Judith Campisi *

* From the Berkeley National Laboratory, Life Sciences Division, Department of Cancer Biology, Mailstop 70A-1118, University of California, Berkeley, CA 94720.

Supported by grants AG09909 and AG11658 from the National Institute on Aging and the U. S. Department of Energy under contract DE-AC03-76SF00098 to the University of California.

Address correspondence to Judith Campisi, Department of Cancer Biology, Mailstop 70A-1118, Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720. Telephone: (510) 486-4419; Fax: (510) 486-4475; Internet: JCAMPISI@LBL.GOV.

Key Words:

Cell Proliferation; Differentiation; Tumorigenesis; Tumor Suppressors; Aging; Cancer; Senescence.

The aging of multicellular organisms is clearly a complex process. Accordingly, several hypotheses have been put forth to explain its causes (see 1 for review). These hypotheses generally fall into two categories. The first invokes extrinsic or intrinsic factors that damage intracellular or extracellular molecules. This damage is thought to compromise the organism because dysfunctional molecules or cells accumulate, or damaged cells die and are not adequately replaced. The major damage-hypotheses of aging propose that errors in protein or DNA synthesis, the generation of reactive oxygen by mitochondria, or non-enzymatic modification of proteins damage cells or tissues. The second category of aging hypotheses invokes programmed or epigenetic changes in gene expression. These changes are thought to compromise the organism because terminally non-dividing cells with altered phenotypes accumulate. The major gene expression-hypothesis of aging proposes that cell division itself is the driving force behind the eventual loss of division capacity and change in phenotype. This intrinsic loss of cell proliferative capacity has been termed replicative or cellular senescence.

Replicative senescence is thought to be a tumor suppression mechanism, and also a contributor to organismic aging. Here, I review what is known about the genetics and molecular biology of cell senescence, and how it might affect aging and cancer. I also discuss the somewhat unorthodox view that senescent cells may actually contribute to carcinogenesis. My intention is not to argue for a particular hypothesis of aging. Aging is very likely due to programmed changes in gene expression in some cells, and damage in others; similarly, aging may be due to an accumulation of cells in altered or later stages of differentiation in some tissues, and loss of cells in others. Replicative senescence is not expected to affect all cells. Rather, it is an intrinsic process that alters the phenotype of mitotic cells. As such, replicative senescence is expected to comprise but one component of aging (Fig. 1). We do not yet understand the molecular bases or physiologic consequences of either the damage or altered gene expression that occur with age. However, this understanding may be critical before we can entertain the possibilities -- and appreciate the limits -- of intervening in the aging process.

It is well-established that most normal somatic cells do not divide indefinitely. This trait is termed the finite replicative life span of cells, and the process that limits the capacity for cell division is termed cellular or replicative senescence. Replicative senescence was first formally described over thirty years ago in cultures of human fibroblasts (2). Since then, many cell types from many animal species have been shown to have a limited replicative life span (reviewed in 3, 4).

Replicative senescence is generally studied in culture. In culture, cell proliferation (used here interchangeably with growth) can be stimulated in a controlled fashion for the many doublings that are needed for most or all cells in a population to senesce. It is much more difficult to manipulate or follow cell proliferation in vivo. Nonetheless, the division potential of some cells has been studied in vivo (reviewed in 3,4). Results from these and other studies discussed below strongly suggest that replicative senescence is not an artifact of cell culture.

Only two, or perhaps three, cell types may fail to senesce -- that is, may have an indefinite or immortal replicative life span. First, the germ line is capable of unlimited replication. Second, many tumors appear to contain cells that are immortal or have an extended replicative life span. Finally, certain stem cells may divide indefinitely, but this has yet to be shown unequivocally. Replicative senescence is particularly stringent in human cells. In contrast to cells from many rodent species, human cells rarely, if ever, spontaneously immortalize (that is, spontaneously fail to senesce) (5-7).

How do cells sense the number of divisions through which they have gone? This number can be substantial -- for example, 60-80 for human fibroblasts of fetal or neonatal origin. Moreover, for a given cell population, the number of doublings at which senescence occurs is fairly reproducible. At present, telomere shortening is perhaps the most viable explanation for a cell division "counting" mechanism (8,9).

Telomeres are the ends of linear chromosomes. They consist of a repetitive sequence, TTAGGG in humans and other vertebrates, and specialized proteins. The telomeric sequence and its binding proteins form a distinctive structure (10) that appears to prevent chromosome fusions, translocations and non-dysjunctions. Thus, telomeres are essential for maintaining the stability of eukaryotic genomes (11). Because DNA polymerases are unidirectional and require a labile primer, each round of replication leaves some 3' bases at the telomere unreplicated (8,9,11). Thus, telomeres shorten with each cell cycle. Telomerase, a multimeric enzyme that adds telomeric repeats to chromosome ends de novo, prevents telomeres from shortening. Most normal somatic cells do not express this enzyme. By contrast, telomerase is expressed by most cells that do not senesce: germ cells, possibly some stem cells, and many tumor cells (8,9,11,12). Thus, progressive telomere shortening is thought to be the mechanism by which somatic cells sense the number of completed divisions. Mammalian telomeres have been shown to shorten with cell division in culture and in vivo (8,9,11-14). It is not known how shortened telomeres induce the senescent phenotype (discussed below). In yeast, a shortened telomere may induce a DNA damage response, change the chromatin structure or redistribute telomere-associated proteins (reviewed in 4), and thereby alter gene expression.

Senescent cells (cells at the end of their replicative life span) are viable and metabolically active. They respond to environmental signals, and, indeed, many genes remain mitogen inducible throughout the replicative life span (15 - 19). Thus, replicative senescence is not programmed cell death or apoptosis. In fact, senescent cells are more resistant to apoptotic death than presenescent cells (20). Two other features distinguish senescent cells from their presenescent counterparts. First, senescent cells fail to proliferate in response to physiologic mitogens. Second, senescent cells show selected changes in differentiated functions.

Irreversible Growth Arrest. Senescent cells stably, and essentially irreversibly, arrest growth with a G1 DNA content (Fig. 2). Once arrested, physiological mitogens cannot induce them to enter the S phase of the cell cycle (reviewed in 4). This failure to respond to mitogens is not due to a general breakdown in growth factor signal transduction. The integrity of general growth factor signaling mechanisms is most clearly established in cultured human fibroblasts (4, 21) where many genes, including at least three protooncogenes (c-jun, c-myc, c-ras-Ha), remain mitogen-inducible (15-19) (Fig. 2).

A subset of mitogen-inducible genes are repressed in senescent cells (reviewed in 4) (Fig. 2). These include genes that encode many of the enzymes needed for DNA replication which are normally induced just prior to the start of S phase. Of particular importance, three essential growth-stimulatory transcription factors are repressed: the c-fos protooncogene (17), the Id1 and Id2 inhibitors of basic helix-loop-helix (bHLH) transcription factors (22), and E2F (23). The mechanisms that repress c-fos and Id are unknown. E2F repression, however, appears to be due to overexpression of p21 and possibly p16, which are inhibitors of cyclin-dependent protein kinases (cdks) (Fig. 2). Cdks phosphorylate the retinoblastoma tumor suppressor protein (pRb). Unphosphorylated pRb binds and suppresses E2F, and pRb phosphorylation by Cdks relieves this suppression. p21 inhibits all Cdks, whereas p16 inhibits a subset of Cdks (24). p21 and p16 are markedly elevated in senescent cells (24,25). p21 suppresses E2F activity (26). E2F activity, in turn, is essential for the expression of cyclin A, cdk1/cdc2, and several enzymes needed for DNA synthesis, all of which are repressed in senescent cells (18,27,28).

Thus, the immediate cause for the growth arrest of senescent cells appears to be repression of a few essential positive regulators (c-fos, Id, E2F; possibly others) (Fig. 2). Failure to express these genes then prevents the expression of other genes needed for growth. The repression of E2F is probably due to the high levels of p21 (and possibly p16), as well as the failure to express two E2F components, E2F1 (23), and E2F5 (29). However, it is unlikely that the cells arrest growth solely due to altered p21, p16 and E2F expression: senescent cells do not synthesize DNA when provided with E2F components (23,30). Moreover, indirect evidence suggests that senescent cells express additional growth inhibitors that interact with Id proteins and cooperate with pRb (30).

Altered Differentiation. Replicative senescence also causes selected changes in differentiated functions. The cell functions that are altered by senescence depend on the cell-type. By the criteria of a stable irreversible growth arrest and change in function, senescent cells resemble terminally differentiated cells. In contrast to terminal differentiation, however, replicative senescence occurs as a consequence of completing a finite number of cell divisions.

The altered differentiation associated with senescence can have rather profound consequences for cell -- and, at least in principle, tissue -- function. For example, upon senescence, adrenocortical epithelial cells lose the ability to express 17a-hydroxylase, a key enzyme in cortisol biosynthesis, in response to physiological inducers (35). Other steroidogenic enzymes are not affected. Thus, senescence alters the spectrum of steroids produced by adrenocortical cells. Similarly, presenescent dermal fibroblasts express low levels of collagenase and stromelysin, metalloproteases that degrade extracellular matrix (ECM) proteins, and relatively high levels of the metalloprotease inhibitors TIMP 1 and 3 (tissue inhibitor of metalloproteinases 1 and 3). Upon senescence, collagenase and stromelysin expression rises and TIMP 1 and 3 expression falls (36-38). Thus, replicative senescence entails a switch in the phenotype of dermal fibroblasts -- from a matrix-producing phenotype to a matrix-degrading phenotype.

Processes that prevent cell proliferation obviously are potential tumor suppressor mechanisms. Indeed, several lines of experimental evidence suggests that cell senescence suppresses tumorigenesis.

First, cells with a finite replicative life span are orders of magnitude less likely to form tumors than immortal cells (39). Moreover, many, if not most, tumors contain cells that are immortal or have an extended replicative life span (9,39,40). This is not to say that immortality is required for tumorigenesis (although, it may be for metastasis). Rather, there appears to be strong selection for cells with an increased replicative potential in the development of cancer. Immortality, of course, permits the extensive cell division that is needed to acquire the many mutations that are the hallmark of malignant tumors.

Second, certain oncogenes act at least in part by immortalizing or extending the replicative life span of cells (see 4,7,40,41 for review). These include mutated or deregulated cellular genes such as p53 or c-myc, as well as the oncogenes of certain viruses that are implicated in a subset of human cancers (for example, the Epstein-Barr or human papilloma viruses). Thus, mutations that lead to tumorigenesis, or the strategies of oncogenic viruses, involve mechanisms that allow cells to escape replicative senescence.

Third, among the genes that are essential for establishing and maintaining the senescent phenotype are two well-recognized tumor suppressors: the p53 and Rb genes. Inactivation of either p53 or Rb extends the replicative life span of human cells. In fact, p53 and/or pRb inactivation is the primary means by which viral oncoproteins extend replicative life span (see 4,7,41 for review). p53 and Rb appear to be essential for cells to irreversibly arrest growth once their telomeres have shortened to a critical length (a terminal restriction fragment of about 4 kb in most somatic cells). In the absence of p53 and Rb, cells continue to proliferate with shortened telomeres; eventually, these cells become genomically unstable and either die or mutate to an immortal phenotype, often, but not always, with an accompanying reactivation of telomerase (7,9,12). p53 and Rb are together the most commonly lost functions in human cancers.

Finally, in addition to the known tumor suppressors p53 and pRb, novel suppressor proteins (as yet unidentified) may also be critical for the growth arrest of senescent cells. One of these may be a bHLH transcription factor that interacts with Id1 and Rb (30). Another may be a repressor(s) of telomerase, which prevents the expression of this enzyme in most normal somatic cells (9).

Taken together, these findings strongly suggest that replicative senescence is controlled by tumor suppressor genes, and that these genes may suppress tumorigenesis at least in part by inducing senescence. The mechanisms that arrest the growth of the senescent cells, rather than those that alter their differentiation, are very likely the important features of replicative senescence vis a vis its role in tumor suppression.

Aging entails changes in both extracellular components and cells; cells can be (very simply) divided into two classes: post-mitotic and mitotic (Fig. 1). Post-mitotic cells, such as mature nerve and muscle cells, have irreversibly arrested growth due to terminal differentiation. Mitotic (or mitotically competent) cells, by contrast, divide when appropriately stimulated. In vivo, mitotic cells may divide frequently (e. g., cells comprising the basal layers of the stomach, intestine or skin), or only when the need for replacement or repair arises (e. g., many liver, kidney or other skin cells). Obviously, replicative senescence occurs only in mitotically competent cells. The idea that the replicative senescence of mitotic cells contributes to organismic aging derives from several lines of evidence.

First, cells from old donors tend to senesce after fewer population doublings (PD) than cells from young donors (reviewed in 3,4,21). Despite considerable scatter in the data, this correlation has been seen in several studies. Thus, in general, human fetal fibroblasts senesce after 60-80 PD, fibroblasts from young adults do so after 30-40 PD, and fibroblasts from old adults senesce after 10-20 PD. These findings suggest that the replicative life span of cells in renewable tissues may be progressively exhausted during the chronological life span of organisms.

Second, limited interspecies comparisons suggest that cells from short-lived species tend to senesce more rapidly than comparable cells from long-lived species (reviewed in 3,4,21). Thus, mouse fetal fibroblasts senesce after 10-15 PD, whereas human fetal fibroblasts proliferate for >50 PD. These results raise the possibility that genes that control the chronological life span of organisms may overlap with genes that control the replicative life span of cells.

Third, cells from humans with hereditary premature aging syndromes senesce prematurely (reviewed in 3,4,21). This is best documented for the Werner’s syndrome (WS). WS is an autosomal recessive, adult-onset disorder in which many, but not all, signs of aging develop in early adulthood. These signs include thin and wrinkled skin, thin and gray hair, atherosclerosis, osteoporosis, type II diabetes and many forms of cancer. The WRN gene was recently cloned and shown to encode a large protein with potential helicase activity (42). Cells from young adults with WS senesce well ahead of cells from age-matched controls. These findings support the idea that organismic life span and cellular replicative life span may be related and controlled by common genes.

Taken together, the evidence suggests that senescent cells may accumulate with age in mitotic tissues. Whether this accumulation contributes to age-related pathology has yet to be determined. In theory, replicative senescence could inhibit tissue repair. However, mitotically competent cells are easily recovered from old tissues, and wounds do heal even in old mammals (although often more slowly). Thus, with the possible exception of the immune response, the arrested growth of senescent cells may not have a major impact on tissue function, although it could compromise the organism by delaying repair. By contrast, the altered function of senescent cells may have a strong impact. As discussed above, senescent cells secrete molecules such as ECM degrading enzymes and inflammatory cytokines, which can have far-ranging, deleterious effects on tissue integrity and/or function. Moreover, relatively few senescent cells would be needed for these effects. Finally, the resistance to apoptosis may explain why senescent cells accumulate and are not cleared. Thus, the most important features of senescent cells vis a vis their role in aging may be their altered differentiation and resistance to death.

As discussed above, one body of evidence suggests that replicative senescence is tumor suppressive, whereas another suggests that senescent cells accumulate and contribute to aging. At first glance, these views may seem unrelated or even at odds. However, evolutionary theories suggest that some traits that are selected to optimize health during the period of reproductive fitness can have unselected and deleterious post-reproductive effects (see 1). Thus, at least in mammals, replicative senescence may contribute to the relative freedom from cancer that is seen during the first half of life. Later in life, it may be deleterious because senescent cells accumulate. Thus, altered differentiation and resistance to apoptosis may be unselected phenotypes that compromise tissue function and integrity as senescent cells accumulate with age.

From the rise in cancer incidence with age, it also appears that replicative senescence fails increasingly with age. This is not surprising since there is no selective pressure to evolve a tumor suppressive mechanism that is efficacious beyond the period of reproductive fitness. It is interesting, however, that at least among mammals, there is a close relationship between the rate of aging and the rate at which cancers arise. In fact, it has been proposed that aging and carcinogenesis may derive from a shared mechanism(s) (45). One possibility is that control over the replicative life span of cells may link the rate of aging with that of tumor incidence. Perhaps fewer genes control replicative senescence in short-lived species, which develop cancer relatively rapidly, than in long-lived species, which have a relatively slow rate of cancer development.

The single largest risk factor for developing cancer is age. In fact, most major forms of cancer rise exponentially with age with rates that range over several orders of magnitude. Traditionally, this exponential rise has been interpreted to indicate that several mutations (estimated from 4 to 12) are required for a particular cancer to develop (see 46). This interpretation supports current models of multi-step carcinogenesis, wherein malignant tumors are proposed to develop as a result of successive mutations. Of course, there is little doubt that mutations are a necessary and critical feature of cancer. And, there is little doubt that tumorigenesis entails multiple mutations that activate protooncogenes and inactivate tumor suppressor genes (47). However, if the rise in cancer with age were due solely to the accumulation of successive independent mutations, it is not clear why the rise should be so sharply exponential. Thus, there must be synergy among genetic and/or cellular processes to promote carcinogenesis. What might these processes be?

First, mutations that increase the susceptibility to additional mutations are certainly expected to contribute to an exponential rise in tumor incidence. Genetic lesions of this sort would lead to a so-called hypermutator phenotype. Mismatch repair genes and p53 are examples of genes that can result in a hypermutator phenotype when appropriately mutated. In addition, processes that decrease the ability of the microenvironment to control cell behavior might be expected to contribute to an exponential rise in tumor incidence. There is now substantial evidence that the microenvironment can exert considerable control over cell growth and differentiation, even in the face of potentially disrupting mutations (48,49). Thus, cells harboring oncogenic mutations may not express a neoplastic phenotype until they are released from the constraints of their microenvironment. The microenvironment includes soluble components such as growth factors and cytokines, as well as insoluble components such as neighboring cells, the basement membrane and the stroma.

Consider, now, the senescent phenotype and the possibility that senescent cells may accumulate with age in mitotically competent tissues. As discussed earlier, cultured senescent cells, particularly senescent stromal cells such as fibroblasts, express ECM degrading enzymes and cytokines that can have long-range effects. Thus, it is quite possible that senescent cells disrupt the microenvironment. A striking example of this may reside in the dermis, where there is a conspicuous age-dependent decline and disorganization of the dermal matrix, of which collagen is a major component. Fibroblasts that express a marker of senescent cells are readily apparent in the dermis of skin from old donors, but much less so in skin from young donors (43). Moreover, senescent, but not presenescent, fibroblasts in culture secrete very high levels of collagenase and stromelysin, which can destroy the dermal matrix (36-38). It is tempting, therefore, to speculate that senescent cells -- whether stromal or epithelial in origin -- can destroy the organization, and hence the constraints, of the basement membrane and underlying stroma in vivo. Should this breach of integrity occur near a cell that is harboring a potentially oncogenic mutation, but has been kept in check by its microenvironment, neoplastic growth may be initiated, or at least favored (Fig. 3).

Soluble factors elaborated by senescent cells may similarly stimulate the growth of pre-neoplastic cells. For example, our preliminary data suggest that cultured senescent, but not presenescent, fibroblasts secrete high levels of heregulin (Lupu and Campisi, unpublished), a ligand for an erbB receptor. Heregulin stimulates the growth of epithelial tumors that have an amplification of the erbB2 gene (50). Similarly, the overproduction of IL-1 by senescent fibroblasts may stimulate local proliferation, albeit indirectly (IL-1 induces the expression of growth factors or growth factor receptors in certain target cells) (51).

Thus, an age-dependent accumulation of senescent cells in mitotic tissues may create a "pro-carcinogenic" microenvironment that can synergize with the accumulation of mutations, which also occurs with age. Hence, replicative senescence, despite its ability to suppress tumorigenesis, may actually contribute to tumorigenesis later in life. This synergy between senescent cell accumulation and mutation accumulation would help explain, at least in part, why tumor incidence increases exponentially with age. According to this hypothesis, the presence of senescent cells might favor only the early stages of tumorigenesis -- local loss of growth control and local invasion. On the other hand, tumor progression, which is a heterogeneous and extremely complex process, may not be favored by an aged tissue environment. In fact, there is some evidence that certain tumors, once established, may grow more slowly in older animals, most likely because angiogenesis is retarded in the old hosts (52).

In summary, if replicative senescence contributes to tumor suppression early in life, late in life we may be faced with a difficult task indeed: how to reverse its hypothesized deleterious effects (compromised tissue function and integrity) without reversing its beneficial effects (tumor suppression). Will this be possible? Optimistically, it may be possible to counteract the altered differentiation of senescent cells, which very likely is responsible for many of their deleterious effects, without reactivating growth. However, we really need a better understanding of whether and how senescent cells affect tissue function and integrity, and how the senescent phenotype is controlled, before we can realistically discuss future possibilities. For the moment, replicative senescence may be considered a double-edged sword. Its potential role in aging and cancer will undoubtedly occupy experimental gerontologists for many years to come.

Figure 1. Aging entails characteristic changes in post-mitotic cells, mitotic cells and extracellular components, which may be caused by extrinsic factors or intrinsic factors. This review focuses on the intrinsic changes that occurs during the aging of mitotic cells that are caused by cell division, or, more precisely, by DNA replication. These changes include a permanent arrest of cell proliferation, resistance to apoptotic death and altered differentiated properties.

Figure 2. Expression of cell cycle-regulatory genes in senescent cells. Many genes remain mitogen-inducible in senescent fibroblasts, but at least three key transcription regulatory factors are not expressed. These are the c-fos protooncogene (fos), the Id1 and Id2 inhibitors of helix-loop-helix transcription factors (Id), and the E2F transcription factor. The lack of E2F activity in senescent cells is very likely due in part to the overexpression of the p21, and possibly the p16, inhibitor of cyclin-dependent kinases (cdks). Cdk activity is needed for the phosphorylation of pRb. In the absence of phosphorylation, pRb suppresses the activity of E2F. Suppression of E2F activity in turn prevents the expression of several genes needed for DNA synthesis, including cyclin A, cdc2 (also known as cdk1), thymidylate synthetase (ts), thymidine kinase (tk), dihydrofolate reductase (dhfr), and the replication dependent histones (histone). Genes that are expressed by presenescent cells but are repressed in senescent cells are indicate by an X beneath their designation.

Figure 3. Senescent cells may disrupt the microenvironment, there by permitting the proliferation of initiated cells harboring potentially oncogenic mutations.