespective of colony size (P ¼ 0.0001).

Ashley et al.




To determine whether inhibition of lumen formation wasdependent onHif transcription factors, we cultured colonies inthe presence or absence of the prolyl-hydroxylase inhibitorDMOG. Prolyl-hydroxylases hydroxylate Hif transcription fac-tors, marking them for degradation and so DMOG interfereswith this process leading to accumulation of Hif1a (24). Wefirst confirmed that DMOG treatment induced expression ofHif1 andHif1-responsive genes in cell lines by immunolabelingfor Glut1 and carbonic anhydrase IX (CAIX) and immunoblot-ting for Hif1 (Supplementary Fig. S3A–S3C). We then treatedcolonies grown from single cells inMatrigel for 7 days and withDMOG for a further 5 days. The effect on F-actin labeling wasdetermined with TRITC-phalloidin. Figure 4C shows thatDMOG inhibited F-actin lumen formation in C80 and LS180

and SW1222. Large F-actin lumens were inhibited and, instead,many colonies exhibited small F-actin foci. There was noobvious effect on colony size. Quantitation of the colonymorphologies showed a significant reduction in the proportionof colonies with large, well-defined F-actin lumens and anincrease in the proportion of colonies with small F-actin foci(Fig. 4D).

We next determined whether DMOG treatment induceddownregulation of CDX1 and its target cytokeratin 20 (16).Immunolabeling of DMOG-treated colonies and vehicle-trea-ted controls showed a marked downregulation of CDX1 inresponse to DMOG, although cytokeratin 20 expressionremained virtually unchanged (Fig. 5A and B). We also lookedat the effect of DMOG on goblet cell lineage differentiation.

Figure 4. A, hypoxia and Hif activation reduces F-actin foci formation. A, single cells of either LS180 or SW1222 were grown in Matrigel under normal 20%oxygen or 1% oxygen for 14 days. The colonies were then labeled with phalloidin (red). Scale bar, 50 mm. B, colony counts of the same experimentshowing total number of colonies of LS180 cells in defined categories, from 3 replicates. Small colony defined as any colony with a radius smaller than250 mm; lumen colony, colony with 1 or more F-actin foci larger than 10 mm. Forty colonies counted from each of 3 wells for hypoxia and normoxia.Hypoxia induced a significant difference in the proportion of lumen to non-lumen colonies (P ¼ 0.0235, Fisher exact test 2 � 2 table), compared withcolonies grown in 20% oxygen. P value represents a Fisher exact test of significance of the change in the proportion of lumen to non-lumen colonies andindicates a significant shift towards non-lumen colony formation in the presence of DMOG.C, lumen F-actin formation is inhibited byDMOG. Single cells weregrown in Matrigel for 1 week in the presence of 2 mmol/L DMOG or vehicle control (DMSO). Colonies were then fixed and F-actin labeled with phalloidin.D, graph showing total colonies with large or small lumens. A total of 40 colonies were counted from each of 3 wells for each condition. Lumen colony,colony with 1 or more F-actin foci larger than 15 mm. For LS180 and C80, DMOG significantly inhibited lumen formation (P � 0.0001, 2 � 2 tables,Fisher exact test of proportion of lumen to non-lumen colonies), compared with colonies grown in DMSO control. SW1222 showed a nonstatisticallysignificant trend towards reduced lumens by DMOG (P ¼ 0.157, Fisher exact test).





Immunolabeling colonies with an anti-goblet cell antibodyindicated that DMOG suppressed goblet cell differentiationin both the C80 and LS180 cell lines (Fig. 5C and D).

Thus, DMOG treatment leads to rapid loss of CDX1 expres-sion but not cytokeratin 20 and inhibits both brush border/enterocyte and secretory goblet cell differentiation.

To confirm that CDX1 downregulation could impair F-actinlumen formation, we used an LS174T-derivative cell lineLS174T-CDX1 with stable expression of a construct encodinga short hairpin RNA targeting CDX1, in which CDX1 wasdownregulated by RNA interference (RNAi; ref. 16). An LS174Tline, LS174T-vector, expressing an empty vector was usedas a control, as these cells maintain normal CDX1 levels.Single cells from either LS174T siCDX1 or LS174T-vector wereembedded in Matrigel and allowed to grow into colonies for10 days, before fixation and phalloidin labeling to visualizeF-actin. Figure 6A and B shows that low CDX1 significantly

reduced the proportion of colonies with well-defined F-actinlumens. We also manually quantified the total number oflumens per colony in the same data and found that CDX1knockdown also reduced this number, that is, CDX1 knock-down inhibited lumens per colony (Fig. 6C). We next tookadvantage of the CDX1 RNAi F-actin data to validate a semi-automatic approach to lumen formation that would facilitatehigh-throughput analysis of stem cell differentiation usingpublic ImageJ software. Figure 6D, i, shows a whole well-composite image of an entire well from a 96-well plate thatcontained F-actin–labeled SW1222 colonies grown in 3D con-ditions. ImageJ software was then used to automaticallyrecognize and analyze the colonies (Fig. 6D, ii). The softwarewas able to detect lumens accurately in a well-differentiatedcolony and was able to distinguish that colony from a poorlydifferentiated colony lacking F-actin foci (Fig. 6D, iii and iv,respectively). To validate this approach, we used our

Figure 5. Prolyl-hydroxylaseinhibition downregulates CDX1and suppresses goblet celldifferentiation. A and B, 10-day-oldcolonies grown in Matrigel fromsingle cells of either C80 (A) orSW1222 (B) treated with 2 mmol/LDMOG or DMSO for 4 days.Colonies were imaged by phasecontrast and coimmunolabeledwith anti-CDX1 and anti-cytokeratin 20 (CK20). Bar, 20 mm.C, single LS180 or C80 cells wereplated on plastic at 500 cells perwell in a 96-well plate and allowedto grow for 1 week in total. On daythree, 2 mmol/L DMOG or DMSO(vehicle control) was added to thewells. The cells were fixed andlabeled with DAPI (blue), phalloidin(red), and anti-goblet cell mucin(PR5D5; green). Bar, 20 mm. D, bargraph showing quantification ofgoblet cell–positive colonies invehicle- or DMOG-treatedwells. Of120 colonies looked at, from a totalof 3 wells for each condition, thenumber that contained goblet cellsand the number that did not aregiven in the bar graph. DMOGcaused a significant decrease inthe proportion of goblet cellcontaining colonies comparedwithDMSO controls for both C80 andLS180 (P ¼ 0.0030, P ¼ 0.0001,respectively, using Fisher exacttest for the relevant 2 � 2 table).

Ashley et al.




semiautomatic method to analyze the LS174T-vector andLS174T-CDX1 colonies manually counted in Fig. 6C and founda good correlation between the automated and manual anal-yses (Fig. 6E). Thus, our approach enables tissue plate wells tobe automatically scanned and analyzed for changes in lumenformation.

Lumen formation is a feature of tumors in vivoTo determine whether lumen formation had relevance to

tumors in vivo, we first examined actin and ezrin polariza-tion in murine xenografts derived from the injection ofHT29, HCT116, or SW1222 cells into the flanks of NOD/SCID mice. Mice were sacrificed after 1 month and resultingtumors removed and processed for formalin-fixed, paraffin-embedded (FFPE) tissue sections, which were stained byhematoxylin and eosin or immunolabeled with anti-actin/ezrin (Fig. 7A). Hematoxylin and eosin staining of thesexenograft tumors showed poorly differentiated high-grade

tumor morphology for HCT116 and HT29 tumors that lackedlumens or actin/ezrin polarization. SW1222 tumors exhib-ited a well-differentiated phenotype with numerous lumenssurrounded by polarized cells. Actin and ezrin immunola-beling showed that neither HT29 nor HCT116 exhibitedpolarization of these markers, whereas SW1222 showedmarked luminal polarization. Thus in vivo, cell lines growin a manner similar to in vitro growth in Matrigel.

Because similar glandular structures (neoplastic glands)are a feature of human colorectal adenocarcinomas and areused by pathologists to grade tumors, we examined whetherthey labeled with the same markers as in vitro lumens. Wecompared 2 colorectal tumors that had been graded previouslyas either moderately differentiated (low-grade) or poorly dif-ferentiated (high-grade). Hematoxylin and eosin staining oftissue sections from the low-grade tumor confirmed the pres-ence of numerous neoplastic glands, which were absent in thepoorly differentiated tumors (Fig. 7B). Immunolabeling of

Figure 6. CDX1 knockdown reduces F-actin lumen formation in LS174T colonies. A, F-actin labeling of LS174T vector and LS174T CDX1 RNAi culturedfrom single cells in Matrigel for 1.5 weeks. B, bar chart showing total counts of 200 colonies in each of three replicate wells, categorized into lumen or non-lumen colonies. Colonies were defined as non-lumen if they contained no F-actin foci with a diameter above 8 mm. Knockdown of CDX1 caused asignificant reduction in the proportion of lumen colonies, compared with vector control colonies (P� 0.0001, Fisher exact test). C,manual counts of averagednumbers of F-actin lumens per colony of LS174T-vector or LS174T-CDX1 colonies from the same experiment as used for B. A total of 120 colonies wereanalyzed (40 each from 3 replicates). The number of lumens in LS174T-CDX1 was significantly reduced compared with the vector control (Student t test, P�0.0001). Error bar, SD. D, semiautomated analysis of F-actin lumen formation using ImageJ. D, i, composite image of an entire well from a 96-well platecontaining F-actin–labeled SW1222 colonies grown in Matrigel from single cells for 10 days. D, ii, masked (outlined) colonies automatically identified by theImageJ algorithm. D, iii, a well-differentiated lumen colony is shown with 7 automatically detected F-actin foci. D, iv, a small poorly differentiatedcolony with only 1 F-actin focus automatically detected. E, automatic F-actin lumen counting was applied to the LS174T data used for C and asignificantly reduced number of F-actin lumens per colony in the LS174T-CDX1 knockdown colonies was detected (Student t test, P ¼ 0.001).





Figure 7. Lumen formation is afeature of tumors in vivo, andhuman primary low- and high-grade tumors show differences inezrin/actin andCEApolarization. A,lumen-forming and non-lumen–forming cell lines form well-differentiated (low-grade) andpoorly differentiated (high-grade)tumors, respectively, as xenograftsin vivo. Hematoxylin and eosin oranti-actin/ezrin immunostaining ofFFPE sections from murinexenograft tumors derived fromSW1222, HCT116, or HT29 cellsinjected into NOD/SCID. Insetshows magnification. B, typicalhematoxylin and eosin staining ofFFPE tissue sections derived froma moderately differentiated (low-grade) primary colorectal tumorfrom a patient, with numerousneoplastic glands visible (left), anda poorly differentiated (high-grade)human colorectal tumor, with noglands (right). C and D, anti-actin/anti-ezrin or anti-CEA/villin labelingof FFPE tissue sections derivedfrom the same patient derivedhigh- and low-grade tumors shownin B. E, characterization of lumenformation in primary culturesderived from a low-grade tumor.Left, phase contrast image oftypical primary cultured spheroidcancer cell colony derived from thesame low-grade human colorectaltumor shown in B–D. Right, F-actinlabeling of these primary coloniesfollowing growth in Matrigel for 7days. F, summary of lumenformation efficiency in Matrigel of apanel of 6 primary colorectalcancer cultures derived from 6separate patients.

Ashley et al.




sections from the same tumors indicated that lumens in thelow-grade tumor were composed of polarized cells expressingactin and ezrin on the apical membranes, whereas no polar-ization of these markers was observed in the poorly differen-tiated tumors (Fig. 7C). Typical of tumors, cellular debris waspresent in the lumens. Actin/ezrin-positive cells present in thestromal surrounding the glands were probably myofibroblasts(Supplementary Fig. S4). Lumen glands were strongly positivefor polarized CEA, whereas CEA was nonpolarized and morecell membrane associated in the high-grade tumor (Fig. 7D).Thus, at least 3 markers of in vitro lumens also mark in vivoglands, indicating that in vitro lumens from cell lines arevirtually identical in nature to lumens in primary tumors.To determine whether lumen formation was a feature of

primary tumor cells in vitro, we derived a primary tumor cellculture from the same low-grade tumor used in Fig. 7B–D.When grown under conditions that promote stem cell popula-tions (25–28), these primary cells grew as well-organizedspheroids (Fig. 7E, left). Following a week-long suspension inMatrigel, the colonies formed large central lumens marked byF-actin, (Fig. 7E, right). Three of 6 primary cultures derivedfromseparate patientswere capable of lumen formation (Table1). Thus, we conclude that lumen formation represents thesameprocess in vivo as it does in vitro and can therefore be usedto identify stem cell populations in primary tumor–derivedcultures.

DiscussionSingle stem cells from colorectal cancer lines, when grown

under 3D conditions, can differentiate into polarized struc-tures with remarkable structural similarity to normal intes-tinal crypts. Lumens express many characteristics of normalintestinal brush borders, including microvilli structural pro-teins (polarized F-actin, villin, ezrin), enzymatic activity(polarized DPPIV), cell surface and secreted glycoprotein(CEA), and tight junctions (TJP1/Zo-1). Thus, colorectalcancer stem cells that produce lumen colonies can beconsidered to be differentiating predominantly along theenterocyte lineage, the most common colonic lineage. Nev-ertheless, goblet cells were also often present in colonieswith F-actin lumens that had been grown from single cells,illustrating their multipotent nature. Importantly, we alsoshow that several of these markers of in vitro lumen brushborder formation are also expressed in a polarized mannerby neoplastic glands present in primary tumors, which aresimilarly composed of polarized cells surrounding a lumen.Thus, it is most likely that these neoplastic lumens alsoindicate stem cell differentiation in the same way as their invitro counterparts. This is supported by our observation thatprimary tumor spheroid cultures derived from moderately/well-differentiated tumors could also form polarized lumencolonies when transferred into Matrigel.Unlike lumen colonies and well-differentiated tumors, non-

lumen cell lines, such as HCT116, totally lacked cellular polar-ization of several polarity markers such as villin, a situationmirrored in vivo by high-grade tumors. Thus, cell lines can alsobe separated into high or low "grades" based on their polar-ization in Matrigel.

One possible explanation for this is that these cells lackexpression of CDX1 (16), and this may disrupt their ability toefficiently polarize and differentiate. Ectopic re-expression ofCDX1 in HCT116 and DLD1 can lead to the formation ofprimitive lumens in Matrigel (14, 15). Correspondingly, CDX1and the related transcription factor CDX2 have been shown tobe important for polarization of intestinal cells (29, 30). A lackof polarization and differentiation will lead to an increase inthe population of stem cells and indeed we have found pre-viously that HCT116 does not contain subpopulations of cellswith different tumor-forming capacity, unlike the lumen-form-ing lines (14). Similarly, high-grade tumors have a significantlyworse clinical prognosis and exhibit a more "stem-like" tran-scription profile (31, 32).

Tumor formation is frequently associated with hypoxia,and our previous studies have shown that prolonged hypoxiainhibits lumen formation (15). Correspondingly, hypoxialeads to a failure to polarize F- actin correctly, as didtreatment of colonies with the Hif1a activator DMOG.DMOG also inhibited CDX1 expression and goblet celldifferentiation, although cytokeratin 20 expression was notaffected, probably due to a difference in turnover ratebetween CDX1 and cytokeratin 20. As the CDX1 promotercontains a number of putative Hif-binding sites (unpub-lished data), it is possible that downregulation of CDX1 isdirectly induced by Hif-mediated transcriptional inhibition.It seems likely that the loss of F-actin lumen expression inresponse to hypoxia/DMOG treatment is triggered by aloss of CDX1. This is supported by the observation thattargeted knockdown of CDX1 in LS174T cells leads to asignificant reduction in the size and intensity of F-actin–labeled lumens. However, we cannot rule out a role forother hypoxia-regulated transcription factors in differenti-ation suppression.

Specific cancer stem cell markers remain elusive. Whilemarkers such as LGR5 (33), ephrin type-B receptor 2 (EphB2;ref. 34), and Lrig1 (35) appear to mark well-defined stem cellpopulations in normal intestine, their expression in tumorsdoes not appear to be entirely specific to cancer stem cells aslow clonogenic cells are also marked, at least for EphB2 (31),and LGR5 (36, 37). F-actin lumen formation should thereforebe useful for defining the cutoff in expression of markerssuch as LGR5 and EphB2 that define cancer stem cellpopulations from non-stem cells. The ability of single cancerstem cells to grow into differentiated lumen colonies pro-vides an unambiguous tool to estimate stem cell popula-tions. Our approach therefore allows for the functionalcharacterization ofCSCs and their differentiation, independentof expression of stem cell surface markers. Furthermore, theF-actin lumen assay can be adapted for high-throughputapplications, which should be useful for identifying drugs thatcan target stem cells.

In summary, we have characterized the formation of lumencolonies from cell line–derived single cells to represent brushborder/enterocyte differentiation and show the extent of het-erogeneity between cell lines and primary tumor cultures.Thus, the cancer stem cells from the cell lines, in their abilityto form lumen colonies, have a striking similarity to normal





intestinal stem cells, which also divide to form the coloniccrypt.

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

Authors' ContributionsConception and design: N. Ashley, T. Yeung, W.F. BodmerDevelopment of methodology: N. AshleyAcquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): N. Ashley, T. YeungAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): N. Ashley, W.F. BodmerWriting, review, and/or revision of themanuscript:N. Ashley, W.F. BodmerStudy supervision: W.F. Bodmer

AcknowledgmentsThe authors thank Kevin Clarke Jenny Wilding and Steve Taylor for technical

and conceptual input.

Grant SupportThis work was funded, in part, by the EpiDiaCan FP7 European Union Grant

no: FP7-241783.The costs of publication of this article were defrayed in part by the

payment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely to indicate thisfact.

Received February 18, 2013; revised May 29, 2013; accepted June 24, 2013;published OnlineFirst July 18, 2013.

References1. Potten CS, Booth C, Pritchard DM. The intestinal epithelial stem cell:

the mucosal governor. Int J Exp Pathol 1997;78:219–43.2. Lacroix B, Kedinger M, Simon-Assmann P, Rousset M, Zweibaum A,

Haffen K. Developmental pattern of brush border enzymes in thehuman fetal colon. Correlationwith somemorphogenetic events. EarlyHum Dev 1984;9:95–103.

3. Holmes R. The intestinal brush border. Gut 1971;12:668–77.4. Holmes R, Lobley RW. Intestinal brush border revisited. Gut 1989;

30:1667–78.5. Carboni JM, Howe CL, West AB, Barwick KW, Mooseker MS,

Morrow JS. Characterization of intestinal brush border cytoskeletalproteins of normal and neoplastic human epithelial cells. A com-parison with the avian brush border. Am J Pathol 1987;129:589–600.

6. West AB, Isaac CA, Carboni JM, Morrow JS, Mooseker MS, BarwickKW. Localization of villin, a cytoskeletal protein specific to microvilli, inhuman ileum and colon and in colonic neoplasms. Gastroenterology1988;94:343–52.

7. Zweibaum A, Hauri HP, Sterchi E, Chantret I, Haffen K, Bamat J, et al.Immunohistological evidence, obtained with monoclonal antibodies,of small intestinal brushborder hydrolases inhumancolon cancers andfoetal colons. Int J Cancer 1984;34:591–8.

8. Hamilton SR, Aaltonen LA. World Health Organization classification oftumors: pathology and genetics of tumors of the digestive system.IARC Press; Lyon, France; 2000.

9. Hinoi T, Tani M, Lucas PC, Caca K, Dunn RL, Macri E, et al. Loss ofCDX2 expression and microsatellite instability are prominent featuresof large cell minimally differentiated carcinomas of the colon. Am JPathol 2001;159:2239–48.

10. O'Connell JB, Maggard MA, Ko CY. Colon cancer survival rates withthe new American Joint Committee on Cancer sixth edition staging. JNatl Cancer Inst 2004;96:1420–5.

11. Ashley N. Regulation of intestinal cancer stem cells. Cancer Lett. 2012Apr 27. [Epub ahead of print].

12. Richman P, Bodmer W. Control of differentiation in human colorectalcarcinoma cell lines: epithelial-mesenchymal interactions. Pathol J1988;156:197–211.

13. Del Buono R, Pignatelli M, Bodmer WF, Wright NA. The role of thearginine-glycine-aspartic acid-directed cellular binding to type I col-lagen and rat mesenchymal cells in colorectal tumour differentiation.Differentiation 1991;46:97–103.

14. Yeung TM, Gandhi SC, Wilding JL, Muschel R, Bodmer WF. Cancerstem cells from colorectal cancer-derived cell lines. ProcNatl Acad SciU S A 2010;107:3722–7.

15. Yeung TM,Gandhi SC, BodmerWF. Hypoxia and lineage specificationof cell line-derived colorectal cancer stem cells. ProcNatl AcadSci USA 2011;108:4382–7.

16. Chan CW, Wong NA, Liu Y, Bicknell D, Turley H, Hollins L, et al.Gastrointestinal differentiation marker Cytokeratin 20 is regulatedby homeobox gene CDX1. Proc Natl Acad Sci U S A 2009;106:1936–41.

17. Wong NA, Britton MP, Choi GS, Stanton TK, Bicknell DC, Wilding JL,et al. Loss of CDX1 expression in colorectal carcinoma: promotermethylation, mutation, and loss of heterozygosity analyses of 37 celllines. Proc Natl Acad Sci U S A 2004;101:574–9.

18. Arango D, Al-Obaidi S, Williams DS, Dopeso H, Mazzolini R, Corner G,et al. Villin expression is frequently lost in poorly differentiated coloncancer. Am J Pathol 2012;180:1509–21.

19. Staloch LJ, Divine JK, Witten JT, Simon TC. C/EBP and Cdx familyfactors regulate liver fatty acid binding protein transgene expression inthe small intestinal epithelium. Biochim Biophys Acta 2005;1731:168–78.

20. Ashley N, Poulton J. Anticancer DNA intercalators cause p53-depen-dent mitochondrial DNA nucleoid re-modelling. Oncogene 2009;28:3880–91.

21. Pittman FE, Pittman JC. Anelectronmicroscopic studyof epitheliumofnormal human sigmoid colonic mucosa. Gut 1966;7:644–61.

22. Bretscher A, Weber K. Localization of actin and microfilament-asso-ciated proteins in the microvilli and terminal web of the intestinal brushborder by immunofluorescence microscopy. J Cell Biol 1978;79:839–45.

23. Polak-Charcon S, Shoham J, Ben-Shaul Y. Tight junctions in epithelialcells of human fetal hindgut, normal colon, andcolonadenocarcinoma.J Natl Cancer Inst 1980;65:53–62.

24. Chan DA, Sutphin PD, Denko NC, Giaccia AJ. Role of prolyl hydrox-ylation in oncogenically stabilized hypoxia-inducible factor-1alpha. JBiol Chem 2002;277:40112–7.

25. Vermeulen L, Todaro M, de Sousa Mello F, Sprick MR, Kemper K,Perez Alea M, et al. Single-cell cloning of colon cancer stem cellsreveals a multi-lineage differentiation capacity. Proc Natl Acad Sci U SA 2008;105:13427–32.

26. Emmink BL, Van Houdt WJ, Vries RG, Hoogwater FJ, Govaert KM,Verheem A, et al. Differentiated human colorectal cancer cells protecttumor-initiating cells from irinotecan. Gastroenterology 2011;141:269–78.

27. Lombardo Y, Scopelliti A, Cammareri P, Todaro M, Iovino F, Ricci-Vitiani L, et al. Bone morphogenetic protein 4 induces differentiationof colorectal cancer stem cells and increases their response to che-motherapy in mice. Gastroenterology 2011;140:297–309.

28. Odoux C, Fohrer H, Hoppo T, Guzik L, Stolz DB, Lewis DW, et al. Astochasticmodel for cancer stemcell origin inmetastatic colon cancer.Cancer Res 2008;68:6932–41.

29. Keller MS, Ezaki T, Guo RJ, Lynch JP. Cdx1 or Cdx2 expressionactivates E-cadherin-mediated cell-cell adhesion and compaction inhuman COLO 205 cells. Am J Physiol Gastrointest Liver Physiol2004;287:G104–114.

30. Gao N, Kaestner KH. Cdx2 regulates endo-lysosomal function andepithelial cell polarity. Genes Dev 2010;24:1295–305.

31. Merlos-Suarez A, Barriga FM, Jung P, Iglesias M, Cespedes MV,Rossell D, et al. The intestinal stem cell signature identifies colorectalcancer stem cells and predicts disease relapse. Cell StemCell 2011;8:511–24.

Ashley et al.




32. de Sousa E, Melo F, Colak S, Buikhuisen J, Koster J, Cameron K, deJong JH, et al. Methylation of cancer-stem-cell-associatedWnt targetgenes predicts poor prognosis in colorectal cancer patients. Cell StemCell 2011;9:476–85.

33. Barker N, van Es JH, Kuipers J, Kujala P, van denBornM,CozijnsenM,et al. Identification of stem cells in small intestine and colon by markergene Lgr5. Nature 2007;449:1003–7.

34. Jung P, Sato T, Merlos-Suarez A, Barriga FM, Iglesias M, Rossell D,et al. Isolation and in vitro expansion of human colonic stem cells. NatMed 2011;17:1225–7.

35. Powell AE, Wang Y, Li Y, Poulin EJ, Means AL, Washington MK, et al.The pan-ErbB negative regulator Lrig1 is an intestinal stem cell markerthat functions as a tumor suppressor. Cell 2012;149:146–58.

36. Kemper K, Prasetyanti PR, De Lau W, Rodermond H, Clevers H,Medema JP. Monoclonal antibodies against Lgr5 identify humancolorectal cancer stem cells. Stem Cells 2012;30:2378–86.

37. Kobayashi S, Yamada-Okabe H, Suzuki M, Natori O, Kato A, Matsu-bara K, et al. LGR5-positive colon cancer stem cells interconvert withdrug-resistant LGR5-negative cells and are capable of tumor recon-stitution. Stem Cells 2012;30:2631–44.





stem cell differentiation and lumen formation in ... · on lumen formation and stem cell...

Documents