Deregulation of cyclins D1 and E and suppression of cdk2 and cdk4 in senescent human fibroblasts

After a restricted number or cell divisions human diploid fibroblasts enter a state of irreversible G₁ arrest, usually referred to as cellular senescence (Hayflick and Moorhead, 1961; Bell et al., 1978; Goldstein, 1990; McCormick and Campisi, 1991). Senescent cells differ from ‘younger’ cells in a number of parameters, the most prominent hallmarks being their dramatically increased size and their inability to initiate DNA synthesis in response to growth factors. Although there is compelling evidence that senescence is neither controlled by a timer measuring the age of the cell nor by a counter recording the number of divisions, the pre-cise mechanisms controlling or inducing the process are not understood. It has been speculated that the entry into senes-cence may be determined by a ‘stochastic’ event, that might arise, for instance, as a consequence of an uneven distrib-ution of cellular components after mitosis with the ensuing differences in the length of G₁. At present, however, basi-cally no data are available to support or contest this hypoth-esis. It has also been postulated that senescence reflects a state of terminal differentiation (Bell et al., 1978). This hypothesis is, however, difficult to test because most studies concerning cellular senescence have been performed with fibroblasts where molecular markers of terminal differen-tiation are poorly defined.

Like the basic biological concepts briefly discussed above, the molecular mechanisms underlying or accompa-nying the senescence process are also not understood. Sev-eral studies performed in the past few years, however, have brought some progress along these lines. A potentially inter-esting observation in this context is the reported decrease of c-fos expression and AP-1 binding activity in senescent human fibroblasts after serum stimulation of growth factor-deprived cells (Seshadri and Campisi, 1990; Riabowol et al., 1992), although these changes are not consistently seen in senescent cells and therefore are presumably not instru-mental (Lucibello et al., 1993). On the other hand, the observed block in the expression of the cell cycle genes cdc2, cyclin A and cyclin B (Stein et al., 1991) is likely to be a consequence of the G₁ arrest rather than its cause, since the products of these genes are believed to act in the second half of the cell cycle (Pines and Hunter, 1990a,b; Walker and Maller, 1991; Girard et al., 1991; Pagano et al., 1992a), i.e. well after the point of G₁ arrest. Of greater significance in this respect is probably the failure of senescent cells to phosphorylate the pRB protein (Stein et al., 1990; Fureal and Barrett, 1991), an event that in actively dividing cells leads to the inactivation of pRB, thereby enabling progres-sion into S-phase (Hinds et al., 1992). In senescent cells this mechanism seems to be impaired, with the consequence of holding pRB in the underphosphorylated inhibitory form.

As mentioned above, it seems conceivable that alteration in the expression of genes acting in G₁ is likely to be part of the mechanism driving cells into replicative senescence. We therefore decided to analyse the expression and inducibility of two recently described cyclins whose func-tions are likely to be associated with G₁→S progression, i.e. cyclins D1 and E (Koff et al., 1991; Lew et al., 1991; Matsushime et al., 1991; Xiong et al., 1991; Dulic et al., 1992; Koff et al., 1992; reviewed in refs, see Hunter and Pines, 1991 and Lew and Reed, 1992). Studies analysing the expression of these genes and/or the activity of their associated protein kinase during the cell cycle have suggested that cyclin E might be active at the time of G₁→S-phase transition (Lew et al., 1991; Dulic et al., 1992; Koff et al., 1992), while cyclin D1 may have a function earlier in G₁ (Matsushime et al., 1991). In addition, we analysed the cdc2-related genes cdk2, cdk4 and cdk5. The cdk2 gene product is thought to act at a similar time to cyclin E, owing to its expression kinetics after serum stim-ulation and its occurrence in complexes with cyclin E in vitro and in vivo (Tsai et al., 1991; Dulic et al., 1992; Faha et al., 1992; Koff et al., 1992; Meyerson et al., 1992; Pagano et al., 1992b; Rosenblatt et al., 1992). The other two cdc2-related proteins, cdk4 (PSK-J3; Meyerson et al., 1992) and cdk5 (PSSALRE; Meyerson et al., 1992), have been shown to associate with and stimulate the kinase activity of cyclin D proteins in vitro (Matsushime et al., 1992; Xiong et al., 1992), but their role in vivo is not clear. While some of our results are not unexpected, i.e. the suppression of cdk2 and cdk4, two other findings came as a surprise: cyclins D1 and E were found to be overexpressed or deregulated in senes-cent cells. Our observations may point to the existence of regulatory loops controlling the expression of cell cycle genes in the G₁ phase of the cell cycle that are not func-tional in senescent cells.

WI-38 cells (ATCC, Rockville, Maryland; obtained at passage 14 corresponding to 19 population doublings; Hayflick, 1965) were cultured in Dulbecco-Vogt modified Eagle’s minimum essential medium (DMEM) supplemented with 10% fetal calf serum (FCS), 0.5% glucose, penicillin (100 i.u./ml) and streptomycin (100 mg/ml). The cells were passaged at a 1:5 ratio.

RNA was isolated according to the method of Chirgwin et al. (1979) and analysed by northern blotting as described (Thomas, 1980). The human cyclin B probe was kindly provided by T. Hunter and J. Pines (Pines and Hunter, 1989). All other cyclin probes were cloned PCR products, in each case comprising the complete coding region (Wang et al., 1990; Koff et al., 1991; Lew et al., 1991; Xiong et al., 1991).

cDNA synthesis was performed with 2–4 μg of total RNA in a reaction volume of 40 μl containing 1.25 μg of oligonucleotide 3′-primer and 1.25 units Taq DNA polymerase (Boehringer, Mannheim, Germany; Saiki et al., 1988; Mumberg et al., 1991). The annealing temperature was 60°C for all primer pairs, the number of PCR cycles used was 20-28 for cyclin and cdk genes, and 16 for GAPDH. The oligonucleotide primer pairs had the fol-lowing sequences (numbers refer to nucleotide positions in the respective sequences quoted; some primers contain, in addition, restriction sites at their 5′-end):

A 10 μl sample of the PCR reaction mixture was separated on a 5% polyacrylamide gel, which was then dried and exposed to Kodak X-ray film overnight. Quantitation was performed by 1-scanning of the gels using a Molecular Dynamics PhosphorIm-ager. Data were corrected using the GAPDH signal as the stan-dard.

The rabbit anti-cyclin D1 (α-CycD1) serum has previously been described (Sewing et al., 1993). Polyclonal antisera against human cyclin E (α-CycE) and human cdk2 (α-Cdk2) were generated by immunising rabbits with the respective bacterially expressed GST fusion proteins as described (Sewing et al., 1993). For immuno-precipitations, 4×105 cells in a 25 cm2 dish were washed twice in PBS and incubated in methionine-free DMEM with 0.2 mCi of [35S]methionine (Amersham, approx. 1400 Ci/mmol) for 60 min. Cells were then washed twice in chilled PBS and lysed by the addition of 0.8 ml RIPA-buffer (1% Triton X-100, 1% sodium deoxycholate, 0.1% sodium dodecyl sulfate (SDS), 0.15 M NaCl and 0.01 M Tris, pH 7.4). Lysates were clarified by centrifuga-tion at 45 K for 30 min at 4°C. Samples were incubated with 8 μl of α-CycD1, α-CycE or α-Cdk2 serum overnight at 4°C, followed by incubation for another 60 min with 40 μl Protein A-Sepharose CL-4B (Pharmacia). Immune complexes were cen-trifuged, pellets were washed five times in RIPA-buffer, resus-pended, boiled in SDS sample buffer and analysed on discontin-uous 12.5% SDS-polyacrylamide slab gels followed by fluorography. Cyclin E-associated kinase activity in immunopre-cipitates was measured using purified histone H1 as the substrate. Immunoprecipitates were washed twice in kinase buffer (50 mM Tris-HCl, pH 7.4, 1 mM DTT, 25 mM MgCl₂). Reactions were carried out in the same buffer in the presence of 0.5 mg histone H1/ml, 0.1 mM ATP and 1 μCi [y-32P]ATP at 30°C for 30 min and stopped by adding SDS sample buffer.

Cells on coverslips were fixed with 3.7% formaldehyde in PBS. Fixed cells were washed twice with cold PBS and permeabilised for 5 min with 0.5% Triton X-100 in PBS. Coverslips were exposed to a 1:200 dilution of α-CycD1 antibodies (Sewing et al., 1993) in PBS/0.5% BSA/0.5% Tween-20 for 1 h at ambient tem-perature. Cells were washed in PBS/0.5% Tween-20 and incu-bated with a 1:500 dilution of α-Cy3-conjugated anti-rabbit anti-body in PBS/0.5% Tween-20 (Dianova, Germany). Counterstaining of DNA was carried out for 5 min with Hoechst 33258 (2 mg/ml in PBS). The cells were observed with a Leitz Diavert microscope and photographed using a Kodak TMX 400 film.

Labelling was accomplished by incubating cells growing on cov-erslips for 1 h in a medium containing 50 mM 5-bromodeoxyuri-dine (BrdU). After washing twice with phosphate buffered saline (PBS) the cells were fixed in ice-cold 75% ethanol overnight at 4°C. The fixed cells were treated with 2 M hydrochloric acid/0.5% Triton X-100 for 30 min to denature the DNA. After washing twice with PBS, coverslips were exposed to a 1:100 dilution of the α-BrdU antibody (Partec, Switzerland) in PBS/0.5% BSA/0.5% Tween-20 for at least 1 h at 37°C. Cells were washed with PBS and incubated with a 1:200 dilution of α-Cy3-conju-gated anti-mouse antibody (Dianova, Germany) in the same buffer as before for at least 30 min. Counterstaining of DNA was car-ried out for 5 min with Hoechst 33258 (2 mg/ml in PBS).

The fibroblast line WI-38 (Hayflick, 1965), obtained from the American Type Culture Collection (ATCC) at passage level 14, was used in the present study. Under our culture conditions, WI-38 cells underwent approximately 32 serial passages (see Fig. 1), corresponding to approximately 60 population doublings, before entering senescence. When continuously exposed to 10% FCS, 34.4% of the cells at passage level 20 showed incorporation of 5-bromo-deoxyuridine (BrdU) during a 1 h labelling period, which means that more than 90% of the cells were in the cell cycle (the S-phase represents approximately 1/3 of the WI-38 cell cycle). The number of BrdU-positive cells decreased to 14.6% at passage 24, 10.9% at passage 29 and less than 0.5% at passage 32. At this stage the cells also dramati-cally increased in size (visible in Fig. 9), as expected for cells entering the senescent state. We also analysed the abil-ity of the cells to leave the quiescent state upon serum stim-ulation, with very similar results. Between 20 and 30% of the cells incorporated BrdU at passage levels 20–29 (measured 17–18 h post-stimulation), but only 1.2% were able to do so after 32 passages.

RNA was isolated from WI-38 cells either continuously exposed to 10% FCS (lanes ‘FCS’) or after FCS stimula-tion of serum-deprived, quiescent cells at passages 20, 24, 29 and 32, and analysed by northern blotting. This exper-iment was carried out in parallel to the BrdU incorporation analysis (Fig. 1) so that the results of both experiments are directly comparable. In ‘young’ cells (passage 20 and 24), the expression of cyclins A, B and D1 was markedly reduced in the quiescent state, and was inducible upon stim-ulation with fetal calf serum. Cyclin A was induced approx. 14 h after serum stimulation, i.e. around the onset of DNA synthesis (see FACS analysis in Fig. 2b), but continued to increase at later time points (Fig. 2a and data not shown). Induction of cyclin A was followed by the G₂/M-specific cyclin B (Fig. 2a and data not shown). In contrast, induc-tion of cyclin D1 occurred early during the G₀→S transi-tion, i.e. approximately 4 h post-stimulation (Fig. 2a). Cyclin E induction followed that of cyclin D1, but reached maximum levels approx. 14 h after serum stimulation, which is before cyclin A expression was at its highest. How-ever, cyclin E mRNA levels could not be accurately quan-titated by northern blotting, due to their very low abun-dance. The order of induction of the cyclin mRNAs seen in several independent experiments (Fig. 2a and data not shown) and confirmed by the quantitative PCR study described later in this study was: cyclin D1 (maximum levels reached in the middle of the G₀→S transition), fol-lowed by cyclin E (late G₁), cyclin A (S-phase) and finally cyclin B (S/G₂).

The expression of the four cyclin genes in asynchronous cells and their inducibility in growth factor-deprived cells was also analysed in cells at later passage levels (Fig. 2a, P29 and P32). Fig. 2a shows that the expression of glyc-eraldehyde phosphate dehydrogenase (GAPDH) RNA did not significantly fluctuate in all samples analysed, apart from an approximately 2-fold higher level in the senescent cells. In addition, the expression of L7 (large ribosomal subunit protein L7) RNA, a known marker of cellular senescence (Seshadri and Campisi, 1990), decreased to approximately 20% in cells at passages 29 and 32 relative to ‘younger’ cell (data not shown). As previously reported by others (Stein et al., 1991), both cyclin A and cyclin B decreased to undetectable levels in the senescent cells, both in cells continuously exposed to serum and in stimulated cells. Interestingly, serum-induction of both mRNAs in quiescent cells at passage level 29 was also greatly dimin-ished, although only a minor fraction of these cells was fully senescent (see Fig. 1). This may indicate that changes in the regulation of cyclin A and B expression might have occurred already in presenescent cells. Surprisingly, the two other cyclins analysed showed the opposite behaviour. The abundance of cyclin D1 mRNA was greatly increased at passage level 32 in both asynchronous and serum-stim-ulated cells, and again these changes were already detectable at passage 29. Likewise, the induction by fetal calf serum of cyclin E was clearly increased in the senes-cent cells, both in cells continuously exposed to serum and in deprived cells, including the early stages of serum stim-ulation.

To confirm and extend the results obtained by RNA blot-ting we measured the abundance of cyclin RNAs in growth factor-deprived and in stimulated cells at different passage levels by quantitative PCR. This was necessary in particu-lar for cyclin E because of its very low expression in WI-38 cells. cDNA was synthesised from each RNA and amplified in the presence of ³²P-labelled dATP, so that a low number of cycles was sufficient to detect the amplified products (Mumberg et al., 1991). This way the only rate-limiting compound was the input cDNA, thus allowing for quantitation of the results by β-radiation scanning. The data were then normalised to the signal obtained with the GAPDH primers to compensate for differences in efficiency of the cDNA synthesis. The results are shown in Fig. 3 and a quantitation of the data obtained for cells at passage levels 20 and 32 is displayed in Fig. 4. It is evident that the results obtained by quantitative PCR measurements are in good agreement with the RNA blotting data. The quantitation in Fig. 4 shows that the 10-fold induction of cyclin B in ‘young’ cells was not seen at passage level 32 when cells had entered senescence. By contrast, cyclin D1 and cyclin E, both induced approx. 10-fold in ‘young’ cells, were deregulated in senescent cells. Both cyclin RNAs showed a high expression in serum-deprived cells that was not inducible upon serum stimulation. Their abundance was approximately 10-fold higher in senescent cells relative to passage 20 cells, which was even above the levels reached after serum stimulation in ‘young’ cells.

The deregulation of cyclin D1 and cyclin E in senescent cells apparently has no effect on these cells’ inability to enter DNA synthesis (Fig. 1). We therefore sought to analyse whether cdk genes might be suppressed in senes-cent cells, with the possible consequence that cyclins D1 and E would lack their catalytic subunits. The results in Figs 3 and 4 demonstrate that the induction of cdk2 and cdk4 (PSK-J3; Meyerson et al., 1992) mRNA expression was indeed impaired in senescent cells. In contrast to the induction observed upon serum stimulation in cells at pas-sage 24, which clearly preceded the induction of cdc2 mRNA, no increase of the very low basal levels of cdk2 and cdk4 was seen in senescent cells. The same was true for cdc2 as previously described by others (Stein et al., 1991). cdc2, however, is believed to act later in the cell cycle in combination with mitotic cyclins so that its sup-pression can probably not explain the G₁ arrest seen in senescent cells. The only cdc2-related gene analysed that was not affected by senescence was cdk5 (PSSALRE; Mey-erson et al., 1992), which showed the same level of expression at all passage levels analysed. In addition, in contrast to cdk2, cdk4 and cdc2, the cdk5 gene was not inducible by serum and showed a relatively high basal expression in both quiescent and stimulated cells.

We next decided to investigate whether the deregulation of cyclin D1 mRNA expression also leads to an overexpres-sion of its gene product. For the immunological detection of cyclin D1 we used an antiserum raised against a bacte-rially expressed glutathione-S-transferase fusion protein (Sewing et al., 1993). The antiserum (α-CycD) is highly specific for cyclin D1 and does not cross-react with any of the other known human cyclins (Sewing et al., 1993). WI-38 cells were metabolically labelled with [³⁵S]methionine and cyclin D1 protein was immunoprecipitated using the α-CycD antibodies. Fig. 5a shows the results of the immunoprecipitation and Fig. 5b quantitation by β-radia-tion scanning. The data clearly show that the synthesis of a cyclin D1 protein of M_r approx. 39,000 in senescent cells continuously exposed to serum is >3-fold greater than in ‘young’ cells kept under the same culture conditions. For comparison, Fig. 5 also includes serum-deprived cells at passage 21, where the synthesis of p39-cyclin D1 is only about 1/10 the level of the synthesis seen in senescent cells. We also attempted to measure cyclin D1-associated kinase activity in vitro using baculovirus-expressed pRB as the substrate, but were unable to detect any phosphorylation with immunoprecipitates from either young or senescent cells (data not shown). This may be attributable to the per-haps neutralising properties of our antibodies, but it has to be pointed out also that cyclin D1-associated kinase activity in mammalian cells has not been demonstrated by other authors (Xiong et al., 1991, 1992; Matsushime et al., 1991, 1992; Won et al., 1992).

We also sought to determine whether the half-life of cyclin D1 protein might be altered in senescent cells. For this purpose, we performed pulse-chase experiments by metabolically labelling cells with [³⁵S]methionine and then incubating them for variable times in an excess of unla-belled methionine, followed by immunoprecipitation using the α-CycD antibodies. The results of this experiment and the quantitations are displayed in Fig. 6. It is evident that the half-lives of cyclin D1 protein in young and senescent cells did not differ significantly (38 and 35 min, respec-tively).

For the analysis of cdk2 and cyclin E proteins in ‘young’ and senescent cells we generated polyclonal antisera against the respective GST fusion proteins expressed in E. coli. These antibodies specifically precipitated cdk2 and cyclin E proteins synthesised by in vitro transcription-translation (data not shown). The cdk2 antibody showed some cross-reactivity with cdc2, but this is of no relevance to the pur-pose of this study, since we performed the immunoprecip-itations with WI-38 cells at a time after serum stimulation when cdc2 mRNA was undetectable (data not shown). These antibodies were then used for the experiments shown in Figs 7 and 8. The in vitro histone H1 kinase assay using cyclin E immunoprecipitates from passage 20 and 32 cells clearly demonstrates that the cyclin E-associated kinase activity was dramatically decreased (by 93%) in the senes-cent cells (Fig. 7). cyclin E is expressed at too low a level in WI-38 cells to be detectable by immunoprecipitation so that we can not formally exclude the possibility that its translation is impaired in senescent cells. Another explana-tion is more likely in view of the high levels of cyclin E mRNA in serum-induced senescent cells and the fact that the cyclin D1 protein is efficiently synthesised in cells at passage 32 (see Figs 5 and 6): the very low level of cyclin E-associated kinase activity in senescent cells may be due to the absence of its catalytic subunit, cdk2 (Koff et al., 1992; Dulic et al., 1992). As shown above, the induction by serum of cdk2 mRNA expression is impaired in senescent cells, and this finding was confirmed by the immuno-precipitation in Fig. 8. While cdk2 synthesis could be read-ily demonstrated in passage 20 cells, no specific immuno-precipitate was seen with extracts from metabolically labelled senescent cells.

Finally, we analysed the expression of cyclin D1 at the single cell level by using the α-CycD antibodies described above in indirect immunofluorescence experiments. The antibodies clearly detected a nuclear protein in WI-38 cells whose binding to the α-CycD antibodies could be competed with recombinant cyclin D1 but not cyclins A, B, C or E (data not shown). We therefore conclude that the α-CycD antibodies are suitable for the detection of cyclin D1 in WI-38 cells. The immunofluorescence experiment was per-formed with asynchronous cells at passage levels 20 and 32 (Fig. 9). To be able to identify the nuclei the cells were also stained with Hoechst 33258 dye. Since the ‘young’ and senescent cells differ in size, a meaningful comparison of the fluorescence signals was only possible by precisely measuring the integrated intensities in each nucleus. For each passage level, the intensities of 100 cells from approx-imately 20 different areas on the slide were determined by digital image analysis (Leica Quantimet), grouped in inten-sity categories and the number of cells in each category was plotted against the intensity (Fig. 10). This evaluation clearly showed that the vast majority of ‘young’ cells expressed cyclin D1 at levels ≤200,000 units, whereas >80% of the senescent cells exhibited greater values, in about 40% >3-fold higher compared to cells at passage level 20. Likewise, the average value for senescent cells was approximately 6-fold higher than the corresponding value for ‘young’ cells, 9.00×10⁴ at passage 20 and 5.95×10⁵ at passage 32.

One of the hallmarks of cellular senescence is an irre-versible arrest of the cells in the G₁-phase of the cell cycle. It is therefore conceivable that alterations in the expression and/or activity of gene products that are normally active in G₀/G₁→S progression is likely to play a pivotal role in this process. To date, however, only very few gene products ful-filling these criteria have been identified. One of this is the pRB protein, whose hyperphosphorylation is impaired in senescent cells (Stein et al., 1990; Fureal and Barrett, 1991). In the present study, we have analysed the expression of three different cyclin genes, which are thought to be asso-ciated with the control of G₀/G₁→S progression (cyclins A, D1 and E), as well as three cdc2-related genes (cdk2, cdk4 and cdk5), which also fall into this category. Expression of these genes was analysed in the human diploid fibroblast cell line WI-38 (Hayflick, 1965) at different passage levels under three different conditions, i.e. in asynchronous cells continuously kept in serum-containing medium, in growth factor-deprived quiescent cells and in serum-stimulated cells.

Our data suggest that, on the basis of their expression kinetics, the cyclins analysed in this study fall into four groups: (i) cyclin D1 is expressed under all the growth con-ditions tested (asynchronous, starved and stimulated cells), but its expression drops in quiescent cells and is inducible by serum-stimulation. It is induced relatively early during the G₀→S progression. (ii) Expression of cyclin E mRNA is hardly detectable in quiescent cells, and its induction fol-lows that of cyclin D1: cyclin E mRNAs appears shortly before and reaches peak levels round the G₁/S border. (iii) The pattern of cyclin A induction resembles that of cyclin E, but reaches maximum levels later in the cell cycle. (iv) cyclin B RNA is also undetectable in quiescent cells, but is induced after cyclins A and E. These results are in good agreement with published data obtained with other cell types that had been released from a chemically induced G₁/S block (Pines and Hunter, 1990a; Lew et al., 1991; Dulic et al., 1992; Koff et al., 1992) or stimulated with a growth factor in a G₀-state (Matsushime et al., 1991).

The cyclins analysed in this study not only differ in the temporal and quantitative regulation of their mRNA levels, but also with respect to their expression in senescent cells. Here, the cyclins fall into two categories, but interestingly these categories do not coincide with the groups defined above. The expression of two cyclins, i.e. cyclins A and B, is suppressed in senescent cells (Figs 2, 3 and 4 and Stein et al., 1991). In contrast, cyclins D1 and E were expressed at higher levels in senescent cells as compared to their younger counterparts. This could be seen in asynchronous cells (Fig. 2), in serum-deprived cells and in stimulated cells (Figs 2, 3 and 4), although the clearest effect was found under conditions of serum starvation (approx. 10-fold higher levels in senescent cells). It is unlikely that the over-expression of cyclins D1 and E plays a causal role in senes-cence, particularly in view of the fact that cyclin D1 is a putative oncogene (Lammie et al., 1991; Motokura et al., 1991; Hinds et al., 1992). This finding is interesting, how-ever, with respect to other aspects, such as the nature of the G ₁ arrest point in senescent cells and the regulation of cyclin expression. One of the most important questions to ask at this point concerns the molecular mechanism under-lying the deregulation of cyclin D1 and E expression in senescent cells. At least two possibilities may account for this phenomenon: (i) the senescent cells are arrested at a point in G₁ where the expression of cyclins D1 and E is at their highest, and (ii) senescent cells might lack a feedback control mechanism that negatively regulates their expression. Clearly, in replicating ‘young’ cells, cyclins D1 and E are induced at different times during the progression from G₀/G₁ to S. In addition, a cyclin which is induced with similar kinetics is not overexpressed, but suppressed, in senescent cells. It seems therefore less likely that the first hypothesis given above is the sole explanation for the observed deregulation of cyclins D1 and E in senescent cells. The second possibility mentioned above is difficult to test at present, as this would require the respective pro-moter regions to be available. It is, however, tempting to speculate that human cells may have evolved a mechanism that is reminiscent of the situation in Saccharomyces cere -visiae where the expression of CLN1 and CLN2 is subject to feedback regulation (Nasmyth, 1990). It is possible that the cyclin D1 and E holoenzymes are engaged in some mechanism of direct or indirect autoregulation. This mech-anism may fail in senescent cells because the (or some of their) respective catalytic subunits are not expressed, such as cdk2 in the case of cyclin E (Koff et al., 1992; Dulic et al., 1992) and cdk4 in the case of cyclin D1 (Matsushime et al., 1992; Xiong et al., 1992). The fact that we do not find a decreased expression of cdk5 in senescent cells does not necessarily argue against such a hypothesis, since the role of cyclin D-cdk5 complexes (Xiong et al., 1992), if they exist in vivo, is unknown. Senescent cells may thus represent an interesting model to elucidate the mechanisms governing the expression of cyclins in human cells.

Both cyclins D1 and E have been described as being able to inactivate the negative regulatory pRB protein (Hinds et al., 1992); in the case of cyclin E presumably through its interaction with cdk2 (Koff et al., 1991; Dulic et al., 1992). Our results, however, suggest that the dereg-ulation of cyclin D1 and cyclin E is not sufficient to drive senescent cells into DNA synthesis (Fig. 1). We attribute this at least in part to the failure of senescent cells to express cdk2 at normal levels, leading to an orphan cyclin E molecule lacking its catalytic subunit. In this respect, cdk2 and presumably other cyclin-associated kinases may play a key role in the process of cellular senescence. As more such genes become available and are characterised in the future, this hypothesis can be tested, for instance, by investigating the effect of their forced expression on replicative senescence.

After finishing this work, Won et al. (1992) reported that, in contrast to our results, cyclin D1 mRNA expression is decreased in senescent human fibroblasts. We do not know the reason for this discrepancy, but our observations were reproducible in five independent experiments, where we consistently saw a deregulation of both cyclin D1 mRNA and protein expression. It is possible that differences in the cell strains or the culture conditions are responsible for this discrepancy, but this question has to remain the subject of future investigations.

WI-38 cells were obtained from the ATCC, Rockville, Mary-land. We are grateful to J. Pines and T. Hunter for the cyclin B cDNA probe, to C. Schalk for FACS analyses and to Dr M. Krause and S. Klingelhöfer for oligonucleotide syntheses. This work was supported by the Deutsche Forschungsgemeinschaft (Mu601/5–2, Mu601/5–3 and Mu601/7–1), and the Dr Mildred Scheel-Stiftung für Krebsforschung. S.B. and A.S. are the recipients of fellowships from the Graduiertenkolleg ‘Zell-und Tumorbiologie’ at the Philipps-Universität Marburg.