Review

Introduction

Free radicals and other 'reactive oxygen (ROS)/nitrogen/chlorine species' (for an explanation of these terms see Table 1) are widely believed to contribute to the development of several age-related diseases, and perhaps, even to the aging process itself (Halliwell & Gutteridge, 1999; Sohal et al., 2002) by causing 'oxidative stress' and 'oxidative damage' (terms explained in Table 2). For example, many studies have shown increased oxidative damage to all the major classes of biomolecules in the brains of Alzheimer's patients (Halliwell, 2001; Butterfield, 2002; Liu et al., 2003). Other diseases in which oxidative damage has been implicated include cancer, atherosclerosis, other neurodegenerative diseases and diabetes (Hagen et al., 1994; Chowienczyk et al., 2000; Halliwell, 2000a, 2001, 2002a, 2002b; Parthasarathy et al., 2000). If oxidative damage contributes significantly to disease pathology (Table 3 lists the criteria needed to establish this), then actions that decrease it should be therapeutically beneficial (Halliwell, 2001; Lee et al., 2002a; Liu et al., 2003). If the oxidative damage is involved in the origin of a disease, then successful antioxidant treatment should delay or prevent the onset of that disease (Halliwell, 1991, 2002a, 2002b; Galli et al., 2002; Steinberg & Witztum, 2002). To establish the role of oxidative damage (Table 3), it is therefore essential to be able to measure it accurately. For example, the failure of interventions with antioxidants such as vitamin E, -carotene or ascorbate to decrease disease incidence in several human intervention trials may have simply been due to the failure of these compounds to decrease oxidative damage in the subjects tested (Halliwell, 1999a, 2000c; Levine et al., 2001; Meagher et al., 2001). In this review, we will examine the methods available to measure reactive species (RS) and oxidative damage, with a particular emphasis on those applicable to human studies.

Table 1 - Nomenclature of reactive species.

Full table

Table 2 - Some key definitions.

Full table

Table 3 - Criteria for implicating RS as a significant mechanism of tissue injury in human disease.

Full table

Measuring RS in vivo: basic principles

Some fascinating techniques such as L-band electron spin resonance (ESR) with nitroxyl probes and magnetic resonance imaging spin trapping are under development to measure RS directly in whole animals (e.g. Berliner et al., 2001; Han et al., 2001; Utsumi & Yamada, 2003), but no probes are currently suitable for human use. Most RS persist for only a short time in vivo and cannot be measured directly. There are a few exceptions: examples include H₂O₂ (discussed below), and perhaps, NO, in the sense that serum levels of NO₂^- have been claimed to measure vascular endothelial NO synthesis (Kelm et al., 1999), despite the fact that NO₂^- is quickly oxidized to NO₃^- in vivo (Kelm et al., 1999; Oldreive & Rice-Evans, 2001). Essentially, there are two approaches to detecting transient RS:

• attempting to trap these species and measure the levels of the trapped molecules and
• measuring the levels of the damage done by RS, that is, the amount of oxidative damage.

Sometimes other approaches are used. They include measurements of erythrocyte antioxidant defences and of total antioxidant activity of body fluids; falls in these parameters are often taken as evidence of oxidative stress. Erythrocytes cannot synthesize proteins, however, and their antioxidant enzyme levels may drop as they 'age' in the circulation (Denton et al., 1975). Thus changes in their levels are more likely to reflect changes in the rates of red blood cell turnover: if this slows down, the circulating erythrocytes will be older on average and so levels of antioxidant enzymes in them will appear to fall. Vice versa, if an intervention accelerates red cell removal or increases erythropoiesis, levels of antioxidants in red cells will seem to rise. Hence, such data should be interpreted with caution.

Depending on the method that is used to measure it, the plasma or serum 'total antioxidant capacity' (TAC) usually involves major contributions from urate, ascorbate and sometimes albumin -SH groups (Benzie & Strain, 1996; Halliwell & Gutteridge, 1999; Prior & Cao, 1999; Rice-Evans, 2000; Bartosz, 2003), although different methods measure different things (Schlesier et al., 2002; Bartosz, 2003). Thus, for example, if plasma albumin levels fall, TAC will fall. If urate levels rise, TAC will rise. The multiple changes in blood chemistry that occur in sick people mean that TAC changes should be interpreted with caution. TAC is also influenced by diet, often because consumption of certain foods may produce changes in plasma ascorbate and/or urate levels (Halliwell, 2003b).

Trapping of RS

The only technique that can 'see' free radicals directly and specifically is electron spin resonance (ESR), because it detects the presence of unpaired electrons. However, ESR detects only fairly unreactive radicals, since reactive ones do not accumulate to high-enough levels to be measured. One solution to this problem is to add 'traps' or 'probes', agents that intercept reactive radicals, reacting with them to form a stable radical that can be detected by ESR. Whole-body ESR techniques are being used on animals (Berliner et al., 2001; Han et al., 2001; Utsumi & Yamada, 2003), but are currently inapplicable to humans due to the lack of any human safety data on the probes. A wide range of traps is available for use in animals and cell culture systems, not only the tried and tested N-tert-butyl--phenylnitrone (PBN) and 5,5-dimethyl-1-pyrroline N-oxide (DMPO) (Khan et al., 2003) but also such relative newcomers as 1,1,3-trimethyl-isoindole N-oxide (TMINO) (Bottle et al., 2003), 5,5-diethylcarbonyl-1-pyrroline N-oxide (DECPO) (Karoui et al., 2004), N-2-(2-ethoxycarbonyl-propyl)--phenylnitrone (EPPN) (Stolze et al., 2003), 5-diethoxyphosphoryl-5-methyl-1-pyrroline N-oxide (DEPMPO) (Liu et al., 1999; Khan et al., 2003) and 5-tert-butoxycarbonyl-5-methyl-1-pyrroline N-oxide (BMPO) (Zhao et al., 2001a). Potential problems in their use are summarized by Rosen et al. (1999), Halliwell & Gutteridge (1999) and Khan et al. (2003). One generally underappreciated problem is that the reaction products giving the ESR signal can be rapidly removed in vivo and in cultured cells, both by enzymic metabolism and by direct reduction by such agents as ascorbate. For example, when DMPO is used to trap OH, any ascorbate present can directly reduce the DMPO-OH adduct to an ESR-silent species (Halliwell & Gutteridge, 1999). Many papers have described alleged antioxidant actions of various compounds evaluated in vitro by measuring decreases in the ESR signal that results when the compound is added to DMPO (or other spin trap) in the presence of radical-generating systems (frequently a Fenton system, H₂O₂+Fe²⁺). These decreases are attributed to scavenging of OH by the added compound, in competition with DMPO for OH. Rarely do these papers report the essential controls (Halliwell, 1995): does the added compound interfere with the radical-generating system (e.g. decomposing H₂O₂ or chelating iron in the Fe²⁺/H₂O₂ system) or does it interact directly with the trap spin adduct, reducing it to an ESR-silent species?

Rizzi et al (2003) have introduced triarylmethyl free radical, TAM OX063, as a probe for the detection of O₂^- in aqueous solution. In this case, the O₂^- reacts with the probe to cause the loss of the ESR signal. One advantage of TAM OX063 is that it is not subject to reduction by such agents as ascorbate or reduced glutathione (GSH).

Ex vivo trapping

Presently available spin traps cannot be administered to humans because of unknown toxicity at the high levels that would be required for radical trapping in vivo, in order for the trap to complete with endogenous molecules for a reactive free radical. However, traps can be used on body fluids and tissue samples. For example, DMPO was used to measure free radicals in human skin biopsies (Haywood et al., 1999). A hydroxylamine probe has been used to measure free radicals in liver biopsies (Valgimigli et al., 2002). PBN has been employed to demonstrate free radical production by adding it to coronary sinus blood drawn during coronary bypass surgery (Tarkka et al., 2000; Clermont et al., 2002). Of course, very highly reactive radicals such as OH, and highly reactive nonradical species (such as peroxynitrous acid) that may be formed in vivo (Sun et al., 1993; Halliwell & Gutteridge, 1999; Ferdinandy & Schulz, 2003), will simply not survive to be detected in ex vivo material. For example, OH reacts with whatever is in its immediate vicinity within microseconds (Halliwell & Gutteridge, 1999) and ONOOH breaks down in seconds (Radi et al., 2001). ESR of ex vivo samples to which traps have been added probably detects secondary radicals resulting from reaction of RS with biomolecules. These secondary radicals will include lipid-derived radicals (alkoxyl, peroxyl, etc.) and possibly also protein radicals (Pantke et al., 1999; Ostdal et al., 2002). Ascorbic acid (vitamin C) reacts with a wide range of free radicals and other RS (Buettner, 1993; Halliwell, 1999c) and one of its oxidation products, semidehydroascorbate radical, can easily be detected by ESR (Buettner & Jurkiewicz, 1993). Measurement of semidehydroascorbate has been used as an indication of free radical production in organs, blood plasma and skin (Buettner & Jurkiewicz, 1993; Sharma & Buettner, 1993; Sharma et al., 1994; Jurkiewicz & Buettner, 1996; Haywood et al., 1999). It is only semiquantitative, since ascorbate radicals quickly react with each other to generate ESR-silent species (ascorbate and dehydroascorbate) (Bielski et al., 1975; Kobayashi et al., 1991). The ascorbate radical is also reduced by enzymes in vivo (May et al., 2001).

Traps, both the ESR 'spin traps' described above and other types of trap (see below), are likely to perturb the biological system under investigation. If traps scavenge RS, and these RS are important mediators of cell or tissue injury, it follows logically that the traps should protect against damage, provided, of course, that they are scavenging enough of the RS (Table 3). Indeed, some compounds derived from spin traps have entered clinical trials, for example, for stroke (Lapchak & Araujo, 2003). However, one must never assume that any beneficial effects that traps exert are due to free radical scavenging without direct evidence for this mechanism, since these compounds can exert other pharmacological effects (Pietri et al., 1998; Floyd et al., 1999; Kotake, 1999).

Aromatic traps

Potentially more useful than spin traps in vivo in humans are aromatic 'free radical' traps known to be acceptable for human consumption, including salicylate and phenylalanine (Kaur et al., 1996; Grootveld & Halliwell, 1986; Kaur & Halliwell, 1994; Halliwell & Kaur, 1997; Coudray & Favier, 2000; Themann et al., 2001). Salicylate is hydroxylated by OH to yield, among other products, 2-3 dihydroxybenzoate, which is not apparently produced enzymically in vivo (Ingelman-Sundberg et al., 1991; Dupont et al., 1999). Both L- and D-phenylalanine are hydroxylated by OH to yield ortho- and meta-tyrosines, which again seem not to be enzymically produced in humans, although the evidence supporting this statement is weak and more studies are required (Kaur & Halliwell, 1994; Halliwell and Kaur, 1997; Li et al., 2003). Both salicylate and phenylalanine have been used to measure ex vivo radical formation in blood from patients with rheumatoid arthritis (Kaur et al., 1996) or in humans exposed to ozone (Frischer et al., 1997; Liu et al., 1997a). Phenylalanine was used to detect OH formation in saliva (Nair et al., 1995). Several human studies have also used salicylate to detect OH in vivo, with some success. Examples include studies on diabetes (Ghiselli et al., 1992), alcoholism (Thome et al., 1997) and myocardial infarction (Tubaro et al., 1992). Endogenous levels of free (Lubec et al., 1997; Ogihara et al., 2003) or protein-bound (Fu et al., 1998; Lamb et al., 1999; Pennathur et al., 2001) ortho- or meta-tyrosines have also been used to implicate OH formation in vivo.

The success of aromatic compounds (or spin traps) in detecting OH will depend on their concentration at sites of free radical generation in relation to that of other OH scavengers (virtually all other biomolecules react fast with OH and will compete), and so it is unlikely that any trap can intercept more than a small percentage of any OH generated. 2,3-Dihydroxybenzoate can be metabolized in the liver, and possibly other organs, causing its loss (unpublished data). It follows that aromatic hydroxylation techniques are not quantitative measures of OH generation, especially if the end products are measured in the plasma, where their origin is unclear. Specificity of the traps for OH is uncertain: ONOO^- can hydroxylate both salicylate and phenylalanine, perhaps, in part, by direct reaction with ONOOH (Kaur et al., 1997). Simultaneous determination of nitration products can distinguish the two (Kaur et al., 1997; Ferger et al., 2001), since OH cannot nitrate aromatic compounds whereas ONOO^- usually does. Myeloperoxidase can also hydroxylate aromatic compounds, although the favored product from salicylate is 2,5-dihydroxybenzoate (Kettle & Winterbourn, 1994; Kawakami et al., 1999). Hypochlorous acid produced by myeloperoxidase can, however, react with O₂^- to produce OH (Candeias et al., 1993).

Another problem is that high-performance liquid chromatography (HPLC) analysis of meta-tyrosine in human samples can be confused by the frequent presence of a coeluting peak of unknown identity (Reddy et al., 1999). Be suspicious if you see an apparent meta-tyrosine peak on HPLC much greater than that of ortho-tyrosine: this is suggestive of an artifact since both products should be formed in comparable amounts when OH attacks phenylalanine. Mass spectrometry (MS) can be used to avoid this problem (Pennathur et al., 2001). As an additional probe, the compound antipyrene is hydroxylated by OH into para- and ortho-hydroxylation products, the latter again not apparently formed enzymically in humans. Antipyrene has been used to measure OH production during exercise (Meijer et al., 2001).

Urate as a trap

An endogenous molecule might also function as a trap (indeed, measurement of meta- and ortho-tyrosines in body proteins as 'endogenous traps' would fit into this category), although it can be argued that measuring specific end products of the trapping of RS by endogenous molecules is the same as measuring 'biomarkers' (see below). Ascorbate reaction with free radicals is one example, as discussed above. Another example is urate, which is readily oxidized by a range of RS (Kaur & Halliwell, 1990), including peroxynitrite (Whiteman et al., 2002b). Several groups have used urate as a 'selective' scavenger of ONOO^- in animal studies, neglecting the fact that it reacts with many species (Kaur & Halliwell, 1990). One of urate's oxidation products, allantoin, can be measured in human body fluids and its plasma levels are elevated in conditions associated with oxidative stress, such as chronic inflammation, diabetes, premature birth, iron overload, chronic heart failure and exercise (Grootveld & Halliwell, 1987; Moison et al., 1997; Ogihara et al., 1998; Benzie et al., 1999; Mikami et al., 2000; Doehner et al., 2002; Pavitt et al., 2002; Yardim-Akaydin et al., 2004). Allantoin can also be measured in urine (Mikami et al., 2000) and cerebral microdialysis fluid (Marklund et al., 2000). Levels of allantoin rise in the human muscle during exhaustive exercise, presumably due to oxidation of urate by RS generated during exercise (Hellsten et al., 2001). Allantoin measurement may be one of the more promising methods for human use, since human urate levels in vivo are high and urate reacts with a wide range of RS.

Measuring stable species in body fluids: hydrogen peroxide

Several laboratories have demonstrated the presence of H₂O₂ in freshly voided human urine (Varma & Devamanoharan, 1990; Kuge et al., 1999; Long et al., 1999; Laborie et al., 2000; Long & Halliwell, 2000; Hiramoto et al., 2002; Kirschbaum, 2002; Yuen & Benzie, 2003). It has also been measured in rabbit urine (Pi et al., 2003). Since assay of urinary H₂O₂ can be performed rapidly and simply (Long et al., 1999), the question has been raised if its levels in urine (after creatinine standardization) can be related to rates of generation of RS in vivo. Caution must be exercised, since drinking coffee artifactually raises urinary H₂O₂ levels (Long & Halliwell, 2000; Hiramoto et al., 2002), apparently because hydroxyhydroquinone in coffee is absorbed into the body, excreted into the urine and autooxidizes therein to produce H₂O₂ (Hiramoto et al., 1998; Halliwell et al., 2004a). Patients with cancer showed elevated urinary H₂O₂ levels (Banerjee et al., 2003). The currently available data are insufficient to support the use of urinary H₂O₂ as a biomarker of oxidative stress and more work is required (Yuen & Benzie, 2003; Halliwell et al., 2004a).

Hydrogen peroxide can also be detected in exhaled air and breath condensate, and the levels appear to increase during inflammation. The origin of the H₂O₂ is unclear: release by respiratory tract cells and oral bacteria are possibilities (Jobsis et al., 2001; Geiszt et al., 2003; Rahman & Kelly, 2003).

Measuring changes in blood pressure and vascular tone

The free radical NO plays a key role in vasodilatation. Its action can be antagonized by O₂^-, which removes NO by reacting fast with it to give ONOO^- (reviewed by Beckman & Koppenol, 1996). In the healthy arterial wall, there is a balance between NO and O₂^-, but increased oxidative stress can diminish the bioactivity of NO (Darley-Usmar et al., 1995; Kojda & Harrison, 1999). Sources of O₂^- in and around the vessel wall include the vascular endothelium (especially in atherosclerotic lesions), fibroblasts, lymphocytes, phagocytes and the enzyme xanthine oxidase (Darley-Usmar & Halliwell, 1996; Reth, 2002; Rey & Pagano, 2002). Upregulation of O₂^--generating NAD(P)H oxidases may play a key role in causing impaired vascular tone and hypertension (Harrison et al., 2003; Lassegue & Clempus, 2003; Murphey et al., 2003).

Many papers have claimed that high doses of infused or orally administrated antioxidants (ascorbate being most commonly used) can scavenge ROS, restore NO bioavailability and ameliorate human endothelial dysfunction (for examples see Plotnick et al., 1997; Duffy et al., 1999; Gokce et al., 1999; Wilkinson et al., 1999; May, 2000; Beckman et al., 2001; Duffy et al., 2001; Pleiner et al., 2002; Heiss et al., 2003), although negative reports are now accumulating (e.g. Darko et al., 2002; Singh et al., 2002; Svetkey & Loria, 2002; Widlansky et al., 2004; our unpublished date). The timing of antioxidant administration and the state of the vascular bed (healthy, mildly diseased or severely diseased) may be critical variables (Carroll & Schade, 2003). Hence it is possible that examination of short-term vascular effects, which is easily achievable in human studies by measurements of forearm blood flow, can gather evidence about localized scavenging of RS in blood vessel walls by antioxidants.

However, it must not be assumed that any beneficial effects of antioxidants on vascular function are necessarily due to free radical scavenging. For example, ascorbate may have a direct stimulating effect on eNOS activity by increasing endothelial tetrahydrobiopterin levels (Huang et al., 2000; d'Uscio et al., 2003).

Fingerprinting of RS

If a RS combines with a biological molecule to leave a unique chemical 'fingerprint', then the presence of that fingerprint (which is of course an example of oxidative damage) can be used to infer that the RS has been formed (Halliwell & Gutteridge, 1999). Such 'biomarkers' can then be used to investigate effects of dietary supplements or synthetic antioxidants on oxidative damage, hopefully leading to experimentally verifiable predictions of the likely effects of these compounds on oxidative stress-related diseases arising subsequently. Indeed, since methods currently available for the direct measurement of RS are of limited applicability to humans (see above), most clinical studies focus on the measurement of oxidative damage. This is to some extent logical, since it is the damage caused by RS that is important rather than the total amount of such species generated. For example, if highly reactive OH radicals are generated within a cell, many of them may react with unimportant targets. The small fraction that reacts with DNA (to cause strand breakage and formation of mutagenic base or sugar oxidation products), with key proteins, or with lipids (initiating lipid peroxidation) may be the important fraction.

Table 4 summarizes the requirements of the ideal biomarker. No currently used biomarker has yet fulfilled the key criterion (A); the experiments have not been carried out. It is sad that biomarkers were never built into all the human intervention trials with antioxidants that have been carried out; they could have been of great value in interpreting the confusing and conflicting results (Halliwell, 1999a, 2000b, 2000c). No biomarker meets all parts of criterion (B), but some are better than others. We have already considered small molecules such as ascorbate and urate, so now will comment briefly on currently available biomarkers of oxidative damage to the three main classes of macromolecules: nucleic acids, proteins and lipids.

Table 4 - Criteria for the ideal biomarker of oxidative damage.

Full table

Lipids

Lipids can be oxidized, chlorinated and nitrated by a range of RS (not including H₂O₂, NO or O₂^-, which are essentially unreactive with lipids; Halliwell & Gutteridge, 1999). Techniques for the measurement of chlorinated and nitrated lipids are being developed, and species such as nitrated linoleate have been detected in human blood plasma (Hazen et al., 1996; Lim et al., 2002; Lima et al., 2002, 2003; Thukkani et al., 2003). We await further studies on these compounds with interest.

Lipid peroxidation is a complex process, and a wide range of products is formed in variable amounts (Halliwell & Gutteridge, 1999). Lipid oxidation can be measured in many ways, but commonly used methods such as diene conjugation and thiobarbituric acid (TBA)-reactive material are of questionable validity (Halliwell & Gutteridge, 1999). In particular, the simple TBA test should be dismissed as unacceptable in modern research, simply because most TBA-reactive material in human body fluids is not related to lipid peroxidation. A significant improvement to the TBA test can be made by using HPLC to isolate the malondialdehyde (MDA)-TBA chromogen before analysis (Halliwell & Gutteridge, 1999). One can also assay MDA directly (e.g. Liu et al., 1997b), but MDA is only one of many aldehydes formed during lipid peroxidation, and MDA can also arise from free radical attack on sialic acid and deoxyribose (Halliwell & Gutteridge, 1999). Unsaturated aldehydes such as 4-hydroxynonenal and acrolein may cause considerably more cytotoxicity in vivo than MDA, so measuring them is perhaps more logical (Ong et al., 2000; Uchida, 1999, 2003). Lipid hydroperoxides and aldehydes can also be absorbed from the diet. For example, some foods contain MDA–amino-acid adducts that can be absorbed through the gut and then excreted in urine (Nelson et al., 1993; Richelle et al., 1999; Draper et al., 2000; Cohn, 2002; Indart et al., 2002; Wilson et al., 2002). It follows that measurements of urinary MDA can be confounded by diet and should not be used as an index of whole-body lipid peroxidation unless diet is controlled.

The isoprostanes

The best available biomarker of lipid peroxidation appears to be the isoprostanes, specific end products of the peroxidation of polyunsaturated fatty acids (Roberts & Morrow, 2002; Fam & Morrow, 2003). Most work has been carried out on the F₂-isoprostanes, which arise from arachidonic acid (Roberts & Morrow, 2002), but some data are also available on isoprostanes from eicosapentaenoic (Nourooz-Zadeh et al., 1997) and docosahexaenoic acids. The latter are often called neuroprostanes or F₄-isoprostanes (Nourooz-Zadeh et al., 1999; Reich et al., 2001; Roberts & Morrow, 2002). Isoprostanes are best measured by MS (Lawson et al., 1999; Roberts & Morrow, 2002). Although some immunoassay kits for F₂-isoprostanes are commercially available, their reliability has been questioned (Proudfoot et al., 1999; Bessard et al., 2001; Roberts & Morrow, 2002). Although isoprostanes can be detected in foods, they do not appear to pass through the gut in sufficient quantities to affect plasma or urinary levels (Richelle et al., 1999; Gopaul et al., 2000). Reliable MS-based techniques have been established to detect isoprostanes and their metabolites in the plasma and urine (Lawson et al., 1999; Li et al., 1999; Fam & Morrow, 2003), although the 'work-up' techniques prior to MS are tedious.

Even the isoprostanes do not qualify as 'ideal' biomarkers. The products commonly measured (e.g. 8-iso-PGF₂) are often minor end products of peroxidation (Lawson et al., 1999), and the amounts formed can be influenced by such parameters as O₂ concentration (Morrow et al., 1998; Fessel et al., 2002; Roberts & Morrow, 2002). Hence, much work is proceeding to identify new products arising from the isoprostane pathway (Roberts & Morrow, 2002; Fessel et al., 2002). Isoprostanes appear to be rapidly metabolized, turning over quickly (Roberts & Morrow, 2002; Basu, 2004). Thus a rise in plasma isoprostane levels could conceivably be due not only to increased formation by oxidative damage to lipids but also by slower metabolism. Measurement of isoprostanes in plasma and urine is, of course, a 'whole-body' measurement; it is not necessarily clear from where the isoprostanes originated. They can also be measured in fluids drawn from specific sites, such as synovial fluid (Basu et al., 2001), pericardial fluid (Mallat et al., 1998), wound exudate (Sim and Stacey, 2003) and breath condensate (Montuschi et al., 1999).

Of especial current interest is the use of isoprostane measurements to study Alzheimer's disease (AD), a condition in which lipid peroxidation and other oxidative damage seem to play significant pathological roles (Montine et al., 1998, 1999; Pratico et al., 2000, 2001, 2002; Halliwell, 2001; Butterfield, 2002). Levels of F₂- and F₄-isoprostanes are elevated in CSF from AD patients (Montine et al., 1998, 2002; Pratico et al., 2000), and may even be elevated prior to the development of AD (Pratico et al., 2001, 2002), consistent with the view that peroxidation may be an important step in progressive neuronal injury leading to clinically manifested disease (Halliwell, 2001; Pratico et al., 2001, 2002). It has been claimed that urinary and plasma levels of isoprostanes are also elevated in AD patients to an extent correlated with the degree of cognitive impairment (Pratico et al., 2000, 2002), although this was not confirmed in other studies (Montine et al., 2002; Bohnstedt et al., 2003). It would be of great advantage in researching neuroprotective antioxidants to have a peripheral biomarker that could report on oxidative events in the brain, and so more work is needed to investigate urinary and plasma isoprostanes in this context and resolve the conflicting data. However, there does seem to be agreement that there are elevated CSF isoprostane levels in AD (Pratico et al., 2000, 2002; Montine et al., 2002).

Increased levels of isoprostanes have been observed in many animal (reviewed by Roberts & Morrow, 2002; Basu, 2004) and human conditions associated with oxidative stress, including cardiopulmonary bypass (Ulus et al., 2003), angioplasty (Guan et al., 1999, 2003; Iuliano et al., 2001) and diabetes (Gopaul et al., 2001; Jialal et al., 2002; Sampson et al., 2002). Hyperglycemia causes oxidative stress (Brownlee, 2001; Gopaul et al., 2001; Jialal et al., 2002), and plasma isoprostane levels in diabetes may be determined to a considerable extent by plasma glucose levels (Sampson et al., 2002).

Isoprostanes have also been used to study effects of dietary antioxidant supplementation on lipid peroxidation. Plasma and urinary isoprostane levels in humans have responded to antioxidant supplementation in some studies (reviewed by Halliwell, 2000a; Roberts & Morrow, 2002), but in general responses are limited or absent in healthy well-nourished subjects, indicating that lipid peroxidation is little affected by supplements. Supplementation of healthy volunteers with vitamins C or E, for example, usually decreases isoprostane levels only slightly, if at all (Levine et al., 2001; Meagher et al., 2001), which might help to explain why these vitamins have had little effect on disease outcomes in many human intervention trials (Halliwell, 2000c). However, some positive effects have been reported, for example, supplementation with vitamins E and C did decrease urinary isoprostanes in one study upon healthy American nonsmoking subjects (Huang et al., 2002). Isoprostane measurements also suggest that obesity and hypercholesterolemia (known risk factors for cardiovascular disease) are associated with elevated rates of lipid peroxidation. This is consistent with the view that lipid peroxidation is an important contributor to cardiovascular disease and can help explain why weight loss and statins decrease cardiovascular events (Roberts et al., 2002; Keaney et al., 2003; Morrow, 2003; Russell et al., 2003; Samuelsson et al., 2003). There is also increasing interest in the link between hypercholesterolemia, oxidative stress and AD (Refolo et al., 2001; Leonarduzzi et al., 2002; Puglielli et al., 2003; Wolfrum et al., 2003; Ong & Halliwell, 2004).

An important question for any biomarker is the coefficient of variation between assays of the same sample and between samples taken from the same subjects at different times. There have been claims that isoprostane levels vary with time of day and from day to day (Helmersson & Basu, 2001; Kanabrocki et al., 2002), possibly because oxidative stress also varies (reviewed by Halliwell et al., 2004a). If there is such a variation, it is essential to take it into account in human studies.

Exhaled hydrocarbons

Exhaled air contains not only isoprostanes (Rahman & Kelly, 2003; Montuschi et al., 1999) and aldehydes but also a range of hydrocarbons, including ethane and pentane (Andreoni et al., 1999; Phillips et al., 2000; Dale et al., 2003). The available data suggest that ethane may be a biomarker of lipid peroxidation, pentane perhaps less so. However, these hydrocarbons are minor end products of lipid peroxidation, formed to variable extents depending on the exact environment of the oxidizing lipids (Halliwell & Gutteridge, 1999). Exhaled hydrocarbons are also difficult to measure routinely in human studies, requiring cumbersome equipment, as compared to, for example, measuring isoprostanes in blood and urine samples.

DNA

Oxidative DNA damage seems to relate to an increased risk of cancer development later in life (reviewed by Halliwell, 2002b). DNA subjected to attack by hydroxyl radical generates a huge range of base and sugar modification products (Dizdaroglu et al., 2002). Such products can be measured by HPLC, gas chromatography (GC)–MS, liquid chromatography (LC)–MS and antibody-based techniques, and arguments rage over which is best (Halliwell, 1999b, 2000c; Dizdaroglu et al., 2002; Kasai, 2002; Collins et al., 2004). None of them is the 'gold standard' we all seek. Initial products of free radical attack upon purines, pyrimidines and deoxyribose undergo transformation into stable end products, whose relative amounts depend on reaction conditions (Dizdaroglu, 1992; Alam et al., 1997; Halliwell, 1999b; Dizdaroglu et al., 2002). Thus it is intrinsically unreliable to measure any single reaction product as an index of oxidative DNA damage, but that is what is usually done, as 8-hydroxy-2'-deoxyguanosine (8OHdG) (Collins et al., 2004). The advantages and artifacts of measuring 8OHdG in cellular DNA have been recently reviewed (Halliwell, 2000c, 2002b; Collins et al., 2004), so they need not be reiterated here. The major problems arise from artifactual oxidative damage to DNA and consequent 8OHdG formation during isolation of DNA, its preparation for analysis and the analysis itself. In general, there is no agreement even on basal levels of 8OHdG in cellular DNA, nor does an agreement seem close (Collins et al., 2004). In particular, the suggestion almost 10 years ago that mitochondrial DNA (mtDNA) is more oxidized than nuclear DNA does not rest on a firm experimental foundation (Beckman & Ames, 1999). Our own studies (unpublished) indicate similar or lower (for some base damage products) levels of oxidative DNA damage in rat liver mtDNA as compared with nuclear DNA. Several RS formed at sites of inflammation, such as HOCl and ONOO^-, can destroy 8OHdG in DNA (Whiteman et al., 1997, 2002a). Nevertheless, significant increases in levels of 8OHdG and other DNA base damage products have been observed in many studies on diseased human material (reviewed by Halliwell, 2000c; Evans et al., 2003), although (as with the isoprostanes) effects of dietary antioxidant supplementation on levels of 8OHdG or other base damage products in vivo seem limited (reviewed by Halliwell, 2000c; Moller & Loft, 2002). These limited effects may explain why antioxidant administration has generally not decreased cancer incidence in human intervention trials in Western countries.

One approach would be to bypass these problems of artifactual DNA oxidation during DNA isolation and analysis by measuring oxidative DNA damage in the intact cell. Antibody methods have been developed for 8OHdG (e.g. Toyokuni et al., 1997) and are useful for visualization of damage, but seem likely to be semiquantitative (Shimoi et al., 2002; Yoshida et al., 2002). The comet assay (Duthie et al., 1996) can be applied directly to cells and measures DNA strand breaks. If a digestion step with DNA repair enzymes is included in the protocol, the increased numbers of DNA strand breaks can be used to estimate the level of oxidized DNA bases in the cell. Studies with the comet assay seem more likely than other methods to show positive effects of antioxidant interventions in human volunteers, and the reasons for this need to be elucidated (Duthie et al., 1996; Toyokuni et al., 1997; Collins & Horvathova, 2001; Moller & Loft, 2002). Is it that the comet assay is more reliable because the baseline is lower, artifactual DNA oxidation having been minimized? Or does the comet assay simply just generate a different kind of artifact, in underestimating damage? It seems very unlikely that all the oxidized bases in compact DNA can be recognized by exogenously applied enzymes. Strand breaks, which are what the comet assay measures, arise not only during oxidative DNA damage but also during repair of that damage (Spencer et al., 1996a). Enzymic DNA cleavage during apoptosis may also generate comets, which can lead to false conclusions about the genotoxicity of apoptosis-inducing agents. An antioxidant that suppressed apoptosis could be misinterpreted to be protecting directly against DNA damage (Choucroun et al., 2001).

None of the analytical methods mentioned above identifies where the oxidative DNA damage is located: is it in expressed genes, inactive genes, merely in 'junk DNA' or even in telomeres (Oikawa & Kawanishi, 1999)? Attempts are underway to address this problem (e.g. Rodriguez et al., 2000; Choi et al., 2003; Sawyer et al., 2003). Another problem in studying damage to DNA by RS is the limited availability of human tissues from which to obtain DNA. Most studies are performed on DNA isolated from lymphocytes or total white cells from human blood, and it is assumed (possibly erroneously: reviewed by Halliwell, 2002b) that changes here are reflected in other tissues. Sperm, buccal cells, placenta, and biopsies of muscle, skin, colon and other tissues are other potential sources of DNA.

Urinary measurements

Measurement of 8OHdG in urine has been used to assess rates of 'whole-body' oxidative DNA damage. This can be achieved by HPLC and MS techniques (Bogdanov et al., 1999; Gackowski et al., 2001; Kasai, 2002; Pilger et al., 2002; Lin et al., 2004), but ELISA techniques have also been described (Shimoi et al., 2002; Yoshida et al., 2002). The presence of interfering peaks in certain urine samples is a major bugbear, and sample 'cleanup' techniques are critical (Bogdanov et al., 1999; Lin et al., 2004). The assays seem reliable (except that ELISA can give high levels sometimes) and data fairly comparable between laboratories (Bogdanov et al., 1999; Gackowski et al., 2001; Cooke et al., 2002; Kasai, 2002; Pilger et al., 2002; Shimoi et al., 2002; Yoshida et al., 2002; Lin et al., 2004), unlike what is seen with the measurement of 8OHdG in isolated DNA (Collins et al., 2004). No urinary 8OHdG appears to originate from diet (Gackowski et al., 2001). However, 8OHdG can arise from degradation of oxidized dGTP in the DNA precursor pool, not just from removal of oxidized guanine residues from DNA by repair processes (Cooke et al., 2002). Hence urinary levels may not truly reflect rates of oxidative damage to DNA. Another reason why one cannot equate 8OHdG in urine with oxidative DNA damage is that there are many other products of this process (Dizdaroglu, 1992; Dizdaroglu et al., 2002). Hence, urinary 8OHdG is a partial measure of damage to guanine residues in DNA and its nucleotide precursor pool. Nevertheless, 8OHdG is the best urinary measure we have at the moment. Other oxidized bases can be measured in urine (e.g. Ravanat et al., 1999), but a confounding by diet cannot yet be ruled out. DNA in foods can be oxidized during storage and cooking, and it is possible that oxidized bases can be absorbed and reexcreted, just as it happens with MDA. This is not a problem with 8OHdG (Gackowski et al., 2001).

Measurement of 8OHdG in urine gives no information about its tissue origin, although it is sometimes possible to sample from specific sites, such as cerebrospinal fluid (Rozalski et al., 2003). The possibility of diurnal variations in urinary 8OHdG levels has been raised, so spot measurements should be interpreted with caution (Kanabrocki et al., 2002). However, our laboratory has not found this to be a major problem (Lin et al., 2004). Products of reaction of DNA bases with end products of lipid peroxidation such as MDA can also be detected in human urine (Hanaoka et al., 2002; Otteneder et al., 2003), but a confounding effect of diet is possible since these adducts may be present in foods, for example, after cooking. DNA base–aldehyde adducts can also be measured in DNA isolated from cells and tissues (Bartsch & Nair, 2000; Marnett et al., 2003), but one must again be cautious because tissue homogenates can undergo rapid peroxidation in vitro (Halliwell & Gutteridge, 1999), leading to possible modification of DNA by aldehydes generated artifactually during the isolation process.

Reactive nitrogen and chlorine species

DNA can also be damaged by reactive nitrogen species, undergoing mainly nitration and deamination of purines (Yermilov et al., 1995; Spencer et al., 1996b; Zhao et al., 2001b; Lee et al., 2002b). It was initially thought that 8-nitroguanine was a specific product of reaction of ONOO^- with DNA, but this does not seem to be the case (Byun et al., 1999). Methods for the measurement of DNA base nitration and deamination products have been developed, but may need more refinement before they can be routinely applied to human material. However, cigarette smoking was observed to raise the level of 8-nitroguanine in peripheral lymphocytes of humans (Hsieh et al., 2002). Antibody-based methods have been used to demonstrate rises in 8-nitroguanine during inflammation in animals (Akaike et al., 2003; Pinlaor et al., 2003). Assays to identify chlorinated DNA bases have also been developed (Whiteman et al. 1997; Henderson et al., 2003).

Ultraviolet light

Ultraviolet light produces a range of lesions in DNA, including thymine dimers (Douki et al., 2000), which can be measured in human urine. Levels increase after exposure to sunlight (Cooke et al., 2001; Le Curieux & Hemminki, 2001) and in psoriasis (Ahmad et al., 1999). Urinary thymine dimers might thus be a useful biomarker to assess effects of antioxidants on sunlight-induced DNA damage.

Proteins

Oxidative damage to proteins may be important in vivo both in its own right (affecting the function of receptors, enzymes, transport proteins, etc., and perhaps, generating new antigens that can provoke immune responses) (Halliwell, 1978; Casciola-Rosen et al., 1997) and because it can contribute to secondary damage to other biomolecules, for example, inactivation of DNA repair enzymes and loss of fidelity of damaged DNA polymerases in replicating DNA (Wiseman & Halliwell, 1996). However, analyzing protein oxidative damage products is an order of magnitude more complex than dealing with DNA: rather than four bases and one sugar there are 20 different amino acids, each of which can be attacked by RS in multiple ways (Davies et al., 1999; Headlam & Davies, 2003). Free radical attack on proteins can generate amino-acid radicals, which may crosslink or react with O₂ to give peroxyl radicals. These may abstract H, triggering more free radicals and forming protein peroxides, which can decompose in complex ways, accelerated by transition metal ions, to generate yet more radicals (Headlam & Davies, 2003). Proteins can also be oxidized during food cooking, meaning that oxidized amino acids could conceivably be absorbed from the diet, which could confound measurements of them in body fluids as putative biomarkers of oxidative damage. Individual amino-acid oxidation products that have been measured in various human tissues include kynurenines (from tryptophan), bityrosine (which appears to be metabolically stable and can be detected in urine), valine and leucine hydroxides, L-dihydroxyphenylalanine (L-DOPA), ortho-tyrosine, 2-oxo-histidine, glutamate semialdehyde and adipic semialdehyde (Uchida & Kawakishi, 1993; Morin et al., 1998; Leeuwenburgh et al., 1999a, 1999b; Giulivi & Davies, 2001; Harth et al., 2001; Headlam & Davies, 2003). Bityrosine, easily detectable in human and other animal urines, formed by free radical attack on a wide range of proteins, and apparently not metabolized, might be a biomarker worth further development for human use (Davies et al., 1999; Leeuwenburgh et al., 1999a, 1999b; Giulivi & Davies, 2001). Patients with sepsis were reported to have increased urinary excretion of bityrosine (Bhattacharjee et al., 2001), as were children with kwashiorkor (Manary et al., 2000), a disease which is believed to involve increased oxidative stress (Golden & Ramdath, 1987; Fechner et al., 2001). An immunohistochemical method to detect bityrosine in human brain has been described (Kato et al., 1998). Sakharov et al. (2003) described the use of an acetyltyramine–fluorescein probe to detect formation of tyrosine radicals in proteins in cells subjected to oxidative stress. If both the tyramine probe and tyrosine residues are oxidized, the probe can crosslink to the protein tyrosyl residues, generating a stable crosslink similar to that in bityrosine. The fluorescein-labeled proteins can then be visualized and identified by proteomic methods.

Theoretically also useful might be the measurement of methionine sulfoxide, since methionine residues appear to be a 'radical sink' in proteins (Levine et al., 1999), that is, oxidative attack at different sites on a protein may eventually appear as methionine sulfoxide. However, methionine sulfoxide residues are usually quickly removed by methionine sulfoxide reductase enzymes (Levine et al., 1999; Sharov et al., 1999), and this amino-acid does not appear to have been used as a biomarker in human studies.

Proteins can also be attacked by reactive chlorine, bromine and nitrogen species (defined in Table 1), giving products such as 3-chlorotyrosine, para-hydroxyphenylacetaldehyde, 3,5-dichlorotyrosine, 3-bromotyrosine, and 3-nitrotyrosine (Hazen & Heinecke, 1997; Wu et al., 1999, 2000; Winterbourn & Kettle, 2000; Greenacre & Ischiropoulos, 2001; Himmelfarb et al., 2001; Aldridge et al., 2002; Gaut et al., 2002a). 3-Nitrotyrosine is often thought to be a specific marker of attack of ONOO^- upon proteins, but in fact it can be formed from tyrosine by a range of reactive nitrogen species and its production in vivo is often dependent on the production of such species as nitrogen dioxide radical (NO₂) by myeloperoxidase or other peroxidases (Halliwell et al., 1999; Greenacre & Ischiropoulos, 2001; MacPherson et al., 2001; Gaut et al., 2002b). 3-Nitrotyrosine has been frequently measured by immunostaining, HPLC and MS in human tissues, but we have been unable to detect it in human urine (unpublished data). Nitrotyrosine is metabolized to 3-nitro-4-hydroxyphenylacetate (NHPA), which some groups have measured in urine as an estimate of nitrotyrosine production in vivo (Ohshima et al., 1990). However, some (and perhaps most) urinary NHPA arises by other mechanisms (Mani et al., 2003), making the assay questionable. Simple HPLC-based assays for 3-nitrotyrosine are prone to artifact (Kaur et al., 1998), and so MS-based techniques are preferred.

Another problem is that exposure of body fluids or tissues to acid during staining or protein hydrolysis (frequently HCl digestion is used to liberate protein-bound nitrotyrosine) can cause artifactual nitration of tyrosine by species derived from nitrous acid (HNO₂), generated when nitrite in the sample is exposed to acid (Frost et al., 2000). Even freezing phosphate- and nitrite-containing samples can cause artifactual nitration (Daiber et al., 2003). In general, the more accurate the measurement methods that are used, the lower the levels of 3-nitrotyrosine detected in human material (Frost et al., 2000; Gaut et al., 2002a; Duncan, 2003; Morton et al., 2003). Nitrotyrosine can be destroyed by hypochlorous acid, generated by activated neutrophils (Whiteman & Halliwell, 1999). Immunostaining for nitrotyrosine is prone to artifacts (Halliwell et al., 1999; Ogino et al., 2002; Ichimori et al., 2003) but, when carried out with appropriate controls, has proved useful in qualitative demonstration of increased formation of reactive nitrogen species in tissues (e.g. MacMillan-Crow et al., 1996; Kooy et al., 1997; Greenacre & Ischiropoulos, 2001). Provided that sensitive and accurate MS-based assays are used, measurements of nitro-, chloro- and ortho-tyrosines can be useful biomarkers of protein damage by RS and how such damage can be modulated by antioxidant or other interventions (Hazen & Heinecke, 1997; Buss et al., 2003). For example, statin treatment decreased levels of these products in the plasma of hypercholesterolemia patients, consistent with the view (discussed above) that hyperchloesterolemia leads to oxidative stress (Shishehbor et al., 2003). Proteomic methods can be used to separate proteins, followed by recognition of modified protein by blotting, and this technique was used to identify targets of nitration in muscle, for example (Kanski et al., 2003). Protein nitrosylation, a reversible modification involving attachment of NO to a metal site or a cysteine residue, can also be measured in human material (Mannick & Schonhoff, 2004).

The carbonyl assay

The most frequently used biomarker of protein damage is the carbonyl assay, measurement of protein carbonyl groups (Chevion et al., 2000; Levine, 2002; Dalle-Donne et al., 2003). Carbonyls can arise as a result of protein glycation by sugars, by the binding of aldehyes (including many of those formed during lipid peroxidation) to proteins and by the direct oxidation of amino-acid side chains by RS to generate such products as glutamate and aminoadipic semialdehydes (Reznick et al., 1992; Calingasan et al., 1999; Chevion et al., 2000; Adams et al., 2001; Requena et al., 2001; Levine, 2002; Dalle-Donne et al., 2003). Carbonyls can be readily measured spectrophotometrically and by ELISA techniques (Buss & Winterbourn, 2002; Levine, 2002), and tissue or plasma levels have been shown to be elevated in many human diseases (reviewed by Dalle-Donne et al., 2003). For example, serum levels were increased during cardiopulmonary bypass (Pantke et al., 1999). Carbonyls can also be measured in other body fluids and in tissues (Chapman et al., 1989; Buss et al., 1997; Chevion et al., 2000; Frank et al., 2000; Schock et al., 2001; Levine, 2002; Tanaka et al., 2003). Of course, carbonyls are not specific as markers of oxidative damage because bound aldehydes and glycated protein are also measured. Indeed, immunochemical assays for protein-bound aldehydes such as acrolein and 4-HNE are widely used (Uchida, 1999, 2003). Data from their use have shown that both aldehydes may be important neurotoxic agents in neurodegenerative diseases (Calingasan et al., 1999; Ong et al., 2000).

The carbonyl assay as applied to tissues and body fluids measures the 'average' extent of protein modification. It is informative to use proteomic techniques to identify the specific proteins damaged. Often only a small selection of proteins is oxidized. For example, in human plasma subjected to oxidative stress, carbonyls appear to reside mostly on fibrinogen (Shacter et al., 1994). In the brains of subjects who died with AD, several specific proteins including enolase, UCH-L1 and DRP-2 seemed to be oxidatively modified (Butterfield et al., 2003).

Direct measurements of glutamate and aminoadipate semialdehyde (major contributors to total protein carbonyl residues) in human plasma proteins have also been used to assess the effects of alterations in dietary antioxidant intake on plasma protein oxidation, but no decreases in their levels were observed after increased intake of flavonoid-rich foods (Nielsen et al., 1999; Young et al., 2000a, 2000b). Whether this is due to the insensitivity of this biomarker or due to a lack of effect of flavonoids on oxidative damage in the human body (Dragsted, 2003; Halliwell et al., 2004b) is uncertain as yet. The possible confounding effect of uptake of oxidized amino acids from the diet also needs to be addressed.

Is there a single biomarker of oxidative stress or oxidative damage?

No, there is not (England et al., 2000), although (as might be expected) biomarkers of damage to several different molecules frequently rise in parallel in cells subjected to severe oxidative stress. Why is there no such correlation in healthy tissues? Levels of 8OHdG in DNA are a steady-state balance between rate of oxidative DNA damage and rate of removal of lesions by DNA repair mechanisms. Isoprostanes, once formed, are metabolized quickly. Oxidized proteins are degraded, mostly by the proteasome system, and appear to turn over more slowly. Thus even if all biomolecules are damaged, the extents and time courses of the biomarkers of such damage can be very different. Hence, no correlation was observed between levels of plasma isoprostanes and oxidative DNA damage products in healthy human subjects (England et al., 2000).

Measurement of RS in cells

In addition to those of us trying to study oxidative stress in humans, many groups are engaged in the apparently simpler task of trying to measure oxidative stress in cells in culture. Many methods allegedly measuring this have become widely used, but precise information on what they really measure is surprisingly sparse, as discussed below. An important first consideration is that the cell culture process itself imposes oxidative stress, both by facilitating generation of RS and by hindering adaptive upregulation of cellular antioxidants (reviewed by Halliwell, 2003a). For example, the 'Hayflick limit' in fibroblasts (replicative senescence after a certain number of cell divisions) may be largely an artifact of oxidative stress during cell culture (Wright & Shay, 2002). In addition, results can be confounded by free radical reactions taking place in the culture media (Grzelak et al., 2000; Roques et al., 2002; Halliwell, 2003a; Wee et al., 2003). For example, many reports of effects of ascorbate and polyphenolic compounds (e.g. flavonoids) on cells in culture are artifacts caused by the oxidation of these compounds in the culture media (Long et al., 2000; Clement et al., 2001; Long & Halliwell, 2001). It is essential to examine what the medium alone might do when you add compounds to it; the results can be surprising (e.g. Clement et al., 2001, 2002). Sometimes, indeed, the presence of cells suppresses free radical reactions occurring in the medium (Halliwell, 2003a).

Some simple principles can be used as guidelines in understanding oxidative stress/oxidative damage in cell culture. Hydrogen peroxide generally crosses cell membranes readily, probably through the aquaporins (Henzler & Steudle, 2000). Thus catalase added outside cells can exert both intracellular and extracellular effects on H₂O₂ level, the former by 'draining' H₂O₂ out of the cell by removing extracellular H₂O₂ and thus establishing a concentration gradient (Halliwell, 2003a). In contrast, O₂^- does not generally cross cell membranes readily (Lynch & Fridovich, 1978; Marla et al., 1997). Thus if externally added superoxide dismutase is protective against an event in cell culture, be wary of what this means; it could be indicative of extracellular O₂^--generating reactions. Similarly, neither the iron-chelating agent deferoxamine (which suppresses most, but not all, iron-dependent free radical reactions) nor the thiol antioxidant GSH enter cells easily, so again be wary if they have protective effects in short-term experiments: this is suggestive of extracellular effects (Meister & Anderson, 1983; Halliwell, 1989; Marla et al., 1997). As an example, Clement et al. (2002) showed that GSH protects against the cytotoxicity of dopamine simply because it reacts with dopamine oxidation products generated in the cell culture medium. Given long enough, however, everything can enter cells, including deferoxamine, superoxide dismutase (SOD) and GSH (Doulias et al., 2003; Rius et al., 2003).

Studies of RS production in cells can, of course, be achieved by using aromatic compounds or spin traps, described above, and oxidative damage can be measured by assaying such end products as oxidized DNA bases, protein carbonyls, isoprostanes and 3-nitrotyrosine. Spin traps have been used successfully in many cell studies, since a wider range of traps at higher concentrations can be used than could ever be employed in vivo. Again, one must be aware of the possibility of rapid reduction of free radical–spin trap adducts to ESR-silent species by nonenzymic antioxidants (such as ascorbate) and cellular enzymatic reducing systems. An interesting combination is 5-((2-carboxy)phenyl)-5-hydroxy-1-(2,2,5,5-tetramethyl-1-oxypyrrolidin-3-yl)methyl-3-phenyl-2-pyrrolin-4-one sodium salt, a nitroxide that is nonfluorescent. When it combines with a RS, the nitroxide is removed, the ESR signal is lost and the fluorescence is restored (Pou et al., 1993).

Let us now comment on the use of probes, usually fluorescence based, that are alleged to detect cellular production of RS.

Dichlorofluorescin diacetate (DCFDA)

DCFDA is the most popular of these probes, frequently being used to detect 'cellular peroxides'. In fact, it is unlikely to do so because it reacts only slowly with H₂O₂ or lipid peroxides (LeBel et al., 1992; Ischiropoulos et al., 1999; Bilski et al., 2002; Ohashi et al., 2002; Wardman et al., 2002) (Table 5). DCFDA enters cells and accumulates mostly in the cytosol. To avoid any cytotoxicity, cells should be loaded with DCFDA at low concentrations. With a variety of cell types, we have found loading at 1–10 M for 45 min–1 h is adequate. Serum-free media must be used since serum will contain endogenous esterase activity and de-esterified dichlorofluorescein (DCF) is less cell permeable and will generate inconsistent data. Higher levels of DCFDA or high light intensities can also result in an artifactual photochemical oxidation to fluorescent products that can be mistaken for ROS generation (Marchesi et al., 1999; Afzal et al., 2003; Bindokas et al., 2003) (Table 5).

Table 5 - Sensitivity of various probes to different reactive species.

Full table

DCFDA is deacetylated by esterases to dichlorofluorescin (DCFH). This nonfluorescent product is converted by RS into DCF, which can easily be visualized by strong fluorescence at around 525 nm when excited at around 488 nm. Many papers have confused dichlorofluorescin and dichlorofluorescein, so it may be best to stick to correct chemical nomenclature (Figure 1). The chemistry of the conversion is complex (Figure 1). Washing cells after loading will allow unreacted DCFDA to diffuse out again. DCFH and DCF can also diffuse out and undergo extracellular reactions (Ohashi et al., 2002; Subramaniam et al., 2002). Thus be careful if you use a plate reader: you may be measuring light emission from the medium as well as from the cells. Neither H₂O₂ nor O₂^- can oxidize DCFH, but peroxyl, alkoxyl, NO₂, carbonate (CO₃^-) and OH radicals can, as can peroxynitrite (LeBel et al., 1992; Ischiropoulos et al., 1999; Bilski et al., 2002; Ohashi et al., 2002; Wardman et al., 2002; Myhre et al., 2003) (Table 5). DCFDA can only detect cellular peroxides efficiently if they are decomposed to radicals, for example, by transition metal ions. For example, in bovine aortic endothelial cells, the generation of a signal from DCFDA upon addition of H₂O₂ required the uptake of extracellular iron from the medium (Tampo et al., 2003). Horseradish peroxidase, myeloperoxidase and other heme proteins can also oxidize DCFH in the presence of H₂O₂ (indeed, DCFDA was first used in biology as a detector for H₂O₂ by adding horseradish peroxidase; Keston & Brandt, 1965; LeBel et al., 1992). Hence cellular peroxidase level and heme protein content are other variables to consider when interpreting studies with this probe (LeBel et al., 1992; Ohashi et al., 2002). For example, Ohashi et al (2002) showed that manipulating the heme content of cells by adding hemin or -aminolevulinic acid affected rates of DCFH oxidation.

Figure 1.

Conversion of DCFDA to a fluorescent product DCFDA is dichlorofluorescin (sometimes written as dichlorofluorescein) diacetate, perhaps better described by its proper chemical name as 2',7'-dichlorodihydrofluorescein diacetate. It is hydrolyzed by cellular esterases to dichlorofluorescin (2',7'-dichlorodihydrofluorescein), whose oxidation by several RS yields florescent DCF (dichlorofluorescein, more correctly called 2'7'-dichlorofluorescein) via an intermediate radial, DCF^-. Peroxidases can also convert it into a phenoxyl radical that can interact with antioxidants such as ascorbate (AH^-), reducing the phenoxyl radical and oxidizing ascorbate, or with GSH. GS resulting from the latter reaction can lead to O₂^- generation. The phenoxyl radical can also be recycled by NADH (not shown), producing NAD radical, which reacts rapidly with O₂ to produce O₂^-:

Full figure and legend (25K)

It follows that DCF fluorescence is an assay of generalized oxidative stress rather than of any particular RS, and is not a direct assay of H₂O₂, NO, lipid peroxides, singlet O₂ or O₂^-. One-electron oxidation of DCFH by various radicals and heme proteins is likely to produce intermediate radicals (Figure 1), including phenoxyl radicals, that can interact with such cellular antioxidants as GSH and ascorbate and with NADH to create more free radicals (Rota et al., 1999). Lawrence et al. (2003) pointed out that cytochrome c is a powerful catalyst of DCFH oxidation, and so use of DCFDA to probe oxidative stress during apoptosis should be approached with caution, since a rise in cytosolic cytochrome c levels could result in a bigger 'signal' without any change in cellular peroxide levels. Chromium (V), pyocyanin, mitoxantrone and ametantrone can directly oxidize DCFH and cause an artifactual signal, and the possibility of such direct oxidations must always be checked before using DCFDA to measure oxidative stress in cells exposed to various toxins (Martin et al., 1998; O'Malley et al., 2004). Variation in cellular esterase content could also conceivably affect the use of DCFDA as a probe, but this issue has not been explored in the literature.

Hempel et al. (1999) have stated that dihydrofluorescein diacetate has certain advantages over DCFDA in some cellular systems, but again it detects a range of RS. 2-[6-4'hydroxy]phenoxy-3H-xanthen-3-on-9-yl]benzoic acid (HPF) and 2-[6-(4'amino)phenoxy-3H-xanthen-3-on-9-yl]benzoic acid (APF) (Figure 2) have been introduced as probes of ROS. Both fluoresce after reaction with OH, ONOO^- and peroxidase-derived species, but only APF emits light after exposure to HOCl. Both appear more stable in cell systems than DCFDA (Setsukinai et al., 2003). Neither responds to ¹O₂, O₂^-, H₂O₂, NO or peroxyl radicals and both are more stable to photochemical events than DCFDA (Table 5).

Figure 2.

Structures of some common probes used for the detection of RS: (a) HPF and APF, (b) luminol, (c) LC and (d) MCLA. (e) Cis-parinaric acid, (f) 4,4-difluoro-5-(4-phenyl-1,3,-butadienyl)-4-bora-3a,4a-diaza-s-indacene-3-undecanoic acid (C11-BODIPY ^581/591), (g) coelenterazine and (h) 8-amino-5-chloro-7-phenylpyridazo [3,4-d] pyridazine-1,4-(2H, 3H) dione (L-O12).

Full figure and legend (18K)

Dihydrorhodamine 123 (DHR)

DHR is a probe widely used to detect several RS (OH, ONOO^-, NO₂, peroxidase-derived species), but is poorly responsive to O₂^-, H₂O₂ or NO (Buxser et al., 1999). DHR is oxidized to rhodamine 123, which is highly fluorescent around 536 nm when excited at about 500 nm (Figure 3). Rhodamine 123 is lipophilic and positively charged, and tends to accumulate in mitochondria, held there by the membrane potential. As a result, once it is formed, little rhodamine 123 leaks out of cells. DHR is also more sensitive at detecting HOCl than DCFDA (Buxser et al., 1999; Ischiropoulos et al., 1999). Rhodamine 123 and ethidium (see below) can be ejected from cells by drug conjugate efflux pumps, so the presence and activity of these membrane transport systems in the cells being studied is a factor that needs to be considered (Buxser et al., 1999).

Figure 3.

Conversion of dihydrorhodamine 123 to rhodamine 123.

Full figure and legend (8K)

At high levels, rhodamine 123 can sensitize singlet O₂ formation in mitochondria and cause NAD(P)H oxidation (Petrat et al., 2003). Palomba et al. (2000) found that exposure of PC12 cells to the synthetic hydroperoxide tert-butylhydroperoxide caused light emission from dihydrorhodamine. This appeared to be due to increased formation of NO leading to ONOO^- formation, which in turn activated the enzyme phospholipase A₂ to raise cellular arachidonic acid levels. Arachidonic acid metabolism to RS was suggested to be the source of the DHR response.

Dihydroethidium (dihydroethidine) (DHE)

This is frequently used as a probe for O₂^-, being oxidized to a fluorescent product. This is usually thought to be ethidium (Figure 4), which tends to intercalate into nuclear DNA. Ethidium fluoresces strongly at around 600 nm when excited at 500–530 nm. Recent work, however, suggests that the product is not ethidium (Zhao et al., 2003). Although somewhat specific for O₂^- (Buxser et al., 1999; Zhao et al., 2003), DHE is readily spontaneously oxidized and often gives a high background (Buxser et al., 1999). It can also undergo a direct redox reaction with ferricytochrome c (Benov et al., 1998).

Figure 4.

Conversion of DHE to ethidium.

Full figure and legend (8K)

Luminol and lucigenin (LC)

These two compounds (Figure 2) are often used to detect the production of RS by activated phagocytes, although they have also been used in other cell types (Faulkner & Fridovich, 1993). A luminol analogue L-012 (Figure 2) was reported to be more sensitive than luminol for the detection of O₂^- and ONOO^- or than dihydroethidine for O₂^- detection (Daiber et al., 2004).The use of luminol to detect O₂^- is a problematic area. It does not react directly with O₂^- but must first be oxidized in a one-electron step (e.g. by OH, ONOO^- or peroxidase plus H₂O₂). The resulting luminol radical reacts with O₂^- to generate a light-emitting product. Unfortunately, the luminol radical can also reduce O₂ to generate O₂^-, that is, the presence of luminol plus an oxidizing agent can lead to artifactual O₂^- generation (Faulkner & Fridovich, 1993). Hence luminol is an unreliable probe; any oxidizing agent that can oxidize luminol by one electron will cause light emission inhibitable by SOD and the luminol is both the source and the detector of the O₂^-. LC is often said to be more specific for the detection of O₂^- than luminol, but again it does not react directly with O₂^- (Faulkner & Fridovich, 1993; Spasojevic et al., 2000). It must first be reduced to LC cation radical (LC⁺), which then reacts with O₂^- to give the fluorescent product. Conversion of LC to LC⁺ cannot be achieved rapidly by O₂^-, and requires other cellular reducing systems (e.g. xanthine oxidase, the mitochondrial electron transport chain or the phagocyte NADPH oxidase), introducing an obvious complexity in interpreting results. LC⁺ can also reduce O₂ to O₂^-, that is, addition of LC can artifactully generate more O₂^- (Tarpey et al., 1999; Sohn et al., 2000; Spasojevic et al., 2000). The extent to which this artifact can interfere with accurate measurement of O₂^- by LC continues to be debated (Munzel et al., 2002), but appears to be significant in some cell systems (Tarpey et al., 1999; Sohn et al., 2000; Spasojevic et al., 2000; Wardman et al., 2002). Alternative probes that might be more specific include coelenterazine (Tarpey et al., 1999; Munzel et al., 2002) and the luciferin analogue 2-methyl-6-(p-methoxyphenyl)-3,7-dihydroimidazo[1,2-]pyrazin-3-one (MCLA) (Figure 2). However, MCLA may also react with peroxyl radicals (Ukamoto and Yonaha, 1998) and coelenterazine with ONOO^- (Tarpey et al., 1999). Both these compounds are highly light sensitive.

Probes of lipid peroxidation/membrane RS

Several membrane-partitioning probes have been introduced to measure lipid peroxidation or RS within membranes. One is cis-parinaric acid (Figure 2). When incorporated into lipids undergoing peroxidation, it is rapidly oxidized, losing its characteristic fluorescence (emission at 413 nm, excitation 324 nm) (Ritov et al., 1996; Drummen et al., 1999). Parinaric acid needs careful handling, because its polyunsaturated structure (Figure 2) makes it highly susceptible to nonspecific oxidation, and so it should be stored in the dark under a N₂ atmosphere. We have found it to work well, although the need to use the UV range is one disadvantage. For example, the required wavelengths are not always available on microplate readers or confocal/flow cytometry systems. It can be added as the free acid or incorporated into specific phospholipids (parinaroyl lipids) and used to study their relative susceptibilities to oxidative damage within membranes (Ritov et al., 1996; Shvedova et al., 2002).

Another probe is diphenyl-1-pyrenylphosphine (Takahashi et al., 2001), reported to react with peroxides to generate a product that fluoresces at 380 nm when excited at 351 nm (Figure 5).We have found it to be light and air sensitive and difficult to obtain reproducible results using it under our laboratory conditions. Another fatty acid probe becoming widely used is C-11-BODIPY^581/591, which upon oxidation shifts its fluorescence excitation and emission from red to green, so that the green/(red+green) ratio can be used as an estimate of membrane oxidation. Use of the ratio helps to decrease variations caused by heterogeneous probe distributions (cell thickness, uneven dye loading, compartmentalization, etc.) (Drummen et al., 2002). C-11-BODIPY^581/591 responds to a range of RS (OH, peroxyl, alkoxyl, ONOO^-) but not to O₂^-, NO, H₂O₂, singlet O₂ or hydroperoxides. Hence it can be used to determine RS within membranes (Drummen et al., 2002; Yoshida et al., 2003). When added to cells, C-11-BODIPY^581/591 seems to enter most membranes, with no preference for any organelle (Drummen et al., 2002). Again, the presence of these various probes can be expected to alter the rate, and perhaps, the mechanistic pathway, of lipid peroxidation. Serum proteins, especially albumin, avidly bind fatty acids and can deter loading of probes into cells, and may also cause slow leaching out when loaded cells are incubated in the cell culture media with added serum. Yet, serum-deprived cells suffer oxidative stress, which can elevate background production of RS (Lee et al., 2001).

Figure 5.

Conversion of diphenyl-1-pyrenylphosphine to a fluorescent product by peroxides.

Full figure and legend (17K)

How should we measure the output of these probes?

The simplest technique is the fluorescence microplate reader, where data are presented as increases or decreases in relative fluorescence. However, the quality and sensitivity of the machines commercially available vary tremendously, and the requirement for additional excitation and emission filters often makes them expensive. Newer models are available that do not require expensive filters but act on a dual monochromator principle and offer greater flexibility. Simultaneous dual wavelength excitation and emission measurements are particularly useful in determining the fluorescence ratio changes required for C-11-BODIPY^581/591 or the product of reaction of O₂^- with DHE. It is important to check if the machine being used is a top-reading fluorescence machine, since this requires the cells to be in suspension. Bottom-reading machines have the advantage that the cells can be measured in situ without the need for trypsinization or cell scraping, processes that themselves generate cellular oxidative stress (Halliwell, 2003a) and result in artifactual changes in fluorescence. Plate readers measure total fluorescence, that is, they do not distinguish between intracellular and extracellular fluorescence from chemical reactions in the culture medium. We have already alluded to this problem in the case of DCFDA.

Flow cytometry offers the advantage of being able to measure the intracellular fluorescence of cells in the culture media. Quantitative data on the numbers of cells emitting fluorescence can be obtained rather than just relative fluorescence units. However, cells are required to be in suspension and require either scraping or trypsinization, which induce oxidative stress. Control experiments to optimize assay conditions must always be conducted to limit this. An additional disadvantage of flow cytometry is that often the cytometer is at room temperature rather than 37°C, which can give variable data with certain cells such as murine primary cortical neurones (Dr Nam Sang Cheung, unpublished observation).

Confocal microscopy is a powerful tool; cells can be loaded with fluorescent dyes and viewed in real time in situ in culture chambers at 37°C. The intracellular location of RS can be visualized, and the role of mitochondrial, endoplasmic reticulum or lysosomal events in oxidative stress may be visualized using counter stains such as MitoTracker, ER Tracker or Lysotracker dyes coloaded with dyes that detect RS. However, caution must be excercised to ensure that the signal from the organelle Tracker stain does not interfere with the measurement of the RS-sensitive dye and vice versa. It should also be stressed that simple overlapping of the Tracker stains with the signal from the RS-sensitive dyes only suggests involvement of a particular organelle, but does not rigorously prove it.

Effects of RS on other probes

A wide range of probes is available to study many cellular events, such as pH changes and ion movements and to identify various cellular organelles (as mentioned above). For example, rises in Ca²⁺ frequently accompany oxidative stress (Halliwell & Gutteridge, 1999). Several papers have indicated that ion probes can be affected by the generation of RS (e.g. the Ca²⁺ probe Fura-2 can be degraded; Sarvazyan et al., 1998) or can themselves contribute to RS formation, such as calcein (Beghetto et al., 2000). MitoTracker Red, a fluorescent probe used to identify mitochondria, can stimulate mitochondrial ROS generation (Minamikawa et al., 1999), for example.

Conclusions

Whatever method you use to trap RS or measure oxidative damage, it is necessary to think carefully about how the method works, what is likely to confound it and how quantitative it can be (how, what and how much). With careful attention to understanding these points, erroneous interpretations can be minimized and the vibrant field of free radical research can continue to move forward.

References

1. ADAMS S., GREEN P., CLAXTON R., SIMCOX S., WILLIAMS M.V., WALSH K. & LEEUWENBURGH C. (2001) Reactive carbonyl formation by oxidative and non-oxidative pathways. Front. Biosci. 6: A17−24. | ChemPort |
2. AFZAL M., MATSUGO S., SASAI M., XU B., AOYAMA K. & TAKEUCHI T. (2003) Method to overcome photoreaction, a serious drawback to the use of dichlorofluorescin in evaluation of reactive oxygen species. Biochem. Biophys. Res. Commun. 304: 619−624. |