Analysis and Sorting of Prostate Cancer CellTypes by Flow Cytometry

Alvin Y. Liu,1* Lawrence D. True,2 Leah LaTray,3 William J. Ellis,1Robert L. Vessella,1 Paul H. Lange,1 Celestia S. Higano,4 Leroy Hood,3 and

Ger van den Engh3

1Department of Urology, University of Washington, Seattle, Washington2Department of Pathology, University of Washington, Seattle, Washington

3Department of Molecular Biotechnology, University of Washington, Seattle, Washington4Department of Medicine, University of Washington, Seattle, Washington

BACKGROUND. Prostate tumor heterogeneity as manifested by differential expression ofmarkers can be attributed to multiple types of cancer cells populating a tumor. Does thecomposition differ between primary tumor and metastasis? How can one isolate the differentcancer cell types to study? What is the relationship among cancer cell types?METHODS. Flow cytometry keying on the prostate epithelial cell surface markers CD57 andCD44 was applied to analyze and sort single cells prepared from tumor tissue samples bycollagenase digestion. In normal tissue, CD57 is found on luminal cells and CD44 on basalcells.RESULTS. CD57+ and CD44+ cells were sorted from various prostate tumor tissue specimens.The CD57+ cancer cell type was found to predominate in primary tumors, while the CD44+

cancer cell type was found to predominate in two visceral metastases. All tumors could becharacterized by a ratio of CD57+ and CD44+ cancer cells.CONCLUSIONS. Two types of prostate cancer cells, CD57+ and CD44+, were identified. Thefinding that most primary tumors contain a predominantly CD57+ cancer cell populationagrees with the argument that cancer cells arise from the transformation of CD57+ luminalcells. However, CD44+ cancer cells are also present in some primary tumors; and in somemetastases, they, and not CD57+ cells, constitute a predominant population. Prostate 40:192–199, 1999. © 1999 Wiley-Liss, Inc.

KEY WORDS: CD44; CD57; prostate cancer cells; flow cytometry

INTRODUCTION

In our search for the genetic cause of prostate can-cer we must find ways to solve the tissue heterogene-ity problem. The relevant cell types need to be siftedfrom various other cell types for experimental analy-sis. The major cell types in the glandular structures ofthe prostate are the epithelial populations of basal andluminal cells, and the stromal population of smoothmuscle cells [1]. A smaller population of epithelialcells with neuroendocrine differentiation is also found[2]. Based on a number of shared characteristics, mostnotably the cytokeratin profile [3] and the synthesis ofprostate-specific antigen (PSA) and prostatic acid

phosphatase (PAP) [4], cancer cells likely arise fromthe transformation of luminal cells. The alternative ofa basal-cell origin of cancer is favored by some inves-tigators [5].

Previously, we showed that it is possible to isolate

Grant sponsor: CaPCURE Foundation; Grant sponsor: Departmentof Veterans Affairs; Grant sponsor: National Science FoundationCenter for Molecular Biotechnology; Grant number: BIR9214821-AM.*Correspondence to Alvin Liu, Ph.D., Department of Urology, Box356510, University of Washington, Seattle, WA 98195. E-mail:aliu@u.washington.eduReceived 8 December 1998; Accepted 17 March 1999

The Prostate 40:192–199 (1999)

© 1999 Wiley-Liss, Inc.

the two epithelial cell populations from tissue by flowcytometry [6]. The reagents used to tag the cell typesare the cell surface molecules CD44 for basal cells andCD57 for luminal cells. We reasoned that if cancer cellsare derived from either luminal or basal cells, theymight also be identified and sorted by CD57 or CD44.As the sorted cells remain viable, we could study theircell biology and gene expression. The isolated homo-geneous cell populations can in turn provide thesource material for the construction of cDNA libraries.We present here our results on the cellular composi-tion as determined by flow analysis of CD57- andCD44-positive cancer cells in primary tumors, a lymphnode metastasis, and two visceral metastases.

MATERIALS AND METHODS

Prostate Specimens

Surgically resected prostates were inked, represen-tative blocks of tissue from the right and left apex,middle, and base were cut, and 5-mm frozen sectionswere prepared to localize tumor by staining. Using thestained sections as templates, tumor tissue was dis-sected from the adjacent tissue block that had not beenfrozen. Samples of tumor$100 mg were processed forflow analysis or sorting. Benign tissue samples weretaken, whenever possible, from the peripheral zone ofthe prostate. Specimens were handled under sterileconditions. The colored inks used to orient the glandswere sterilized by ultraviolet (UV) irradiation.

Single-Cell Suspension

Tissue specimens were minced and digested withcollagenase in RPMI-1640 medium supplementedwith 5% bovine serum albumin (BSA) and 10−8 M di-hydrotestosterone, as described elsewhere [6]. The di-gestion medium was saved for later detection of se-creted cancer cell proteins. A discontinuous densitygradient of Percoll (Amersham Pharmacia, Piscata-way, NJ) solutions was used to partition the stromaland epithelial cell populations prior to flow sorting[6].

Flow Cytometry

The procedure was described elsewhere [6]. Thecells were resuspended in 0.1% BSA-Hank’s balancedsalt solution and divided into two aliquots of 50–100ml for labeling by antibody-dye conjugates. A smallfraction without antibody was used to delineate theunstained and autofluorescent populations. The anti-body-fluorescent dye conjugates used were aCD57-PE(PE is R-phycoerythrin, 575-nm peak fluorescence) for

luminal cells and luminal cell-like cancer cells, andaCD44-PE for basal cells and basal cell-like cancercells. The CD44 antibody was clone G44-26 (IgG2b, K,PharMingen, San Diego, CA), and the CD57 antibodywas clone NK-1 (IgM, K, Sigma, St. Louis, MO). FITC(fluorescein isothiocyanate, 530-nm peak fluores-cence)-conjugated CD44 or CD57 antibodies wereused together with the PE-conjugated antibodies intwo-color analysis and sorting. Cells were collected inRPMI-1640 medium.

Gene-Expression Analysis

The technique of reverse transcriptase-polymerasechain reaction (RT-PCR) and cDNA amplification wasapplied to detect the expression of marker genes in thesorted cell populations, and has been previously de-scribed [7]. PCR products were resolved by agarosegel electrophoresis. The oligonucleotide primers AT-CACCGACAGCACAGACAGAATCCCT and ATCT-GATTCAGATCCATGAGTGGTATGG were used todetermine the CD44 isoforms expressed by CD44+ tu-mor cells. The first primer matches a sequence in exon5 of the CD44 gene, and the second primer matches asequence in exon 15. Alternatively spliced exons areinserted between theses two exons. This strategy wassimilar to that used for the analysis of CD44 isoformsin prostate cancer cell lines [8].

RESULTS

Primary Tumors Contain Mainly CD57+

Cancer Cells

Representative examples of the flow analyses of tu-mor samples taken from radical prostatectomy speci-mens are presented in Figure 1. In these analyses, cellswere stained separately by aCD57-PE and aCD44-PE.CD57-positive cells (events) are represented by greendots, and CD44-positive cells by red dots in the cyto-grams (Fig. 1). The population of unstained (CD57− orCD44−) and autofluorescent cells is represented byblack dots. Depending on the debris content of thesample, the size of this population could be consider-ably inflated. The amount of CD57+ luminal or CD44+

basal cells seldom scored the expected percentage ofthe total events registered in flow analyses of benigntissue [6]. Cancer specimen RM was predominantlyCD57+, as shown in Figure 1A. The finding that therewere only a small number of CD44+ cells was consis-tent with the histologic assessment that the tissuespecimen was cancerous. A benign specimen wouldcontain CD44+ basal cells, as shown in Figure 1B. Theabsence of basal cells is a diagnostic feature of prostatecancer [4]. The sorted CD57+ and CD44+ cell numbers

Prostate Cancer Cell Types 193

A

B

C

from specimen RM and three other specimens (VA,AB, and DT) are listed in Table I. Unlike RM and VA,a higher proportion of CD44+ cells was sorted fromspecimens AB and DT. Figure 1C shows the flowanalysis of specimen DT, in which three times as manyCD44+ cells as CD57+ cells were sorted. The flow-sorted specimen consisted of a single, well-circumscribed nodule of cancer. There was little be-nign tissue, and hence the CD44+ cells were not con-taminating basal cells. Given this type of result, a moreapt descriptive parameter for each tumor would be theratio of CD57+ and CD44+ cells. The ratios for the fourprimary tumors along with other tumor features aregiven in Table II. None of the patients in this cohorthad evidence of regional lymph node metastasis. Tu-mor AB with more CD44+ cells had invaded the semi-nal vesicles. Regarding many more of the other pros-tatectomy tumors analyzed to date, the majority werelike RM in that they contained a predominantly CD57+

population (our unpublished findings). In no case wasthere a pure CD44+ cancer, in the way that RM was apure CD57+ cancer.

Visceral Metastases Contain More CD44+

Cancer Cells

Due to their scarcity, only three metastases werestudied in the same time period. They included speci-men JL from a left external iliac lymph node, specimenJW from a necrotic tumor mass in the abdominal cav-ity, and specimen BR from a tumor mass along thesmall bowel containing poorly differentiated cancercells with an unusually high mitotic rate. The nodalmetastasis, by flow cytometry, contained three timesas many CD57+ cells as CD44+ cells (Table I). In con-trast, the two visceral metastases contained a pre-dominantly CD44+ cell population (for example, fivetimes as many CD44+ as CD57+ cells for JW). TheCD57 and CD44 cytograms of specimen BR are shownin Figure 2. Cells were not sorted from BR. A two-color analysis indicated that the two populations weredistinct, though the possibility of double positive cellscould not be entirely ruled out. The sorted popula-tions of CD44+, CD44−, and CD57− cells from speci-men JW were processed for RNA isolation, and theirrespective gene-expression pattern was analyzed byRT-PCR. Not enough CD57+ cells were collected to beincluded in this analysis. The PCR products resolvedby agarose gel electrophoresis are shown in Figure 3.The b2-microglobulin (B2M) and epithelial cell-specific epithelial glycoprotein (EGP) genes were usedas positive controls. The efficiency of flow sorting wasjudged from the CD44 result. The CD44+ population,as expected, contained CD44 sequences, whereas theCD44− population did not (compare lanes 5 and 6 inFig. 3). Such a correlation between CD44+ cells and thepresence of CD44 sequences (see also Liu et al. [6])showed that there was no nonspecific binding of theantibodies. The major CD44 isoform in these cancercells was found to be the standard form. This wasdetermined by RT-PCR amplification using primersflanking the exon junction into which alternativelyspliced variant isoforms were inserted, followed bycloning and sequencing of the PCR products. The ma-jor product size was 94 bp, with the following se-quence at the spliced junction (↓) of exons 5 and 15:GCTACCAC↓AGACCAA… The CD44+ cancer cellswere also scored positive for PSA, PAP, transcriptionfactor ETS2, androgen receptor (AR), and the cell di-vision-associated marker Ki67. Unlike sorted CD57+

luminal cells [6], these cancer cells apparently contin-ued, after sorting, their synthesis of PSA and PAP.Furthermore, the CD44+ cancer cells differed fromCD44+ basal cells in that basal cells do not synthesizePSA and PAP [6], so that they were not likely to be thedirect descendent of basal cells. Note that the CD44−

population contained cells positive for neuron-specificenolase (NSE), a marker often encountered in high-

<

Fig. 1. A: Flow analysis of primary tumor RM. Percoll gradient-purified epithelial cells (epi) of the dissected tumor tissue weredivided into two aliquots for labeling with aCD57-PE and aCD44-PE. A small fraction without added antibody (no Ab) was analyzedfirst to delineate the negative population (left). Positive popula-tions were scored outside this negative population. The CD57+

population is colorized green (middle), and the CD44+ populationcolorized red (right). B: Flow analysis of benign tissue. A two-color analysis of the cells is shown. C: Flow analysis of primarytumor DT. A histologic section of the tumor is shown in themicrograph. In this case, Percoll was not used before flow, as thetumor tissue contained minimal amounts of stromal elements. Thethree cytograms show the flow analysis of tumor cells preparedfrom this sample.

TABLE I. CD57+ and CD44+ Cells Sorted From TumorTissue Samples*

Specimen No. of cells Specimen No. of cells

RM CP VA CPCD57+ 77,000 CD57+ 81,000CD44+ 2,000 CD44+ 16,000AB CP DT CPCD57+ 246,000 CD57+ 160,000CD44+ 158,000 CD44+ 428,000JL LN mets JW metsCD57+ 175,000 CD57+ 120,000CD44+ 63,000 CD44+ 624,000

*Specimens RM, VA, AB, and DT are primary tumors (CP), JLLN mets is a node metastasis, and JW mets is an abdominalmetastasis. Listed are the numbers of cells sorted from the samespecimens.

Prostate Cancer Cell Types 195

grade diseases. A small percentage of NSE-positivecells is found in normal tissue, and these cells areCD57+ [2] (also, our unpublished data). Whether theseNSE+ cancer cells are CD57+ or not remains to be de-termined.

DISCUSSION

Flow cytometry is the single most powerful tech-nique in cell population analysis. A lack of cell-specificmarkers and the extra effort to prepare single cellshave kept it from being applied to cell analysis of solidtumors. We showed that it was possible to preparesingle cells from minced prostate tissue by collagenasedigestion, and to sort epithelial cells by antibodiesagainst the differentially expressed cell surface anti-gens CD44 and CD57 [6]. Similarly, this procedurecould be employed to analyze cell populations of tu-mor tissue.

We showed here that both CD57+ and CD44+ cancercells can be identified, and their ratio as determined byflow cytometry in tumor specimens can be informa-tive. They constitute two distinct types, and both typescan be found in the same primary tumors. We subse-quently verified the presence of CD57+ and CD44+

cancer cells in those positive samples by immunohis-tochemistry (unpublished data). Nonetheless, theCD57+ cells of the primary tumors analyzed to dateconstituted the predominant population. In metasta-ses (save perhaps nodal ones), the CD44+ cells ap-peared predominant. Their prominence in metastasescould be due to natural selection, since CD44+ cancercells, like basal cells, have proliferative capacity (asindicated by ETS2 positivity [7]) and a survival advan-tage over CD57+ cancer cells. These observations sug-gest that CD44+ cancer cells are more malignant. Wewill assay the malignant potential of these cells by invitro tests, using sorted cells. We will also use follow-up of patients to determine if a higher ratio of CD44+

cancer cells in primary tumors correlates with treat-

ment failure. As can be seen in Table II, the amount ofCD44+ cells can range from minimal to significantlylarger proportions. If our hypothesis is correct, it fol-lows that the odds of surgical failure would increasefor those tumors with a higher proportion of CD44+

cancer cells. The standard CD44 isoform is the majorspecies expressed by these cancer cells. It is also thespecies detected in cells established from micrometas-tases [9].

Cancer specimens had previously been screened byother investigators for CD57 or CD44 reactivities sepa-rately, but not simultaneously. The question of wheth-er CD57+ and CD44+ cancer cells represent differenttypes, as shown here, was not addressed. On the onehand, well-differentiated tumors were shown to con-tain the highest percentage of CD57+ cancer cells andthe strongest reactivity, whereas poorly differentiatedtumors showed the lowest percentage and the weakestreactivity. Patients with more CD57+ (vs. CD57−) cellsin their tumor had higher survival and nonprogres-sion rates [10]. A loss of CD57 staining was noted inprogression to hormone independence [11,12]. CD57was not detected in PC3, DU145, and LNCaP [13] (alsoour unpublished data). On the other hand, increasedCD44 reactivity in poorly differentiated tumors [14]and a positive correlation of CD44 expression to highGleason scores [15] have been reported. CD44 was de-tected in cell lines PC3, DU145, TSU-Pr1, PPC-1, andALVA-31, but not LNCaP [6,8,16–18]. The double-nullphenotype of LNCaP would suggest a third cancer celltype, that of CD57−/CD44−, as well as those ofCD57+/CD44− and CD57−/CD44+. Alternatively, thisnull phenotype may be the result of cell selection oradaptation to long-term culture. Some xenografts thatwere established from metastatic lesions were alsofound to be CD44-negative (unpublished data). TheCD44− phenotype may explain the decrease in CD44expression in disease progression reported by a num-ber of investigators [14,19–22]. Other CD markers aretherefore needed to type these cancer cells. A hypo-

TABLE II. CD57+/CD44+ Cancer Cell Ratio and Tumor Characteristics*

Specimen CD57/CD44 ratio Gleason score Volume Tumor stage

RM 38.5 3 + 4 = 7 7.3cc Organ-confinedVA 5.1 4 + 3 = 7 4 cc Organ-confinedAB 1.6 4 + 4 = 8 9 cc Seminal vesiclesDT 0.37 3 + 3 = 6 2 cc Organ-confinedJL LN mets 2.7 3 + 3 = 6 [Left iliac node]JW mets 0.19 4 + 5 = 9 [Rectum]

*The CD57/CD44 ratio is calculated from the number of cells sorted from the tumorsamples processed, and as none were sorted from specimen BR, its ratio was notincluded. Volume is the estimated tumor volume in the resected prostate. Organsites from which metastases were taken are in brackets.

196 Liu et al.

thetical lineage of the epithelial cell types identified byCD57 and CD44, which we hope to test, is outlined inFigure 4.

Lastly, Isaacs et al. [23] generated data in the Dun-ning rat model suggesting that CD44 was a suppressorof prostate cancer metastasis. In that study, the human

CD44 gene was transfected into metastatic Dunningtumor cells, and the resultant hCD44+ cells becamenonmetastatic. However, our results obtained fromthe analyses of human cancer specimens showed thatCD44+ cancer cells could constitute a major popula-tion in visceral metastases. If CD44 were a prostate

Fig. 2. Flow analysis of visceral metastasis BR. The cytograms show the reactivity of cells in the epithelial fraction of BR to aCD44-PE,aCD57-PE, and aCD44-PE/aCD57-FITC. CD44-positive cells constituted 37%, while CD57-positive cells 3% of the events (cells andparticles) scored. Labeling with both antibodies showed that the CD57+ population (green) appeared to be distinct from the CD44+

population (red). mets, metastasis; no Ab, no antibody.

Prostate Cancer Cell Types 197

cancer metastasis suppressor in man, one would notexpect to find CD44+ cancer cells in any metastasis.Recently, another group also reported in a preliminarystudy the detection of CD44+ cells in metastatic lesions[24]. Several prostate cancer xenografts in our LuCaPseries [25] that were established from samples takenfrom metastatic lesions were also found to be positivefor CD44 expression (unpublished data).

CONCLUSIONS

In summary, our results support the argument thatprostate cancer cells arise from the transformation ofluminal cells. This is inferred from the observationthat most primary tumors contain CD57+ cancer cells,CD57 being a luminal cell-specific marker. The lumi-nal cell-like cancer cells may progress to a type that isbasal cell-like, as the number of CD44+ cancer cellsseems to increase with disease severity. The cancerprocess is characterized by a loss of cellular differen-tiation, since the CD44+ cancer cell type is consideredless differentiated than the CD57+ cancer cell type,analogous to the difference between normal CD44+

basal and CD57+ luminal cell types. Tumor heteroge-neity can be attributed to the presence of more than

one type of cancer cells. CD57+ and CD44+ representtwo types.

REFERENCES

1. Cunha GR, Hayward SW, Dahiya R, Foster BA. Smooth muscle-epithelial interactions in normal and neoplastic prostatic devel-opment. Acta Anat (Basel) 1996;55:63–72.

2. di Sant’Agnese PA. Neuroendocrine differentiation in carci-noma of the prostate. Diagnostic, prognostic, and therapeuticimplications. Cancer 1992;70:254–268.

3. Okada H, Tsubura A, Okamura A, Senzaki H, Naka Y, KomatzY, Morii S. Keratin profiles in normal/hyperplastic prostatesand prostate carcinoma. Virchows Arch [A] 1992;421:157–161.

4. Mostofi FK, Sesterhenn IA, Davis CJ. Prostatic carcinoma: prob-lems in the interpretation of prostatic biopsies. Hum Pathol1992;23:223–241.

5. Bonkhoff H, Remberger K. Differentiation pathways and histo-genetic aspects of normal and abnormal prostatic growth: astem cell model. Prostate 1996;28:98–106.

6. Liu AY, True LD, LaTray L, Nelson PS, Ellis WJ, Vessella RL,Lange PH, Hood L, van den Engh G. Cell-cell interaction inprostate gene regulation and cytodifferentiation. Proc Natl AcadSci USA 1997;94:10705–10710.

7. Liu AY, Corey E, Vessella RL, Lange PH, True LD, Huang GM,Nelson PS, Hood L. Identification of differentially expressedprostate genes: increased expression of transcription factorETS-2 in prostate cancer. Prostate 1997;30:145–153.

8. Liu AY. Expression of CD44 in prostate cancer cells. Cancer Lett1994;76:63–69.

9. Putz E, Witter K, Offner S, Stosiek P, Zippelius A, Johnson J,Zahn R, Riethmuller G, Pantel K. Phenotypic characteristics ofcell lines derived from disseminated cancer cells in bone mar-row of patients with solid epithelial tumors: establishment ofworking models for human micrometastases. Cancer Res 1999;59:241–248.

10. Liu X, Zhan B, Tomoyoshi T, Masanori T. Immunohistochemicalstudy of HNK-1 (leu-7) antigen in prostate cancer and its clinicalsignificance. Chin Med J [Engl] 1995;108:516–521.

11. Tanahashi T, Namba K, Murao T. Studies on relationship be-tween histology, tumor markers and clinical course in prostatecancer. Nippon Hinyokika Gakkai Zasshi [Jpn J Urol] 1990;81:680–685.

12. Mai KT, Commons S, Perkins DG, Yazdi HM, Collins JP. Ab-sence of serum prostate-specific antigen and loss of tissue im-munoreactive prostatic markers in advanced prostatic adeno-

Fig. 3. Gene-expression analysis of sorted cell populations of visceral metastasis JW. Analyzed are the sorted CD44+, CD44−, and CD57−

cancer cell populations. The gene markers tested include B2M, CD44, EGP, NSE, PSA, AR, PAP, ETS2, and Ki67. RT-PCR products wereresolved by gel electrophoresis and their expected sizes are: B2M, 560 bp; CD44, 600 bp; EGP, 890 bp; NSE, 1,010 bp; PSA, 460 bp; AR,780 bp; PAP, 670 bp; ETS2, 850 bp; and Ki67, 860 bp. lHindIII is the DNA size marker.

Fig. 4. Hypothetical epithelial cell lineage. The four epithelial celltypes depicted are characterized by their CD44, CD57, PSA, andETS2 expression.

198 Liu et al.

carcinoma after hormonal therapy: a report of two cases. HumPathol 1996;27:1377–1381.

13. Rokhlin OW, Cohen MB. Expression of cellular adhesion mol-ecules on human prostate tumor cell lines. Prostate 1995;26:205–212.

14. Murant SJ, Handley J, Stower M, Reid N, Cussenot O, MaitlandNJ. Co-ordinated changes in expression of cell adhesion mol-ecules in prostate cancer. Eur J Cancer 1997;33:263–271.

15. Zhang X, Sakamoto H, Takenaka I. Accumulation of p53 andexpression of CD44 in human prostatic cancer and benign pros-tatic hyperplasia: an immunohistochemical study. Br J Urol1996;77:441–444.

16. Welsh CF, Zhu D, Bourguignon LYW. Interaction of CD44 vari-ant isoforms with hyaluronic acid and the cytoskeleton in hu-man prostate cancer cells. J Cell Physiol 1995;164:605–612.

17. Lokeshwar BL, Lokeshwar VB, Block NL. Expression of CD44 inprostate cancer cells: association with cell proliferation and in-vasive potential. Anticancer Res 1995;15:1191–1198.

18. Stevens JW, Palechek PL, Griebling TL, Midura RJ, Rokhlin OW,Cohen MB. Expression of CD44 isoforms in human prostatetumor cell lines. Prostate 1996;28:153–161.

19. Kallakury BVS, Yang F, Figge J, Smith KE, Kausik SJ, Tacy NJ,Fisher HAG, Kaufman R, Figge H, Ross JS. Decreased levels of

CD44 protein and mRNA in prostate carcinoma. Cancer 1996;78:1461–1469.

20. Nagabhushan M, Pretlow TG, Guo Y, Amini SB, Pretlow TP, SyM. Altered expression of CD44 in human prostate cancer duringprogression. Am J Clin Pathol 1996;106:647–651.

21. Noordzij MA, van Steenbrugge G, Verkaik NS, Schroder FH,van der Kwast TH. The prognostic value of CD44 isoforms inprostate cancer patients treated by radical prostatectomy. ClinCancer Res 1997;3:805–815.

22. De Marzo AM, Bradshaw C, Sauvageot J, Epstein JI, Miller GJ.CD44 and CD44v6 downregulation in clinical prostatic carci-noma: relation to Gleason grade and cytoarchitecture. Prostate1998;34:162–168.

23. Gao AC, Lou W, Dong J, Isaacs JT. CD44 is a metastasis sup-pressor gene for prostatic cancer located on human chromo-some 11p13. Cancer Res 1997;57:846–849.

24. Takahashi S, Kimoto N, Orita S, Cui L, Sakakibara M, Shirai T.Relationship between CD44 expression and differentiation ofhuman prostate adenocarcinomas. Cancer Lett 1998;129:97–102.

25. Ellis WJ, Vessella RL, Buhler KR, Bladou F, True LD, Bigler SA,Curtis D, Lange PH. Characterization of a novel androgen-sensitive, prostate-specific antigen-producing prostatic carci-noma xenograft: LuCaP 23. Clin Cancer Res 1996;2:1039–1048.

Prostate Cancer Cell Types 199