Down-Regulation of CD9 Expression during Prostate Carcinoma Progression Is Associated with CD9 mRNA Modifications Free

In North America, adenocarcinoma of prostate is the most frequently diagnosed cancer and the second leading cause of cancer death in men (1). Contrasting with these alarming figures, the mechanisms implicated in the progression of prostatic adenocarcinoma are not completely understood. In addition, because the natural history of prostatic adenocarcinoma is highly variable and unpredictable, more accurate markers to predict tumor progression are urgently needed for optimal therapy.

Using microcell-mediated chromosome transfer, we reported previously that the transfer of a fragment of a normal chromosome 12 (from human foreskin fibroblasts) was sufficient to suppress tumorigenicity of prostate carcinoma DU145 cell–derived hybrids (2). The fact that this suppression was lost on subsequent removal of the introduced human chromosome 12 fragment from these hybrids (2) argues strongly for the presence of one or more tumor suppressor gene(s) and/or other genes whose inactivation plays a role in prostate tumorigenesis. Human chromosome 12 had already been implicated in prostate cancer progression. Indeed, a small region of the long arm of this chromosome was sufficient to inhibit prostate cancer metastatic colonization of the Dunning AT6.1 rat prostate cancer cell line (3). Moreover, loss of heterozygosity (4) as well as reduced expression of specific genes present in a 2-Mb region of chromosome 12p12-13 has been associated with advanced prostate adenocarcinoma (5). Using irradiation, we further fragmented chromosome 12, and radiation hybrids exhibiting either tumorigenic or nontumorigenic phenotypes, depending on the portion of chromosome 12 retained, were generated. Using differential display reverse transcription-PCR, cDNA fragments whose expression varied between tumorigenic and nontumorigenic hybrids were identified. One of these cDNA fragments was derived from the cluster-of-differentiation antigen 9 (CD9) gene, located on human chromosome 12p13.3 (6, 7).

CD9, also referred to as MRP-1 (for motility related protein 1, not to be confused with the multidrug resistance protein 1), is a 25- to 27-kDa cell membrane glycoprotein (8–10) and a member of tetraspanins (11). Tetraspanins have been identified as a superfamily of transmembrane proteins containing over 32 members (12) expressed in many cell types. Members of this protein family associate with each other, with adhesion molecules, and other membrane proteins to form multimolecular complexes (11). It is believed that through these associations, tetraspanins can form a “tetraspanin web” (13) and thus serve as “molecular facilitators” or “organizers” of multimolecular complexes on the cell surface (11, 14, 15). The tetraspanin web, as the name suggests, provides a dynamic network for molecular interactions. Accordingly, tetraspanins are implicated in a variety of normal and pathologic processes, such as tissue differentiation (16), egg-sperm fusion (17, 18), virus-induced syncytium formation (19), and tumor-cell metastasis (11). Tetraspanins have also been involved in cell adhesion and motility (20). Interestingly, transfection of CD9 reduces metastasis in vivo and this effect has been related to the suppression of tumor cell growth and motility (21).

In cancer, decreased expression of CD9 protein has been implicated in progression of breast (22), non–small cell lung (23), and colon cancers (24). Moreover, decreased CD9 expression correlated with poor prognosis in several human malignancies, such as colon (25), lung (26), breast (22, 27), and ovarian carcinomas (28).

Based on our findings in tumorigenic and nontumorigenic human prostate cancer cells, our objective was to determine the role played by the CD9 protein in prostate cancer progression. Using immunohistochemistry, CD9 expression was characterized in 167 primary tumors (107 localized and 60 advanced), 65 lymph node metastases, and 23 bone metastases. Our results show that CD9 expression is greatly reduced and even lost during prostate cancer progression. To understand the mechanisms responsible for the loss of CD9 protein expression, CD9 cDNA was amplified, cloned, and sequenced from prostate cancer cell line DU145-N19 and its derived nontumorigenic hybrid DBM9-7 and from PC-3 and its derivative cell lines. CD9 cDNA was also cloned from prostatic adenocarcinoma, prostatic intraepithelial neoplasia (PIN), and normal prostate tissues obtained from 11 patients. Our results indicated that cDNAs derived from four prostatic adenocarcinoma cell lines, one PIN, and four prostatic adenocarcinoma specimens harbored one of three repeatedly occurring deletions affecting the CD9 coding region. Moreover, one prostatic adenocarcinoma cell line, one PIN, and seven prostatic adenocarcinoma specimens contained at least one and up to three of four point mutations, including a stop codon.

Cell lines and culture conditions. DU145-N19 is a tumorigenic, G418-resistant derivative of DU145. DBM9-7 is a nontumorigenic DU145-N19–derived microcell hybrid containing an extra fragment [del(12)(q13)] of a normal human chromosome 12 (2). DU145-N19 was grown in MEM supplemented with 10% fetal bovine serum (Canadian Life Technologies, Burlington, Ontario, Canada) and 800 μg/mL G418 (Roche Diagnostics, Laval, Quebec, Canada). DBM9-7 was grown in the same medium supplemented with 400 μg/mL hygromycin B (Roche Diagnostics) to ensure the retention of the tagged chromosome 12 (2). Both cell lines were grown in the absence of other antibiotics and were free of Mycoplasma as determined by Hoechst 33342 staining.

PC-3 was obtained from the American Type Culture Collection (Manassas, VA), whereas its derivatives, PC-3M, PC-3M-Pro4, and PC-3M-LN4 (29), were kindly provided by Dr. I.J. Fidler (University of Texas M. D. Anderson Cancer Center, Houston, TX); they were cultured in MEM supplemented with 10% fetal bovine serum.

Cloning and analysis of candidate cDNAs. We have used differential display (differential display reverse transcription-PCR; ref. 30) to isolate cDNA fragments expressed at higher level in DBM9-7 (the nontumorigenic hybrid) compared with DU145-N19 (the tumorigenic parental cell line). The expression of candidate fragments was assessed by Northern blot hybridization on membranes containing DU145-N19 and DBM9-7 polyA⁺ RNA. Interesting candidates were cloned and sequenced.

Western blotting. Protein samples (30 μg) extracted from two benign prostatic hyperplasia specimens, two localized, and three advanced human prostatic adenocarcinomas were separated on 13% SDS-PAGE and transferred to nitrocellulose membrane (0.45 μm). CD9 protein expression was detected on a 2-h incubation with an antihuman CD9 monoclonal mouse antibody (NCL-CD9; 1:500; Novocastra Laboratories/Vector Laboratories, Burlington, Ontario, Canada); the signal was amplified with a 1:2,000 dilution of a horseradish peroxidase–coupled goat anti-mouse antibody and revealed using enhanced chemiluminescence Western blotting protocol (GE Healthcare, Baie d'Urfe, Quebec, Canada) as described previously (31).

Patients and tissue materials. Patients provided inform consent and were investigated and treated at the McGill University Health Center. One hundred and seven patients had clinically localized disease and were treated with radical prostatectomy. Of these, 42 patients had received 3 months of preoperative combined androgen blockade hormone therapy. Formaldehyde-fixed and paraffin-embedded whole-mount sections of prostatectomy specimens were used for immunohistochemical analysis. Twenty-seven patients were at pathologic stage T₁N₀M₀, 32 at stage T₂N₀M₀, 47 at stage T₃N₀M₀, and 1 at stage T₂N₁M₀. The average age was 62 years (range, 44-76) at diagnosis, and the average follow-up interval was 62 months (range, 1.6-102).

Sixty patients had clinically advanced disease; in these cases, needle biopsy or transurethral resection of prostate material was used for immunohistochemical study. These patients were palliatively treated with bilateral orchiectomy or medical castration. Eleven patients were at stage T₃N₀M₀, 6 were at stage T_2-3N_1-3M₀, and 43 were at stage T_2-3N_1-3M₁. The average age was 71 years (range, 47-87) at diagnosis and the average follow-up interval was 39 months (range, 1-112). Furthermore, 23 bone and 65 lymph node specimens with metastatic involvement were available for immunohistochemical analysis; however, no clinical data were available.

Immunohistochemical technique. All prostatic tissues were fixed in 10% buffered formaldehyde and paraffin embedded. Four-micrometer-thick sections were obtained and deparaffinated in xylene. Tissues were rehydrated in decreasing graded ethanol and TBS. CD9 antigen was retrieved by heating at high pressure in a pressure cooker with 1 mmol/L EDTA (pH 8.0) for 10 to 15 min. Endogenous peroxidase was blocked by 1.5% of hydrogen peroxide in 50% methanol. An immunohistochemistry kit, HistostainTM-SP kit (Zymed, South San Francisco, CA), was used to detect CD9 signals (32). Tissues were first incubated with nonimmune goat serum for 25 min to minimize nonspecific binding and then incubated with monoclonal antihuman CD9 mouse antibody (NCL-CD9; dilution of 1:500) for 1 h. The CD9 antibody recognizes an epitope located in the extracellular region of the protein. The signal was detected with a goat anti-mouse IgG for 30 min and streptavidin-peroxidase conjugate for 15 min. The staining was revealed with the 3,3′-diaminobenzidine substrate (Sigma-Aldrich Canada Ltd., Oakville, Ontario, Canada). Tissues were counterstained with hematoxylin for 20 s and dehydrated in increasing graded ethanol and xylene. Staining intensity for CD9 expression was assessed under standard microscope.

Scoring of immunohistochemical staining. Two independent reviewers (J-C.W. and L.R.B.) analyzed all tissue sections, without knowledge of patient's clinical data, and a consensus was achieved about the final score. The effectiveness and specificity of the antibody (NCL-CD9) for immunohistochemical analysis was verified on a mammary fibroadenoma. As shown previously (22), cytoplasmic expression of CD9 protein was observed on epithelial cells of breast tissue. Ten whole-mount sections from 10 cases of prostatic adenocarcinoma were randomly selected and used to standardize CD9 scoring. The intensity of immunoreactivity (cytoplasmic staining often with apical and/or membranous enhancement) was defined as negative (score 0) versus positive (on a scale of 3; score 1 for weak staining, score 2 for moderate staining, and score 3 for strong staining). The percentage of tumor cells with immunoreactivity for CD9 was defined as ±5% approximation (e.g., 25-30%). The range of intensity within positive tumor cells was also marked. For instance, if one third of cells were stained weakly, whereas two thirds of cells were stained moderately, the range would be 1 to 2; therefore, a predominant intensity staining would be 2. Gleason pattern(s) present on the representative slide was also analyzed at the same time. The cutoff standard for CD9 immunohistochemical staining was adapted from previous analysis (27). In brief, when the total number of tumor cells with immunoreactivity (scores 1-3) for CD9 protein is >50% in a given specimen, the sample was classified as CD9 positive. When there were <50% of tumor cells with immunoreactivity, the sample was classified as reduced.

Statistical analysis. The χ² test and the Student's t test were done using Excel software (Microsoft Corp., Redmond, WA). Statistical significance was defined as P value <0.01.

Analysis of CD9 cDNA.CD9 cDNA was synthesized using Expand Reverse transcriptase according to the manufacturer's instructions (Roche Diagnostics) from total RNA isolated from 5 prostatic carcinoma cell lines, 10 prostatic carcinoma tissue specimens, 2 PIN tissue specimens, and 6 normal prostate tissue specimens (grossly dissected from frozen slides). CD9 cDNA was PCR amplified using high-fidelity Pfu DNA polymerase (Stratagene, La Jolla, CA). The primer sequences were derived from CD9 cDNA sequence, where position 1 was assigned to the first nucleotide of the coding sequence (33). Forward CD9 primer (5′-ATGCCGGTCAAAGGAGGCAC-3′) starts at nucleotide 1 and reverse CD9 primer (5′-GACCATCTCGCGGTTCCTGC-3′) starts at nucleotide 684. PCR products were purified and cloned using PCR-Script Amp cloning kit (Stratagene).

We also directly sequenced some PCR products (without cloning) using a similar set of primers (forward primer at position 3-22 and reverse primer at position 658-677). For the cloned CD9 cDNAs, sequencing was done using ABI Prism DNA analyzer 3700 (Applied Biosystems, Foster City, CA) with T7 and T3 primers. CD9 cDNA sequence analysis was done using Blast 2 nucleotide alignment.

CD9 mRNA is overexpressed in nontumorigenic hybrids. Differential display reverse transcription-PCR analysis identified 21 cDNA fragments, for which expression levels were higher in the nontumorigenic DBM9-7 cell line when compared with the parental tumorigenic DU145-N19 cell line (data not shown). Northern blot analysis confirmed that one of these candidates hybridized to a 1.4-kb mRNA, which was expressed in DBM9-7 at 3-fold higher levels compared with DU145-N19. Sequencing of this fragment revealed that it was derived from exon 8 of the CD9 gene. Because loss or decreased expression of the CD9 protein had already been associated with poor prognosis in breast and lung cancers (23, 26, 27), we pursued the characterization of CD9 protein expression and the analysis of its transcript in human prostate cancer.

Expression of CD9 in prostatic adenocarcinoma. Using Western immunoblotting, the 27-kDa CD9 protein was detected in two benign prostatic hyperplasia and two localized prostatic adenocarcinoma specimens. However, the CD9 protein was absent in two of three advanced prostatic adenocarcinoma specimens (Fig. 1). Using immunohistochemistry, CD9 expression was analyzed in 255 specimens (167 prostate specimens with primary adenocarcinoma, 65 lymph node specimens with metastasis, and 23 bone specimens with metastasis). Most sections from the 167 prostate specimens contained normal and/or hyperplastic epithelium in the background, thus providing an internal control for CD9 detection.

In nonneoplastic glands of all prostate specimens, CD9 was predominately expressed as diffuse cytoplasmic staining within secretory (luminal) cells, including focal apical enhancement (Fig. 2A). CD9 immunoreactivity was also observed in a few scattered basal cells (Fig. 2A). In specimens of patients who had preoperative combined androgen blockade, CD9 was predominantly expressed as diffuse cytoplasmic staining within the hyperplastic basal cell constituent of nonneoplastic glands. Intraluminal prostatic secretions had a strong and uniform immunoreactivity for CD9 (Fig. 2A). In agreement with the CD9 distribution in myelin of the central and peripheral nervous systems (34), prostatic peripheral nerves were strongly stained and served as an additional internal positive control for whole-mount sections. CD9 staining was not seen in the stroma.

In the adenocarcinomatous component, CD9 expression was characterized by diffuse cytoplasmic staining that was homogeneous (within one scale, 0-3) in the majority (88%) of specimen. In localized prostate cancers, 29% of the cases revealed uniform intensity without any changes, whereas differences in staining intensity (such as scales 1-2 or 2-3) occurred in 51% of the cases. The remaining 20% of the cases harbored more staining variation, ranging from scale 1 to 3 (Fig. 2B). There was no sample that crossed the staining cutoff (scales 0-1); thus, there was no discrepancy for the classification of CD9 status for any given specimen. A very similar staining spectrum was observed for bone specimen with metastases, in which 83% of the cases showed homogeneous staining. In these specimens, 43% had completely lost CD9 staining (scale 0), whereas an additional 22% had a weak staining (scale 1). Similarly, a fairly uniform staining was seen in 98% of advanced prostate cancers, of which 57% had no CD9 staining (scale 0), whereas 27% had weak staining (scale 1). Uniform staining was also detected in 95% of lymph node metastases, with an impressive 64% harboring no CD9 staining (scale 0) and 21% having weak CD9 staining (scale 1).

Eighty-two percent of the localized prostate cancers without hormone pretreatment (Table 1) still preserved CD9 expression with >50% of cells having positive immunoreactivity (Fig. 2B). Thus, only 18% of cases had reduced or negative CD9 expression (Fig. 2C). This percentage was not significantly different (P = 0.08) from the one (33%) obtained in cases with hormone pretreatment (Fig. 2D; Table 1). However, compared with localized prostate cancers, advanced prostate cancers had very significant differences in CD9 expression (85% with reduced or lost CD9 expression), independent of hormone pretreatment or not (P < 1 × 10⁻⁷ and P < 1 × 10⁻¹², respectively; Table 1). This finding was also true for lymph node (Fig. 2E) and bone (Fig. 2F) metastases, in which 85% and 65%, respectively, had reduced or lost CD9 expression. There was no statistical difference between advanced prostate cancers, lymph node metastases, and bone metastases (Table 1).

Inverse correlation of clinicopathologic variables with CD9 expression. In the 167 primary tumors with available clinicopathologic data, CD9 expression was inversely correlated with prostate-specific antigen level at diagnosis (P = 0.0058), tissue differentiation (by Gleason score; P = 0.0027), pathologic stage (P < 1 × 10⁻⁸), and metastasis status (P < 1 × 10⁻⁸; Table 2). Prostate-specific antigen level in CD9-reduced cases was much higher than in positive cases (mean, 64.4 and 16.0 ng/mL, respectively; P = 0.0058). As histologic differentiation was lost, more CD9-reduced cases were observed (8% for well-differentiated tumors, 40% for moderately differentiated tumors, and 55% for poorly differentiated tumors; P = 0.0027). As the clinical stage was more advanced, more CD9-reduced and/or negative cases were observed (25%, 36%, and 82% for stage T₂, T₃, and N+ or M+, respectively; P < 1 × 10⁻⁸). In addition, 82% (41 of 50) of prostate specimens obtained from patients with metastatic disease had reduced or lost CD9 expression compared with only 31% (36 of 117) for patients without metastasis (P < 1 × 10⁻⁸). However, when analyzed with Kaplan-Meier curve using WinSTAT from R.FINCH software, decreased or loss of CD9 immunostaining did not predict biochemical recurrence (as defined by a prostate-specific antigen value >0.1; P = 0.31) in our population of 107 patients, with a median follow-up of 62.7 months.

Mutations of CD9 cDNA in prostatic adenocarcinoma. To further understand the mechanisms implicated in the loss of CD9 protein expression during prostatic adenocarcinoma progression, we PCR amplified, cloned (up to 12 clones per amplification reaction), and sequenced its mRNA coding region. No alteration of the CD9 mRNA transcript was detected in 36 clones isolated from six normal prostate specimens. Although wild-type CD9 cDNA clones were obtained in all samples analyzed, clones derived from four PC-3–derived prostatic carcinoma cell lines, four specimens of prostatic adenocarcinoma, and one specimen of PIN also harbored one of three major deletions (Table 3). These deletions removed nucleotides 115 to 487, 190 to 585, and 120 to 619 of the 684 bp CD9 coding sequence. Thus, from the 228 amino acid CD9 protein, amino acids 39 to 163, 64 to 195, and 40 to 207 were eliminated by these deletions. These deletions affected the large extracellular and intracellular domains of the protein. The presence of the PC-3M-LN4 deletion (del 64-195) was confirmed on direct sequencing of the mRNA amplification product (without cloning). These deletions were not detected in genomic DNA derived from some of these samples, arguing for the existence of transcriptional CD9 mRNA modifications. Another deletion was detected in the DU145 cell line, whereas an in-frame insertion was present in mRNA derived from PC-3M-Pro4 (Table 3).

Furthermore, common missense point mutations were observed in one prostatic carcinoma cell line (PC-3M-LN4), one specimen of PIN, and seven specimens of prostatic adenocarcinoma (Table 3). Some specimens were harboring more than one missense mutation. Interestingly, CD9 protein expression was not detected in most of these cases (except in one specimen of prostatic adenocarcinoma). A bp substitution resulting in a new stop codon, located in the second cytoplasmic domain (amino acid 83), was also present in one PIN and in two prostate cancer patients where we did not detect the CD9 protein (Table 3). Overall, clones harboring CD9 mRNA modifications (either deletions or point mutations) represented 10% to 25% of the clones analyzed. Figure 3 summarizes the CD9 mRNA alterations and their predicted effect on the protein.

The present study indicates that, like in other human malignancies, loss of CD9 protein expression is associated with prostatic adenocarcinoma progression and metastasis. Indeed, our data showed that CD9 is well expressed in nonmetastatic disease but less expressed or absent in metastatic prostate cancer. In fact, the majority (76%) of our localized prostate cancers, 99% of which are nonmetastatic, were CD9 positive. In contrast, only 15% of our advanced prostate cancers, 82% of which are metastatic, still retained CD9 expression. The prostate-specific antigen level at diagnosis is significantly higher (P < 1 × 10⁻⁵) in advanced than localized prostate cancers. Advanced prostate cancers are also less differentiated (P < 0.001) and more advanced in pathologic stage (P < 10⁻²⁶). Therefore, the inverse correlation of CD9 with clinical pathologic findings is consistent with a putative role for CD9 at some critical point in the metastatic cascade.

There was no significant difference in CD9 expression when cases were stratified according to whether patients had received combined androgen blockade therapy or not (Table 1). CD9 protein was highly expressed in hyperplastic basal cells of nonneoplastic constituents of the prostate from patients with combined androgen blockade therapy. In the adenocarcinomatous component, no significant correlation between CD9 expression and the response rate to combined androgen blockade therapy (35) was detected. Our results contrast with the small induction (1.8-fold) of CD9 mRNA seen in LNCaP prostate cancer cells following 8 h of incubation with the synthetic androgen R1181 (36); the same authors also reported that prostate tissues from one patient treated with androgen ablation therapy did not show any CD9 staining in its luminal cells. As reported in Table 1, we also observed such reduced or absent CD9 staining in primary prostate tumors of patients undergoing combined androgen blockade. However, such treatment did not result in statistically significant difference in CD9 staining (P = 0.08) because some untreated patients with localized primary prostate cancer also had reduced CD9 expression. Although we cannot rule out the possibility that CD9 could be regulated by androgens, our results strongly suggest that such regulation is unlikely to influence disease progression.

The divergent distribution of CD9 expression in localized and advanced prostate cancer hinders the disease-free survival rate and prognosis analysis. There were 47 of 107 cases with cancer relapse in the localized prostate cancers. However, we could not find a statistical correlation with CD9 expression because 25 (53%) of them were CD9 positive. It should also be noted that many patients were lost on follow-up, as shown by the median follow-up of 62.7 months, although all prostatectomies were done before 1996. Conversely, the situation was similar in advanced prostate cancers because most cases (51 of 60) showed a significant reduction or loss of CD9 expression. Because they represent two totally different entities of the disease, we could not combine and analyze these two databases together. Indeed, only 43% of our patients with localized cancers relapsed compared with 100% of those with advanced cancers. It is also true for the mortality rate after 5 years of follow-up because only one patient (1%) died of localized prostate cancer, whereas 42% of patients died from advanced prostate cancer. Although reduced expression of CD9 protein has been associated with cancer progression in different tumor types, this is the first report implicating CD9 mRNA alterations in CD9 protein inactivation. Although wild-type CD9 mRNA was present in all samples analyzed, four deletions, one insertion, and four base substitutions were also detected in tumorigenic and precancerous (PIN) samples. No such alterations were detected in normal samples. The deletion (amino acids 23-59) found in DU145 cells and the 81 bp insertion seen in PC-3M-Pro4 cells are likely the result of alternative splicing. The DU145 deletion completely eliminates exon 2. Consequently, the new encoded protein will have a frameshift producing a stop codon (UGA), thus generating a truncated protein. For the PC-3M-Pro4 insertion, both splicing donor (CT) and acceptor (AG) sites in CD9 genomic DNA are localized at each end of the insertion. Moreover, when analyzed for their strength using a splice-site model accounting for pairwise dependencies (37), the new splicing sites bordering the 81 bp insertion have a maximum enthropy of 7.19 for 3′ splicing site and 6.64 for 5′ splicing site. These values are as good as others found on the boundary of other CD9 exons (maximum enthropy of 4.87 for 3′ splicing site of exon 7 or 6.66 for 3′ splicing site of exon 2), arguing that this region is likely to be inserted.

Three of four deletions were simultaneously found in CD9 cDNAs derived from primary tumor samples and in mRNA of highly tumorigenic and poorly differentiated prostatic carcinoma cell lines of the PC-3 lineage (Table 3). One deletion (amino acids 64-195) was found in four different samples, arguing against random PCR amplification errors. These three deleted regions could not have been generated by conventional alternative splicing due to absence of donor and acceptor splicing sites at the deletion boundaries. However, we noticed that nucleotides present at the beginning of the deleted region were the same as the ones present when the RNA sequence resumed. These repeats varied from three nucleotides (TCA) in deletion (120-619) to eight nucleotides (TCTTCAGACA) in deletion (115-487). The presence of such repeats could be implicated in the generation of the deleted mRNA molecules we have identified and could represent a novel mechanism for RNA alterations. These three deletions affect the large extracellular and intracellular domains of the protein. Such deletions are thus likely to interfere with CD9 functions because these domains are required for CD9 interaction with other proteins (11, 14, 15). Interestingly, all four prostatic adenocarcinomas containing these deletions did not express CD9 protein as shown by immunohistochemical analysis.

The large extracellular domain of CD9, and more precisely amino acids Gly¹⁵⁸, Val¹⁵⁹, and Thr¹⁷⁵, are the major residues responsible for enhancing the binding of diphtheria toxin to its receptor (38, 39). It is worthy to note that one of these residues was substituted (Thr¹⁷⁵Ala; A to G) in two different prostatic adenocarcinomas that had lost CD9 expression. In these two cases, the same major deletion (amino acids 64-195) was also detected in their derived cDNAs. An obvious effect of these mutations (deletions and point mutations), if these mRNAs are translated, will be to generate altered and/or inactivated CD9 protein. A similar situation was found in another CD9-negative tumor sample where a stop codon (Gln⁸³Stop), located in the cytoplasmic domain, and a deletion (amino acids 64-195) were detected. The mechanisms responsible for the generation of the deleted mRNAs are still unknown; however, base substitutions in RNA molecules have been reported in mammalian cells as a result of RNA editing (40). Although RNA editing has not yet been reported in prostate cancer cells, all bp substitutions detected in CD9 mRNA are compatible with such mechanism.

The presence of CD9 mRNA substitutions in prostatic adenocarcinomas suggests that the function of the CD9 protein may be hindered by such amino acid substitutions. Thus, CD9-positive staining seen in a high proportion (82%) of localized prostate cancer (and perhaps also in bone metastases, 35% being positive) should be interpreted with caution. It is indeed possible that in cases where CD9 base substitution occurs, the encoded protein, although still detectable with an antibody, is inactive and unable to adequately interact with other proteins. This could also explain why CD9 expression is not a good predictor of recurrence; indeed, such inactivation of CD9 will still be seen as CD9-positive staining in patients with recurrence. Alternatively, the fact that the epitope recognized by the CD9 antibody used for our immunohistochemical staining is located in the large extracellular loop of the protein will preclude CD9 detection if this domain is deleted in the mutated protein. This could explain why we could not detect CD9 in cases where we identified the three major deletions. In these cases, a truncated protein, undetectable with our antibody, could still be present inside the cells.

In conclusion, we have shown that CD9 protein expression is inversely correlated with prostatic adenocarcinoma progression and metastasis. As tumors become less differentiated and more advanced in stage, CD9 expression is greatly reduced and even lost. Moreover, although never detected in normal prostate samples, mRNA modifications could play a role in regulating CD9 function during prostate cancer progression. Combined with previous reports on CD9 expression in other tissues, these findings strongly suggest that CD9 would act as a tumor suppressor protein.

We thank France Landry for her technical assistance, Hassan Behlouli for his help in statistical analysis, Than-Vinh Dam for his assistance in image processing and the preparation of frozen prostate slides, Laurent Guy for the establishment of a localized prostate cancer (whole-mount slides) bank, Eltaher A. Omar for the establishment of an advanced prostate cancer bank, Hazem Ismael for his assistance in updating the clinical information of the prostate tissue bank, and Val Zvereff for helpful discussions and revision of the manuscript.