‘Reactive species’ (RS) of various types are formed in vivo and many are powerful oxidizing agents, capable of damaging DNA and other biomolecules. Increased formation of RS can promote the development of malignancy, and the ‘normal’ rates of RS generation may account for the increased risk of cancer development in the aged. Indeed, knockout of various antioxidant defence enzymes raises oxidative damage levels and promotes age-related cancer development in animals. In explaining this, most attention has been paid to direct oxidative damage to DNA by certain RS, such as hydroxyl radical (OH•). However, increased levels of DNA base oxidation products such as 8OHdg (8-hydroxy-2′-deoxyguanosine) do not always lead to malignancy, although malignant tumours often show increased levels of DNA base oxidation. Hence additional actions of RS must be important, possibly their effects on p53, cell proliferation, invasiveness and metastasis. Chronic inflammation predisposes to malignancy, but the role of RS in this is likely to be complex because RS can sometimes act as anti-inflammatory agents.
In 1996, I co-authored in the Biochemical Journal an article speculating on the role of oxygen radicals and other ‘reactive species’ (RS; defined in the legend to Figure 1) in cancer . Some 10 years later, have we established their role? Yes, the currently available data show them to be important players. But how do they act? We have greatly expanded our knowledge of the molecular mechanisms by which RS can modulate cellular processes relevant to cancer (reviewed in ), but have not really ascertained which are their most important actions. In this review, I will first present the evidence that RS are important in cancer, ask what RS are capable of doing, consider the evidence that some or all of these actions occur in vivo, and finally attempt to assess the importance of these events to the origin and progression of cancer.
How cells respond to increasing exposure to reactive species
EVIDENCE THAT REACTIVE SPECIES ARE IMPORTANT IN CANCER
It was shown as early as 1984 that exposure of mouse fibroblasts to ROS (reactive oxygen species) can lead to transformation [3,4]. However, just the right level of exposure is needed; too many RS can injure or kill cells (Figure 1). Indeed, ROS have a huge range of potential actions on cells (discussed further below; see also Figure 1), and one could easily envisage them as anti-cancer (e.g. by promoting cell-cycle stasis, senescence, apoptosis, necrosis or other types of cell death, and inhibiting angiogenesis) or as pro-cancer (promoting proliferation, invasiveness, angiogenesis, metastasis, and suppressing apoptosis). The same can be said about other types of RS, such as RNS (reactive nitrogen species) (Figure 1). For example, NO (nitric oxide) and other RNS can promote DNA damage and cause mutations, and inhibit caspases, delaying apoptosis. On the other hand, they can inhibit cytochrome oxidase and slow mitochondrial ATP formation, impairing cell proliferation and slowing tumour growth. Indeed, that might be the host's ‘purpose’ in NO• production by iNOS (inducible nitric oxide synthase) in macrophages that sometimes gather around tumours [9,10].
So which predominates in cancer – good or bad effects of RS? One way to approach this question is to raise RS levels by decreasing antioxidant defences, and see what happens. The results are clear-cut. Knockout mice lacking CuZnSOD (copper- and zinc-containing superoxide dismutase), a major cellular scavenger of superoxide radicals (O2•−), have increased rates of liver cancer development later in life . Although CuZnSOD is present in most parts of the cell , the most important scavenger of O2•− in the mitochondrial matrix is MnSOD (manganese-containing superoxide dismutase) [12,13]. Absence of this enzyme in mice is lethal soon after birth, indicating the potential of excess mitochondrial O2•− to cause severe damage [12,13]. Heterozygous animals (50% of normal mitochondrial MnSOD) survive, but show increased risk of developing lymphomas, adenocarcinomas and pituitary adenomas as they age . MnSOD and CuZnSOD catalyse the reaction:
The H2O2 generated by SOD enzymes in mitochondria and cytosol is probably largely removed by peroxiredoxins, which are thioredoxin-dependent peroxidase enzymes, although GPxs (glutathione peroxidases) also contribute [2,5]. There are four GPx enzymes in animals; simultaneous knockout of two of them (GPx1 and GPx2) in mice led to intestinal cancers , whereas elimination of some of the peroxiredoxins promotes development of lymphomas, sarcomas and adenomas . Embryonic fibroblasts from mice lacking peroxiredoxin-1 showed elevated ROS production, evidence of elevated c-myc expression, and increased susceptibility to transformation by ras overexpression . Mice lacking catalase sometimes show increased incidence of ‘spontaneous’ mammary tumours .
The data seem clear: elevated RS can lead to more cancer. But how?
DIRECT DAMAGE TO DNA BY REACTIVE SPECIES
This is the most commonly offered explanation. Indeed, exposure to ionizing radiation has long been known to favour cancer development. Radiation-induced carcinogenesis appears to involve initiation, promotion, activation of proto-oncogenes and inactivation of stability- and tumour-suppressor genes. Some genetic damage by radiation occurs by direct absorption of energy by DNA, but much is mediated by the formation of the highly reactive hydroxyl radical (OH•) [19,20]. Thus OH• could be involved in all stages of radiation-induced carcinogenesis. Hydroxyl radical attack upon DNA generates multiple mutagenic purine, pyrimidine and deoxyribose oxidation products [1,2,19–22]. One of the most studied is 8OHdG (8-hydroxy-2′-deoxyguanosine).
The incidence of most cancers rises with the fourth or fifth power of age in animals, e.g. approx. 35% of humans have cancer by age 85. It has been hypothesized that this is largely due to a lifetime of attack by RS such as OH• [23,24], generated endogenously and sometimes in addition by certain exogenous carcinogens, such as cigarette smoke . Are endogenous RS production and oxidative DNA damage rates large enough to account for the age-related development of ‘spontaneous’ cancers? The steady-state levels of DNA base oxidation products in human cells vary according to the measurement method and between laboratories (reviewed in [20,22,24,26]). The average value for all lesions may be approx. 1 per 105 DNA bases or more. Is this enough to cause cancer? By comparison with the levels of adducts of known carcinogens that are detected in carcinogen-exposed animals, it could be (for a detailed analysis of the numbers, see ). Of course, one can always debate the relative mutagenicity of xenobiotic carcinogen adducts compared with that of RS-derived DNA lesions. However, many of the latter are significantly mutagenic, they are formed in our DNA since conception, and the levels present could have even more impact if damage is located in specific genes that are relevant to cancer, as has been suggested in some studies. For example, oxidative DNA damage in human gastric mucosa infected with Helicobacter pylori was unevenly distributed between genes, being especially elevated in the p53 gene, for example . Reactive chlorine, bromine, sulfur and nitrogen species can also attack DNA [2,9]. Some RNS (e.g. N2O3) deaminate DNA bases to mutagenic lesions, and peroxynitrite (ONOO−, a product of reaction of O2•− with NO•), forms 8-NG (8-nitroguanine) (plus several other products) in DNA. 8-NG rapidly detaches from the DNA, leaving potentially mutagenic apurinic sites [9,28]. RNS have been suggested to be especially important in causing DNA damage (measured as 8-NG formation) in chronic inflammation induced by the Opisthorchis (liver fluke) parasites and in the stomachs of humans infected with H. pylori [9,28]. Reactive chlorine species can chlorinate cytosine and adenine (reviewed in ). Attack of RS on proteins and lipids might also aggravate DNA damage (Figure 2).
Some of the ways in which reactive species could facilitate cancer development
DNA need not always be attacked directly to raise levels of mutagenic oxidation products. Defects in some of the enzymes that repair oxidative DNA damage lead to increased levels of 8OHdG and other mutagenic bases in cellular DNA and are associated with increased age-dependent cancer incidence in animals [34–36]. This evidence is less compelling, in general, than that provided by knockout of antioxidant defence enzymes, because many of the enzymes involved in base and nucleotide excision repair act on several lesions in DNA, not just oxidative damage, and so it is possible that non-oxidative mutagenic lesions accumulate as well. However, some data are clear-cut. Just as DNA can be attacked by RS, so can its precursors. Human and other animal cells contain an enzyme (MTH1) that hydrolyses 8OHdGTP, 8- and 2-OHdATPs and 8-chlorodGTP, preventing their incorporation into DNA. Mice lacking this enzyme show an increased rate of spontaneous tumorigenesis with age, especially in the lung, stomach and liver [34,35].
However, it is unlikely that the pro-cancer effects of RS are all due to elevated direct oxidative damage to DNA bases (Figures 1 and 2), simply because there are situations where 8OHdG levels can be elevated, but cancer rates do not increase (reviewed in [26,34], but also see ). Thus oxidative DNA damage may be necessary, but not sufficient, for cancer development, and RS must then act by additional mechanisms. Let us examine what these might be.
REACTIVE SPECIES AND CELL PROLIFERATION
As well as causing direct oxidative damage, RS affect cell behaviour in many ways. Depending on the type and level of RS present, time of exposure, the quantity of cellular antioxidant defences that remove RS and the activities of cellular repair systems that clear oxidative damage, cells exposed to RS can show increased proliferation, halted cell cycle, senescence, apoptosis or necrosis (Figure 1). Participation of RS in the initiation, promotion and progression of cancer could thus involve their effects on the cell cycle, gene expression, direct or indirect damage to DNA, and apoptosis and other forms of cell death (Figure 1). Angiogenesis, carcinogen metabolism and metastasis may also be affected by RS, as will be discussed below. The levels of RS at which cells ‘switch’ from proliferation to halted cell cycle, or undergo any of the other events, are cell-type-specific, e.g. fibroblasts proliferate rapidly at RS levels that impair chondrocyte proliferation . For example, transfection of NIH-3T3 fibroblasts in culture with a ras oncogene led to production of excess superoxide radical (O2•−) by the action of an NOX (NADPH oxidase) enzyme complex, and this O2•− promoted abnormal proliferation . In some other cells, the same experiment (NOX expression in response to ras and formation of more O2•−) can result in senescence . Communication via gap junctions is generally decreased in tumour cells, which may facilitate their proliferation, and excess RS can aggravate this by decreasing gap-junctional communication .
p53 and reactive species
The actions of p53 are intimately linked with cancer and seem similarly intimately linked  with RS (Figure 3). First, the normal low levels of active p53 are involved in maintaining the cellular antioxidant defence network, promoting the transcription of genes encoding MnSOD, GPx1 and proteins that regenerate oxidized peroxiredoxins, among others .
The p53 protein has both pro- and anti-oxidant functions
Secondly, RS can increase p53 activity. Thus the expression of ras oncogenes can accelerate O2•− production, which under some circumstances can increase p53 activity and trigger senescence (although the O2•− can be pro-proliferative instead if p53 function is impaired) . Thirdly, higher p53 activity can cause RS production by several mechanisms, and these RS might sometimes contribute to the cytostatic and pro-apoptotic effect of p53 (Figure 3). Thus the anti-proliferative effects of RS can involve activation of p53, not only indirectly (by producing DNA damage), but also by promoting p53 phosphorylation . One protein involved in the actions of p53 in increasing RS production is p66Shc, a protein that interacts with the mitochondrial electron-transport chain to promote ROS generation . Indeed, mice lacking p66Shc live longer and show decreased levels of oxidative damage . Other mechanisms include p53-dependent increases in the levels of proline oxidase, a mitochondrial enzyme that produces H2O2 as it converts proline into pyrroline 5-carboxylate . Fourthly, p53 seems to promote mitochondrial respiration, by maintaining cytochrome oxidase activity (at least in cell lines) , and mitochondria are a major source of ROS . Indeed, loss of p53 may account for the accelerated rates of glycolysis in some malignant cells .
However, if too many oxidizing RS are generated, they can inhibit p53 action. This protein has ten cysteine residues, some of which are involved in binding a zinc ion required for activity. Oxidation of cysteine inhibits the action of p53 as a transcription factor. For example, p53 can be oxidized and nitrated by ONOO−, and nitrated p53 has been detected in human glioblastoma, a highly malignant brain cancer . Oxidized cysteine residues on p53 can be re-reduced by Ref-1 and the thioredoxin system [5,40]. Oxidative inactivation of p53 will in turn raise cellular ROS levels (Figure 2). Thus the relationship of p53 with RS is complex and intimate, and it is hard to predict which way it will turn in a particular tumour.
RS are involved in the actions of at least some of the other proteins involved in cell-cycle regulation. One is the ATM (ataxia telangiectasia mutated) protein. Patients lacking functional ATM, as well as ATM-knockout mice, show higher levels of oxidative damage, and some of the pathological defects in the mice can be diminished by antioxidant administration [45–48]. One of these effects, bone marrow failure, illustrates one consequence of too many RS; they induce excessive proliferation of haemopoietic stem cells, leading to eventual exhaustion of the stem cell pool .
REGULATING APOPTOSIS BY REACTIVE SPECIES
Moderate levels of RS (and what is ‘moderate’ depends on the cell type) tend to promote apoptosis (Figure 1), or cell death that has at least some of the features of apoptosis, in most ‘normal’ cells [49,50]. However, in some malignant cells, RS can have the opposite effect. For example, in the melanoma cell line M14, decreasing O2•− levels by overexpressing CuZnSOD promoted apoptosis, and decreasing CuZnSOD levels inhibited apoptosis . Generation of ROS by an NOX enzyme system was found to be an anti-apoptotic, pro-proliferative stimulus in pancreatic cancer cell lines , and the ability of mitochondrial AIF (apoptosis-inducing factor) to oxidize NADH and make O2•− contributes to the viability of certain carcinoma cell lines .
How are these ‘pro-survival’ effects achieved? Oxidative inactivation of caspases might sometimes occur (Figure 1). In some cells, elevated O2•− leads to a higher cytosolic pH that deters caspase activation . In contrast, H2O2 can promote apoptosis not only by attacking DNA (usually by its conversion into DNA-damaging OH• by reacting with transition metal ions ) and by damaging mitochondria (Figure 1), but also by lowering the cytosolic pH. The actions of O2•− and H2O2 on cellular pH seem to involve increasing or decreasing respectively the activities of a plasma membrane Na+/H+ exchange system . Another mechanism is for RS to inactivate PTEN (phosphatase and tensin homologue deleted on chromosome 10), leading to increased activation of the Akt pathway, which tends to promote cell survival . However, the relationship of Akt signalling to the whole carcinogenic process is complex, e.g. some studies indicate that activation of this pathway could suppress invasion and metastasis .
A caution on cell-culture studies
Before going further, I must raise an important caveat. One must be very careful in extrapolating from experiments on malignant (or other) cells in culture to the in vivo state. The cell-culture process increases cellular ROS formation. Intracellular O2 levels are low for most cells in the human body, usually <10 mmHg (1 mmHg=0.133 kPa), which diminishes RS formation . Yet most animal cells are cultured under 95% air and 5% CO2. This is grossly hyperoxic (about 159 mmHg) and increases ROS production by mitochondria and cellular oxidase enzymes . To make it worse, cell-culture media are often deficient in antioxidants (e.g. vitamins C and E) and antioxidant precursors (e.g. selenium) and contain ‘free’ iron ions, present as contaminants  or even added deliberately [e.g. iron(III) salts are added to DMEM (Dulbecco's modified Eagle's medium)]. These iron salts can catalyse RS formation in the medium. Indeed, the mild oxidative stress of cell culture may explain why many cell types proliferate so well in culture (Figure 1) [54–56]. Given the speed of ‘natural selection’ in rapidly dividing cultured cells, it is possible that, if such cells are grown in a pro-oxidant environment over several generations, they may evolve to use ROS for signalling pathways that promote proliferation and suppress cell death by mechanisms that may not normally operate in vivo [54,57]. The pro-oxidancy of cell culture has especially confused studies of cellular ‘senescence’ after repeated cell division in culture. Indeed, in mouse embryonic fibroblasts in culture, oxidative stress seemed to be more important than telomere shortening in causing senescence. Culture under hypoxic conditions preserves the replicative potency of many cells, i.e. the ‘Hayflick limit’ is in part an artefact of culture conditions [58–60]. Culture under ambient air may also confuse studies of sensitivity to anti-cancer agents. For example, when Ehrlich ascites cells taken from mouse peritoneum were transferred to cell culture under 21% O2, their sensitivities to such drugs altered radically within a few days, an effect prevented by culture under 4% O2 .
Many papers have appeared claiming that ascorbate and polyphenolic compounds such as flavonoids are toxic to malignant cells in culture and may have therapeutic potential. At least some of these claims are artefacts; the added compounds undergo in vitro oxidation in the culture medium (especially in DMEM) to produce H2O2 and other oxidation products that are the true toxic agents [54,62–64].
METASTASIS AND REACTIVE SPECIES
Are the cellular effects of RS, mostly demonstrated in laboratory-grown cells, really relevant to cancer? We do not really know, but some studies suggest that they are, in the content of invasiveness and metastasis [65–71]. One mechanism of RS action in this context is by modulating integrin expression; another is by suppressing anoikis, the apoptosis that normally results when cells separate from their normal environment. Facilitation of metastasis may also involve changes in cell-surface morphology, intercellular communication, cell mobility and increased vascular permeability. In addition, degradation of extracellular matrix involves MMPs (matrix metalloproteinases), some of which are secreted in latent forms (by both tumour cells and phagocytes, especially macrophages ) that can be activated by RS . MMP enzymes can also liberate growth factors bound to matrix constituents. For example, levels of stromelysin-1 (MMP3) are increased in many breast tumours. The possibility of a ‘vicious cycle’ of events is suggested by the observation that exposing mouse mammary epithelial cells to this enzyme increased their ROS production, causing oxidative DNA damage and genomic instability . Indeed, human metastatic prostate cancer cells seemed to be producing more RS than primary malignant cells , as is evident from higher levels of 8OHdG, 4-HNE (4-hydroxynonenal; an end-product of lipid peroxidation ), and 3-nitrotyrosine (a biomarker of attack on proteins by RNS ). H2O2 can stimulate not only the proliferation but also the migration of human prostate cancer cells, in part by modulating their expression of heparin affin regulatory peptide .
But are these effects important in vivo? Possibly; several studies indicate that experimentally increasing the levels of peroxide-metabolizing enzymes in or around malignant cells injected into animals decreases their metastatic potential [75–78]. The level of lysyl oxidase, an H2O2-producing enzyme, was increased in distant metastatic human breast cancer tissues as compared with primary cancer tissues, and levels were in both cases greater than in normal breast tissue . In mice, weakly tumorigenic and non-metastatic cells became malignant and invasive after co-implantation with a gelatine sponge to induce inflammation, an effect abolished by instilling a SOD preparation . The pro-metastatic effect was also diminished if the mice lacked NOX activity, also consistent with a role for RS in vivo .
Another possibility is that ROS could sometimes promote angiogenesis, e.g. by increasing cellular production of VEGF (vascular endothelial growth factor) [66,79,80]. In addition, the actions of VEGF (and angiopoietin-1) on endothelial cells may themselves involve increased O2•− generation by a NOX [79,81,82]. In bladder cancer, elevated thymidine phosphorylase activity has been shown to correlate with increased angiogenesis and poor prognosis, an effect that may involve ROS-mediated effects on VEGF production . There is the usual biphasic effect of RS, however: too much oxidative damage can impair angiogenesis . Thus we do not know as yet the significance of RS in modulating angiogenesis in vivo.
ARE MALIGNANT CELLS IN A PRO-OXIDANT STATE?
It has been argued that some malignant cells use ROS to suppress apoptosis, accelerate proliferation, metastasis and angiogenesis, and possibly to promote genetic instability, e.g. by increased oxidative DNA damage [2,7,52,66–71,74,85]. Many malignant cells produce high levels of ROS in culture . But is this a cell culture artefact or do malignant cells really make more ROS in vivo? Several experiments reveal that many of them do (reviewed in [20,28,87]). For example, chronic lymphocytic leukaemia cells freshly taken from patients showed increased ROS production compared with normal lymphocytes , as did B-cell lines from patients with Burkitt's lymphoma associated with Epstein–Barr virus infection  and malignant B-cells from patients with hairy cell leukaemia . For solid tumours, however, demonstrating increased RS production in vivo is difficult to achieve owing to methodological inadequacies [2,91], so most investigators have examined oxidative damage levels rather than ROS production. Many such studies have shown increased levels of 8OHdG in human and other animal tumours [2,20,21,92–94]. This is seen both in ‘spontaneous’ cancers and in those induced by a range of carcinogens, despite the fact that the primary DNA-damaging mechanisms of many carcinogens do not appear to involve RS [2,20]. There are exceptions of course, such as the direct involvement of ROS in the carinogenicity of arsenic and of several metals [2,95,96]. In the few cases where it has been examined, elevation in the levels not only of 8OHdG but also of multiple other oxidative DNA base-damage products has been shown; such a pattern is diagnostic of DNA damage by OH• [2,20].
Increased oxidative damage levels in malignant cells could result from: (i) more RS formation with unaltered antioxidant defences; (ii) unaltered RS formation with decreases in antioxidant defences; (iii) failure to repair oxidative damage, so that levels rise; or (iv) any combination of the above. Which is responsible?
Increased formation of reactive species?
Data suggest that at least some malignant tumours in vivo do generate more ROS than normal cells [65,71,77,85,87]. Sources of ROS include lysyl oxidase , proline oxidase , increased mitochondrial O2•− formation and O2•−-generating NOX systems . For example, some human prostate cancers contain mutations in mitochondrial DNA that may increase mitochondrial ROS production . A caveat: do not assume that the increased RS production is an early event, e.g. in primary leukaemia cells, the mitochondrial DNA mutations seemed more associated with damage due to chemotherapy, an effect again perhaps due to increased ROS generation . NOX systems are homologues of the phagocyte O2•−-generating components, and are found in normal as well as malignant tissues. For example, NOX1 is normally present in colon, prostate, uterus and vascular smooth muscle, and appears to be constitutively active . Excitement rose when overexpression of NOX1 in NIH-3T3 cells or prostate cancer cell lines was found to increase O2•− formation, followed by more cell proliferation and the development of some features of malignancy . But is NOX1 more active in human colon cancer? It is not clear as yet; reports are variable (some even say that levels decrease) and its level may depend on the differentiation state of the tumour [39,65,100,101]. In contrast, NOX1 levels are elevated in human prostate cancers, hairy cell leukaemia and some skin cancers [69,90]. Indeed, several authors have suggested that RS may play an especially important role in the origin and progression of prostate cancer [102,103], although this remains a suggestion. Another caveat; NOX activities are usually carefully regulated by cytosolic proteins , so more protein does not necessarily equate to more O2•− production.
Altered DNA-repair rates?
Defects in DNA repair, and possibly polymorphisms in DNA-repair enzymes, contribute to the origins of some cancers . However, for several cancers, some studies report increased activities of DNA-repair enzymes that can act on oxidative DNA damage, whereas others report the opposite [34,35,104,105]. DNA repair may sometimes be down-regulated in the later stages of malignancy, perhaps promoting genetic instability and tumour progression [104,106,107]. Indeed, variability in rates of DNA repair is an important determinant of the success of chemotherapy . But be wary of drawing conclusions based on gene expression or protein levels for repair enzymes, since mRNA or protein do not always equate to activity, e.g. some repair enzymes can be inactivated by RS [2,21]. DNA-repair systems are complex and overlapping, so changes in a single enzyme may have little overall effect. Thus, in general terms, malignant cells do seem able to repair DNA, but it is hard to determine what contributions alterations in repair systems make to increased oxidative DNA-damage levels. Overall, the ability to repair oxidative DNA damage in many human cancers does not seem to be markedly impaired, and may sometimes be increased.
Changes in antioxidant defences?
Tumour cells in culture often show low MnSOD activity. Transfection of a MnSOD gene can suppress the malignant phenotype in some (but not all) of these cell lines [108,109]. Some authors suggest that elevated MnSOD generates more H2O2, which causes inhibition of proliferation, sometimes via p53 activation. Consistent with this, overexpression of catalase can often prevent the effects of excess MnSOD in cell lines . Low activities of CuZnSOD, catalase and GPx1 are also often reported in transformed cell lines [109–111].
Again, one must be careful when extrapolating from cultured cells to those in vivo. Thus, if cells in culture ‘select for’ high levels of RS to help them proliferate, there might be a powerful selection pressure to decrease antioxidant levels (up to a point, that is; that point being when oxidative damage rises to a level that impairs cell function). What do direct measurements of antioxidants in malignant human tissues show? Results are variable [2,31,94,109,111–114], but, in general, there is no clear pattern of major decreases in MnSOD or other antioxidant enzymes in freshly obtained human cancerous tissue. Indeed, levels of MnSOD are sometimes (e.g. in mesothelioma, neuroblastoma, melanoma, stomach, ovarian and breast cancer) elevated [94,111–114]. Some investigators propose that high MnSOD activity is associated with increased tumour invasion, metastasis and a poor prognosis, possibly because a cellular redox imbalance induced by MnSOD overexpression leads to increased activities of MMPs . The levels of peroxiredoxins may also be increased in some (but not all ) malignant cells [111–114], but these enzymes are easily inactivated by ROS , so again remember that protein level or gene expression do not necessarily equate to activity. Probably what is important in vivo is not the level of any one antioxidant enzyme, but the balance of RS production, antioxidant defence (including the ratio of SOD to H2O2-consuming enzymes) and the repair systems in malignant cells.
Overall, the evidence supports the view that at least some malignant cells make more RS in vivo. Antioxidant defence and repair activities may sometimes rise, but not enough to cope with the extra RS. It is thus possible that malignant cells produce more RS for their own benefit, but must be careful not to exceed the level where oxidative damage becomes so severe that cell function is impaired (Figure 1).
AN ANTIOXIDANT PARADOX?
If RS are important in causing cancer, it appears paradoxical that consumption of antioxidant supplements does not decrease cancer incidence in humans, unless one starts with a nutrient-deficiency state [2,115–118]. The paradox is more apparent than real; these antioxidant supplements simply fail to significantly decrease levels of oxidative damage in vivo, and so their lack of effect is predictable [2,119–122]. On a related note, should cancer patients consume antioxidants? The clinical trials conducted to date to address this issue have given little evidence of benefit [2,123–125], again possibly because of failure of the administered antioxidants to decrease oxidative damage [2,119–122]. Indeed, in one study of head and neck cancer, patients were given α-tocopherol (400 units/day) and β-carotene (30 mg/day) during radiotherapy (which acts by increasing DNA damage, probably mostly oxidative, in the tumour cells to an intolerable level ) and for up to 3 years later. There was a tendency to have less severe acute adverse effects during therapy in the test group as compared with the placebo group. Unfortunately, the test group also showed a trend for cancer recurrence rates to be higher .
CACHEXIA AND REACTIVE SPECIES
Advanced malignant disease is accompanied by body wasting, including loss of muscle mass. Cytokines, especially TNFα (tumour necrosis factor α), play an important role, in part by inducing oxidative stress involving increased mitochondrial ROS production (reviewed in [2,127]). Indeed, levels of lipid peroxidation, protein oxidation and 3-nitrotyrosine are increased in the wasted muscles of mice with hepatoma .
CHRONIC INFLAMMATION, CANCER AND REACTIVE SPECIES
Overall, the data from knockout animals show that elevated RS levels can lead to cancer, probably by multiple mechanisms (Figures 1–3). Chronic inflammation is a well-known risk factor for cancer development. Examples include asbestosis, inflammatory bowel disease, pancreatitis, oesophagitis, bile duct inflammation, H. pylori infection, infection with the parasite Schistosoma haematobium (which produces chronic bladder inflammation with increased risk of bladder cancer) and hepatitis, often associated with chronic infection and inflammation caused by hepatitis B and C viruses, or liver flukes of the Opisthorchis genus (reviewed in [2,9,28]). Indeed, a transgenic mouse strain expressing a hepatitis B virus protein in the liver (provoking an immune response) showed chronic hepatitis and elevated liver 8OHdG levels, which preceded the development of hepatoma . Where it has been measured, all of these inflammatory conditions in humans are accompanied by increased levels of oxidative (and sometimes nitrative) DNA damage [9,20,28]. In human melanoma patients, the presence of iNOS and nitrotyrosine was correlated with poor outcome . Oxidative damage is usually measured as 8OHdG levels, although sometimes other oxidatively damaged bases are measured and found to be elevated as well. Methodological caution is needed in such studies, since a biopsy sample may include not only tumour cells but also inflammatory cells , which may be activated to damage their own DNA  and so raise bulk 8OHdG levels in the whole tissue sample. Nevertheless, the available evidence overall supports the occurrence of elevated DNA damage by RS in inflammation-related cancer [2,20].
Just as inflammation can promote cancer development, a tumour often induces an inflammatory response, e.g. it may secrete cytokines that recruit phagocytes and lymphocytes. Many tumours are infiltrated by macrophages . Whether this response is good (helping to suppress tumour growth) or bad (facilitating more mutations in the tumour cells and promoting metastasis, e.g. by secreting proteolytic enzymes) is uncertain and may differ from tumour to tumour [72,130,132]. Animal studies have suggested that progression to malignancy often involves activation of NF-κB (nuclear factor κB), which (among its multitude of possible effects) can suppress apoptosis [133,134]. The effects of RS on NF-κB activation are complex; in some (but by no means all) cells, low to moderate RS levels activate NF-κB (often by increasing net protein phosphorylation), whereas higher RS levels tend to interfere with its action as a transcription factor by oxidizing the protein [2,135–137]. What is defined as ‘moderate’ or ‘high’ depends on the cell type . For example, mice lacking uncoupling protein 2, which modulates mitochondrial O2•− generation, showed increased NF-κB activation, oxidative damage and more tumours in the colon after exposure to the carcinogen azoxymethane , consistent with a link between these events. However, it is not that simple: in a colitis model in mice, knockout of the IKKβ (inhibitory κB kinase β) gene, whose product is needed for NF-κB activation in intestinal epithelium, did not decrease the severity of inflammation (and presumably not therefore RS formation, although it would have been good to measure this), but tumour incidence was decreased by 80%. Interestingly, knockout of the NF-κB system in macrophages in these animals also decreased tumour incidence (by 50%), implying that (in this system at least) the host inflammatory response was pro-carcinogenic . So, although some link of increased RS production to chronic inflammation-induced cancers seems likely, be wary of assuming that it is as simple as inflammation→more RS→cancer.
Mice lacking phagocyte NOX showed less metastasis after injection of fibrosarcoma cells , whereas, in contrast, some studies suggest that iNOS and ONOO− might be anti-metastatic in melanomas . One complicating factor is that there is a feedback loop; whereas RS at the right level might promote carcinogenesis and contribute to inflammation, RS can sometimes suppress their own production. Thus too many RS can damage the phagocytes generating them and can also attack lymphocytes to damp down the immune response . There are several cases in animal models where lowering RS levels leads to increased inflammation, as well as cases where such lowering is anti-inflammatory (reviewed in ). Indeed, it has been suggested that phagocyte RS production can impair T-cell functions in patients with advanced cancer [139,140]. These various contradictions may help to explain the generally very limited effect of antioxidants in the treatment of human chronic inflammations or in the prevention of the cancers that arise from them; also contributing is their ineffectiveness in decreasing oxidative damage, as discussed above (reviewed in ).
So here we have a paradox. The commonly used antioxidant supplements do not scavenge RS well in vivo, and, even if they did, that might sometimes be pro-inflammatory (but it still could conceivably deter later cancer development, i.e. de-link cancer from the inflammation). So what interventions might work better? Anything that stops chronic inflammation, so we need better anti-inflammatory agents.
IMPLICATIONS FOR CHEMOTHERAPY
If malignant cells in vivo are under self-inflicted (and/or host-inflicted) oxidative stress, and that stress contributes to malignancy, then antioxidants that really do act to decrease oxidative damage in vivo might have anti-cancer effects. The opposite approach would be to stress these cells further by providing agents that generate RS. These should have limited effects on normal cells (provided that RS levels do not rise too high), but might, when added to their already elevated RS generation, push malignant cells ‘over the edge’, increasing oxidative damage to a point that the cell cannot cope [87,88,141]. Indeed, it was observed that human leukaemia cells with high ROS levels are more sensitive to 2-methoxyoestradiol than normal cells . This agent appears to increase oxidative stress in cells , although exactly how it works is not clear .
Does increased RS formation thus contribute to the anti-tumour effects of chemotherapy and radiotherapy? Almost certainly so in radiotherapy . Most agents used in chemotherapy act in specific ways, e.g. cytosine arabinoside and cisplatin by directly interfering with DNA replication, taxol by blocking mitotic spindle formation, methotrexate by interfering with the supply of DNA precursors, irinotecan and the anthracyclines by inhibiting topoisomerases (among other mechanisms), farnesyltransferase inhibitors by interfering with the action of small GTPases, adaphostin by inhibiting protein kinases, and proteasome inhibitors by interfering with cell division. Many (and possibly all) of these agents also cause oxidative stress, which has often been suggested to contribute to their side effects [142,144–156]. For example, it seems that mutations in the Bcr/Abl kinase domain involved in the resistance of leukaemic cells to imatinib also result in increased cellular ROS production, leading to DNA damage that contributes further to the development of resistance . Does oxidative stress also aid the anti-cancer effects of the above compounds? It seems likely.
Of course, interfering with the RS–antioxidant balance in cancers has risks. If one gives too many RS generators, normal cells will be at risk. For example, arsenic trioxide (As2O3) may act against acute promyelocytic leukaemia by causing oxidative stress  (and indeed certain antioxidants can block its action in cell culture), but As2O3 can also be carcinogenic and RS are probably involved in the mechanism .
The data with knockout animals that lack antioxidant enzymes, supported by data from some animal knockouts of repair enzymes, strongly support the view that RS contribute to the age-related development of cancer. Direct damage to DNA is probably one key event, but is insufficient alone to produce cancer, suggesting that the ability of RS to suppress apoptosis, and promote proliferation, invasiveness and metastasis (and possibly angiogenesis) are also important. The relative contributions of these various mechanisms are unclear. Cancer associated with chronic inflammation may also involve RS, although they are by no means the only important element. Thus we understand more now than about the importance of RS than we did 10 years ago, but not exactly how they are acting. Our improved knowledge has not yet helped us to deliver better strategies for prevention and treatment (except to discourage healthy people and cancer patients from gobbling huge doses of ‘antioxidant supplements’). Indeed, RS are both good and bad, so they may be difficult to modulate using an ‘antioxidant’ dosage that is expected to work on all subjects. The trend now in cancer treatment is to tailor-make therapies based on the gene expression, cell signalling and proteomic profiles of a tumour. Perhaps we need to do the same for its ‘oxidative stress status’. If it is high, attack either with more ROS or with powerful antioxidants that actually work, perhaps? If it is low, seek another approach. Indeed, levels of gene expression for antioxidant defence enzymes and other proteins related to cellular redox balance have been said to constitute a ‘redox signature score’ that seemed to be predictive of outcome in patients with diffuse large B-cell lymphoma . Techniques to measure oxidative stress and oxidative damage have improved enormously in the past few years (reviewed in [2,91]), although there is still a way to go. Thus development of such a score by combining parameters of oxidative damage with those of redox state, DNA-repair activity and antioxidant defence might be a useful therapeutic tool.
ataxia telangiectasia mutated
copper- and zinc-containing superoxide dismutase
Dulbecco's modified Eagle's medium
inducible nitric oxide synthase
manganese-containing superoxide dismutase
nuclear factor κB
reactive nitrogen species
reactive oxygen species
vascular endothelial growth factor