Distinct physiological strategies are used to cope with constant hypoxia and intermittent hypoxia in killifish (Fundulus heteroclitus)

Variations in oxygen availability influence the quantity and quality of habitat available for fish (Graham, 1990; Burnett, 1997; Breitburg et al., 2009). Relatively stable, constant hypoxia can develop in ice-covered or stratified lakes, or following eutrophication events (Diaz, 2001; Diaz and Rosenberg, 2008). Intermittent patterns of hypoxia exposure are common in tide pools and estuaries due to a variety of factors, including daily cycles of respiration and photosynthesis (Breitburg, 1992; Diaz, 2001; Tyler et al., 2009). Overall, the incidence of aquatic hypoxia is expected to rise due to global climate change, urbanization, pollution, and other anthropogenic causes (Diaz, 2001; Ficke et al., 2007).

The major challenge presented by hypoxia is the potential development of a cellular ATP supply–demand imbalance (Hochachka et al., 1996; Boutilier, 2001). Fish encountering hypoxia often attempt to maintain cellular ATP supply either by increasing O₂ transport to support aerobic respiration, or by increasing the use of anaerobic energy metabolism. Tissue O₂ supply can be improved by increasing branchial O₂ uptake (e.g. increasing ventilation, lamellar perfusion) or the rate of circulatory O₂ transport (e.g. increasing haemoglobin content or blood flow, changes in the concentration of allosteric modifiers) to counteract the effects of O₂ limitation on aerobic metabolism in hypoxia (Holeton and Randall, 1967; Hughes, 1973; Greaney and Powers, 1977, 1978; Nikinmaa and Soivio, 1982; Claireaux et al., 1988; Weber and Jensen, 1988; Perry et al., 2009). At any oxygen tension (P_O2) above the critical oxygen tension (P_crit), tissue O₂ supply is sufficient to meet metabolic demands, and support some aerobic scope for activity (Burton and Heath, 1980; Pörtner and Grieshaber, 1993; Lefrancois et al., 2005; Pörtner, 2010). Below P_crit, aerobic metabolism becomes dependent upon and decreases with environmental P_O2, and anaerobic metabolism is often used to help supplement ATP supply (Dunn and Hochachka, 1986; Pörtner and Grieshaber, 1993; Scott et al., 2008; Richards, 2009). Anaerobic metabolism can be favoured by increasing the activity and gene expression of glycolytic enzymes (e.g. lactate dehydrogenase), or by reducing carbohydrate flux into the tricarboxylic acid cycle and decreasing aerobic enzyme activity (e.g. cytochrome c oxidase, citrate synthase) (van den Thillart et al., 1980, 1994; Almeida-Val et al., 1995; Martínez et al., 2006; Richards et al., 2008). In addition to mechanisms that help maintain ATP supply, some tolerant organisms can also reduce ATP demands through active depression of resting metabolic rate (van Waversveld et al., 1989; van Ginneken et al., 1997). This is associated with reductions in energetically costly processes, such as ion transport, protein synthesis and mitochondrial proton leak (Bickler and Buck, 2007; Wood et al., 2007, 2009; De Boeck et al., 2013). The reliance on each of these strategies to balance ATP supply and demand varies between species, and could foreseeably be altered by the pattern or severity of hypoxia exposure.

The effects of intermittent hypoxia on fish physiology are poorly understood. This is starkly contrasted by the extensive literature on intermittent hypoxia in mammals, which has uncovered widespread physiological, developmental and genomic consequences that are distinct from continuous hypoxia exposure (Neubauer, 2001; Douglas et al., 2007; Farahani et al., 2008). There is evidence that exposure to repeated bouts of hypoxia compromises growth in some fish species (Atlantic salmon, Salmo salar, and southern catfish, Silurus meridionalis) but not others (spot, Leiostomus xanthurus, and killifish, Fundulus heteroclitus) (Stierhoff et al., 2003; McNatt and Rice, 2004; Burt et al., 2013; Yang et al., 2013). Exposure to daily oxygen cycles has been observed to increase hypoxia tolerance and aerobic swimming performance in hypoxia in southern catfish (Yang et al., 2013), to increase resting metabolism measured in normoxia in summer flounder (Paralichthys dentatus) (Taylor and Miller, 2001), and to reduce red blood cell GTP concentration and increase plasma bicarbonate concentration in carp (Cyprinus carpio) (Lykkeboe and Weber, 1978). Repeated 2 h bouts of hypoxia have also been shown to amplify some of the transcriptional responses to hypoxia compared with a single bout in the epaulette shark (Hemiscyllium ocellatum), including some genes in oxygen and energy homeostasis pathways (Rytkönen et al., 2012). However, the extent to which the physiological effects of intermittent hypoxia differ from those of constant hypoxia is unclear. Intermittent hypoxia differs from constant hypoxia as a stressor because it potentiates the production of reactive oxygen species (ROS), which may cause oxidative stress, but it also provides opportunities for recovery during the oxygenated periods between hypoxia bouts – distinctions that could favour divergent coping mechanisms between these two patterns of hypoxia exposure.

The objectives of this study were (i) to compare the effects on hypoxia tolerance of acclimation to constant hypoxia versus intermittent diel cycles of nocturnal hypoxia, and (ii) to investigate the physiological mechanisms underlying the acclimation response to each pattern of hypoxia exposure. We examined the killifish F. heteroclitus, an estuarine species that copes with both seasonal and daily fluctuations in dissolved oxygen content in its native habitat (Stierhoff et al., 2003; Burnett et al., 2007; Tyler et al., 2009). These fluctuations can be sudden, severe, and are mediated by factors such as the daily interplay between photosynthesis and cellular respiration, tidal movements, temperature, and wind patterns (Tyler et al., 2009). We integrated whole-animal respirometry with measurements of subordinate physiological traits dictating oxygen transport and utilization, including gill morphology, haematology, capillarity and fibre composition of skeletal muscle, and the activities of metabolic enzymes in the skeletal muscle, liver, heart and brain.

We exposed killifish to constant hypoxia or diel cycles of nocturnal hypoxia (12 h hypoxia at night, 12 h normoxia during the day) for 7 or 28 days at a moderate O₂ tension (P_O2=5 kPa) or for 7 days at a severe P_O2 (2 kPa). Hypoxia acclimation tended to reduce resting rates of oxygen consumption (Ṁ_O2). Because Ṁ_O2 did not scale isometrically (Ṁ_O2=12.061M ^0.491, where M is body mass), as previously observed in the closely related F. grandis (Everett and Crawford, 2010), we corrected Ṁ_O2 for body mass using the residuals from an allometric regression (Fig. 1A). Exposure to 7 days of either pattern of severe hypoxia substantially reduced Ṁ_O2 compared to normoxic controls (Fig. 1B). Though not significant, 7 days of moderate hypoxia also tended to reduce Ṁ_O2 (Fig. 1B). There was no effect of moderate hypoxia on Ṁ_O2 after 28 days of exposure (Fig. 1C), but the Ṁ_O2 across all 28 days groups appeared to be lower on average than the same 7 days treatment groups.

Hypoxia acclimation increased hypoxia tolerance, indicated by a lower P_crit and in some acclimation groups, a lower P_O2 at loss of equilibrium (LOE) during progressive hypoxia, relative to normoxia-acclimated controls (Fig. 2). Exposure severity, but not pattern, influenced the magnitude of the effect of 7 days of hypoxia acclimation on P_crit, which was lowest after acclimation to severe hypoxia but was also reduced by moderate hypoxia (Fig. 2C). In contrast, only constant hypoxia significantly reduced P_O2 at LOE, after acclimation to either 7 days of severe hypoxia or 28 days of moderate hypoxia (Fig. 2D,F). We corrected P_crit and P_O2 at LOE for body mass using the same residual approach that we used for Ṁ_O2 (Fig. 2A,B), but very similar results were obtained without using this mass correction (data not shown).

Hypoxia acclimation also affected blood lactate concentration (Fig. 3). Unlike other treatment groups, resting blood [lactate] was higher after acclimation to severe intermittent hypoxia, relative to normoxic controls. Blood [lactate] rose substantially at LOE across all treatments compared with levels in each group at rest in their acclimation condition (P<0.001 for both the 7 and 28 days data for the main effect of sampling point, i.e. rest versus LOE, in two-factor ANOVA) (Fig. 3). Furthermore, there was a significant interaction between sampling point and severity of hypoxia during acclimation in the 7 days group (P=0.042), and a similar trend was seen in the 28 days groups (P=0.090). This implies that acclimation to severe hypoxia for 7 days, whether constant or intermittent, blunts the rise in blood [lactate] at LOE. Indeed, the fold increase in blood [lactate] at LOE was lower on average in fish acclimated to severe hypoxia (approximately 6.6- and 4.6-fold above resting levels in the constant and intermittent groups, respectively) compared with that in normoxic controls (∼11.1-fold). Considering the concurrent differences in the P_O2 at LOE, it suggests that hypoxia acclimation changed the relationship between P_O2 and blood lactate accumulation.

Constant hypoxia had numerous effects on gill morphology. Because body mass has a significant effect on overall gill size, we again used the residuals from an allometric regression of gill morphometric measurements to compare these variables statistically (Fig. 4). Constant but not intermittent hypoxia acclimation led to a minor (<10%) but significant decrease in the total length of gill filaments relative to normoxia, after both 7 days of severe hypoxia or 28 days of moderate hypoxia (Fig. 4). Both 28 days hypoxia acclimations also decreased the average length of the gill filaments slightly (<8%) (Table 1). Clearly, hypoxia acclimation does not lead to morphometric changes that increase the surface area for branchial O₂ diffusion in this species.

The cell composition of the gill epithelium also changed following 28 days acclimation to moderate constant hypoxia (Fig. 5). In the 4 ppt brackish water in which we held the killifish, most of the mitochondrion-rich cells (MRCs) of normoxia-acclimated animals exhibited a typical seawater morphology with deep apical crypts. The proportion of freshwater-type MRCs increased after acclimation to constant hypoxia, as many more MRCs exhibited a transitional (wide and shallow apical crypts) or typical freshwater (convex surface) morphology (Fig. 5D). There were no differences in pavement cell (PVC) surface area or MRC density on the trailing edge of the filaments (Fig. 5C; Table 1).

There were substantial differences in the O₂ carrying capacity of the blood between fish acclimated to constant and intermittent hypoxia. Fish acclimated to either duration of constant hypoxia significantly increased haematocrit and whole-blood haemoglobin concentration, whereas fish acclimated to intermittent hypoxia did not differ from normoxic controls (Fig. 6). The increases in constant hypoxia acclimation were greater with severe hypoxia (∼1.9-fold) than with moderate hypoxia (∼1.4-fold). There were no changes in mean cellular haemoglobin concentration (P=0.876 for 7 days groups, and P=0.371 for 28 days groups; data not shown).

Hypoxia acclimation influenced the oxidative phenotype of the axial (swimming) muscle. Relative to normoxia, killifish acclimated to 28 days of constant hypoxia had significantly less total oxidative muscle area [as reflected by succinate dehydrogenase (SDH) positive staining], fewer oxidative fibres, and a 27% lower density of oxidative muscle as a proportion of the entire axial musculature (Fig. 7; Table 2). The muscle phenotype of fish acclimated to intermittent hypoxia did not differ significantly from normoxia-acclimated animals. Oxidative muscle area declined exclusively due to a decrease in the number of modestly oxidative fibres (MOx) at the interface between the oxidative and glycolytic regions (Table 2), with no change in the number of highly oxidative (HOx) fibres or in the average size of either fibre type (see Materials and methods for details). There were no significant differences in the myoglobin content of the entire axial musculature, but the non-significant pattern of variation after 28 days of acclimation mirrored the variation in muscle oxidative phenotype (Tables 3 and 4). There was also a significant 15% decrease in cytochrome c oxidase (COX) activity in the entire axial musculature (sampled to include all fibre types) following 28 days acclimation to both hypoxia patterns. However, there were no other differences in the activities of metabolic enzymes in the muscle (Tables 3 and 4).

Only intermittent hypoxia significantly altered muscle capillarity. Capillary density and capillary-to-fibre ratio in the glycolytic (SDH-negative) muscle increased 30% following acclimation to 28 days of intermittent hypoxia (Fig. 7; Table 2). In contrast, there were no changes in the capillarity of the oxidative muscle with intermittent hypoxia, or in capillarity for either muscle fibre type in fish acclimated to constant hypoxia.

Intermittent hypoxia, but not constant hypoxia, increased the biochemical capacities for oxidative energy metabolism and gluconeogenesis in the liver. Acclimation to severe intermittent hypoxia for 7 days increased the maximal activities of COX, citrate synthase (CS), lactate dehydrogenase (LDH) and phosphoenolpyruvate carboxykinase (PEPCK), without affecting pyruvate kinase (PK) or hydroxyacyl-coA dehydrogenase (HOAD) activities (Fig. 8). In contrast, acclimation to severe constant hypoxia for 7 days increased only LDH activity. There was no significant variation in liver mass or protein contents in the 7 days groups. Acclimation to moderate intermittent hypoxia for 28 days also increased liver COX activity and liver protein content, and acclimation to either of the hypoxia patterns for 28 days reduced liver mass (Table 4).

Capacities for oxidative phosphorylation and substrate oxidation appeared to remain unchanged by hypoxia acclimation in the heart and brain, as there were no differences in the activities of COX, CS, LDH, PK or HOAD (the latter two measured in heart only) (Tables 3 and 4). PEPCK activity was not detected in muscle, heart or brain. Interestingly, brain mass was larger in fish acclimated to 7 days of severe intermittent hypoxia, but heart mass did not vary between treatments (Tables 3 and 4).

Killifish are routinely exposed to fluctuating conditions in their native estuarine environment. Our results show that the responses of killifish to intermittent hypoxia during the course of acclimation are distinct from those to constant hypoxia. Although acclimation to both patterns of hypoxia exposure reduced resting Ṁ_O2 and P_crit, there were considerable differences between patterns in several physiological traits that dictate oxygen transport and utilization. Constant hypoxia reduced gill surface area, increased blood haemoglobin content, and led to greater reductions in muscle oxidative capacity. These changes did not occur in response to intermittent hypoxia, which was alone in amplifying the oxidative and gluconeogenic capacities of the liver and in increasing the capillarity of glycolytic fibres in the swimming muscle, both of which should improve recovery and lactate clearance between hypoxia bouts. Our results suggest that there are different mechanisms of acclimation that depend upon the pattern of exposure, and that the strategies for coping with constant and intermittent hypoxia may differ.

Acclimation to the same duration and magnitude of constant or intermittent hypoxia had similar effects on P_crit (Fig. 2C,E), suggesting that both patterns of exposure somehow improve the extraction and transport of O₂ in hypoxia, and thus broaden the functional P_O2 range for sustaining resting metabolism (Chapman et al., 2002; Søllid et al., 2003; Fu et al., 2011). This was at least partly explained by reductions in the resting Ṁ_O2 measured in normoxia (Fig. 1), which has been shown to be related to P_crit in many previous studies (van Ginneken et al., 1997; Mandic et al., 2009; Speers-Roesch et al., 2010). This supports the notion that low routine O₂ demands are associated with hypoxia tolerance, but there are clearly other changes in physiology with hypoxia acclimation that are also important (including those observed here).

Although constant and intermittent hypoxia led to comparable reductions in P_crit and resting Ṁ_O2, there is only modest evidence that this is caused by similar underlying physiological mechanisms. Increases in plasma lactate concentration with acute hypoxia are well documented in several fish species (Dunn and Hochachka, 1986; Cochran and Burnett, 1996; Virani and Rees, 2000; Scott et al., 2008). However, severe hypoxia acclimation, whether constant or intermittent, partially blunted this rise in blood lactate concentration at LOE (Fig. 3). This suggests that severe hypoxia acclimation reduces lactate accumulation at a given level of hypoxia, in association with lower rates of metabolism and ATP demand. Nevertheless, the capacity for lactate production by LDH is either similarly unaffected (heart, brain and muscle) or elevated (liver) (Fig. 8) by acclimation to both patterns of hypoxia. The general absence of any changes in metabolic capacity in the heart or brain with hypoxia acclimation, and the increases in LDH in the liver, are similar to previous observations in killifish (Greaney et al., 1980; Martínez et al., 2006). As the organs that are most sensitive to oxygen limitation, the heart and brain are probably protected from hypoxia by a preferential redistribution of blood flow (Axelsson and Fritsche, 1991; Gamperl et al., 1995; Soengas and Aldegunde, 2002). Otherwise, the physiological responses to constant and intermittent hypoxia were largely distinct.

Acclimation to constant hypoxia leads to the greatest reduction in the absolute lower P_O2 limit for acute survival, as reflected by a significant decrease in P_O2 at LOE, relative to normoxic controls (Fig. 2D,F). The potential causes of LOE in hypoxia are numerous, and could include metabolic acidosis or the cascade of cellular events that result from ATP supply–demand imbalance (Boutilier, 2001; Bickler and Buck, 2007). Variation in the time to or P_O2 at LOE has been suggested to arise from differences in total glycogen stores, the capacity for anaerobic metabolism, tolerance of metabolic acidosis, and the capacity for metabolic depression (Almeida-Val et al., 2000; Nilsson and Östlund-Nilsson, 2008; Mandic et al., 2013). As discussed above, anaerobic capacity probably did not distinguish constant hypoxia from intermittent hypoxia because LDH activities were generally similar between groups (Fig. 8; Tables 3 and 4). However, it is possible that fish acclimated to constant hypoxia have a greater capacity for depressing metabolism or tolerating metabolic acidosis in acute hypoxia than those acclimated to normoxia or intermittent hypoxia.

The reductions in gill filament length and the changes in cell composition on the gill epithelium in response to constant hypoxia may be mechanisms for depressing the metabolic costs of ion transport. As the first step of the O₂ transport cascade, the gills are crucial for O₂ uptake, and gill surface area often increases in water-breathing fishes with hypoxia acclimation (Hughes, 1966; Chapman et al., 1999; Søllid et al., 2003; Evans et al., 2005). However, the large gas exchange area of the gills facilitates passive ion loss in a hypo-osmotic environment, which necessitates active ion pumping to maintain ionic homeostasis. This underlies the ‘osmorespiratory compromise’ that leads to trade-offs between respiratory gas exchange and osmoregulation (Randall et al., 1972; Nilsson, 2007). Correspondingly, some species reduce gill surface area and/or ion permeability in response to hypoxia exposure, presumably to minimize ionic disruption rather than facilitate O₂ uptake (McDonald and McMahon, 1977; Wood et al., 2007, 2009; Matey et al., 2011; De Boeck et al., 2013). The killifish acclimated to constant hypoxia in this study decreased gill surface area and shifted their gill epithelium towards a freshwater morphology (Figs 4 and 5; Table 1). Because the freshwater gill is less permeable to ions than the seawater gill due to the presence of deep tight junctions between cells (Chasiotis et al., 2012), this transition should have reduced passive ion loss to the surrounding hypo-osmotic brackish water (Sardet et al., 1979; Scott et al., 2004; Chasiotis et al., 2012). The structural changes in the gills with constant hypoxia acclimation may then act to minimize the costs of ion transport and facilitate metabolic depression. This probably reduced O₂ diffusion capacity across the gill epithelium, and may have limited the ability of killifish acclimated to constant hypoxia to support high metabolic rates (e.g. exercise). However, as the fish in this study were inactive and their oxygen demands were quite low (Fig. 1), there may have been excess gill surface area and uptake capacity that could be done without in hypoxic fish.

Fish acclimated to constant hypoxia increased whole-blood haemoglobin content and haematocrit (Fig. 6). A similar response has been observed in many species following hypoxia acclimation, and is often accompanied by increases in haemoglobin–O₂ affinity that are mediated by changes in allosteric effectors (Wood and Johansen, 1972; Greaney and Powers, 1977, 1978; Claireaux et al., 1988; Weber and Jensen, 1988; Chapman et al., 2002; Silkin and Silkina, 2005). Circulatory O₂ carrying capacity would have been enhanced in killifish acclimated to constant hypoxia if the measured increase in haemoglobin content was reflective of the entire blood volume. It is also possible that changes in peripheral vasoconstriction and blood flow, which are known to occur in some species in hypoxia (Axelsson and Fritsche, 1991), reduced the entrance of erythrocytes into the capillaries, thus reducing capillary haematocrit and concentrating the erythrocytes in the major vessels (from which haemoglobin content and haematocrit were measured). Interestingly, we and others (Lykkeboe and Weber, 1978; Taylor and Miller, 2001) have shown that the same haemoglobin and haematocrit responses do not occur with intermittent hypoxia acclimation. This suggests that cumulative exposure duration, and not only absolute P_O2 during each hypoxia bout, may control haemoglobin content by promoting erythropoiesis and erythropoietin release by the kidneys, or by changing blood volume and/or the proportion of plasma and erythrocytes in the capillaries and secondary circulation (Lai et al., 2006; Rummer et al., 2014).

Constant hypoxia reduced the abundance of oxidative fibres in the muscle, due to a reduction in the number of modestly oxidative fibres at the interface between the oxidative and glycolytic regions, where fast oxidative muscle fibres are normally situated (Fig. 7; Table 2) (Scott and Johnston, 2012; McClelland and Scott, 2014). Because muscle recruitment proceeds from slow oxidative, to fast oxidative, to fast glycolytic as swimming intensity increases (Rome et al., 1984), a general reduction in swimming activity with hypoxia could have reduced the neural stimulation of fast oxidative fibres. Reduced neural activation is a key stimulus initiating the transition of fast fibres from an oxidative to a glycolytic phenotype (Bassel-Duby and Olson, 2006), so it is foreseeable that this process was induced in the fast oxidative muscle region due to a reduction in activity levels during hypoxia acclimation. Hypoxia could also regulate mitochondrial abundance within individual fibres, as hypoxia inducible factor stimulates mitochondrial autophagy in mammals (Zhang et al., 2008), and thus reduce overall muscle oxidative capacity even further. However, there is some interspecific variability in the effects of hypoxia on muscle phenotype, as tench (Tinca tinca), but not crucian carp (Carassius carassius), have been observed to reduce muscle oxidative capacity by decreasing mitochondrial content with hypoxia acclimation (Johnston and Bernard, 1982, 1984).

Acclimation to intermittent hypoxia appears to improve the capacity to recover from each hypoxia bout during the intervening periods of normoxia. The use of anaerobic metabolism during hypoxia, reflected by increases in lactate production (Fig. 3) and metabolic acidosis (Johnston, 1975; Dunn and Hochachka, 1986; Scott et al., 2008), incurs an oxygen debt that must later be repaid (Heath and Pritchard, 1965). Increases in the activity of several enzymes occurred in the liver in response to intermittent hypoxia acclimation, including COX, CS, LDH and PEPCK (Fig. 8). These changes should have augmented the capacity for gluconeogenesis and lactate oxidation in the liver, possibly to increase this organ's capacity for metabolizing the lactate produced during each hypoxia bout. Constant hypoxia did not affect PEPCK activity (Fig. 8), consistent with previous observations (Martínez et al., 2006).

The capillarity of the glycolytic (but not oxidative) muscle also increased in fish acclimated to intermittent hypoxia (Fig. 7), which should increase the capacity for lactate clearance from the muscle (the largest tissue in the body) during the daily normoxic periods between hypoxia bouts. Capillarity did not increase in response to constant hypoxia in this study (Fig. 7) or in previous studies of other fish species (Johnston and Bernard, 1982; Jaspers et al., 2014). As fibre size and number were unaffected by intermittent hypoxia, the increased capillarity appears to be caused by angiogenesis, and not a reduction in fibre size due to muscle atrophy. Angiogenesis could have been stimulated by an increase in muscle lactate (Constant et al., 2000; Gladden, 2004) or by high blood flows during recovery from hypoxia that may be needed to clear a lactate load (Egginton, 2011). Angiogenesis could have also occurred as a response to a decline in intracellular P_O2 (Mathieu-Costello, 1993; Hoppeler and Vogt, 2001). However, we did not observe any significant variation in the myoglobin content of the muscle, which would have increased cellular O₂ supply and has been observed to occur in response to hypoxia (Fraser et al., 2006). Regardless of the cause of this increase in capillarity, it is possible that it represents part of a general strategy to enhance the overall capacity for lactate turnover. This could even involve an increased use of the Cori cycle, in which the liver resynthesizes glucose from lactate through gluconeogenesis and then returns glucose to the muscle via the circulation, although the existence of the Cori cycle in fish is uncertain (Milligan and Girard, 1993).

Acclimation to intermittent hypoxia increases hypoxia tolerance in killifish, and the mechanisms involved appear to be distinct from those for constant hypoxia. Intermittent hypoxia is also known to have different effects than constant hypoxia on the control of ventilation and circulation in mammals (MacFarlane et al., 2008; Prabhakar and Semenza, 2012). On the one hand, it is possible that comparable changes occur in fish and mammals in response to intermittent hypoxia. On the other hand, fish that are routinely exposed to intermittent hypoxia in their native environment, such as estuarine killifish, might employ uniquely evolved strategies for coping with intermittent hypoxia. As the occurrence of hypoxia increases worldwide, it will be important to better appreciate how the pattern of exposure influences the impacts of hypoxia on aquatic organisms.

Adult, wild-caught Fundulus heteroclitus Linnaeus 1766 of both sexes were purchased from a commercial supplier (Aquatic Research Organisms, NH, USA), shipped to McMaster University, and held for at least one month in brackish (4 ppt) water at room temperature (∼21°C) before experimentation. Water quality (pH, nitrates, nitrites and ammonia) was maintained with regular water changes. Fish were fed commercial flakes (Big Al's Aquarium Supercentres, Mississauga, ON, Canada) 6 days per week, and were not fed for 24 h prior to respirometry or sampling.

Exposures were carried out in 35 l glass aquaria with the same water chemistry as described above, in which normoxia, constant hypoxia, or nocturnal (intermittent) hypoxia was sustained for 7 or 28 days (Table 5). Constant normoxia (20 kPa, 8 mg O₂ l⁻¹) was maintained by continuously bubbling the water with air. Constant hypoxia was maintained by bubbling the water with nitrogen gas, and the appropriate P_O2 was maintained by a feedback loop using a galvanic oxygen sensor that automatically controlled the flow of nitrogen with a solenoid valve (Loligo Systems, Tjele, Denmark). Nocturnal (‘intermittent’) hypoxia was maintained using the same O₂ controller as for constant hypoxia, but gas flow was alternated between air (08:00 h to 20:00 h local time) and nitrogen (20:00 h to 08:00 h) with an additional solenoid valve controlled by a photoperiod timer that was synchronized with the light cycle (12 h:12 h light:dark). During the hypoxic periods, the set-point (i.e. 5 or 2 kPa) was tightly regulated with a hysteresis of 0.02 kPa. Larger deviations from set-point were infrequent and never exceeded 0.4 kPa. Two levels of hypoxia were used for 7 days exposures, either moderate (5 kPa, 2 mg O₂ l⁻¹) or severe (2 kPa, 0.8 mg O₂ l⁻¹), but only moderate hypoxia was used for 28 days exposures. Fish were prevented from respiring at the water surface with a plastic grid barrier. The body masses and standard body lengths of fish used for each series of experiments are shown in Table 5.

Stop-flow respirometry was used to determine Ṁ_O2, P_crit and the P_O2 at LOE. Fish were held overnight in normoxia in a respirometry chamber (90 ml cylindrical glass) that was situated in a darkened buffer tank and continuously flushed with normoxic water (flushing circuit). The chamber was also connected to a recirculating circuit that flowed past a fibre-optic oxygen sensor (PreSens, Regensburg, Germany). Both circuits were driven by pumps controlled by AutoResp software (Loligo Systems).

Oxygen consumption measurements began the following morning in normoxia, with two sequential flush and measurement periods. During flush periods, both the flush and recirculating pumps were active, and the chamber received a steady flow of water from the buffer tank (i.e. the chamber and buffer tank were equilibrated). During measurement periods, the flush pump was turned off, isolating the chamber from the buffer tank, and the change in oxygen concentration due to fish respiration was measured. Fish were then subjected to a progressive hypoxia protocol, in which buffer tank P_O2 was reduced in 2 kPa steps using the O₂ control system described above. Each level of hypoxia was maintained for 10 min, and oxygen consumption rate was measured as in normoxia. After measurement at an ambient P_O2 of 2 kPa, the chamber was closed such that the fish consumed the remaining oxygen until it lost equilibrium. Oxygen consumption was calculated from the change in chamber oxygen concentration over time as previously recommended (Clark et al., 2013). P_crit was calculated using a program developed by Yeager and Ultsch (1989).

Immediately after losing equilibrium, fish were euthanized by a blow to the head followed by pithing. The tail was bisected at the base of the anal fin, and blood was collected from the caudal blood vessels in a heparinized capillary tube. Whole blood was used to measure haemoglobin content (using Drabkin's reagent following manufacturer’s instructions; Sigma-Aldrich, Oakville, ON, Canada) or was frozen in liquid nitrogen and stored at −80°C for later determination of lactate concentration. Some blood was also centrifuged for 5 min at 12,700 g to measure haematocrit.

A separate set of fish from those used for respirometry were acclimated and sampled at rest (rather than at LOE). Sampling was done at a consistent time of day (between 13:00 h and 16:00 h) to minimize the effect of circadian rhythms and other diurnal variations on our results. Sampling was therefore during the normoxic period for the intermittent hypoxia acclimation groups. Fish were euthanized and blood was collected and analysed as described above. A transverse steak of the trunk was cut at the anterior base of the anal fin, coated in embedding medium (Fisher Scientific Company, Ottawa, ON, Canada), frozen in liquid N₂-cooled isopentane, and stored at −80°C until use for muscle histology. An adjacent hemi-section of the axial muscle (containing the entirety of the red and white fibres) and the entire intact liver, heart and brain were removed, frozen immediately in liquid nitrogen, and stored at −80°C for later measurement of enzyme activities. The gill baskets were removed intact and fixed (2% paraformaldehyde, 2% glutaraldehyde) at 4°C.

After fixing, the four arches on one side of the gill basket were isolated and cleaned of excess tissue such that individual filaments and lamellae were visible. Images at ×10 magnification were taken on each side of the arches using a stereomicroscope to determine the length and number of filaments. Total gill filament length for an entire fish was calculated by doubling the sum of all individual filament lengths measured, and then multiplying the result by 1.15 to account for the approximate effects of curling, as recommended previously (Hughes, 1966).

Six fish from each 28 days treatment group that had a similar body mass (3–5 g) were selected for scanning electron microscopy. Filaments from the middle portion of one side of the second gill arch were removed to enable viewing of both the leading and trailing edges of the remaining filaments. The gills were post-fixed in 1% OsO₄ for 1 h and then dehydrated in progressively higher concentrations of ethanol (from 50 to 100%). After critical point drying with liquid CO₂, samples were sputter-coated and viewed in an ESEM 2020 scanning electron microscope (Electroscan Corporation, Wilmington, MA, USA). Images were taken at ×350 and ×500 magnification to determine gill filament depth (parallel to water flow), maximum lamellar height (perpendicular to the filament and to water flow), and lamellar depth (parallel to filament depth). High-magnification images were taken at ×2000 and ×5000 to quantify MRC density, MRC and PVC size, and MRC phenotype (seawater-type apical crypts or freshwater and transitional-type MRCs for which the cell surface is visible) (Scott et al., 2004; Laurent et al., 2006). PVC and MRC measurements were restricted to the trailing edge. All gill morphometric images were analysed using ImageJ software (Rasband, 2014).

Muscle blocks were cut into 10 μm sections at −20°C with a cryostat (Leica Microsystems, Wetzler, Germany), mounted on Superfrost Plus slides (Fisher Scientific Company), air dried, and stored at −80°C until staining. Staining of sections was performed using standard methods that have been previously described (Egginton, 1990; Scott and Johnston, 2012). Sections were stained for SDH activity to identify oxidative muscle fibres, using the following assay conditions: 41.7 mmol l⁻¹ Na₂HPO₄, 8.3 mmol l⁻¹ NaH₂PO₄, 80 mmol l⁻¹ sodium succinate, 0.1% NBT, pH 7.6. Alkaline phosphatase activity was used to stain for capillaries, using the following assay conditions (in mmol l⁻¹): 28 NaBO₂, 7 MgSO₄, 1 NBT, 0.5 BCIP, pH 9.3. Sections were stained for 1 h at room temperature in each protocol, after which slides were mounted with Aquamount (Fisher Scientific Company) and stored at 2°C until they were imaged with a Nikon Eclipse E800 light microscope (Nikon Instruments, Melville, NY, USA).

The total transverse area and number of SDH-positive (oxidative) muscle fibres were determined for each fish from the SDH activity stains. Average oxidative fibre size was calculated by dividing the area by the number of fibres within that area. We observed two distinct patterns of SDH-positive staining intensity – one indicative of high oxidative capacity (HOx; dark SDH staining throughout the fibre) and one indicative of modest oxidative capacity (MOx; less intense but still positive SDH staining, usually near the fibre periphery) (Fig. 7) – so we also quantified the total number of fibres exhibiting each staining pattern. We quantified from only one lateral side of the fish, but we multiplied the results by two to calculate the total oxidative fibre area and number of fibres in the entire trunk. Based on their location in the axial muscle, the fibres we characterized as having a high oxidative capacity included both slow oxidative and a subset of fast oxidative fibre types (Scott and Johnston, 2012; McClelland and Scott, 2014).

Average glycolytic fibre size, capillary density, and capillary-to-fibre ratios were determined from the alkaline phosphatase stains. Average glycolytic (SDH-negative) fibre size was determined from six to eight images taken throughout the white muscle of each fish, by dividing the known area of the image by the unbiased number of fibres within the image (Egginton, 1990). Capillary density (number of capillaries counted per unit of area) and capillary-to-fibre ratio (the number of capillaries relative to the number of fibres counted in sections) were quantified in the entirety of the highly oxidative region on one lateral side of the fish. The same capillarity indices were determined from six to eight images taken throughout the glycolytic region. Preliminary assessments verified that a sufficient number of images were analysed to account for heterogeneity across the axial musculature (determined by the number of replicates necessary to yield a stable mean value). For all histological measurements, the average value for each fish was calculated and used for statistical comparison between treatment groups. All muscle histology images were analysed using ImageJ by an observer that was blind to treatment group.

The maximal activities of several enzymes were assayed using standard methods (Bears et al., 2006; Schnurr et al., 2014). The frozen muscle, liver, heart and brain samples were weighed and homogenized on ice in 20 volumes of homogenization buffer (20 mmol l⁻¹ Hepes, 1 mmol l⁻¹ sodium EDTA and 0.1% Triton X-100) at pH 7.0. Assays were performed to determine the maximal activity of each enzyme in the tissue homogenate at 25°C, by measuring the rate of change in absorbance at 550 nm (COX), 340 nm (HOAD, LDH, PEPCK, PK), or 412 nm (CS) for at least 5 min. The COX assay was performed immediately following homogenization, after which the homogenate was stored at −80°C. Other enzymes were assayed after a consistent number of freeze–thaw cycles for each enzyme. Assay conditions were as follows: COX, 100 μmol l⁻¹ (brain, liver and heart) or 50 μmol l⁻¹ (muscle) of fully reduced cytochrome c in 50 mmol l⁻¹ Tris containing 0.5% Tween-20 at pH 8.0; CS, 1.0 mmol l⁻¹ oxaloacetate and 0.15 mmol l⁻¹ acetyl-CoA (brain), or 0.5 mmol l⁻¹ oxaloacetate and 0.3 mmol l⁻¹ acetyl-CoA (other tissues), each in 50 mmol l⁻¹ Tris containing 0.1 mmol l⁻¹ DTNB at pH 8.0; HOAD, 0.1 mmol l⁻¹ acetoacetyl-CoA and 0.3 mmol l⁻¹ NADH in 50 mmol l⁻¹ imidazole at pH 7.2; LDH, 0.3 mmol l⁻¹ NADH and 0.5 mmol l⁻¹ pyruvate (heart, liver and muscle) or 0.15 mmol l⁻¹ NADH and 1 mmol l⁻¹ pyruvate (brain), each in 50 mmol l⁻¹ Hepes at pH 7.4; PEPCK, 1.1 mmol l⁻¹ PEP, 0.15 mmol l⁻¹ NADH, 0.5 mmol l⁻¹ dGDP, 20 mmol l⁻¹ NaHCO₃, 1 mmol l⁻¹ MnCl₂.4H₂O, and excess coupling enzyme (malate dehydrogenase) in 50 mmol l⁻¹ imidazole at pH 7.4; PK, 10 mmol l⁻¹ PEP, 0.15 mmol l⁻¹ NADH, 5 mmol l⁻¹ ADP, 100 mmol l⁻¹ KCl, 10 mmol l⁻¹ MgCl₂.6H₂O, 10 μmol l⁻¹ fructose-1,6-bisphosphate, and excess coupling enzyme (LDH) in 50 mmol l⁻¹ Mops at pH 7.4. Preliminary assays determined the lowest possible substrate concentrations that would stimulate maximal activity. All enzyme assays were run in triplicate in a 96-well microplate spectrophotometer (Molecular Devices, Sunnyvale, CA, USA) with temperature control. Activities were determined by subtracting the background reaction rate without a key substrate (COX, cytochrome c; CS, oxaloacetate; HOAD, acetoacetyl-CoA; LDH, pyruvate; PK, phosphoenolpyruvate; PEPCK, dGDP) from the rates measured in the presence of all substrates. We used extinction coefficients (ε) of 28.5 and 13.6 optical density (mmol l⁻¹) cm⁻¹ for COX and CS assays, respectively. We calculated ε empirically for the remaining assays by constructing standard curves of absorbance versus NADH concentration in the buffers appropriate for each assay.

Undiluted skeletal muscle homogenate was assayed for myoglobin concentration using a modification of the method described by Reynafarje (Reynafarje, 1963). Homogenates were centrifuged at 13,700 g for 100 min at 4°C (Eppendorf, Hamburg, Germany). The supernatant was completely reduced by rotating for 8 min in a tonometer containing pure carbon monoxide gas, followed by addition of sodium dithionite and a further 2 min of rotation in CO. Reduced samples were diluted and absorbance was read at 538 and 568 nm, and tissue myoglobin content was calculated as described (Reynafarje, 1963).

Whole-blood lactate concentrations were measured by thawing the frozen samples and acidifying them using an excess of 8% HClO₄ solution. Acidified extracts were incubated for 40 min at 37°C in assay buffer (0.6 mol l⁻¹ glycine, 0.5 mol l⁻¹ hydrazine sulphate with excess β-NAD⁺ and LDH), and then absorbance was read at 340 nm in duplicate.

A residual approach accounted for the influence of body mass on O₂ consumption rate (Fig. 1), critical oxygen tension and oxygen tension at loss of equilibrium (Fig. 2) and total gill filament length (Fig. 4). Data were first regressed to body mass (M) using the general allometric equation Y=aM ^b (where a and b are constants), and residuals from the regression were then calculated for each individual. The calculated residuals were used for statistical comparisons (see below). Data are reported graphically as both residuals and as the sum of the residual and the expected value for an average-sized killifish (see Figs 1, 2 and 4 for details).

Data are reported as means±standard error (except where data for individual fish are shown). For most data, hypoxic acclimation treatments were compared with normoxic controls with a one-way ANOVA. A two-way ANOVA was used to assess the effects of hypoxic acclimation treatment, sampling point (i.e. rest versus LOE), and their interaction on blood [lactate]. Bonferroni multiple comparisons post-tests were used for paired comparisons. A significance level of P<0.05 was used throughout.

The authors would like to thank Klaus Schultes and Marcia Reid for technical support, as well as Meghan Schnurr, Paras Patel, Sherry Du, Kyle Crans and Sajeni Mahalingam for help and advice. We would also like to thank two anonymous reviewers for their excellent comments on a previous version of this manuscript.