On the methodological limitations of detecting oxidative stress: effects of paraquat on measures of oxidative status in greenfinches

Aerobic metabolism inevitably creates reactive oxygen species (ROS). These hyper-reactive molecules are essential for many important physiological functions, provided that their production and neutralisation by antioxidant systems are properly regulated by an organism. If the regulatory system is overwhelmed, distracted or fails, uncontrolled production of ROS can lead to biomolecular damage, characterised as a state of oxidative stress, OS (Sies, 1991; Halliwell and Gutteridge, 2007). Managing OS is now widely believed to be a key physiological factor generating life history trade-offs and constraining investment in a range of traits, such as growth, reproduction, lifespan and communication (von Schantz et al., 1999; Dowling and Simmons, 2009; Monaghan et al., 2009; Metcalfe and Alonso-Alvarez, 2010; Garratt and Brooks, 2012).

One of the most urgent problems requiring clarification in the realm of ‘oxidative stress ecology’ (McGraw et al., 2010) is the question of whether and how increased OS impinges on fitness (Hõrak and Cohen, 2010; Selman et al., 2012). Although ROS are involved in many pathological conditions (Halliwell and Gutteridge, 2007), demonstration of such effects in wild animals has remained a challenging task and to date only a few convincing examples of the causal effects of OS on variation in fitness-related traits exist (reviewed by Monaghan et al., 2009; Buttemer et al., 2010; Metcalfe and Alonso-Alvarez, 2010; Garratt et al., 2011; Selman et al., 2012). Similar problems have been described in laboratory animal models (Cohen et al., 2010; Speakman and Selman, 2011).

Yet, some studies on wild animals have demonstrated connections between measures of antioxidant capacity and some components of fitness, such as survival (Saino et al., 2011; Noguera et al., 2012), reproductive effort (Morales et al., 2008; Markó et al., 2011), telomere length (Beaulieu et al., 2011), sperm quality (Bonisoli-Alquati et al., 2011) or expression of signal traits (Markó et al., 2011). Many more have demonstrated various connections between measures of antioxidant capacity and different types of stress (Cohen et al., 2007; Cohen et al., 2008; Beaulieu et al., 2011; Costantini et al., 2011). Such findings are, however, difficult to interpret without proper knowledge about the causes and consequences of natural and experimentally generated variation in individual antioxidant levels. Thus, despite the exponential increase of publications related to OS ecology during the past decade (McGraw et al., 2010), progress in understanding the true role of OS in ecology and evolution has remained elusive. There is disagreement about the proper sampling schemes to use, metrics to obtain and assays to run that capture proper antioxidant functions and balances relative to OS demands. Many of the papers published so far on free-living animals suffer from methodological problems, and even the conceptual issues, such as functional definitions of OS, damage and oxidative balance systems, are still under debate (Halliwell and Gutteridge, 2007; Costantini and Verhulst, 2009; Pamplona and Costantini, 2011; Garratt and Brooks, 2012; Selman et al., 2012).

Proper interpretation of previously established associations between measures of antioxidant capacity and components of fitness requires experiments on model organisms of animal ecology that involve generation of systemic OS in vivo (e.g. Pérez-Rodríguez, 2009; Hõrak and Cohen, 2010). We are currently aware of only a few similar experiments, performed in wild bird species. For instance, Isaksson and Andersson generated systemic OS in great tits (Parus major) by administration of the pro-oxidant compound paraquat in order to test for the in vivo antioxidant properties of carotenoids (Isaksson and Andersson, 2008). Galván and Alonso-Alvarez used a related compound, diquat, for testing the effects of OS on the expression of melanin-based colouration and various markers of oxidative status in red-legged partridges, Alectoris rufa (Galván and Alonso-Alvarez, 2009; Alonso-Alvarez and Galván, 2011).

Diquat (1,1′-ethylene-2,2′-bipyridylium) and paraquat (1,1′-dimethyl-4,4′-bipyridylium, PQ) are widely used agricultural chemicals and environmental contaminants. Redox cycling and consequent ROS generation are thought to be key cytotoxic mechanisms induced by these bipyridyl herbicides (reviewed by Fussell et al., 2011). During redox cycling, diquat and PQ undergo an enzymatic one-electron reduction, forming radical cations. Under aerobic conditions, these radicals react rapidly with molecular oxygen, forming superoxide anion and regenerating the parent compounds. Dismutation of superoxide anion generates H₂O₂ and, in the presence of redox-active transition metals, highly toxic hydroxyl radicals. These ROS can cause OS, leading to damage to lipids, proteins and nucleic acids (Halliwell and Gutteridge, 2007). PQ does not naturally occur in animal circulation. However, as the main mechanism underlying PQ toxicity is OS due to overproduction of ROS (Dinis-Oliveira et al., 2008), its administration has become a standard tool for generation of OS in biological systems (e.g. Halliwell and Gutteridge, 2007; Knasmüller et al., 2008). Because the prime purpose of the current study was to test the sensitivity of a number of markers of redox status to experimentally induced OS, we considered induction of OS by administration of PQ relevant for our research purposes.

Here, we report a study of PQ administration to greenfinches, Carduelis chloris (Linnaeus 1758), in order to test how experimentally generated OS affects different parameters of antioxidant protection [total antioxidant capacity (TAC) and oxygen radical absorbance (OXY)], individual antioxidants (glutathione, uric acid and carotenoids) and indicators of oxidative damage such as malondialdehyde (a marker of lipid peroxidation), protein carbonyls (a marker of protein oxidation) and measures of DNA damage, assessed in single cell gel electrophoresis (comet assay). DNA damage is expectedly one of the most informative measures of OS because it combines both free radical production and attack as well as antioxidant and repair mechanisms (Collins, 2009). Measuring markers of oxidative damage is important because most definitions of OS involve damage. However, perhaps because of their methodological complexity and requirement for specialised equipment (in the case of proper quantification of malondialdehyde, MDA), such measures have been regrettably under-represented in ecological studies of animals. We predicted that in the case of successful generation of systemic OS by PQ treatment, all our measures of oxidative damage should increase. If higher levels of antioxidant capacity and individual antioxidants reflect a beneficial redox state, these should decrease in PQ-exposed birds. Alternatively, if OS induces protective upregulation, antioxidant levels may also increase after exposure to OS (Prior and Cao, 1999; Costantini et al., 2010). Finally, we highlight the need for reconsideration of over-simplistic models of OS and present the framework providing some conceptual aid for planning and interpreting further experiments in OS ecology.

Thirty-six wild female greenfinches were captured in mist-nets at bird feeders in a garden in the city of Tartu, Estonia (58°22′N, 26°43′E) on 9 January 2012. The birds were housed indoors in individual cages (27×51×55 cm) with sand-covered floors. Mean (±s.d.) temperature in the aviary during the experiment was 11.1±1.5°C and humidity was 39±3%. The birds were supplied ad libitum with sunflower seeds. Until the start of the PQ treatment and after the treatment had ended, all the birds received 10 μg ml⁻¹ carotenoid solution in their drinking water to compensate for the naturally low carotenoid content of sunflower seeds. Carotenoid supplementation consisted of lutein and zeaxanthin (20:1, w/w), prepared from Oro Glo liquid solution of 11 g kg⁻¹ xanthophyll activity (Kemin AgriFoods Europe, Herentals, Belgium). Birds were held on the natural day-length cycle using artificial lighting by luminophore tubes. The birds were released into their natural habitat on 9 March 2012. The study was conducted under licence from the Estonian Ministry of the Environment (licence no. 1-4.1/11/100, issued on 23 March 2011) and the experiment was approved by the Committee of Animal Experiments at the Estonian Ministry of Agriculture (decision no. 95, issued on 17 January 2012).

On the morning of 26 January, before the lights had been switched on, all the birds were weighed and their blood sampled (ca. 200 μl of blood from the tarsal or brachial vein) for recording the baseline values of haematological parameters. Subsequently, the birds were divided into three groups of 12 individuals on the basis of similar age and body mass. From 27 January onwards, birds in one of the experimental groups (PQ1) started to receive a solution of PQ (Sigma-Aldrich 856177, St Louis, MO, USA) at a dose of 0.1 g l⁻¹ in their drinking water. The second experimental group (PQ2) received 0.2 g l⁻¹ PQ solution; the control group received just water. Doses of PQ were selected on the basis of a previous study (Isaksson and Andersson, 2008) where administration of similar doses to great tits (a passerine about 2/3 the size of a greenfinch) proved to be non-lethal during the 6 week experimental period. PQ treatment lasted for 7 days and was terminated after the discovery of a decline in body mass in the PQ2 group. Admittedly, administration of PQ with drinking water does not enable us to control for possible between-individual differences in water consumption, which might obscure the results. However, we chose this way of administration because it has been shown to be effective in previous studies (Galván and Alonso-Alvarez, 2009; Alonso-Alvarez and Galván, 2011). Additionally, we were concerned about causing extra stress to our experimental birds by administration of PQ via daily injection or oral intubuation. Between the evening of day 7 and the evening of day 8, five birds in the PQ2 group had died. Because we were unable to predict such an outcome on the basis of body mass recorded on the evening of day 7, we considered that termination of PQ administration would be sufficient to alleviate the suffering of experimental animals. No mortalities occurred before the ending of PQ treatment. One more bird in PQ2 group and one bird in the PQ1 group died 2 days after the end of PQ treatment. No more mortalities occurred up to the release of the birds on 9 March, by which time all of them had recovered their pre-experimental body mass. Birds were blood sampled again on the morning of 6 February, i.e. 3 days after the end of the PQ treatment. We did not collect blood samples immediately after terminating PQ administration mainly because we expected that ecologically relevant effects of severe OS on studied physiological markers would last longer than 1 day. An additional reason was that we were concerned that stress associated with blood sampling immediately after the end of PQ treatment might be lethal to all individuals in the high PQ group.

Blood was collected into 200 μl Microvette tubes with lithium heparin as an anticoagulant. Immediately after blood collection, tubes were placed into a cooled and light-protected box on snow and centrifuged within 1 h of sampling for 5 min at 6700 g at 4°C to separate plasma from erythrocytes. Plasma and erythrocytes were stored at −80°C until analysed within 2 months. Blood for the comet assay was stored at room temperature until processed later the same day.

The concentration of carotenoids was determined spectrophotometrically from 15 μl of plasma, diluted in acetone as described elsewhere (Sild et al., 2011). The concentration of uric acid was determined from 5 μl plasma samples by an enzymatic colorimetric test with lipid clearing factor (uric acid liquicolor, HUMAN, Wiesbaden, Germany).

Plasma TAC was measured from 5 μl plasma samples according to a method described previously (Erel, 2004) with minor modifications (see Sepp et al., 2010). The assay is based on the capacity of antioxidants in the solution to decolorise ABTS⁺ [2,2-azinobis(3-ethylbenzothiazoline-6-sulphonate)] according to their concentration and antioxidant capacity. The main contributors to TAC are plasma uric acid and free sulphydryl groups of proteins (Erel, 2004; Sepp et al., 2010). The results are quantified in mmol l⁻¹ Trolox (water-soluble vitamin E analogue) equivalents.

Plasma OXY was measured with the OXY-adsorbent test (Diacron International, Grosseto, Italy) from 5 μl plasma samples according to the manufacturer's instructions (Costantini, 2011). This test quantifies the ability of the plasma non-enzymatic antioxidant compounds to cope with the in vitro oxidant action of hypochlorous acid (HOCl, an endogenously produced oxidant). The concentrations of OXY are expressed as mmol l⁻¹ of HOCl neutralised. Previous studies in birds, including greenfinches, have shown that OXY does not correlate with TAC (Sepp et al., 2012a).

Total glutathione (GSH) levels in erythrocytes were determined within a week of sampling as described elsewhere (Galván and Alonso-Alvarez, 2008; Rahman et al., 2007) with modifications (see Hõrak et al., 2010). The results are given in μmol g⁻¹ pellet. Plasma MDA was assessed from 5 μl plasma samples by liquid chromatography mass spectrometry analysis based on a modified protocol (Andreoli et al., 2003) as described elsewhere (Hõrak et al., 2010). The repeatability of plasma MDA concentration was 0.82 (F_47,48=10.3, P<0.0001).

Determination of carbonyl level in proteins is used as an index of the extent of protein oxidative damage. The assay was carried out as previously described (Chiou et al., 2012). Briefly, after centrifugation, erythrocytes were washed twice with 0.9% sodium chloride and were centrifuged again at 6700 g. A 5% erythrocyte suspension in 0.15 mol l⁻¹ NaCl, 10 mmol l⁻¹ sodium phosphate buffer (pH 7.4) was stored at −20°C; 200 μl of the haemolysate was mixed with 600 μl of 8 mmol l⁻¹ 2,4-dinitrophenylhydrazine in 3 mol l⁻¹ HCL. The samples were incubated for 50 min at 37°C in the dark and vortexed every 10 min. Then, 600 μl of 25% (w/v) trichloroacetic acid was added and the tube was left on ice for 8 min and subsequently centrifuged for 3 min at 6000 g to collect the protein precipitates. This pellet was washed using 200 μl of 15% trichloroacetic acid. The pellet was washed twice with 800 μl of ethanol–ethyl acetate (1:1, v/v). The final precipitate was dissolved in 300 μl of 5 mol l⁻¹ guanidine hydrochloride solution and was incubated for 15 min at 37°C with mixing on a thermoshaker at 275 r.p.m. Any insoluble materials were removed by repeated centrifugation. Carbonyl level was calculated from the peak absorbance of the spectra at 355–390 nm, using an absorption coefficient of 22,000 l mol⁻¹ cm⁻¹. The results are expressed as μmol mg⁻¹ protein. The protein content of the sample was assessed using Bio-Rad Protein Assay Kit (Bio-Rad Laboratories, Hercules, CA, USA) according to the manufacturer's instructions. The repeatability of protein carbonyl concentration was 0.65 (F_51,92=6.2, P<0.0001).

The comet assay technique, modified for the detection of oxidised bases, enables oxidative damage to DNA to be assessed. The alkaline version of the assay measures DNA strand breaks and alkali-labile sites, i.e. apurinic/apyrimidinic sites or baseless sugars, while the use of bacterial repair endonucleases enables detection of oxidised DNA bases (mainly 8-oxo-7,8-dihydroguanine, 8-oxoGua). In order to calculate the amount of oxidised DNA bases, we ran simultaneously both the basic alkaline version of the comet assay (which detects strand breaks and alkali-labile sites) and the same comet assay on cells incubated with a bacterial endonuclease (formamidopyrimidine DNA glycosylase, FPG). This enzyme converts oxidised DNA bases to alkali-labile sites, which can be detected with the alkaline comet assay. Subtracting the result of one assay from the other gives a reliable estimate of the total amount of oxidised DNA bases (Collins, 2009).

The intactness of the DNA in erythrocytes was evaluated using a Trevigen Comet Assay kit (Trevigen, Gaithersburg, MD, USA) according to the instructions in the Trevigen FLARE Assay Kit manual, with slight changes in electrophoresis conditions. Briefly, immediately after sampling, 1 μl of whole blood was serially diluted in PBS (Ca²⁺ and Mg²⁺ free) to reach a concentration of roughly 1×10⁵ cells ml⁻¹; 10 μl of the suspension obtained was then mixed with 105 μl of molten LMAgarose (1% low-melting agarose, Trevigen) and 50 μl of this was immediately pipetted and evenly spread onto both wells of the comet slide. The slide was then incubated at 4°C in the dark for 60 min to accelerate gelling of the agarose disc and then transferred to pre-chilled Lysis solution (Trevigen) with 10% DMSO for 4±1 h at 4°C. Next, the slides were washed three times for 10 min each in buffer (10 mmol l⁻¹ Hepes, 0.1 mol l⁻¹ KCl, 10 mmol l⁻¹ EDTA, 0.2 mg ml⁻¹ BSA, pH 7.4) and then incubated in a humidity chamber at 37°C for between 30 and 45 min. For each slide, one well received 80 μl⁻¹ buffer containing 28 U ml⁻¹ bacterial FPG (Sigma F3174) while the same amount of buffer only was pipetted on the other well. After incubation, a denaturation step was performed in alkali solution (300 mmol l⁻¹ NaOH, 1 mmol l⁻¹ EDTA, pH 13) at room temperature for 20 min, in the dark. The slide was then transferred to pre-chilled alkaline electrophoresis solution (300 mmol l⁻¹ NaOH, 1 mmol l⁻¹ EDTA, pH 13) and subjected to electrophoresis at 1 V cm⁻¹, for either 10 min (for the blood collected before PQ treatment) or 5 min (for the blood collected after treatment) in the dark at room temperature. The electrophoresis time was shortened to avoid comet tails detaching from their heads. Hence, in order to calculate the change in oxidative DNA damage between the first and second blood sampling, individual values of the DNA damage markers were standardised to z-scores (mean=0, s.d.=1) within each blood sampling event. At the end of the electrophoresis, the slides were washed twice for 10 min with distilled water, immersed in 70% ethanol at room temperature for 5 min and air dried at 37°C. DNA was stained with 50 μl of SYBR Green I dye (Trevigen, 1:10,000 in Tris-EDTA buffer, pH 7.5) for 20 min in the dark and immediately analysed using an Olympus digital camera attached to an Olympus BX41 epifluorescence microscope.

For each slide well, 85±29 randomly chosen comets were analysed using an Olympus BX41 epifluorescence microscope with an excitation filter of 470–490 nm and a barrier filter of 510–550 nm. Fluorescent images of single cells were captured at 100× magnification and images were scored for percentage of DNA in the tail using Tritek CometScore Freeware v1.5 image analysis software. After scoring, 30 of the most aberrant comets (i.e. 15 cells with the highest and 15 cells with lowest scores) for each individual were excluded from the dataset in order to reduce manual scoring error. The amount of oxidised purines in DNA was calculated by subtracting the mean value of buffer-only control wells (representing DNA strand breaks and alkali-labile sites) from the mean values of enzyme-treated wells (representing strand breaks, alkali-labile sites and oxidised purines in the DNA) as described elsewhere (Collins, 2009). The repeatability of the percentage of DNA in the tail of buffer-only control wells was 0.79 (F_55,3027=209.3, P<0.0001) and that of enzyme-treated wells was 0.80 (F_59,3234=217.6, P<0.0001).

Mean values of the studied parameters before and after treatment were compared using one-way ANOVA with treatment as a factor with three levels (no PQ, low PQ and high PQ). The effects of treatments in terms of changes in the studied parameters between the first and second blood sampling were tested in mixed-model ANOVA (PROC MIXED, SAS v. 9.2 software), including individual identity as a random factor. Treatment, time (first versus second blood sampling) and time×treatment interaction were included as fixed factors. A significant time×treatment interaction term was considered as indicative of the effect of treatment on the change in parameter value. Denominator degrees of freedom for type 3 tests of fixed effects were estimated using the Kenward–Roger method. Directions of the treatment effects are indicated in Table 1 and Fig. 1. Assumptions for parametric models (normality of residuals, homogeneity of variances) were met for all the models. All tests are two-tailed with an α-level below 0.05 as a criterion for significance. Sample sizes vary between different analyses because of our inability to collect sufficient good quality blood samples from all birds.

None of the studied parameters differed between treatment groups at first blood sampling, i.e. prior to PQ treatment (Table 1). PQ treatment for 7 days resulted in 50% mortality among the birds in the high PQ (0.2 g l⁻¹) group. The survivors in that group exhibited a significant (14%) drop of body mass, a 184% increase in oxidative DNA damage and a 13% increase in GSH levels in erythrocytes (Fig. 1, Tables 1, 2). None of these effects could be detected in the low PQ (0.1 g l⁻¹) group. Treatment with PQ had no effect on our measures of lipid (MDA) or protein (carbonyl) peroxidation, parameters of antioxidant protection (TAC and OXY), plasma uric acid or carotenoids. Body mass was the single parameter showing significant repeatability between blood sampling events in the pooled data (R=0.64, F_28,29=4.5, P<0.0001), as well as in the control (R=0.92, F_11,12=51.4, P<0.0001) and PQ1 group (R=0.84, F_10,11=11.7, P=0.0002) separately. Among the control birds, GSH (R=0.50, F_10,11=3.0, P=0.043) and OXY (R=0.71, F_6,7=5.9, P=0.018) also appeared significantly repeatable between blood sampling events. None of the other measured variables revealed significant individual consistency in time (all P>0.1 for both pooled data and treatment groups separately). As indicated in Table 1, all birds lost mass during the experimental period while at the same time their levels of carbonylated proteins increased and TAC decreased irrespective of treatment. Other variables did not show any consistent changes during the experimental period (Table 2).

Our experiment resulted in inadvertent generation of 50% mortality in the treatment group receiving 0.2 g l⁻¹ PQ in their drinking water. Survivors exhibited a remarkable (transient) drop in body mass and an increase in erythrocyte GSH levels and oxidative damage to DNA. The main mechanism of PQ toxicity is through the production of the superoxide anion, which catalyses the formation of additional ROS, such as hydrogen peroxide and the most noxious hydroxyl radical (Dinis-Oliveira et al., 2008). PQ may also affect other physiological processes, such as insulin metabolism (e.g. Kimura et al., 2010), that indirectly may influence oxidative status. It is thus most likely that our experiment resulted in the induction of OS in the high PQ treatment group.

Proper assessment of OS requires measurement of damage to the most important classes of biomolecules, i.e. lipids, proteins and DNA (Selman et al., 2012). To our knowledge, this is the first study in a wild bird species where all these markers have been assessed simultaneously in the context of experimentally induced OS. Use of the modified comet assay technique enabled us to detect oxidatively damaged purines, the major substrate being 8-oxoGua (Collins, 2009). 8-oxoGua is highly mutagenic and it is considered to be one of the most deleterious of ROS-induced adducts, having a key role in pathophysiological processes such as cancer and ageing (e.g. Bruner et al., 2000). More generally, maintenance of genetic integrity is needed for the correct expression of many enzyme-dependent mechanisms and to prevent a number of diseases (Jackson and Bartek, 2009). Our experiment thus resulted in the detection of a biologically meaningful measure of oxidative damage. This finding suggests the high potential utility of measurement of oxidised purines in avian erythrocytes. The relevance of this marker is further supported by a study in yellow-legged gull (Larus minchellis) chicks where growth rate correlated positively with oxidative damage to DNA bases and lipids (Noguera et al., 2011).

Contrary to our predictions, we failed to detect oxidative damage to lipids and proteins on the basis of MDA and protein carbonyls. Neither did we record any changes in parameters related to total antioxidant protection (TAC, OXY), uric acid or carotenoids. This result could be partly ascribed to the low test power due to small sample sizes. However, similar sample sizes enabled us to detect the experimental effects on DNA damage and GSH, so we can assert that if the damage to lipids and proteins or changes in other parameters occurred, the effect was smaller than the effect on DNA damage and GSH. It should also be noted that none of the studied parameters was affected by the low PQ treatment despite sample sizes comparable to those of most biomedical studies. Similar to the findings of the current study, PQ treatment did not affect plasma MDA in Japanese quail (Coturnix coturnix japonica) (Galvani et al., 2000). However, in that study an increase in lung MDA content was detected, so we cannot exclude the possibility that PQ treatment in greenfinches might have increased lipid peroxidation in tissues other than blood.

Mean TAC decreased and protein carbonyls increased between blood samplings in all experimental groups. Currently, we lack explanations for these patterns. None of the measured parameters, except body mass, showed significant repeatability over the 11 day period between blood samplings in the pooled data. However, GSH and OXY were significantly repeatable among the control birds. A previous study in captive greenfinches in our lab established that plasma carotenoids, TAC and erythrocyte GSH showed significant individual consistency over an 8 day period, while no individual consistency was detected in plasma OXY, MDA and uric acid (Sepp et al., 2012a). Thus, only the results for GSH in control birds are consistent with the findings of Sepp and colleagues. The discrepancies between their results and ours can be explained by a lower test power due to smaller sample size in the current study and a longer time lag between the measurements (11 versus 8 days). However, the possibility that our treatments had disrupted the normal homeostasis of measured parameters cannot be excluded either, especially in the case of GSH, which increased significantly in the PQ2 group.

The major limitation of the current study was our inability to measure the parameters of oxidative status in the blood of birds treated with a high dose of PQ before the occurrence of mortality. The sample of survivors obviously contained only the birds that managed to tolerate the PQ-induced OS, so we cannot be confident that we could have detected a similar increase in GSH (or a lack of increase of MDA and protein carbonyls) among non-survivors. Furthermore, as PQ was administered via drinking water, we cannot exclude the possibility that the survivors ingested less PQ than non-survivors as a result of lower water consumption. It is thus noteworthy that despite these problems we still documented increased DNA damage among survivors of high PQ treatment.

One possible explanation for the patterns emerging in the current study would be that the effects of PQ treatment on our measures of lipid and protein peroxidation, as well as on most of the parameters related to antioxidant protection, had vanished by the time of measurement. Second blood sampling occurred 3 days after termination of the PQ treatment and it might seem possible that the birds had restored the homeostasis of most of the physiological measures by that time. However, only two of the measured haematological parameters in the control group were significantly repeatable between the two blood sampling events and none of these parameters was repeatable among PQ-treated birds. This may mean that at least under current experimental conditions, the majority of studied markers reflected only short-term changes or random fluctuations in redox balance. If such findings apply more generally, it would imply that the chances of detecting OS on the basis of the majority of markers typically measured in ecological studies of animals are poor. For instance, it would be difficult to imagine that any naturally occurring source of OS (such as senescence or voluntarily increased work load) would induce systemic oxidative insults comparable in magnitude to the high PQ treatment applied in the current study.

The hypothesis that animals are capable of restoring their redox homeostasis just 3 days after the cessation of a major oxidative insult would be difficult to reconcile with the idea of OS being a major mechanism responsible for the generation of physiological trade-offs involving life-history and signal traits. However, we cannot exclude the hypothetical possibility that (quick) restoration of redox homeostasis per se might appear costly in terms of some ecologically relevant currency. Finally, it should be noted in the context of the current study that two previous experiments conducted on greenfinches in our lab have detected increased lipid peroxidation (plasma MDA) in response to the experimental blocking of GSH synthesis (Hõrak et al., 2010) and coccidian infection (Sepp et al., 2012b). One possible explanation for discrepancies between the results of the current study and those cited above might be the different timing of blood sampling with respect to treatments between studies. In a study by Hõrak and colleagues, the birds were bled 1 day after the last injection of buthionine sulphoximine, an inhibitor of GSH synthesis (Hõrak et al., 2010). In the study by Sepp and colleagues, the birds were bled during the peak phase of experimental coccidian infection (Sepp et al., 2012b).

The lack of treatment effects on TAC and OXY is noteworthy because both parameters have been repeatedly shown to be sensitive to various physiological manipulations such as induction of restraint stress (Cohen et al., 2007), immune challenges (reviewed by Costantini and Møller, 2009), dietary manipulations (e.g. Arnold et al., 2010; Costantini, 2010; Bourgeon et al., 2011), strenuous exercise (e.g. Neubauer et al., 2010; Beaulieu et al., 2011), administration of testosterone (Alonso-Alvarez et al., 2009; Tobler and Sandell, 2009) or glucocorticoid hormones (reviewed by Costantini et al., 2011). Yet, chemical induction of OS had no effect on these parameters in our experimental protocol, which involved bleeding the birds 3 days after termination of the 7 day PQ treatment. Similarly, neither plasma TAC nor one of its correlates (uric acid) was influenced by the administration of diquat in red-legged partridges (Galván and Alonso-Alvarez, 2009) or PQ in American kestrel (Falco sparverius) nestlings (Hoffman et al., 1987). Taken together, these findings raise the question of whether the popular and handy assays of antioxidant protection, such as TAC and OXY, are biologically relevant markers of oxidative status in the first place (see also Sies, 2007; Garratt and Brooks, 2012). Alternatively, the lack of an effect on the two measures of antioxidant capacity might be the result of mobilisation of antioxidants from tissues where they are stored to plasma, which might have masked the depletion in circulating antioxidants. If so, this would mean that these assays are sensitive to OS but that the effect may appear after a longer experimental period than that considered in the present study or quickly if the availability of antioxidants (stored in tissues or from food) is limited.

Regarding the lack of PQ-induced effects on plasma carotenoids, our results are similar to those of Isaksson and Andersson (Isaksson and Andersson, 2008) where administration of PQ at a dose of 0.09 g l⁻¹ for 21 days had no effect on carotenoids in plasma, liver and feathers in great tits. Another experiment manipulating oxidative status in greenfinches by blocking the synthesis of GSH also failed to detect any effects on circulating carotenoids or carotenoid-based feather colouration, although plasma MDA levels were increased (Hõrak et al., 2010). These findings are consistent with the idea that circulating carotenoids do not contribute much to the antioxidant defences of birds (Hartley and Kennedy, 2004; Costantini and Møller, 2008; Pérez-Rodríguez, 2009). However, administration of a related pro-oxidant, diquat, to growing red-legged partridges (0.5 ml l⁻¹ for 33 days) reduced plasma carotenoid levels and allocation of carotenoids into red beaks and eye rings (Alonso-Alvarez and Galván, 2011). That experiment also found that diquat treatment reduced GSH content and lipid peroxidation in erythrocytes (Galván and Alonso-Alvarez, 2009). These findings again contrast with our results showing no effect on lipid peroxidation and a considerable increase in erythrocyte GSH. GSH is a major endogenous antioxidant, participating directly in the neutralisation of free radicals, as well as recycling exogenous antioxidants such as vitamins C and E. GSH is a cofactor in GSH peroxidases, which reduce H₂O₂ and peroxidised fatty acid residues (Halliwell and Gutteridge, 2007). It has been reported that GSH concentration in different tissues decreases as a consequence of oxidative challenges in birds (reviewed by Galván and Alonso-Alvarez, 2009). However, OS can also increase GSH synthesis (reviewed by Alonso-Alvarez et al., 2008; Rodríguez-Estival et al., 2010). In the current study, the high PQ dose increased erythrocyte GSH levels while the low dose had no effect on GSH.

More generally, such contrasting patterns illustrate the difficulties of assessment of OS because of compensatory upregulation of antioxidant protection mechanisms in response to oxidative insults (e.g. Prior and Cao, 1999; Hõrak and Cohen, 2010; Garratt and Brooks, 2012). We propose that the chances of detection of OS are particularly hampered by the time-lagged nature of such processes because the markers of antioxidant protection may remain elevated long after any measures of damage have dropped back to initial levels (Fig. 2A). Such a model clearly indicates that the chances of detecting the occurrence of past OS depend on the timing of the measurement with respect to the dynamics of both damage and protection. An example is provided by the induction of exhaustive exercise (incremental treadmill running until volitional exhaustion) in trained young men (Vider et al., 2001). Immediately after exercise, increased markers of lipid peroxidation and elevated blood total GSH levels were detected. However, 30 min after the exercise the effects on lipid peroxidation had vanished, so that changes in redox state could only be detected by elevated plasma total antioxidant status and blood GSH. In the case of such delayed responses, it also becomes obvious that correlations between different measures of oxidative damage, levels of antioxidant protection and individual antioxidants can vary in sign and magnitude, again depending on the dynamics of protective responses. Indeed, such a pattern in correlations is evident in a previous meta-analysis (Dotan et al., 2004). All this undermines the practical applicability of simplistic ‘seesaw’ models of OS (see also Garratt and Brooks, 2012).

Notably, the scenarios outlined in Fig. 2A seem more widespread than those depicted in Fig. 2B, which represents an increase in oxidative damage in parallel with a decrease (or increase) in antioxidant protection. For instance, in a meta-analysis of the effects of pollution on parameters of oxidative damage and antioxidant protection in wild animals, the single parameter with significant effect size was an increase in GSH levels (Isaksson, 2010). Another meta-analysis (Costantini and Møller, 2009) showed that induction of immune responses explained only 4% of the variation in the increase of markers of OS (i.e. increase in damage and reduction in protection). However, when the direction of the effect was removed (any relevant parameter was considered just as a marker of oxidative status), the explanatory power of the model rose to 15%. Similarly, signed effect size (reflecting more damage and less protection) was lower (0.55) than non-signed effect size (0.75) in a meta-analysis of the effects of glucocorticoid hormones on oxidative balance (Costantini et al., 2011).

In conclusion, this study has demonstrated that even chemical induction of severe OS can yield unexpected results: not all measures of oxidative damage may increase and measures of protection may remain unchanged. This study also stresses the importance of timing for measurement of the markers of oxidative status with respect to the occurrence of oxidative insults. On the positive side, our experiment underpins the diagnostic value of measurement of oxidative damage to DNA bases and assessment of GSH levels. We hope that this study will highlight the need for reconsideration of over-simplistic models of OS and that the framework sketched in Fig. 2 might provide some conceptual aid for planning and interpreting further experiments in OS ecology. In particular, we encourage the experimental induction of OS in order to confirm whether TAC and OXY respond to the generation of oxidative damage. Such an approach is indispensable for understanding and interpreting the function and origin of variation in these increasingly popular and easily measurable indices of antioxidant protection.

We thank Tuul Sepp, Ulvi Karu, Marju Männiste, Pirko Jalakas and Martin Pent for help with bird maintenance, experiments and biochemical analyses. We thank Ants Kaasik for help with statistics and Tuul Sepp and two anonymous reviewers for constructive comments on the manuscript. Stefaan Van Dyck (Kemin Agrifoods Europe) kindly donated OroGlo carotenoid supplement.