Characterization of Epithelial Senescence by Serial Analysis of Gene Expression: Identification of Genes Potentially Involved in Prostate Cancer¹

Mammalian somatic cells have a limited proliferative capacity when cultivated in vitro. For instance, human fibroblasts stop dividing after 50–70 population doublings and enter a terminal arrest state termed replicative senescence (1). Senescent fibroblasts are strongly enlarged and refractory to mitogen stimulation. However, they are metabolically active and survive in culture for several month. A similar limitation of proliferative capacity has been observed for most other cell types (2, 3). Replicative senescence is induced by progressive telomere shortening, which occurs during each cell division (4). Telomere erosion presumably generates a DNA damage signal, which leads to activation of p53 and subsequent transcriptional induction of the cdk³ inhibitor p21^CIP1. Therefore, prevention of telomere shortening by ectopic expression of the catalytic subunit of telomerase (hTERT) is sufficient to immortalize primary cells in vitro provided they are cultivated under the appropriate conditions.

Induction of a senescence-like phenotype also occurs after aberrant mitogenic signaling and after environmental and genotoxic insults. This form of senescence has been termed cellular senescence as opposed to telomere-associated replicative senescence (5). Similar to apoptosis, cellular senescence is thought to be a mechanism of tumor suppression because it prevents the outgrowth of cells that have acquired mutations in genes rendering them cancerous (6). Consistent with this model, several tumor suppressor genes (e.g., p16) or their products (p53) are activated at the onset of cellular senescence. In addition, mice engineered to display elevated p53 activity show premature aging and a drastically decreased incidence of cancer, supporting a role of cellular senescence as a tumor suppressive mechanism relevant for the whole organism (7).

Recently, it has been shown that mammary epithelial cells have the capacity to spontaneously escape replicative senescence and enter a phase of genomic instability, which may give rise to immortal cells (8). According to calculations by Morris (9), a similar evasion of replicative senescence has to occur for the development of any epithelial cancer.

Complicating the issue, the presence of senescent fibroblasts promotes the proliferation of premalignant and malignant but not normal epithelial cells presumably by generating an altered microenvironment (10). Therefore, Krtolica et al. (10) suggested that senescence may promote carcinogenesis in aged organisms while it protects against cancer early in life.

Prostatic cancer is the most frequent malignancy in the United States and the second leading cause of cancer deaths in men today (11, 12, 13, 14). Among a variety of environmental and genetic factors favoring the development of prostatic cancer, aging is the most significant risk factor. It has been estimated that 15–30% of males over the age of 50 and as many as 80% of the males over the age of 80 harbor clinically undetected foci of prostate cancer (15). On the basis of the in vivo expression of pH 6.0 specific β-galactosidase, a marker of cellular senescence (16), it has been suggested that the accumulation of senescent prostate epithelial cells within prostatic glands might play a role in the development of prostatic diseases (17).

The characterization of senescence in epithelial cells is still in its beginning. However, a detailed characterization of senescence in epithelial cells is necessary to understand how carcinoma circumvent this program. This approach may allow to identify genes involved in the development of prostate cancer, a disease for which relatively few causal genetic events are known. Furthermore, changes in gene expression during senescence of PrECs may provide insights into the aging mechanisms of the prostate. To characterize genome-wide expression during senescence of PrECs, we used SAGE, a quantitative method developed by Velculescu et al. (18). Here, we describe differentially expressed genes identified by SAGE, which presumably represent components of pathways and mechanisms involved in the induction and maintenance of senescence. Genetic inactivation or deregulation of these genes may lead to immortalization and neoplastic transformation of PrECs.

PrECs used for SAGE were derived from a 17-year-old accident victim (Clonetics, San Diego, CA). PrECs were cultivated in PrEC growth medium (Clonetics) on collagen type I vented flasks (BioCoat; BD Falcon, Bedford, MA) according to the supplier’s instructions. PrECs were passaged at ∼70% confluence by splitting 1:3 using collagenase 1S (Sigma, Deisenhofen, Germany).

For qPCR analysis, additional PrEC samples were obtained from two prostate cancer patients (patient 1: 56 years old; patient 2: 63 years old). After radical prostatectomy, tissue wedges free of malignant cells were removed from the transition zone. These explants were minced into organoids of 1 mm³ and seeded on collagen I-coated plates in PrEC growth medium, allowing a homogeneous epithelial cell population to grow out. These cytokeratin-positive cells were passaged until senescence.

Antibodies specific for p21 (clone: 6B6) were obtained from BD PharMingen (Bedford, MA). p16- (clone: C-20), p15- (clone: C-20), p53- (clone: Pab 1801), p27- (clone: C-19), or α-tubulin- (clone: TU-02) specific antibodies were obtained from Santa Cruz Biotechnology (Santa Cruz, CA). Enhanced chemiluminescence signals were detected on a Image Station 440CF (Kodak-Perkin-Elmer, Boston, MA).

MicroSAGE was performed according to a protocol (version 1.0e) accessible online.⁴ We included additional purification steps of DNA intermediates after each PAGE step using Sephadex G-25 columns to ensure complete removal of contaminants. In brief, mRNA was isolated from human PrECs with the Dynabeads mRNA Direct Kit (Dynal, Smestad, Norway), and cDNA was synthesized on magnetic particles using the Superscript Choise System (Invitrogen, Groningen, the Netherlands). After cleaving the cDNA with NlaIII, linkers containing recognition sites for BsmFI were ligated to the cDNA. Linker tags were released by BsmFI digestion from the magnetic particles, ligated, and a 102-bp fragment was amplified with biotinylated primers. Ditags (∼26 bp) were released by NlaIII cleavage. Biotinylated linkers were completely removed using streptavidin-linked magnetic beads and subsequent PAGE purification. The ditags were concatenated, and concatemers of 500–800 bp were subcloned into pZERO (Invitrogen). After colony PCR, products >500 bp were sequenced using BigDye-terminator V2.0 reagents (Applied Biosystems, Lincoln, CA). Products were purified using Sephadex G-50 in filter plates (Millipore, Bedford, MA) and analyzed on an automated capillary DNA Sequencer (3700; Applied Biosystems). The results were analyzed using the SAGE2000 software provided by Dr. Ken Kinzler (Johns Hopkins University Medical School, Baltimore, MD). To exclude tags generated by sequencing errors, only tags that occurred at least twice were included in the analysis. After statistical analysis using Monte Carlo simulations, SAGE tags (P < 0.05; differential regulation >2.5-fold) were assigned to cDNAs using the Unigene database (release 03/01). Tags with multiple matches were additionally analyzed after retrieving the 11^th base of the tag. Matching and position of all tags listed in Tables 1 and 2 were confirmed using the tag-to-gene-mapper function provided online.⁵

qPCR analysis was performed as described in detail by Menssen and Hermeking (19). The Light Cycler, software version 3.5.2, FastStart DNA Master SYBR Green I, and cDNA synthesis reagents were used according to the manufacturer’s instructions (Roche Applied Science, Mannheim, Germany). A total of 1 μg of RNA was reverse transcribed into cDNA. PCR efficiency of different primer pairs was determined using logarithmic dilutions of cDNA templates. After determining the slope of the reaction over a range of 20–32 cycles, PCR primer efficiency was calculated according to the equation: E = 10^−1/slope. Specificity of PCR products was confirmed by melting curve analysis, gel analysis, and direct sequencing (DNA Analyzer 310; Applied Biosystems). Primer sequences and examples of qPCR raw data are available as supplemental material. qPCR determinations were normalized with primers specific for eukaryotic elongation factor 1 α-1, which was equally represented in both SAGE libraries (316:369 tags). The difference in gene expression was calculated incorporating the efficiency (E) of each primer pair according to Pfaffl (20):

Human PrECs derived from a 17-year-old accident victim were cultivated until they ceased to proliferate (10. passage, ∼30 population doublings). During serial cultivation, the frequency of cells showing markers of senescence as cellular enlargement (Fig. 1,a) and positive staining for β-galactosidase at pH 6.0 (data not shown) increased to >90% of the population. The protein levels of the cdk inhibitor p16^INK4A increased during senescence of PrECs (Fig. 1,b). Furthermore, a minor increase in the protein levels of the cdk inhibitor p15^INK4B could be detected (Fig. 1,b). However, no significant changes in the protein levels of p53, its transcriptional target, the cdk inhibitor p21^CIP1, and the cdk inhibitor p27^KIP1 could be detected as PrECs became senescent (Fig. 1,b). qPCR allowed to detect induction of p16^INK4A mRNA in senescent PrECs (Fig. 1,c). Expression of p15^INK4B mRNA was induced in senescent PrECs, whereas p21^CIP1 and p27^KIP1 mRNA levels were not significantly altered (Fig. 1,c). The qPCR results were in accordance with the Western blot analysis shown in Fig. 1,b. Connective tissue growth factor, a gene previously found to be induced during senescence of PrECs (21), was induced as detected by qPCR (Fig. 1 c). Induction of p16^INK4A and unchanged p21^CIP1 expression in senescent PrECs has been reported previously (22, 23). The lack of p53/p21^CIP1 activation suggests that the cessation of proliferation observed in senescent PrECs was not because of shortening of telomeres and a subsequent DNA damage-mediated cell cycle arrest.

To acquire a comprehensive, unbiased picture of changes in gene expression during senescence, SAGE was used. With SAGE, sequence tags of 10–11 bp from the 3′-end of each transcript are isolated, concatenated, and sequenced to generate so-called SAGE libraries (18). The abundance of a specific tag in a SAGE library is proportional to the expression level of its corresponding mRNA. Because of the limited number of early passage PrECs, a protocol adapted to small amounts of mRNA was used (MicroSAGE, for details see “Materials and Methods”). A SAGE library of 72,068 tags was generated from subconfluent, terminally arrested, senescent PrECs at passage 10 and compared with a library of 72,069 tags derived from PrECs arrested because of confluence at passage 3. Both cell populations were cultivated in the presence of growth factors. A comparison of senescent to exponentially proliferating, early passage PrECs was avoided because many genes would have been differentially regulated because of the drastically different growth state and cell cycle distribution of arrested versus proliferating cells. The 144,137 SAGE tags collected in total correspond to 25,645 unique mRNA species. Using statistical analysis by Monte Carlo simulation, we determined that 273 tags showed significant differential expression (P < 0.05; fold difference >2.5). The complete set of SAGE data can be accessed online.⁶ This web site allows the analysis and comparison of the differentially expressed tags/genes identified here with numerous other SAGE studies (24).

To estimate the accuracy of the MicroSAGE analysis and to determine whether differential regulation of the identified transcripts occurs generally during senescence, qPCR was used to determine the abundance of exemplary transcripts in PrECs from three different donors (Fig. 2): differential regulation during senescence as detected by MicroSAGE was confirmed for all genes tested and is discussed in detail below. Therefore, the MicroSAGE results accurately reflect the levels of gene expression in the two cell populations analyzed. Furthermore, the changes in gene expression observed by SAGE in the senescent PrECs from one donor were also observed in senescent PrECs from two additional donors (Fig. 2). These results suggest that most of the changes in gene expression detected by SAGE in this study generally occur in senescent human PrECs.

Differentially expressed SAGE tags were matched to the cDNAs of the Unigene database (release 03/01). Those transcripts, which corresponded to known genes and unambiguously contained the SAGE tag next to the most 3′ NlaIII-site, were sorted according to their function (Tables 1 and 2). Seventy genes induced in senescent PrECs are listed in Table 1, whereas 65 genes repressed during senescence are depicted in Table 2. Of the 157 tags significantly induced during senescence, 87 tags matched to functionally uncharacterized transcripts or had no matches in the database. Among the 116 repressed tags were 51 tags that had no functional assignment or matches in the database.

Because we reasoned that transcripts up-regulated during senescence may be targets for down-regulation during tumor progression (and vice versa), the expression data were compared with previously published studies or public SAGE data analyzing differential gene expression in prostate cancer (Refs. 25, 26; Tables 1 and 2, right column). For a number of genes, e.g., DKK3, the proposed correlation could be confirmed. These interesting cases are discussed below.

We also compared the SAGE results obtained here with other profiling studies on senescence performed with PrECs (27), human diploid fibroblasts (28, 29), and muscle cells (30). Indeed, we identified several examples where similar senescence-specific gene expression could be found (Tables 1 and 2, right column): e.g., PAI-1 appears as a gene universally up-regulated during senescence in different cell types (Table 1). However, most of the changes detected here are specific for PrECs (Tables 1 and 2). The tag for β-galactosidase (TTACTTTTTT, Hs. 79222) was not significantly differentially regulated (8:13 tags), which is consistent with the hypothesis that an increase in lysosomal mass is responsible for the increase in β-galactosidase activity observed in senescent cells.

Irreversible cell cycle arrest in the presence of otherwise mitogenic growth factors is a hallmark of senescence. The changes in gene expression we detected by MicroSAGE suggest the involvement of several key regulators in the establishment and maintenance of cell cycle arrest in senescent PrECs (Tables 1 and 2). It has been shown previously that ectopic expression of the helix-loop-helix factor Id1 is able to reactivate the cell cycle in senescent human fibroblasts (31), presumably by inhibitory association with ets transcription factors, which are required for elevated expression of p16^INK4A (32). A similar mechanism may be operating in PrECs: in early passage PrECs, Id1 was expressed at high levels (Table 2). However, during senescence, Id1 decreased dramatically, which may explain the induction of p16^INK4A mRNA and protein (Fig. 1, b and c). Elevated p16^INK4A may then lead to a reduction of G₁ phase-specific cdk activity and hypophosphorylation of pRb. Active pRb binds and inactivates members of the mitogenic E2F transcription factor family, which subsequently leads to inhibition of G₁-S cell cycle progression (reviewed in Ref. 33). In addition to this mode of E2F inactivation, we observed a decreased expression of E2F4 (Table 2), which may additionally contribute to the inability of senescent PrECs to traverse the G₁-S phase. E2F4 is expressed at elevated levels in immortal prostate cancer cells (25), suggesting that E2F4 may be involved in immortalization of PrECs. The reduced levels of CKS-1 expression in senescent PrECs may also contribute to the permanent cell cycle arrest observed in senescent PrECs (Fig. 2,B). CKS-1 knockout mice have a profound defect in cell proliferation, suggesting that CKS-1 is necessary for full activity of Cdk2 (34). CKS proteins affect cdk activity by directly binding to cdk complexes and facilitating ubiquitin-mediated proteolysis of associated inhibitors like p27^KIP1 (34). However, the levels of p27^KIP1 protein do not increase significantly during senescence of PrECs (Fig. 1,b). Therefore, it is likely that CKS-1 targets other proteins for degradation in early passage PrECs. Down-regulation of cyclin B1 expression was confirmed by qPCR (Fig. 2 B) and may lead to a cell cycle arrest in the G₂ phase. Consistent with this observation, cell cycle arrest of senescent cells is not restricted to arrest in the G₁ phase but also occurs in the G₂ phase (28). Paradoxically, expression of cyclin D1 was increased in terminally arrested prostate cells. Senescent human fibroblasts show a similar increase in cyclin D1 mRNA and protein levels (35).⁷ This may constitute a compensatory up-regulation, which results from inhibition of the cdk4/cyclin D1 pathway by p16^INK4A.

Elevated levels of enzymes involved in remodeling of the ECM has been observed previously during senescence of fibroblasts (28, 29). The deregulation of these genes during senescence may contribute to the altered ECM observed in aged tissues. In senescent PrECs, elevated levels of matrix metalloproteinase MMP-14 and cathepsin B expression were detected by SAGE (Table 1). On the other hand, PAI-1, an inhibitor of a matrix-degrading protease, was induced significantly (confirmed by qPCR, Fig. 2,A). Up-regulation of PAI-1 has also been observed in other cell types undergoing senescence and presumably leads to disruption of ECM maintenance (36). Expression of the gene encoding the adhesion molecule fibronectin 1, which contains multiple binding sites for diverse ECM and cell surface molecules, was increased in senescent PrECs (confirmed by qPCR, Fig. 2,A). Consistent with an antiproliferative role of fibronectin, its expression is generally reduced in transformed cells (37). Senescent PrECs showed increased expression of β4 integrin, a transmembrane receptor, which mediates cell-matrix interactions (confirmed by qPCR, Fig. 2,A, Table 1). Down-regulation of β4 integrin is characteristic for prostate cancer and prostatic intraepithelial neoplasia (38).

Senescent PrECs undergo dramatic changes in size and shape (Fig. 1,a). These changes could be because of the up-regulation of several key regulators and components of the cytoskeleton (Table 1): e.g., the gene encoding the intermediate filament forming protein vimentin was induced in senescent PrECs (Table 1). Senescent human fibroblasts also show elevated expression of vimentin (39). In addition, the elevated expression of gelsolin (confirmed by qPCR, Fig. 2,A), which fragments actin networks in a calcium-regulated manner, may be involved in the altered morphology of senescent cells. Interestingly, levels of gelsolin are diminished in breast cancer (40) and ectopic expression of gelsolin suppresses tumorigenicity (41). Expression of intermediate chain I of cytoplasmatic dynein (DNClI) was increased in senescent PrECs. Among several components of cytoplasmic dynein, up-regulation during senescence was shown to be unique for DNClI (42). The expression of tropomyosin 1-α (confirmed by qPCR, Fig. 2,A) and fibulin-1 was increased in senescent PrECs. Interestingly, down-regulation of human epithelial tropomyosin has been observed in prostate carcinoma cells (43), and ectopic fibulin-1 expression inhibits motility and invasion of human ovarian and breast cancer cells (44). RAC1 and cdc42, which both encode GTP-binding, ras-like molecules, were induced in senescent PrECs and have been implicated in the reorganization of actin filaments during wound healing processes in fibroblasts: cdc42 expression is sufficient to induce filopodia, whereas RAC1 is required for lamellipodia formation (45). The induction of both genes is presumably involved in the characteristic spreading of senescent PrECs (Fig. 1 a).

As discussed above, transcription factors like Id1 and E2F4 may have central roles in regulating or antagonizing cellular events, which are part of the senescence program. The transcription factor EGR1 is the product of an immediate early growth-response gene and directly induces TGF-β expression (46). Consistent with our finding that EGR1 is repressed in senescent PrECs, expression of TGF-β is diminished concomitantly (Table 2). The transcriptional repressor ATF4 (CREB2) was repressed, suggesting that genes regulated via cyclic AMP-response elements may be derepressed during senescence of PrECs. p8, which encodes a basic-helix-loop-helix transcription factor, was significantly repressed in senescent PrECs. Interestingly, mitogenic and metastatic potential has been assigned to p8 (47, 48). SKI-interacting protein (49), which was induced in senescent PrECs, functions as an antagonist of the oncogene product SKI, thereby presumably contributing to the terminal arrest of PrECs.

Senescent PrECs are refractory to stimulation of proliferation by external growth factors, implying that repression of receptors or mediators of signaling events should be detectable. However, of the detected changes in mRNA levels only the repression of PDGFα fulfills this criterion. On the other hand, induction of negative regulators of signaling could lead to the unresponsiveness of senescent cells to mitogens: we observed induction of DKK3 (confirmed by qPCR, Fig. 2 A), which presumably represents an antagonist of wnt-signaling. Up-regulation of DKK3 during senescence was also observed in human diploid fibroblasts.⁷ Ectopic expression of DKK3 inhibits tumor cell proliferation and expression of DKK3 is significantly down-regulated in non-small cell lung carcinomas (50). Interestingly, DKK3 is localized on 11p15, a locus often deleted in human cancer (50).

IFNs are capable of generating a variety of cellular responses, e.g., cell cycle arrest, thereby having antitumor and antiviral effects. Senescent PrECs displayed elevated levels of IFN regulatory factor 3 mRNA. IFN regulatory factor 3 transactivates IFN-responsive genes through sequence specific binding of IFN response elements (51). IFN-α-inducible protein IFI-6-16 was dramatically increased in senescent PrECs as determined by MicroSAGE and qPCR analysis (Fig. 2 A). Furthermore, IFN-induced transmembrane protein 2 (1–8D) was induced in senescent PrECs. Although the function of these genes is unknown, their induction could be involved in the dominant cell cycle arrest observed in senescent PrECs.

Senescent PrECs showed increased connexin 26 expression. Connexin 26 protein forms intercellular channels present in gap junctions, which allow the transfer of ions and small signaling molecules between basal and luminal cells of the human prostate (52). Consistent with a role of connexin 26 in regulation of cell proliferation and differentiation, prostate cancer cell growth can be suppressed by ectopic expression of connexin 26 (53).

Senescent cells acquire an increased resistance towards apoptotic insults (54, 55). In senescent PrECs, this may, in part, be because of the down-regulation of the proapoptotic gene Bad detected in this study (Table 2). On the other hand, we observed induction of TRAIL (APO2L, confirmed by qPCR, Fig. 2), which binds to the TRAIL receptors TRAILR1/DR4 and TRAILR2/DR5 (56). TRAIL induces apoptosis in prostate cancer cells but also in normal PrECs (57). It will be interesting to determine whether increased TRAIL expression, which may occur in the prostate because of accumulation of senescent PrECs (17), contributes to suppression of tumor formation in the aging prostate.

In the future, it will be important to analyze whether the differential regulation of genes identified in this study is required for the induction or maintenance of senescence in PrECs. During neoplastic transformation, genes required for the senescent phenotype may be inactivated through genetic (mutation, deletion) or epigenetic alterations (e.g., methylation). Furthermore, transcriptional repression of senescence-inducing genes may occur. Therefore, genome-wide analyses of changes in gene expression patterns and of genetic alterations, which occur during formation of prostate cancer, will be complementary to this study. Recent examples of gene expressing studies on prostate cancer cells (25, 58, 59, 60) include a report by Shou et al. (60), which shows that down-regulation of several IFN-regulated genes is characteristic for the transition from nontumorigenic benign prostatic hyperplasia to tumorigenic prostatic hyperplasia: one of these genes is IFI-6-16, which is strongly induced during senescence (Table 1, Fig. 2). These examples suggest that genes induced during senescence are good candidates for genes, which are inactivated/down-regulated during cancer initiation and/or progression. On the other hand, genes down-regulated during cellular senescence may represent potential therapeutic targets for inhibition of prostate cancer cell proliferation because specific inhibition of such gene products may lead to reactivation of the senescence program in immortal cancer cells.

We thank Peter Palm for help with, and Dieter Oesterhelt for access to, automated sequencing, and Holger Rumpold and members of the lab for discussion and comments. Heiko Hermeking’s laboratory is supported by the Max-Planck-Society and the Deutsche Krebshilfe.