Cocultures.

Presenescent (5 × 10⁴) and senescent (1 × 10⁵) fibroblasts were allowed to attach to 6-well culture dishes overnight and were incubated in serum-free medium for 1–3 days to generate lawns with similar cell numbers. Epithelial cells were incubated in growth factor-deficient medium for 2–3 days, plated (2 × 10⁴/well) on fibroblast lawns, and maintained in growth factor-deficient medium for 8 days, unless noted otherwise. Growth factor-deficient medium contained KBM, 1.8 mM CaCl₂, 5 μg/ml insulin, 0.5 μg/ml hydrocortisone for HaCAT and normal keratinocytes, DME/F12, 5 μg/ml insulin, 1.4 μM hydrocortisone, 5 μg/ml prolactin for SCp2 cells, DME/F12, 250 ng/ml insulin, 10 μg/ml transferrin, 2.6 ng/ml Na₂SeO₃, 1 ng/ml epidermal growth factor, 0.1 nM estradiol, 1.4 μM hydrocortisone, 5 μg/ml prolactin for S1 cells, and DME/F12, 5 μg/ml insulin for MDA231 cells. Cultures were fixed in 4% paraformaldehyde and were stained with 1% Rhodanile blue (Sigma) or 1 μg/ml 4,6-diamidino-2-phenylindole (DAPI).

Soluble and Matrix Factors.

Epithelial cells were seeded in upper chambers of Millicells-PCF (Millipore) containing presenescent or senescent lawns in the lower chambers, or onto matrices prepared by culturing fibroblasts in serum-free medium for 2–3 days, removing the cells with EDTA (Versene, GIBCO/BRL) at 37°C or 0.05% Nonidet P-40 at 23°C (with similar results), and washing with serum-free medium.

Quantification of Epithelial Cells.

Fluorescent images from five random fields/well were captured at 400× and analyzed by using a modification of SCIL_IMAGE 1.3 (Netherlands Organization for Applied Scientific Research, Institute of Applied Physics, University of Amsterdam). The program was customized to recognize differences in DAPI fluorescence between epithelial (smaller, more intense) and fibroblast (larger, less intense) nuclei and calculate epithelial fluorescence/field. We verified that the epithelial DAPI signal is proportional to cell number by expressing enhanced green fluorescence protein (EGFP) by using a retrovirus (LXSN) (22) and comparing epithelial DAPI and EGFP fluorescence in cocultures. EGFP was expressed in >90% of cells.

Senescence Induced by p14^ARF, RAS-Ha(V12), and H₂O₂.

Fibroblasts infected with p14^ARF (22) or RAS-Ha(V12) retroviruses (23) were selected in puromycin (22), incubated without puromycin for 2–3 (p14^ARF) or 3–4 (RAS) days (22, 23), and replated for coculture as described above. Confluent presenescent fibroblasts were given 550 μM H₂O₂ in serum-containing medium for 2 h, left to recover overnight, then replated at subconfluence (24). After 5–7 d, cells developed a senescent phenotype and were replated for coculture.

Senescent Fibroblasts Do Not Stimulate Normal Epithelial Cells.

In striking contrast to preneoplastic and neoplastic epithelial cells, genetically normal keratinocytes grew equally well on presenescent and senescent fibroblasts. After 8 days in coculture, there was no statistical difference between growth on presenescent and senescent lawns. This was true for neonatal and adult human keratinocytes (Fig. ​(Fig.2)2) and human mammary epithelial cells (not shown). Thus, although senescent fibroblasts stimulated preneoplastic and malignant epithelial cells, they did not differentially stimulate normal epithelial cells.

Open in a separate window

Figure 2

Effects on normal epithelial cells. Normal adult (Adult) or neonatal (Neonatal) human keratinocytes were seeded onto WI-38 lawns, cultured for 8 days, and quantified by DAPI fluorescence. Error bars = SEM from duplicate wells from one of two experiments. Gray bars, presenescent lawns; black bars, senescent lawns.

Kinetics.

To follow the kinetics of the growth stimulation, we seeded EGFP-expressing HaCAT cells onto presenescent or senescent fibroblast lawns and monitored EGFP fluorescence with time (Fig. ​(Fig.33A). Growth on senescent fibroblasts surpassed that on presenescent fibroblasts within 4 days (Fig. ​(Fig.33A), and continued to do so until the epithelial cells reached confluence (after 9–10 days; not shown). Unless noted otherwise, subsequent experiments were terminated within 8 days.

Contribution of Secreted Factors.

Senescent fibroblasts might stimulate epithelial cell growth by direct cell–cell interaction or by secreting diffusible factors or an insoluble extracellular matrix.

To test these possibilities, we cultured cells in two-chamber dishes. These separated fibroblast lawns from the epithelial cells by a porous membrane (0.4 μm pore size), preventing direct contact but permitting exchange of soluble diffusible factors. Soluble factors secreted by senescent fibroblasts were 2- to 3-fold more potent in stimulating HaCAT (not shown) and SCp2 (Fig. ​(Fig.33B) growth than those secreted by presenescent fibroblasts. Overall, however, senescent fibroblast-derived soluble factors were 10-fold less potent than direct cell contact (Fig. ​(Fig.33B, soluble vs. cells, black bars), even though epithelial cells attached equally well to the membrane and fibroblasts (not shown).

To determine the contribution of secreted matrices, we allowed fibroblasts in serum-free medium to deposit extracellular matrix onto culture dishes for 2–3 days. We then removed the cells by calcium chelation or mild detergent. Immunocytochemistry showed that fibroblast-depleted dishes contained abundant fibronectin (not shown), suggesting that at least this matrix component survived cell removal. We plated SCp2 cells onto the matrices and quantified cell number by DAPI fluorescence. Matrix produced by senescent fibroblasts was 3- to 4-fold more stimulatory than matrix produced by presenescent fibroblasts (Fig. ​(Fig.33B). This difference was not caused by differences in epithelial cell attachment (not shown).

Together, these results indicate that about 10% of the growth stimulation caused by senescent fibroblasts was caused by secreted soluble factors, whereas 40% was caused by secreted extracellular matrix. These are minimal estimates because cells may experience higher levels of soluble factors in direct coculture, and/or matrix components may be lost or inactivated during cell removal. Thus, at least 50% of the growth stimulation was attributable to the secretory phenotype of senescent fibroblasts.

Growth Stimulation Is Independent of the Senescence Inducer.

Overexpression of certain oncogenes or tumor suppressor genes or DNA damage can induce a phenotype that closely resembles replicative senescence (2, 12). We therefore asked to what extent the ability to stimulate preneoplastic epithelial cells depended on replicative exhaustion.

First, we induced senescence by overexpressing p14^ARF, a tumor suppressor that mediates a senescence response (22). We seeded HaCAT cells onto lawns of presenescent or p14^ARF-arrested fibroblasts and assessed HaCAT number by EGFP fluorescence. p14^ARF-arrested fibroblasts stimulated HaCAT cell growth 2- to 3-fold relative to presenescent fibroblasts (Fig. ​(Fig.33 C and D).

We next tested 82-6 fibroblasts that we immortalized with hTERT (21), the catalytic subunit of human telomerase (26, 27). The immortal cells had completed >80 doublings, or twice the replicative lifespan of unmodified control cells. HaCAT cells grew equally well on immortal and presenescent control fibroblasts [Fig. 3D (−, ARF) vs. (−, TERT-ARF)]. Thus, cell division per se was not responsible for the stimulation caused by replicatively senescent fibroblasts. In response to p14^ARF, the immortal fibroblasts arrested growth with a senescent morphology (not shown), whereupon they stimulated HaCAT growth 2- to 3-fold [Fig. 3D (+, ARF) vs. (+, TERT-ARF)].

We also induced senescence by expressing oncogenic RAS [RAS-Ha(V12)] (23) and seeded HaCAT cells onto presenescent or RAS-arrested fibroblasts. RAS-arrested fibroblasts stimulated HaCAT growth 1.5-fold compared with presenescent fibroblasts (Fig. ​(Fig.33D, RAS). This stimulation may be an underestimate because we analyzed the cocultures after only 5 days, owing to poor survival of RAS-expressing cells after >5 days in growth factor-deficient medium.

Finally, we treated fibroblasts with a sublethal dose (550 μM) of H₂O₂, allowed them to develop a senescent phenotype (24), and seeded HaCAT cells onto untreated or H₂O₂-treated fibroblasts. HaCAT cells proliferated 2- to 3-fold more readily on the H₂O₂-arrested fibroblasts (Fig. ​(Fig.33D, H₂O₂).

Thus, fibroblasts induced to senesce by several means stimulated the proliferation of preneoplastic epithelial cells. This stimulation was 50–70% of that caused by replicatively senescent fibroblasts (Fig. ​(Fig.33D, R), indicating that, at least qualitatively, the growth stimulation was independent of the senescence inducer.

Senescent Fibroblasts Stimulate Tumorigenesis.

To test the idea that senescent cells create a microenvironment that promotes the growth of potentially or frankly neoplastic cells in vivo, we injected epithelial cells, alone or with fibroblasts, into immunocompromised (nu/nu) mice (Fig. ​(Fig.4).4).

HaCAT cells alone did not form tumors after 40 (Fig. ​(Fig.44Ai) or 180 (not shown) days, as reported (17). When injected with an equal number of presenescent fibroblasts, three of 20 animals developed small tumors (<20 mm³); one regressed and two were excised for histology (Fig. ​(Fig.44Aii). Most of the animals (17 of 20), however, were tumor-free after >100 days. By contrast, when HaCAT were injected with senescent fibroblasts, seven of 15 animals developed tumors that averaged twice the size of those formed by presenescent fibroblasts (P < 0.01) (Fig. ​(Fig.44Aiii). Most of the tumors began to regress before excision, suggesting they were not malignant. This was not the case with SCp2 cells. SCp2 cells also did not form tumors (Fig. ​(Fig.44Bi), as reported (25). Moreover, when injected with presenescent fibroblasts, they did not form tumors after 120 (Fig. ​(Fig.44Bii) or 160 (not shown) days. However, when injected with senescent fibroblasts, they formed very large (400–2,000 mm³) tumors in four of six animals (Fig. ​(Fig.44Biii) with a latency of >80 days. These tumors would have killed the animals if not excised. Thus, senescent fibroblasts stimulated hyperproliferation and neoplastic progression, respectively, of HaCAT and SCp2 cells in vivo.

To determine effects on malignant epithelial cells, we used Ha(Pk), a weakly tumorigenic HaCAT derivative (unpublished data), and MDA231, an aggressive human breast cancer cell line (20). Ha(Pk) cells alone did not form tumors after 40 days (Fig. ​(Fig.44Ci) but did form tumors in four of five animals after 90 days (not shown). Presenescent fibroblasts accelerated tumorigenesis; within 25 days, three of five mice developed small (30–50 mm³) tumors that grew slowly over the next 2 weeks (Fig. ​(Fig.44Cii). Senescent fibroblasts accelerated tumorigenesis further; four of five mice developed tumors, two of which appeared in <10 days and grew rapidly (>100 mm³) (Fig. ​(Fig.44Ciii). More striking results were obtained with MDA231 cells. Small numbers (2 × 10⁵) of MDA231 cells produced tumors that reached 300–400 mm³ in two of five mice in 45 days (Fig. ​(Fig.44Di). hTERT-immortalized fibroblasts slightly accelerated tumorigenesis, causing tumors that reached 100–500 mm³ in four of five mice (Fig. ​(Fig.44Dii). Senescent fibroblasts greatly accelerated tumorigenesis. Four of five mice developed tumors that reached 300–1,600 mm³ (Fig. ​(Fig.44Diii). Thus, at least for HaCAT, Ha(Pk), and MDA231, presenescent or hTERT-immortalized fibroblasts did not inhibit tumorigenesis. More important, senescent fibroblasts strongly stimulated or facilitated tumorigenesis.

We thank R. Weinberg for the hTERT cDNA, S. Lowe for the RAS-Ha(V12) vector, D. Yip for preliminary experiments, W. Chou for help with computer analyses, the S. Fisher laboratory for help with tumor analyses, V. Band and R. Lupu for advice on tumor studies, L. Frappart for pathological evaluation of the tumors, and S. Hayward, G. Cunha, and T. Tlsty for discussions. This work was supported by Department of Defense Fellowship DAMD17-98-1-8063 (to A.K.) and National Institutes of Health Grants AG09909 (to J.C.) and CA82548 (to P.-Y.D.).