The study of aging, by nature multidisciplinary, has been characterized by a dizzying variety of theories, a huge phenomenological literature, and the absence of firmly established primary causes. The diverse life histories of animal species, which manifest aging in very different ways, has been an obstacle to testing unified theories. For experimental gerontology to provide more than a catalog of age-related changes, it has been necessary for biologists to define the alterations that are common to most old cells, tissues, and animals, while simultaneously respecting that there is not a single phenomenon of aging or a single cause. This has taken some time, and from the outside it may have appeared that the field has been mired in phenomenology. Perhaps for this reason, the study of aging was until recently avoided by molecular biologists, who naturally favored clear-cut phenomena. As the molecular details of development, cancer, and immunology yielded to modern tools during the 1970s and 1980s, the field of aging lagged, and mechanisms responsible for aging failed to emerge.
Throughout this time, though, there has never been a shortage of unified theories attempting to reduce aging to something more tractable. In fact, gerontologists have been prolific in this regard (83, 205). Whereas some researchers have believed that a small number of random, deleterious mechanisms could explain degenerative senescence, others have opted for theories of "programmed" aging, in which senescence is the final destination in a developmental pathway. In the course of these debates, a number of scientists have rallied around a set of ideas called the free radical theory of aging: loosely, the belief that damage by reactive oxygen species is critical in determining life span. This theory inspired many experiments in which evidence of oxidative damage in aged animals was sought.
In the last 10 years or so, the nature of aging research has changed dramatically; one might say that the field has entered early adulthood. The tools of molecular biology are now sophisticated and accessible enough that researchers within gerontology have adopted them. At the same time, molecular biologists situated on the edge of aging research have made inroads and have discovered that fruit flies and nematodes are amenable to the study of aging. Also, medical researchers investigating human diseases of aging, such as Alzheimer's disease (AD) and inherited progerias, have overcome long-standing roadblocks.
It has been gratifying, therefore, that many of the preliminary studies in what might be called "molecular gerontology" lend credibility to the free radical theory. Results from disparate experimental systems have recently shown that oxygen radicals play a role in degenerative senescence, and the pace of discoveries is quickening. The likely result of this collision of scientific approaches will be the unraveling of the physiological tangle of aging, and it seems safe to say that one of the important knots will turn out to be oxidative stress.
However, there is a danger that in the excitement of theoretical confirmation, certain nuances are lost. For instance, the revelation that oxygen radicals may be involved in neurodegeneration does not mean that oxidative stress determines life span. The free radical theory has sought not only to explain the mechanisms of degenerative senescence, but it has also attempted to explain differences between species' life spans in terms of oxidants. So, although many recent studies indicate that oxygen radicals play some kind of role in aging, only a small number of these support the more ambitious version of the free radical theory. On the other hand, there is no reason to cling to such a stringent version of the free radical theory, and it is becoming apparent that whether or not they determine life span, oxygen radicals are certainly important players in aging's pathophysiology. In other words, the scope of the free radical theory of aging should include aging-associated oxidative stress in general, rather than limiting itself to those oxidative events that may determine life span. In fact, many current articles indicate that such a blurring of distinctions has already occurred and that as it is commonly used, "free radical theory" encompasses a broad set of ideas. Therefore, our first purpose is to delineate these different conceptions of the free radical theory, as a prerequisite to its critical evaluation.
Because of the recent popularity of free radical research, a large number of reviews have addressed various aspects of the interplay between oxidants and aging (6, 15, 18, 33, 51, 57, 58, 60, 65, 82, 85, 87, 103, 106, 123-126, 130, 161, 166, 196, 203, 227, 257, 273, 275, 288, 293, 297, 307, 312, 315, 335, 338, 348, 353, 357). Rather than merely updating this literature, our aim is to provide a systematic categorization of the types of experiments that have been performed. The phenomenon and study of aging are incredibly diverse, encompassing organisms from rotifers to mammals and techniques from physiology to genetics. Although it is precisely the broad sweep of evidence that lends the free radical theory its appeal, the menagerie of animals and techniques sometimes obscures the logic. By breaking the literature down into smaller pieces (a practice we find necessary ourselves), we hope to make it easier for readers to judge the theory. Moreover, by imposing a structure, we aim to highlight novel and definitive approaches, because it is these that will replace the phenomenology of past decades.
In this review, then, we briefly outline the evolution of the free radical theory and then delineate the different areas of evidence. We focus on recent experiments and point to the areas that we feel are most likely to provide future insights. The way in which we have categorized the literature is outlined in the table of contents (sects. IV-XVII) and in Table 1. Although any such system is somewhat arbitrary, we hope that ours will make it easier both to assimilate the existing literature and to envision future experiments. In writing a review on as broad a topic as the free radical theory, we have been forced to limit both the content and the number of references cited. Although we have done our best to include recent work, omissions were inevitable. We apologize to all authors whose work we have not managed to include, and direct readers to other recent reviews for material we have left out.
B. Sources of Oxidants
Ground-state diatomic oxygen (3
g
O2 or more commonly, O2), despite being a radical species and the most important oxidant in aerobic organisms, is only sparingly reactive itself due to the fact that its two unpaired electrons are located in different molecular orbitals and possess "parallel spins." As a consequence, if O2 is simultaneously to accept two electrons, these must both possess antiparallel spins relative to the unpaired electrons in O2 , a criterion which is not satisfied by a typical pair of electrons in atomic or molecular orbitals (which have opposite spins according to the Pauli exclusion principle). As a result, O2 preferentially accepts electrons one at a time from other radicals (such as transition metals in certain valences). Thus, in vivo, typical two- or four-electron reduction of O2 relies on coordinated, serial, enzyme-catalyzed one-electron reductions, and the enzymes that carry these out typically possess active-site radical species such as iron. One- and two-electron reduction of O2 generates O2· and hydrogen peroxide (H2O2), respectively, both of which are generated by numerous routes in vivo, as discussed below. In the presence of free transition metals (in particular iron and copper), O2· and H2O2 together generate the extremely reactive hydroxyl radical (·OH). Ultimately, ·OH is assumed to be the species responsible for initiating the oxidative destruction of biomolecules. In addition to O2·, H2O2 , and ·OH, two energetically excited species of O2 termed "singlet oxygens" can result from the absorption of energy (for instance, from ultraviolet light). Designated by the formulas 1
gO2 and 1g+O2 , both of these species differ from the triplet ground state (3gO2) in having their two unpaired electrons in opposite spins, thereby eliminating the spin restriction of ground-state O2 and enabling greater reactivity. The chemistry of oxygen and its derivatives has been extensively discussed elsewhere (115, 342). Because all of these species (O2·, H2O2 , ·OH, 1gO2 , and 1g+O2), by different routes, are involved in oxygen's toxicity, we will collectively refer to them as "oxidants."
It is now beyond doubt that oxidants are generated in vivo and can cause significant harm (20, 37, 60, 91, 112, 351). There are numerous sites of oxidant generation, four of which have attracted much attention: mitochondrial electron transport, peroxisomal fatty acid metabolism, cytochrome P-450 reactions, and phagocytic cells (the "respiratory burst"). Before a discussion of the potential contributions of different sources of oxidants, it is worthwhile briefly to outline them.
In the textbook scheme of mitochondrial respiration, electron transport involves a coordinated four-electron reduction of O2 to H2O, the electrons being donated by NADH or succinate to complexes I and II, respectively, of the mitochondrial electron transport chain (ETC). Ubiquinone (coenzyme Q, or UQ), which accepts electrons from complexes I (NADH dehydrogenase) and II (succinate dehydrogenase), undergoes two sequential one-electron reductions to ubisemiquinone and ubiquinol (the Q cycle), ultimately transferring reducing equivalents to the remainder of the electron transport chain: complex III (UQ-cytochrome c reductase), cytochrome c, complex IV (cytochrome-c oxidase), and finally, O2 (115). However, it appears mitochondrial electron transport is imperfect, and one-electron reduction of O2 to form O2· occurs. The spontaneous and enzymatic dismutation of O2· yields H2O2 , so a significant by-product of the actual sequence of oxidation-reduction reactions may be the generation of O2· and H2O2 .
How much O2· and H2O2 do mitochondria generate? In classic experiments during the 1970s, measurements of H2O2 generation by isolated mitochondria indicated that it is maximal when ADP is limiting and the electron carriers are consequently reduced ("state 4" respiration) (26). Estimates of state 4 H2O2 generation by pigeon and rat mitochondrial preparations amounted to 1-2% of total electron flow (26, 229). One problem with this estimate of mitochondrial H2O2 generation is its reliance on the use of buffer saturated with air (20% O2). In vivo, the partial pressure of O2 is ~5%, so these calculations may overestimate the flux of oxidants in vivo. Even disregarding the use of air-saturated buffer, the initial estimate of percentage ETC flux leading to H2O2 can be challenged on the grounds that in these experiments the concentrations of substrates fed to mitochondria were higher than occurs physiologically (118, 146). When H2O2 is measured with more physiological concentrations, the flux is ~10-fold lower (118), and experiments using subcellular fractions of SOD-deficient Escherichia coli suggest in vivo leakage of 0.1% from the respiratory chain (146).
What proportion of mitochondrial H2O2 ultimately derives from ETC O2· generation? Unfortunately, the measurement of O2· generation by intact mitochondria is prevented by the presence of mitochondrial SOD (mSOD). Therefore, the isolation of submitochondrial particles from which mSOD has been removed (by sonication of the intact organelles followed by extensive washing) was used for the detection of ETC O2·. In these experiments, stoichiometric estimates of the ratio of O2· generation (by submitochondrial particles) to H2O2 generation (by the intact organelles) fell between 1.5 and 2.1 (24, 25, 69, 89, 191); because two O2· molecules dismutate (either spontaneously or with the help of mSOD) to form one molecule of H2O2 , such results suggest that virtually all mitochondrial H2O2 may originate as O2· (27). Moreover, because most cellular H2O2 originates from mitochondria, O2· from the ETC may be a cell's most significant source of oxidants (37).
In a recent discussion of the classic in vitro work (88), some of the original experimenters take issue with the idea that free O2· exists in mitochondria as a result of normal flux through the ETC. They point out that in addition to having removed mSOD from mitochondria, the sonication they employed also resulted in the loss of cytochrome c, which rapidly scavenges O2· in vitro and is present in mitochondria at local concentrations from 0.5 to 5 mM. In mitochondria, they argue, mSOD and cytochrome c rapidly scavenge O2· (in the matrix and intermembrane spaces, respectively). More to the point, the authors stress that unless the ETC was poisoned with inhibitors such as antimycin A, O2· generation was not detected in their experiments (89, 191). Arguing that mSOD should act to increase the rate of O2· generation in vivo (by accelerating product removal by dismutation to H2O2), they suggest that the actual role of mSOD in vivo may be to increase H2O2 generation (with O2· as a rapidly consumed intermediate) (88). Ultimately, there remains a good deal of uncertainty surrounding the mechanisms, quantity, and meaning of mitochondrial O2· generation in vivo (228), a mystery which is deepened by recent reports of enzymatic nitric oxide (NO·) generation in mitochondria (C. Giulivi and C. Richter, personal communication). Because O2· and NO· react to form the oxidant peroxynitrite (ONOO), mitochondrial O2· generation may soon need to be considered in the light of its ability to destroy NO· and form ONOO, as discussed in section XVIIIB.
A second source of oxygen radicals is peroxisomal 
-oxidation of fatty acids, which generates H2O2 as a by-product. Peroxisomes possess high concentrations of catalase, so it is unclear whether or not leakage of H2O2 from peroxisomes contributes significantly to cytosolic oxidative stress under normal circumstances. However, a class of nonmutagenic carcinogens, the peroxisome proliferators, which increase the number of hepatocellular peroxisomes and result in liver cancer in rodents, also cause oxidative stress and damage (7, 157, 177, 230). Interestingly, during the regeneration of the liver after partial hepatectomy, there exist peroxisomes that do not stain for catalase activity (232), hinting that during rapid cell proliferation, oxidant leakage from peroxisomes may be enhanced.
Microsomal cytochrome P-450 enzymes metabolize xenobiotic compounds, usually of plant origin, by catalyzing their univalent oxidation or reduction. Although these reactions typically involve NADPH and an organic substrate, some of the numerous cytochrome P-450 isozymes directly reduce O2 to O2· (105, 168) and may cause oxidative stress. An alternative route for cytochrome P-450-mediated oxidation involves redox cycling, in which substrates accept single electrons from cytochrome P-450 and transfer them to oxygen. This generates O2· and simultaneously regenerates the substrate, allowing subsequent rounds of O2· generation (115). Although it is unclear to what extent cytochrome P-450 side reactions proceed under normal conditions, it is possible that such chronic O2· generation by cytochrome P-450 is the price animals pay for their ability to detoxify acute doses of toxins (6).
Finally, phagocytic cells attack pathogens with a mixture of oxidants and free radicals, including O2·, H2O2 , NO·, and hypochlorite (38, 220, 262). Although the massive generation of oxidants by immune cells differs from the above three sources of free radicals to the extent that it is the result of pathogenesis, it is nevertheless a normal and unavoidable consequence of innate immunity. Chronic inflammation is therefore unique among the endogenous sources of oxidants, because it is mostly preventable (49, 231, 245).
In addition to these four sources of oxidants, there exist numerous other enzymes capable of generating oxidants under normal or pathological conditions, often in a tissue-specific manner (115). To give a single relevant example, the deamination of dopamine by monoamine oxidase generates H2O2 , in some neurons, and has been implicated in the etiology of Parkinson's disease (80). Finally, the widespread catalytic generation of NO·, achieved by various isozymes of nitric oxide synthase and central to processes as diverse as vascular regulation, immune responses, and long-term potentiation, increases the potential routes for destructive oxidative reactions (187). The interaction between O2· and NO· results in ONOO, which is a powerful oxidant.
As originally articulated by Harman (120), the free radical theory of aging did not distinguish between these different sources of oxidants. However, the rate of living hypothesis clearly singled out mitochondrial O2· and H2O2 generation, since it is the mitochondrial respiration rate that negatively correlates with MLSP. Also, as many other established sources of oxidants are tissue specific (associated with hepatic, neuronal, and other specialized functions), they are less likely to explain aging across a broad range of species. For this reason, mitochondrial O2· and H2O2 have captured the lion's share of attention. However, it may turn out that for some age-associated disorders, nonmitochondrial oxidants are critical. In the expanded sense of the free radical theory, any oxidants, mitochondrial or not, may play a role. Therefore, despite the great number of intracellular sources of oxidant that have been identified in a qualitative way, in terms of ranking their relative importance, the field is in its infancy.
C. Targets of Oxidants
What are the targets of endogenous oxidants? The three main classes of biological macromolecules (lipids, nucleic acids, and proteins) are susceptible to free radical attack, and there is plentiful evidence that all suffer oxidative damage in vivo. Although it is well beyond the scope of this review to treat the biochemistry of oxidative damage in any great depth, the area has been expertly reviewed (115). A synopsis of the better known pathways of oxidative damage, however, is warranted; the most familiar end products are described here.
The earliest research on the destruction of biological molecules by oxidants involved lipids (109). Food chemists have long understood that the rancidity of fats results from peroxidative chain reactions in lipids ("autoxidation"); a lipid hydroperoxyl radical abstracts a hydrogen atom from the double bond of a neighboring unsaturated lipid, forming a hydroperoxide and an alkyl radical, the latter which combines with O2 to regenerate a lipid hydroperoxyl radical capable of initiating another round of oxidation. Ultimately, intramolecular reactions and decomposition yield cyclic endoperoxides and unsaturated aldehydes, the latter of which are reactive and may act as mutagens (194) or inactivate enzymes (39, 322), or operate as endogenous fixatives, reacting with proteins and nucleic acids to form heterogeneous cross-links (42). Moreover, a primary effect of lipid peroxidation is decreased membrane fluidity, which alters membrane properties and can significantly disrupt membrane-bound proteins (324).
Oxidative damage to nucleic acids includes adducts of base and sugar groups, single- and double-strand breaks in the backbone, and cross-links to other molecules. The spectrum of adducts in mammalian chromatin oxidized in vitro and in vivo includes more than 20 known products, including damage to all four bases and thymine-tyrosine cross-links (70, 71, 113). The electrochemical properties of the adduct 8-oxo-guanine (oxo8gua) and the deoxynucleoside 8-oxo-2,7-dihydro-2'-deoxyguanosine (oxo8dG), which have permitted the coupling of extremely sensitive electrochemical detection to high-performance liquid chromatography (HPLC), have resulted in hundreds of studies of its formation, accumulation, and excretion (17). The identification of specific enzymatic repair of oxidative lesions has recently provided both proof of the significance of oxidative DNA damage as well as tools to manipulate the load of damage in vivo by genetic knockout (17, 23, 78, 192, 266, 291).
The oxidation of proteins is less well characterized, but several classes of damage have been documented, including oxidation of sulfhydryl groups, reduction of disulfides, oxidative adduction of amino acid residues close to metal-binding sites via metal-catalyzed oxidation, reactions with aldehydes, protein-protein cross-linking, and peptide fragmentation (317, 318). A particularly intriguing recent development has been the realization that a number of enzymes possessing active-site iron-sulfur clusters are acutely sensitive to inactivation by O2· (86, 176). For example, E. coli aconitase is inactivated by O2· with a rate constant of 109 M1·s1 (95, 96). Mammalian mitochondrial aconitase is inactivated in vitro and in vivo by treatments that increase mitochondrial O2· generation, such as growth under hyperbaric conditions (97, 98). Because aconitase participates in the citric acid cycle, its inhibition would be expected to have pleiotropic effects. Moreover, the mechanism of aconitase inhibition by O2· has been demonstrated to involve the release of free iron from the enzyme (86). Free iron atoms catalytically exacerbate oxygen stress (see below), and it has been proposed that superoxide's genotoxicity is a function of its ability to liberate protein-bound iron (159, 184).
Unlike lipids and nucleic acids, proteins represent a very diverse target for oxidative damage. Although protein oxidation has been demonstrated at the level of the peptide backbone and amino acids, there has been relatively little scrutiny of differences between proteins in their sensitivities. A detailed quantitative comparison of bovine serum albumin and glutamine synthase has shown susceptible residues of the former (methionine and the aromatic amino acid residues) to be oxidized about twice as fast as those on the latter, implicating all four levels of protein structure in relative susceptibility (19). A study of the oxidation sensitivities of a various cloned K+ channels from T lymphocytes, cardiac cells, and neurons revealed that whereas five of the cloned channels were highly sensitive to oxidation, an equal number were resistant (74). Differential sensitivities raise the possibility that the loss of homeostasis that is a hallmark of aging could result from the selective oxidation of proteins.
In the context of aging, a particularly relevant aspect of oxygen's toxicity is its promotion by some metals and by elevated O2 partial pressure. Iron and copper catalyze the homolytic cleavage of ROOH (the Fenton reaction), leading to the generation of ·OH (115). It is ·OH that is the most reactive oxidant, reacting at diffusion-limited rates. The catalytic properties of iron and copper explain why cells possess metal-chelating proteins such as ferritin and transferrin, which reduce the concentration of redox-active metals (114, 211). In humans, the body's content of iron increases with age (in men throughout their lives, and in women after menopause), and it has been suggested that this accumulation may increase the risk of oxidative damage with age (169, 331). Finally, oxidative stress in vivo is aggravated by increasing O2 partial pressure, due to a more pronounced flux of mitochondrial O2· (37). Consequently, the manipulation of O2 partial pressure is a relatively simple tool that has been used to test the free radical theory.
D. Antioxidant Defenses
Cells are equipped with an impressive repertoire of antioxidant enzymes, as well as small antioxidant molecules mostly derived from dietary fruits and vegetables (5, 351). These include 1) enzymatic scavengers such as SOD, which hastens the dismutation of O2· to H2O2 , and catalase and glutathione peroxidase (GPX), which convert H2O2 to water; 2) hydrophilic radical scavengers such as ascorbate, urate, and glutathione (GSH); 3) lipophilic radical scavengers such as tocopherols, flavonoids, carotenoids, and ubiquinol; 4) enzymes involved in the reduction of oxidized forms of small molecular antioxidants (GSH reductase, dehydroascorbate reductase) or responsible for the maintenance of protein thiols (thioredoxin reductase); and 5) the cellular machinery that maintains a reducing environment (e.g., glucose-6-phosphate dehydrogenase, which regenerates NADPH). The complement of defenses deployed differs not only between organisms or tissues, but even between cellular compartments. For instance, GPX plays an important role in mammals but is absent from flies and nematodes (298, 330), and there exist in humans three forms of SOD (cytosolic Cu,Zn-SOD, mitochondrial Mn-SOD, and extracellular SOD), encoded and regulated independently (91).
As far as supporting the free radical theory of aging is concerned, the universality of antioxidant defenses is good news. Although the nature of these defenses varies between species, the presence of some type of antioxidant defense is universal. In fact, some antioxidants, such as SOD, are very highly conserved. Clearly, an indifference to oxygen free radicals is inconsistent with life, underlining the centrality of oxidative damage. Moreover, the fact that antioxidant defenses are not uniform has been incorporated into the free radical theory; differences in antioxidant defenses between species have been put forth to explain differences in life span. Although there is something uncomfortably ad hoc in these two different interpretations of the data, they are not inconsistent. Whereas aerobic life requires organisms to cope with oxidation to some extent, different evolutionary pressures appear to have selected for more or less investment in these defenses, as is discussed in section III.
Finally, a persistent problem in testing the free radical theory is that antioxidants are both parallel (different antioxidants can play similar roles, e.g., catalase and GPX) and serial (enzymes operate in tandem to decompose radicals to harmless products, e.g., SOD and catalase). Consequently, measurements of individual antioxidant activities do not have great relevance. In fact, as is discussed below, measurements of age-related changes in individual antioxidants have led to conflicting results (297). For this reason, aggregate assays have been devised, such as the susceptibility of crude cellular homogenates to in vitro oxidation by ionizing radiation (2, 300). Although these assays do not provide any information about the specific mechanisms of defense, they conveniently measure overall effectiveness.
F. Synthesis: Interaction of Oxidant Generation, Oxidative Damage, and Repair
The existence of multiple intracellular sources of oxidants and complex defenses has led to refinements of the free radical theory. For example, it is clear that the metabolism of oxygen radicals is dynamic, with damage resulting from an increase in oxidant generation or a decrease in antioxidant defenses. Consequently, a difference in life span between species or individuals could be due to different rates of living, or to different "rates of scavenging" (57, 82). The picture has been further complicated by the discovery of specific enzymatic repair of oxidative damage, leading to "repair" or "fidelity" versions of the theory, in which life span is determined by the failure to correct oxidative damage (137, 239).
The relationship among these three components of oxidative stress: oxidant generation, antioxidant protection, and repair of oxidative damage, and the way in which they have been investigated in testing the free radical theory, is illustrated schematically in Figure 1. Increases in oxidant generation, and decreases in antioxidant protection and repair systems, are among the theory's testable predictions and have been examined both as a function of age in individuals of the same species, as well as between species of differing MLSP.
View larger version (43K):
[in this window]
[in a new window]
| FIG. 1.
The ultimate outcome of oxidative stress is a function of 1) oxidant generation, 2) antioxidant defenses, and 3) repair of oxidative damage. Bolds arrows denote oxidative damage, and dashed arrows denote routes for its prevention or repair. Because of the ways in which these processes may interact, multiple positive- and negative-feedback loops are possible. Aging (A) is situated at intersection of these processes. In testing the free radical theory, changes in processes 1-3 have been measured both as a function of age and as a function of species' maximum life span potential. The similarity of the figure to the international emblem of radiation is not a coincidence; the free radical theory has its roots in radiation biology.
|
|
Finally, an extremely important (if experimentally recalcitrant) aspect of the interactions between oxidants, antioxidants, and repair are feedback loops, positive and negative, between them. Antioxidant defenses and cellular repair systems have been shown to be induced in response to oxidative challenges (67, 119, 320) and are of course potential targets of oxidative destruction (145). Also, the generation of oxidants may be enhanced by the malfunctioning of oxidatively damaged molecules (28, 303). Therefore, with the examination of Figure 1, it is not difficult to envision ways in which primary oxidative destruction of any target (e.g., the components of the mitochondrial ETC, scavenging enzymes such as SOD, or DNA repair enzymes) might promote further oxidative damage in what is frequently called a "catastrophic vicious cycle."
Although such cycles are intuitively appealing, their documentation awaits future work and will be extremely difficult from a technical standpoint. An alternative to lab-based approaches, namely, the computational modeling of these complex interactions in what has been termed a "Network Theory of Ageing," is being pursued by theoretical gerontologists (170). Ultimately, a question of obvious importance is whether or not such cycles, if they exist, could be broken, and modeling may help pinpoint weak links for therapeutic intervention.
As the free radical theory has gained ground, it has incorporated other ideas. For example, as mentioned above, the rate of living hypothesis dovetailed with the free radical theory once mitochondrial free radical generation was confirmed. Three other ideas that have been influential are the evolutionary concept of antagonistic pleiotropy, the somatic mutation theory of aging, and the mitochondrial theory of aging.
The phenomenological approach to the free radical theory has involved looking for traces of oxidative damage in vivo. Phenomenology is not well suited to critically testing the free radical theory, since the data (which are voluminous and generally supportive) mainly represent correlations. Documented increases in oxidative damage, no matter how impressive, may be a consequence of a primary nonoxidative event. Nevertheless, phenomenology is the foundation upon which more powerful experiments depend, since the analytical methods developed for it have been used to compare species, genetic mutants, and populations with differing life spans. In fact, almost all of the biomarkers of oxidative stress described in this section have been found to accumulate at a faster rate in short-lived species, and in many cases, this rate correlates with O2 consumption. Familiarity with the most frequently measured end points is a prerequisite to assessing the free radical theory.
If oxidative damage is a significant cause of cellular degeneration, then one expects to see more of it in older individuals. Oxidative damage has been described in terms of "accumulation, modification, and depletion": accumulation of end products of oxidative damage (such as lipofuscin), modification of existing structures (such as oxidative adducts in DNA), and depletion (such as the loss of enzymatic activity or reduced thiols).
A. Accumulation of Oxidative End Products
The gradual and steady accumulation of intracellular yellow-brown fluorescent pigments, referred to as lipofuscin, occurs in numerous phyla. Lipofuscin arises prominently in postmitotic cells (where, it is argued, it remains undiluted by rounds of cell division; Ref. 296) and is located in small granules in secondary lysosomes. Lipofuscin is structurally complex and variable, consisting mostly of cross-linked lipid and protein residues (251, 296, 327), and is ubiquitous, documented in species as diverse as nematodes, fruit flies, rats, bees, crab-eating monkeys, and crayfish. Most important, it is abundant in aged tissues, where it may occupy more than one-half of the volume of the cell (347, 348).
Early on, it was discovered that incubation of amino acids with the lipid peroxidation product malonaldehyde under acidic conditions leads to the formation of lipofuscin-like fluorophores (42). The plausibility of such a reaction, given the contents and low pH of lysosomes, suggested that lipid peroxidation in vivo leads to the formation of lipofuscin (324). Other in vitro studies of lipid peroxidation have since uncovered a great number of routes to fluorescent, cross-linked products via promiscuous oxidative chemistry, which suggests that lipofuscin is a biomarker of lipid peroxidation (160, 296, 348).
Despite extensive in vitro experiments, it is not known with certainty how lipofuscinogenesis occurs in vivo, nor how lipofuscin comes to accumulate with age. Lipid peroxidation could occur throughout the cell and be followed by lysosomal phagocytosis and cross-linking of peroxidative by-products, in which case an age-related increase in lipofuscin content could be seen as the result of oxidative damage. Alternatively, an age-associated decline in lysosomal activity (due to something besides oxidation) might increase the residence time of phagocytosed material enough to enhance lipofuscinogenesis in situ from constant amounts of peroxides (135). In support of the latter possibility, infusion into rat brains of lysosomal proteinase inhibitors leads to the rapid accumulation of lipofuscin-like granules (149). This scenario suggests that lipofuscinogenesis may be a consequence, not a cause, of aging.
Experiments with cultured cardiac myocytes have established roles for both oxidative damage and lysosomal turnover (28). Lipofuscin accumulates in these cells in culture, and growth under increasing O2 partial pressure from 5 to 40% markedly enhances its accumulation (306). Inclusion of iron in the growth medium further increases lipofuscinogenesis, and the iron chelator desferal depresses it, suggesting that Fenton reaction-generated ·OH is an initiator (200). Finally, antioxidants inhibit lipofuscin formation in cultured cardiomyocytes, whereas lysosomal protease inhibitors increase it (199, 201, 202).
Even if oxidative damage is primarily responsible for depositing lipofuscin in the lysosomes of senescing animals, is it more than a biomarker of aging? It has been theorized that lipofuscin accumulation is likely to impair autophagy, as more lysosomal volume is occupied by the indigestible material (28). Because lysosomes are responsible for the recycling of materials and organelles, their failure may include the following: 1) a delay in mitochondrial turnover (with a concomitant decrease in mitochondrial efficiency or an increase in mitochondrial oxidant generation), 2) an accumulation of oxidatively modified proteins and lipids in the cytosol awaiting degradation (potentially aggravating cytosolic lipid peroxidation), 3) an accumulation of lipofuscin-bound iron in a redox-active form (which might promote further intralysosomal lipid peroxidation), and 4) the disruption of lysosomal membranes (and the spillage of hydrolytic enzymes into the cytosol). Although these speculations (28) remain to be substantiated, it has been shown that when treated with sublethal doses of H2O2 , cultured cells display lysosomal disruption and leakage of the lysosomal compartment into the cytosol (29). Also, it has been demonstrated that the sensitivity of cultured primary hepatocytes to oxidation, which was associated with a loss of GSH and an influx of Ca2+, was prevented by the iron chelator desferal (235). What is intriguing about these results is the fact that whereas desferal stabilized the lysosomes, it did not prevent the loss of GSH or the increase in intracellular calcium, so it may be that it is lysosomal leakage per se, rather than peroxidative damage, which is the actual lethal step in this model of oxidative killing.
B. Steady-State Levels of Oxidative Modification
Unlike cytosolic proteins whose half-lives are measured in minutes or hours, some extracellular proteins are rarely recycled, and oxidative modification of these old macromolecules occurs. A class of fluorescent cross-linked molecules that is distinct from lipofuscin forms on long-lived proteins such as collagen and lens crystallin (216). These modifications are initiated by the reaction of reducing sugars with free amino groups (glycation), a chemical sequence that is unrelated to oxidation and results in a molecule known as an Amadori product. Further nonoxidative rearrangements result in stable, cross-linked advanced glycation end products (AGEs) (35), whose absolute abundance appears to be an excellent biomarker of age (217). Recently, it was discovered that oxidation is one fate of the Amadori product. Pentosidine, the name given to a cross-link involving arginine, lysine, and pentose moieties, is one such "glycoxidation product," the formation of which requires the presence of O2 (326). It appears as if Amadori products themselves are a source of H2O2 in vitro, which then accelerates glucose-mediated fluorogenic collagen cross-linking in a catalase-sensitive fashion, although it is unclear to what extent this occurs in vivo (77, 217). As is the case with other AGEs, the tissue burden of pentosidine is elevated in diabetics, as a consequence of hyperglycemia.
Pentosidine has been found to accumulate as a function of age in shrews, rats, dogs, cows, pigs, monkeys, and humans, yielding equivalently shaped curves in all cases (284). It is not clear, however, how glycooxidative modifications might contribute to degeneration. It has been proposed that cross-linking in cartilage is related to its decreased elasticity and relative resistance to proteolysis in old animals (9). However, the absolute amount of collagen pentosidine cross-links attained at death is much higher in long-lived than in short-lived species: 6-7 pmol/mg in 3.5-yr-old shrews, 15-18 pmol/mg in 25-yr-old monkeys, and 50-100 pmol/mg in 90-yr-old humans (284). In other words, it appears that the rate of pentosidine accumulation may merely be a measure of more rapid oxidative damage in short-lived species rather than an actual cause of dysfunction.
An intriguing twist to this story has been the cloning of a specific cellular receptor for AGE, called RAGE (receptor for AGE), that belongs to the immunoglobulin superfamily and is expressed by mononuclear cells and the vascular endothelium (277, 278). One of the effects of AGE binding by RAGE is the generation (in mononuclear cells) of intracellular oxidants, the activation of the oxidant-sensitive transcription factor NF
B, and the induction of downstream events linked to atherogenesis (279, 280). As discussed in section XVB, it has recently been shown that RAGE, which is highly expressed by microglial cells in the brain (142), is a receptor for amyloid -peptide (A). The RAGE binding of A results in oxidant generation, implicated in the etiology of AD (293, 344).
Several amino acid residues in proteins are susceptible to oxidative modification, forming side chain carbonyl derivatives (317). The development of sensitive methods for the analysis of protein carbonyls by Stadtman and co-workers (181) enabled them to study oxidative modification in human brain tissue and cultured fibroblasts, and in rat liver. They found a two- to threefold rise in protein carbonyl content between young and old age, an increase from 10 to ~30% of the total protein pool (315). The increase was exponential and correlated well with decreased activity of the oxidation-sensitive enzyme glucose-6-phosphate dehydrogenase (G-6-PD). In comparison, the rise in protein oxidation in the mongolian gerbil was less dramatic, increasingly significantly in brain, heart, and testis, but not in kidney. As in human tissues, trends for the activity of G-6-PD correspond to the increased damage, falling in brain and heart but not in kidney (300). Similar results have been reported in an insect model. An age-associated 2.5-fold increase in the protein carbonyl content of old versus young houseflies has also been documented (299), and as in humans, the increase occurs exponentially during the life span. The similarity of the degree and pattern of increase in insects and mammals is striking, considering the enormous difference in their MLSP (40 days vs. 100 yr). Moreover, protein carbonyl levels increase similarly in mitochondrial extracts from the thoracic flight muscles of these animals (303). Mitochondrial aconitase is particularly prone to oxidative modification during aging in vivo and was identified by the immunoblotting of housefly mitochondrial protein extracts with a monoclonal antibody designed to detect protein carbonyls (343b). Carbonylation of this key citric acid cycle enzyme increased in parallel with a decline in its activity.
Somewhat stronger evidence that protein oxidation may play a causative role in senescence comes from comparisons of "crawlers" versus "fliers" of the same aging cohort. Although the two groups share the same chronological age, crawlers are phenotypically senescent individuals that have lost the ability to fly and have a shorter remaining average life span than do fliers (e.g., 9.0 days vs. 13.3 days for 10-day-old crawlers and fliers, respectively). The protein carbonyl content of crawlers was 29% higher than that of fliers (299), reflecting their greater phenotypic age, as was the degree of carbonyl modification of mitochondrial (but not cytosolic) aconitase (343b). Humans suffering from Werner's syndrome, a disease characterized by premature senescence, are individuals whose phenotypic aging is also accelerated, and they too appear to have more extensive protein oxidation. Fibroblasts from Werner's patients of all ages have a level of protein carbonyls equivalent to that in 80-yr-old controls (233). In a creative study attempting to correlate protein oxidation to a physiologically relevant end point, it was shown that in old mice, interanimal variation in protein carbonyl content of two different areas of the brain (cerebral cortex vs. cerebellum) was associated with parallel interanimal variation in memory and motor function deficits (90).
Are protein carbonyls physiologically relevant, or are they merely markers? What are the actual consequences of protein modification? Unfortunately, there are few quantitative data with which to answer this question, although qualitative data exist. The fate of oxidized proteins may depend on the form of damage. For example, metal-catalyzed oxidation of G-6-PD by iron/citrate results in a thermolabile enzyme that is a better substrate for proteolysis than is the native enzyme (93). Rapid turnover of metal-oxidized G-6-PD may therefore proceed efficiently. On the other hand, G-6-PD modification by 4-hydroxy-2-nonenal, a lipid peroxidation product, also inactivates the enzyme but does not render the enzyme thermolabile or increase its degradation by proteases (322). This difference exists despite the fact that in both cases, the same lysine residue is affected. To make matters more complex, the cross-linking of G-6-PD multimers by 4-hydroxy-2-nonenal (which predictably results in a product with lipofuscin-like fluorescence) produces a molecular species that actually inhibits the multicatalytic protease (92). The physiological cost of protein oxidation is presently an unknown quantity.
The appearance of protein-bound 3,4-dihydroxyphenylalanine (DOPA) on ·OH-damaged proteins has been characterized; when converted to a quinone, protein-bound DOPA can undergo redox cycling, generating O2·. It has therefore been proposed that protein oxidation may contribute to the progression of aging not merely by the loss of protein function, but also by an acceleration of the flux of oxidants (61, 63, 64, 69, 101, 102).
The oxidative modification of DNA has also been studied in animals of different ages, with conflicting results. Although some studies have reported a modest increase in specific oxidative adducts, single-strand breaks, and abasic sites, others have been negative (23, 132, 156, 221, 337). The failure to detect an age-related increase in oxidative adducts by the analytical chromatographic techniques typically employed may have been due to the difficulty of working close to the limit of sensitivity (17). In fact, it has become apparent that the measurement of the adduct oxo8dG is frequently plagued by artifacts (29a, 44b, 127c, 156a, 248a) and that these may have compromised some published experiments. Of particular concern are measurements of oxo8dG in mtDNA (16), which have generally been higher than in nuclear DNA, but which may be particularly prone to artifacts associated with the analysis of small samples (16, 127c). Moreover, it is noteworthy that even among the highly variable published estimates of oxo8dG in mtDNA are values that are equivalent to the lowest measured values of oxo8dG in nuclear DNA (131a). Because of the small number of studies of mtDNA and the high variability between the measured values, it is not yet possible to conclude that mtDNA is, in fact, more heavily oxidized than nDNA. Encouragingly, alternative PCR-based methods for measuring oxidative damage have recently been used to compare oxidation of mtDNA and nDNA by exogenous oxidants, with the result that the former appears more sensitive than the latter (270a, 343a), although these studies could not quantify baseline values of damage. With methodological improvements, future experiments may be more conclusive. For instance, the use of single-cell gel electrophoresis (the comet assay) to measure single-strand breaks and abasic sites in whole rat hepatocytes in situ revealed a statistically significant 1.5-fold increase in old rats compared with young rats (131) (although this experiment did not distinguish between oxidative and nonoxidative damage).
In any case, even if the burden of oxidative adducts does increase with age, there is virtually no information about the likely effect of oxidative DNA damage in vivo, apart from the knowledge that it leads to mutations and cancer. The fact that there is active DNA repair in postmitotic tissues (in which the danger of mutation due to replication is nonexistent), and that such repair is often targeted to transcribed regions of the genome, suggests that DNA damage itself interferes with gene expression and is not tolerated (116, 117). This important question deserves more attention.
D. Age-Associated Trends in Antioxidant Defenses and Repair
What is the cause of age-related oxidative damage? It could result from less active antioxidant defenses and repair, but studies that have measured age-related changes in antioxidant defenses have generated conflicting results. Recent measurements of antioxidants in mongolian gerbils (300) and mice (215) are representative of the types of patterns that have been uncovered in many other studies (65, 248, 254, 263, 274, 301, 302, 305, 329). In various tissues of gerbils, there was not a consistent pattern of change; increases in SOD and decreases in GSH were observed, whereas GPX was equivalent at different ages and catalase increased or decreased, depending on the tissues and the age at analysis. In mouse brain, on the other hand, significant decreases in SOD, catalase, and GSH reductase were observed, although GPX levels were unchanged.
Another complication is that defenses are induced in response to stress. Therefore, a higher level may indicate better protection, or alternatively, greater need for antioxidant defenses due to an increase in oxidant generation. Studies of antioxidants in rat heart and skeletal muscles illustrate this point. In heart, decreases in cytosolic SOD and GPX and increases in mitochondrial SOD and GPX were noted in older animals, and several indexes of oxidative damage were also elevated (151). From these results, it was concluded that although overall myocardial antioxidant defenses were weakened in the older animals, they were induced in mitochondria as a compensatory response. In skeletal muscles, in contrast, increases were observed in both cytosolic and mitochondrial forms of all of the enzymes studied (150), despite the fact that indexes of lipid peroxidation were again elevated; in this case, it was concluded that both cytosolic and mitochondrial antioxidants were induced. The credibility of these hypotheses is not in question, but it is hard to see how they could be disproved. When these and similar studies of age-related antioxidant levels are combined, what remains is a confusing assemblage of ambiguous trends.
Of course, interactions between antioxidants are complex, which aggravates the problem. To avoid the problems posed by assays of individual antioxidants, aggregate measures of antioxidant defenses have been devised. A crude but integrative measure of antioxidant defenses, for instance, is the susceptibility of a homogenate to induced oxidation. X-irradiation of a whole body homogenate of houseflies results in a linear, dose-dependent increase in protein carbonyls. When homogenates of old and young flies are compared, the rate of induction of protein carbonyls by X-irradiation is 45% higher in 14- than 5-day-old flies. This suggests that the antioxidant defenses in older flies are less able to cope with oxidative stress. Moreover, the activity of G-6-PD, an enzyme known to be sensitive to oxidation, decreases upon X-irradiation of living flies, and does so to a greater extent in old than young animals (2). When this assay was applied to the gerbil samples described above, in which no overall change in antioxidants was seen, a clear difference between young and old tissues emerged. Whereas 6 krad of X-irradiation induced a 20-38% increase in protein carbonyls in 5-mo-old animals, it induced a 152-211% increase in 26-mo-old animals (300). Similarly, although synaptosomes from young and old mice contain equivalent amounts of ATP and GSH, those of old mice were far more sensitive to GSH depletion by the diethyl maleate than those from young mice (197). Lastly, reperfusion injury is a well-established model of oxidative stress associated with the reestablishment of blood flow following ischemia, and it causes greater oxidative damage to heart tissues of old rats than young ones (192a). The use of a polyclonal antiserum specific for adducts between lipid peroxidation end products and proteins detected such covalent modifications of mitochondrial proteins from old but not young animals, which was associated with a more dramatic loss of respiratory capacity in the former. Whereas the baseline mitochondrial respiratory parameters (before ischemia-reperfusion) did not differ between young and old animals, the administration of a physiologically relevant stress revealed a probable age-related decline in antioxidant defenses.
Another alternative to measuring absolute levels of antioxidants in old versus young animals is to investigate the ability of animals of different ages to induce antioxidants, an approach that has been applied to the analysis of SOD in the nematode Caenorhabditis elegans (59). Whereas in young animals challenge with hyperoxia or the redox cycling compound plumbagin resulted in an increase in SOD activity, in middle-aged or old animals it actually resulted in a net loss of activity.
What about repair of oxidative damage? Does its activity decrease with age? The bulk of evidence suggests that there is probably not an overall age-associated change in the intrinsic ability of cells to degrade damaged proteins (94, 270). Although a dramatic decrease in the activity of the oxidized protein-specific alkaline protease has been reported in old rat hepatocytes (318), no change in this activity was measured in the heart or brain of 25- versus 5-mo-old gerbils (300). In a separate work, a 50% age-related decline in a single activity (peptidylglutamyl-peptide hydrolase activity) of the hepatic multicatalytic protease was associated with its selective sensitivity (relative to the multicatalytic protease's other activities) to metal-catalyzed oxidation, suggesting that resistance to oxidants may (logically) characterize the proteases responsible for degrading damaged proteins (46).
Although the intrinsic protease activity may not decrease with age, there is evidence that repair of oxidized proteins may be less easily induced in response to an oxidative insult in old animals. For instance, exposure of young and old rats to 100% O2 increased the content of protein carbonyls in both groups over a 48-h period. Between 48 and 54 h of exposure, however, alkaline protease activity was induced in young animals, with a corresponding decrease in protein carbonyls to initial levels. In old animals, on the other hand, no increase in activity was observed, and protein carbonyl levels continued to rise throughout the time course (318).
There is circumstantial evidence from mutagenesis studies that either antioxidant defenses or repair of oxidative DNA damage (or both) is less efficient in old mice. The induction of somatic mutations in mice by 
-irradiation is from 2.3- to 3.6-fold higher in old than in young animals, depending on the dose (99). The induction of mutations in young and old animals was reduced by feeding the animals a cocktail of dietary antioxidants, confirming that oxidants played a mutagenic role in these experiments. Therefore, the more pronounced induction of mutations in older mice is indirect evidence of decreased antioxidant defenses and repair (99). Later experiments employing peripheral lymphocytes from young and old human subjects resulted in similar results (100). The ability of human peripheral lymphocytes to repair oxidative DNA damage induced by H2O2 has also been found to be less efficient in cells from older donors (14).
Altogether, the results above suggest that older cells may be less able to prevent oxidative damage from occurring, and less effective at removing the damage once it has occurred. There is a clear need for more and better data about age-related trends in defenses and repair.
All of the data discussed in the previous section identify age-related differences within a given species. A complementary, and in some ways more powerful approach, is to compare species that have different MLSP. Comparative biochemistry and physiology, inspired by the rate of living hypothesis, has played a key role in establishing the free radical theory. Several representative studies are described below.
Top
References
The free radical theory of aging, conceived in 1956, has turned 40 and is rapidly attracting the interest of the mainstream of biological research. From its origins in radiation biology, through a decade or so of dormancy and two decades of steady phenomenological research, it has attracted an increasing number of scientists from an expanding circle of fields. During the past decade, several lines of evidence have convinced a number of scientists that oxidants play an important role in aging. (For the sake of simplicity, we use the term oxidant to refer to all "reactive oxygen species," including O
