Energetic constraints on electric signalling in wave-type weakly electric fishes

How much energy an organism allocates to its various functions is often hard to determine, but trade-offs in energy allocation may become apparent under extreme conditions (reviewed in Stearns, 1992; Roff, 1992). Given the challenges associated with breathing underwater (e.g. Lindsey, 1978; Wootton, 1990) and the growing body of literature describing negative impacts of hypoxia (low oxygen) on fishes (e.g. Pollock et al., 2007; Wu, 2009; Chapman and McKenzie, 2009), aquatic hypoxia is a good example of one such energy-limiting condition. Therefore, testing the effect of hypoxia on communication behaviours is a promising way of assessing the energetic costs of generating communication signals. Several studies suggest that hypoxia may reduce the occurrence of communication behaviours. For example, Siamese fighting fish reduce the frequency of aggressive opercular displays when exposed to hypoxia (Abrahams et al., 2005); the African cichlid Pseudocrenilabrus multicolour victoriae shows fewer aggressive and mating displays after acclimation to hypoxic conditions (Gotanda et al., 2011), similarly, sailfin mollies reduce the frequency of reproductive displays (Timmerman and Chapman, 2004); schooling behaviour may also be disrupted under low oxygen (Domenici et al., 2007); and dominance hierarchies in populations of three-spined sticklebacks appear to be less stable under hypoxia than normoxia (Sneddon and Yerbury, 2004). Electric fishes are an excellent group for an in-depth exploration of the impacts of hypoxia on communication systems because their communication signals, i.e. electric signals, are easy to measure relative to the signals used by other fishes.

Gymnotiformes are a group of weakly electric fishes that are widespread throughout neotropical freshwater systems, such as the Amazon Basin, where hypoxia can be a common occurrence (Carter and Beadle, 1931; Kramer et al., 1978; Val and de Almeida-Val, 1995; Crampton, 1998). These fishes generate electric organ discharges (EODs) and sense perturbations of the resulting electric field for purposes of orientation, prey detection and communication (Bullock and Heiligenberg, 1986; Heiligenberg, 1991). One group of gymnotiforms are called pulse-type fishes, because their EODs are brief pulses separated by longer, and often variable, pauses. The remaining species produce oscillatory (‘wave-type’) EODs at frequencies of up to 2 kHz (Crampton and Albert, 2006). These high EOD frequencies have been proposed to be energetically expensive (Crampton, 1998; Salazar and Stoddard, 2008). If high-frequency EODs are expensive, then fish may modulate their EOD signal (frequency and amplitude) in response to hypoxic stress and/or they may compensate for costs of signalling through other adaptations that maximize oxygen uptake efficiency.

Only a handful of studies have estimated energetic costs associated with EOD production, with contrasting results (e.g. Crampton, 1998; Hopkins, 1999; Julian et al., 2003). For example, Crampton (Crampton, 1998) studied hypoxia tolerance of weakly electric gymnotiforms and characterized the oxygen profile of their habitats. He found that 16 out of the 17 species occurring in severely hypoxic or anoxic habitats were pulse-type gymnotiforms. Thus, he suggested that wave-type EOD frequency may impose greater energy demands (Crampton, 1998). Using pharmaceutical methods and respirometry, Salazar and Stoddard (Salazar and Stoddard, 2008) estimated the cost of electrical signalling relative to the total energy budget of the pulse-type gymnotiform Brachyhypopomus gauderio (at the time assumed to be B. pinnicaudatus) (Giora and Malabarba, 2009). Their results suggest a high cost of signalling in this species, with males devoting a greater percentage of the total energy budget (11–22%) to EOD signal generation than females (3%). Additional indirect evidence for a high cost of maintaining EOD is that female Apteronotus leptorhynchus prefer males with a higher EOD frequency (O. Bargelletti, J. F. Gogarten and R.K., unpublished) and that EOD frequency is positively related to the body size of males (Dunlap, 2002; Triefenbach and Zakon, 2003). In a related species, it was recently shown that the fish evaluate EOD frequency to assess the dominance status of conspecifics and that higher-frequency fish are dominant over fish with lower EOD frequency (Fugère et al., 2011). In contrast, Julian et al. (Julian et al., 2003) found no relationship between EOD type (pulse-type or wave-type) and metabolic rate in a survey of 23 gymnotiform species (34 individual fish in total), suggesting that energy costs associated with generating and processing the EOD are similar between pulse-type and wave-type fish, even though pulse-type fish have low EOD frequencies and many wave-type fish produce high EOD frequencies. Furthermore, it appeared unlikely that EOD generation is expensive because the metabolic rates found for gymnotiform fish were much lower than those for non-electric fish from temperate zones extrapolated to tropical temperatures. Although these studies contribute to our understanding of the cost of electrical signalling, none have directly explored the question using an energetic stressor to identify energetic costs nor have these studies examined potential trade-offs between EOD and maintenance metabolism in wave-type electric fishes.

The goal of this study was to explore the energetic costs of electrical signalling by identifying potential trade-offs between EOD (frequency and amplitude) and oxygen uptake during hypoxia stress in two wave-type gymnotiform fishes thought to differ in low oxygen tolerance. Apteronotus leptorhynchus (Ellis 1912) is typically found in well-oxygenated whitewater systems in South America and is thought to be relatively intolerant of hypoxia (Crampton, 1998; Crampton and Albert, 2006). This species has an electric organ that is derived from nerve cells (Kirschbaum and Schwassmann, 2008) and that produces baseline EOD frequencies of 800 to 1000 Hz in males and 600 to 800 Hz in females. It is thought that individual A. leptorhynchus do not modulate EOD frequency or amplitude in response to environmental changes (except temperature) (see Dunlap et al., 2000), even though they use frequency modulations in a communication context (Hagedorn and Heiligenberg, 1985; Zakon et al., 2002). In contrast, Eigenmannia virescens (Valenciennes 1836) is found in both high- and low-oxygen systems such as várzea (floodplains), and whitewater and blackwater systems of South America, and appears to be more hypoxia tolerant than A. leptorhynchus, both in its natural habitat and under experimental conditions (Crampton, 1998; Julian et al., 2003; Kirschbaum and Schwassmann, 2008). The electric organ of E. virescens is derived from muscle tissue and produces a lower EOD frequency compared with A. leptorhynchus, ranging from approximately 250 to 600 Hz (Hopkins, 1974). Males tend to have lower EOD frequencies than females, although there is a large degree of overlap (Hopkins, 1974; Hagedorn and Heiligenberg, 1985). Eigenmannia virescens are also able to modulate their EOD frequencies transiently during social interactions, but will eventually return to their intrinsic baseline frequency (Heiligenberg, 1991).

To explore costs of signalling, we measured EOD frequency and amplitude while quantifying routine metabolic rate, critical oxygen tension (P_crit) and aquatic surface respiration (ASR) thresholds for the two species. P_crit and ASR thresholds are commonly used indices of hypoxia tolerance. Fishes will regulate their metabolic rate over a range of oxygen partial pressures (P_O2). The P_crit is the point at which a further reduction of P_O2 causes a shift in metabolic rate from oxygen independent to oxygen dependent (Beamish, 1964; Ultsch et al., 1978). When fish have access to the surface, many species use ASR under hypoxia, whereby they skim the surface film of water where diffusion maintains a higher level of oxygen than in the hypoxic water column. The ASR₉₀ threshold represents the oxygen level at which a fish spends 90% of its time skimming the surface film (Kramer, 1983; Kramer et al., 1983). A lower ASR₉₀ threshold and lower P_crit suggest higher tolerance to hypoxic stress (Wen-Chi Corrie et al., 2008; Reardon and Chapman, 2010). A positive relationship between EOD (frequency and amplitude) and metabolic rate would suggest that maintenance of a high EOD frequency/amplitude is energetically expensive, as would a decrease in EOD frequency and/or amplitude after reaching the P_crit and ASR₉₀ threshold. We therefore predicted a lower ASR₉₀ threshold and lower P_crit in E. virescens than in A. leptorhynchus. We also predicted a stronger effect of hypoxic stress on EOD frequency and amplitude in A. leptorhynchus.

Wild-caught fish of the two species were obtained from two commercial suppliers, Belowwater and MIRDO Importations (Montreal, QC, Canada). Fish were individually housed in 10 litre aquaria at 27°C and a conductivity of 150 μS, and exposed to a 12 h:12 h light:dark photoperiod. Fish were fed live black worms (Lumbriculus variegatus) or frozen bloodworms (chironomid larvae) twice weekly. Experimental protocols were approved by the McGill University Animal Care Committee (protocol no. 5408).

To quantify routine metabolic rate and P_crit, we used one intermittent closed respirometer, whereby oxygen consumption was estimated by oxygen decline in a closed container in response to fish respiration (Julian et al., 2003). The total water volume of the respirometer was 0.215 l. Low constant water current was maintained at 0.833 ml s^–1 throughout each trial to ensure proper water mixing. To calculate oxygen consumption by the fish, oxygen (% saturation) and temperature data (°C) were recorded every 30 s over the course of the trial using Ocean Optics FOXY probes and thermistors, respectively (Dunedin, FL, USA). Decreases in oxygen partial pressure (converted to kPa from % saturation) were used to calculate metabolic rate and P_crit. Chlorided silver-wire electrodes mounted on either end of the fish respirometry chamber were used to record EOD (frequency and amplitude) throughout the trial, concurrently with the measurements of water temperature and P_O2. The signal was amplified (AC/DC Differential Amplifier, Model 3000, A-M Systems, Carlsborg, WA, USA) and digitized at a sampling rate of 20 kHz (PCI-6052E, National Instruments, Austin, TX, USA) using MATLAB R2007b (The MathWorks, Natick, MA, USA). Experimental conditions were maintained similar to those of the aquarium facility holding environment, with water temperature and conductivity during the respirometry trials averaging 26.9°C and 150 μS, respectively.

Routine metabolic rate (RMR) and P_crit were quantified on a total of 22 fish (A. leptorhynchus N=11; E. virescens N=11). Experiments were conducted during the day, when these nocturnal fish are least active. To ensure a post-absorptive state, fish were not fed for at least 24 h prior to use. Prior to each trial, fish were acclimated for 2 h in the experimental setup, with the respirometer in flow-through mode, circulating aerated water into the fish chamber from the reservoir. After the acclimation period, the respirometer was closed off from the outside reservoir. Following Julian et al. (Julian et al., 2003), oxygen levels were allowed to drop to 80%, then the system was returned to open mode and the fish chamber was flushed with aerated water. Once the P_O2 levels in the water returned to saturation, the system was closed again. A total of five trials were performed consecutively for each fish to produce an average RMR. Within each individual, the five successive measurements of oxygen consumption differed from each other by less than 8%, suggesting that a stable rate of oxygen consumption had been achieved. After the fifth trial, the respirometer was switched into closed mode and the P_O2 levels were allowed to drop until P_crit was reached. For each fish, activity during each run was noted, dividing activity levels into two categories: resting with small fin undulations or swimming. Any fish characterized as swimming was omitted from the analysis to ensure measures were focused on RMR. At the end of each trial, mass, total length and gender were recorded, and then the fish was returned to its home tank. Trials were conducted using Millipore-filtered tank water to minimize biological activity. However, levels of background respiration were estimated from control runs in the empty respirometer.

ASR trials were conducted in a 20 litre tank divided into three compartments as described in Wen-Chi Corrie et al. (Wen-Chi Corrie et al., 2008). Each of the side compartments housed an air stone, a nitrogen diffuser, a heater and a Fluval 1+ Underwater pump (Hagen, Montreal, QC, Canada). Throughout each trial, temperature was maintained at 27.0±0.1°C (mean ± s.e.m.) using a LoliTemp controlling unit (Loligo Systems, Tjele, Denmark). The fish was placed in the central compartment (34 cm long) with an opaque tube to hide in for a 30 min acclimation period, during which P_O2 was held at a mean of 18.45±0.1 kPa. Oxygen levels were dropped incrementally by gently bubbling nitrogen gas into the tank controlled by a LoliOxy oxygen controlling unit (Loligo Systems) to the following levels: 18.45, 14.5, 10.4, 6.2, 4.2, 3.7, 3.3, 2.9, 2.5, 2.1, 1.7, 1.2 and 0.8 kPa (0.1 kPa accuracy).

After each drop in oxygen, there was a 10 min acclimation period followed by 3 min of observation, during which activity was noted every 5 s. The levels of activity were recorded as: (1) motionless aside from gill ventilations, (2) undulating fins but not moving in tube, (3) swimming in tube, (4) swimming outside of tube or (5) performing ASR. Oxygen levels were dropped progressively until the fish spent 90% of the observation time performing ASR or until the fish lay on the bottom of the aquarium. Using a digital video recorder (TroubleShooter™ High-Speed Camera, Fastec Imaging, San Diego, CA, USA), gill ventilation rates, measured as the number of ventilations per 10 s, were determined at the beginning and at the very end of the trial (Chapman et al., 1995). Carbon rod electrodes (38.1 cm length) mounted opposite each other diagonally at either end of the fish chamber were used to record EOD (frequency and amplitude) simultaneously with P_O2 and water temperature using custom-written routines in MATLAB R2007b. The large size of the electrodes relative to the size of the fish chamber was chosen to minimize the issue of amplitude changes due to changes in the fish position relative to the electrodes. Within each individual fish at each P_O2 level, the mean standard error on raw 30 s median amplitude was 0.0007 even though the fish position was changing. Given that the standard errors across P_O2 levels ranged between 0.045 and 0.153, any recorded amplitude shifts due to changes in fish position relative to the electrodes were small relative to changes in amplitude due to changing P_O2 levels. Once the experiment was stopped, the oxygen was slowly returned to normal, and the fish was allowed to recover. After recovery, the mass, total length and gender of the fish were recorded, and the fish was returned to its home tank. ASR₉₀ thresholds were quantified on a total of 26 fish (A. leptorhynchus N=14; E. virescens N=12).

For the ASR trials, the average percent time spent performing each behavior at each P_O2 level was calculated for the two species. All activities with exception of ‘motionless’ were summed to estimate the percent time that the fish were active. For each species, Pearson correlation was used to test for relationships between body mass and performance thresholds (P_crit and ASR₉₀), and between body mass and EOD (mean pre-threshold frequency and amplitude). There was no relationship between body size and performance thresholds or between body mass and EOD frequency (P>0.273 for all). Similar to other studies, there was a strong correlation between body mass and mean EOD amplitude across individuals (Pearson correlation: P_crit trials, R=0.452, P=0.034, N=22; ASR₉₀ trials, R=0.555, P=0.007, N=26). Amplitude was thus normalized by dividing each amplitude value by the maximum amplitude value for each fish to make the changes in the EOD amplitude comparable among subjects. Decreases in normalized EOD amplitude indicate decreases in absolute EOD amplitude. Separate ANOVAs were used to test for an effect of species and gender on ASR₉₀ threshold, mass-adjusted and P_crit. Repeated-measures ANOVA was used to test for an effect of species, gender and their interaction on gill ventilation rate before and after the ASR₉₀ threshold and before and after the P_crit. Repeated-measures ANOVAs were also used to test for a change in EOD (frequency and normalized amplitude) before and after the ASR₉₀ threshold and before and after the P_crit. ‘Before’ refers to mean EOD frequency and amplitude values calculated up to 2 min before each fish reached the performance threshold (ASR₉₀ and P_crit). ‘After’ refers to mean EOD frequency and amplitude values calculated from the onset of the fish reaching each performance threshold until the end of the experiment. Then, for each of the ‘before’ and ‘after’ data sets, ANOVAs were used to test for an effect of species, gender and their interaction on the change in EOD (frequency and normalized amplitude). Individual analyses of covariance (ANCOVAs) were used to test for an effect of species, gender and their interaction on mean ‘before’ EOD frequency and amplitude with either mass-adjusted , P_crit or ASR₉₀ as covariates. This set of analyses was used to detect relationships between EOD and performance thresholds while controlling for effects of species and gender. There was no interaction between any of the three covariates and either species or gender (ANCOVA: 0.108<P<0.865); therefore, the fixed effect by covariate interaction was removed from the final models.

Across species and genders, mean mass-adjusted was 0.197 mg O₂ g^–1 h^–1, similar to values reported for these species by Julian et al. (Julian et al., 2003). In addition, when mass-adjusted was included as a covariate for EOD (frequency or amplitude) with species and gender as fixed effects, there was no effect of the covariate on EOD frequency, and no effect of mass-adjusted on normalized EOD amplitude (Table 1, Fig. 1). Regardless of oxygen consumption, A. leptorhynchus showed higher EOD frequencies and lower absolute EOD amplitudes than E. virescens.

P_crit was lower in E. virescens (mean 1.46 kPa) than in A. leptorhynchus (mean 2.14 kPa; ANOVA: F_1,19=32.507, P<0.001; Fig. 2). Across species and genders, there was a drop in both EOD frequency and normalized amplitude after the P_crit (repeated-measures ANOVA: EOD frequency, F_1,22=6.840, P=0.016; EOD normalized amplitude, F_1,22=17.522, P<0.001). For each species, EOD frequency was 3% lower after the P_crit compared with before (Table 2, Fig. 3A), whereas normalized EOD amplitude declined by 31% (Table 2, Fig. 3B).

There was no effect of gender or the interaction between gender and species on P_crit (ANOVA: gender, F_1,19=1.128, P=0.304; species×gender, F_1,19=2.222, P=0.156). Also there was no effect of species, gender or their interaction on the change in EOD frequency (Table 2), the magnitude of the amplitude decline (Fig. 3B) or mass-adjusted (Table 1).

When P_crit was included as a covariate for EOD frequency and amplitude with species and gender as fixed effects, there was no effect of the covariate on EOD frequency (ANCOVA: P_crit, F_1,18=0.234, P=0.476; species, F_1,18=44.740, P<0.001; gender, F_1,18=1.990, P=0.179; species×gender, F_1,18=2.575, P=0.129) and no effect of the covariate on normalized EOD amplitude (ANCOVA: P_crit, F_1,18=0.084, P=0.777; species, F_1,18=1.407, P=0.255; gender, F_1,18=3.598, P=0.079; species×gender, F_1,18=2.017, P=0.177).

Consistent with the pattern for P_crit, E. virescens exhibited a lower mean ASR₉₀ threshold than A. leptorhynchus (ANOVA: F_1,24=37.222, P<0.001; Fig. 4A), suggesting higher tolerance to hypoxic stress in E. virescens. However, there was no effect of gender or the interaction between gender and species on the ASR₉₀ threshold (ANOVA: gender, F_1,24=0.010, P=0.922; species×gender, F_1,24=0.085, P=0.773). All fish increased their gill ventilation rates after crossing the ASR₉₀ threshold (Table 2, Fig. 4B). Despite reduced overall activity levels, A. leptorhynchus spent more time using ASR as P_O2 decreased (Fig. 5A). Shortly after the ASR₉₀ threshold, A. leptorhynchus stopped using ASR and became motionless at the bottom of the tank. Eigenmannia virescens remained active throughout the ASR trials, allocating more time to performing ASR as the P_O2 dropped (Fig. 5B).

Across species and genders, there was a drop in both EOD frequency and normalized amplitude after the ASR₉₀ threshold (repeated-measures ANOVA: EOD frequency, F_1,23=5.061, P=0.034; EOD normalized amplitude, F_1,23=131.743, P<0.001). For both species, mean EOD frequency was 2% lower after the ASR₉₀ threshold (Table 2, Fig. 6, Fig. 7A). In contrast to the P_crit trials, E. virescens showed a larger drop in amplitude after the ASR₉₀ threshold (61%) than A. leptorhynchus (29%) (Table 2, Fig. 6, Fig. 7B). In both species, males showed a 1–3% larger drop in normalized EOD amplitude than females (Table 2, Fig. 7B). However, there was no effect of gender, species or their interaction on gill ventilation rates or the change in EOD frequency measured either before or after the ASR₉₀ threshold (Table 2).

When the ASR₉₀ threshold was included as a covariate for EOD frequency and amplitude with species and gender as fixed effects, there was a weak, non-significant positive relationship between the covariate (ASR₉₀ threshold) and EOD frequency (ANCOVA: ASR₉₀, F_1,21=4.003, P=0.059; species, F_1,21=221.167, P<0.001; gender, F_1,21=13.032, P=0.002; species×gender, F_1,21=35.972, P<0.001), but no effect of the covariate on mean normalized EOD amplitude (ANCOVA: ASR₉₀, F_1,21=0.007, P=0.933; species, F_1,21=9.015, P<0.007; gender, F_1,21=3.473, P=0.076; species×gender, F_1,21=0.304, P=0.587).

Although Crampton (Crampton, 1998) provided the first interspecific comparisons of hypoxia tolerance in gymnotiform fishes, this is the first study to quantify P_crit and ASR₉₀ thresholds in wave-type gymnotiforms. Under normoxic conditions, we found no evidence for a relationship between metabolic rate and either EOD frequency or amplitude, suggesting that there is no detectable metabolic cost associated with producing a higher-frequency EOD (Fig. 1A). However, when exposed to progressive hypoxia, fish responded, in general, by reducing EOD amplitude with only a modest decrease in frequency (Figs 3, 6, 7). The weak drop in frequency associated with hypoxic stress suggests that the two species in this study may be incapable of reducing frequency effectively under short-term hypoxic stress as postulated earlier (Crampton, 1998; Markham et al., 2009), or the ability to modulate frequency may require a longer acclimation period (days instead of hours). In the wave-type fish Sternopygus macrurus, Markham et al. (Markham et al., 2009) suggested that amplitude may be the only means of EOD modulation, and provided evidence that amplitude can be modulated quickly by ion channel trafficking in electrocytes (within minutes) in response to environmental changes, such as social stimulation or nightfall. Pituitary peptide hormones control the insertion of sodium channels into the excitable membrane of the electrocytes of the electric organ. With a larger number of sodium channels available, more sodium current can flow during the electrocyte action potential, which increases action potential amplitude at the level of individual cells and overall EOD amplitude at the organismal level because of the combined effect of many electrocytes. Because sodium needs to be pumped back out of the electrocytes by the ATP-dependent sodium–potassium pump, additional ion flux is expected to be energetically costly (Attwell and Laughlin, 2001). In support of Markham and colleagues (Markham et al., 2009), the results of the present study suggest that amplitude reduction indeed may be the main option for saving energy in wave-type electric fish when confronted with hypoxia, at least in the short term (minutes to hours). More work is necessary to identify whether the drop in amplitude is only a consequence of ATP depletion in electrocytes, which then become unable to maintain normal ionic balances with the Na⁺/K⁺-ATPase, or whether it follows from hypoxia-induced downregulation of the Na⁺/K⁺-ATPase and/or the inactivation of sodium channels (reviewed in Hochachka and Lutz, 2001), or whether at least some species of weakly electric fish can balance their energy needs under hypoxic conditions by controlling ion channel trafficking (Markham et al., 2009).

Although both species exhibited modest drops in EOD frequency during hypoxic stress, E. virescens demonstrated a greater flexibility in EOD via larger drops in amplitude during the ASR₉₀ trials (ASR₉₀: 61%) relative to A. leptorhynchus (ASR₉₀: 29%; Figs 6, 7). In addition, E. virescens exhibited more consistent amplitude reduction patterns and thus seems to have more control over their amplitude. Compared with E. virescens, A. leptorhynchus exhibited more variable reductions in amplitude, with some fish reducing amplitude at very high oxygen levels and more than half of the fish maintaining amplitude down to a level close to the ASR₉₀ threshold, after which amplitude dropped quickly. Although it is possible that individual variation in activity levels may play a role in the variability of A. leptorhynchus amplitude reductions, this observation also provides some support for the hypothesis that E. virescens can actively regulate their EOD amplitude in order to maintain their metabolic rate under hypoxic stress, whereas A. leptorhynchus may be less capable of doing so. It is not clear at this point whether the differences in EOD flexibility may derive, at least in part, from the difference in the type of electric organ. The electric organ of Eigenmannia spp. is derived from muscle tissue, whereas the electric organ of all apteronotids is composed of the terminals of spinal motor neurons (Bennett, 1971). Greater flexibility of EOD amplitude in Eigenmannia spp. is also suggested by circadian modulation of EOD amplitude in the related species S. macrurus (Markham et al., 2009); this has also been observed in Eigenmannia but not in Apteronotus (Goldina et al., 2007). This difference in amplitude flexibility may, at least partially, explain the difference in hypoxia tolerance between the two species (Fig. 2, Fig. 4A).

ASR is an adaptive behavioural response to hypoxia (e.g. Kramer, 1983; Kramer et al., 1983; Val et al., 1998). By performing ASR, fish are accessing the oxygenated surface layer of water, although swimming to the surface is energetically costly in itself. Nevertheless, gill ventilation rates are expected to decrease once ASR is utilized if oxygen demands are being met (Chapman et al., 1995). In this study, the overall increase in gill ventilation rates after the ASR₉₀ threshold indicates that for both species (Fig. 4B), the use of ASR is not particularly efficient (Chapman et al., 1995) and may be associated with a need to offload aquatic CO₂ (Brauner and Randall, 1998). In contrast to the ASR experiment, fish were quiescent in the respirometer during the P_crit trials, even though they did not have access to the surface. The larger drop in amplitude and the difference between species during the ASR₉₀ trials (Fig. 7) relative to the P_crit trials (Fig. 3) may be due, at least in part, to the relatively unconstrained position of the fish relative to the recording electrodes. In the P_crit trials, the fish were always in the same orientation and at approximately the same distance from the recording electrodes, which were positioned in front of the head and behind the tail, respectively. It is also likely that the large amplitude drop and the difference between species may be driven by the difference in activity levels between the two species during the ASR₉₀ trials. Chabot and Claireaux (Chabot and Claireaux, 2008) suggest that energetic limitations due to oxygen will impact metabolic scope more so than metabolic rate. They provide evidence in four marine species that metabolic scope is proportional to P_O2 (Chabot and Claireaux, 2008). Although metabolic rate was not recorded during the ASR₉₀ trials, we observed shifts in both the EOD and activity levels as P_O2 dropped. Opposite to A. leptorhynchus, E. virescens remained active throughout the ASR trials and thus their relatively large reduction in amplitude may have allowed them to allocate more energy to activity as P_O2 dropped, especially if their aerobic capacity was becoming more limited. If indeed metabolic scope decreases with decreases in P_O2, we might expect to see a link between EOD frequency and/or amplitude and metabolic scope. An important next step in resolving the energy demands of electrocommunication may be to quantify active metabolic rate and metabolic scope in conjunction with EOD in the context of hypoxia.

Although there was no evidence for metabolic cost to scale with EOD frequency, the results of this study suggest that wave-type fish reduce their EOD amplitude to reduce direct energetic costs, and thus maintain metabolic rate under hypoxia. The strength of the voltage perturbation caused by a nearby object on the skin of the fish is proportional to the strength of the fish's electric field (Rasnow, 1996). Thus, for an object, such as prey, a given distance from the fish, the perturbation experienced by electroreceptor organs in the skin will decrease with decreasing EOD amplitude. As a consequence, the downside of amplitude reduction is a reduced range of detection of the electrosense. Amplitude reduction during the daytime, when the fish are resting and hiding, has been described in various pulse-type fish species (Hagedorn, 1995; Franchina and Stoddard, 1998) and in S. macrurus (Markham et al., 2009), which is closely related to Eigenmannia. Metabolic measurements in Brachyhypopomus gauderio (previously called B. pinnicaudatus) suggest that EOD generation in this species is very costly with a maximal proportion of the total energy budget of 22% in sexually mature males at night (Salazar and Stoddard, 2008). Therefore, amplitude reduction likely constitutes a mechanism for saving energy in addition to rendering the fish less conspicuous to electroreceptive predators (Stoddard, 2002).

Although EOD frequency in most pulse-type electric fish is highly variable (Crampton and Albert, 2006), it has been shown to be quite constant in wave-type fish. In fact, the EOD pacemaker nucleus of A. leptorhynchus is the most regular biological oscillator known with coefficients of variation of the EOD period as low as 2×10^–4 (Moortgat et al., 1998). Even though EOD frequency of wave-type fish can be extremely stable on time scales of hours, it can be modulated rapidly and transiently under the control of pre-pacemaker nuclei for purposes of communication (Metzner, 1999; Zakon et al., 2002). Slower changes on the time scale of development and sexual maturation have been shown to be under the control of steroid hormones (Meyer et al., 1983; Meyer et al., 1987; Zakon et al., 1991; Dunlap and Zakon, 1998), affecting voltage-gated sodium and potassium channels in the electrocytes (Ferrari et al., 1995; McAnelly and Zakon, 2007). Given that the kinetics of sodium and potassium channels in electrocytes are tightly correlated with each other and with EOD frequency (McAnelly and Zakon, 2007), it appears likely that the kinetics are matched for a given EOD frequency to minimize sodium waste current (Hasenstaub et al., 2010). Short-term reduction of EOD frequency, on the time scale of our experiments, may therefore provide little, if any, energy savings.

In the present study, the fish were exposed to progressive hypoxia for only a few hours. If the hypoxic episode were to persist for days or weeks (as in nature), it is unclear whether the levels of amplitude reduction and/or activity levels would be maintained or further reduced because of weaker hypoxia tolerance under long-term hypoxia exposure (Plante et al., 1998). Based on evidence from earlier hormone studies, longer-term exposure to hypoxia may also influence the endocrine system, which might in turn impact EOD frequency and possibly amplitude (Dunlap and Zakon, 1998). Another possibility is that hypoxia-adaptation mechanisms, such as changes in blood chemistry or gill structure, would allow the EOD amplitude to return to normoxic levels. Thus, an important next step will be to quantify EOD modulations in response to long-term hypoxia (days or weeks).

The apparent inability of wave-type weakly electric fish to change their EOD frequency substantially even when challenged with hypoxia supports the notion that these fish are ‘locked in’ at their individual-specific baseline frequencies by tightly matched kinetics of sodium and potassium channels in their electrocytes (McAnelly and Zakon, 2000). There is evidence that EOD frequency is related to body size in males of several species and that it is used as a signal of dominance (Hagedorn and Heiligenberg, 1985; Dunlap, 2002; Triefenbach and Zakon, 2003; Fugère et al., 2011). An inability to change the baseline EOD frequency on short time scales could thus be a biophysical mechanism that prevents the animals from bluffing and enforces the honesty of EOD frequency as a signal. However, it is currently unclear which mechanism should prevent a small male A. leptorhynchus from expressing ion channels with kinetics tuned to high frequencies. Short-term deviations from the baseline do occur in communication interactions under the control of the pacemaker nucleus in the brainstem (jamming avoidance response, chirps, rises and long-term frequency elevations) (Zakon et al., 2002), but they may be expensive because of the resulting mismatch between ion channel kinetics and EOD frequency. A signalling function of EOD amplitude has not been demonstrated, but it appears likely, because amplitude of the EOD is positively correlated with body size and thus the size of the electric organ (Dunlap, 2002; Curtis and Stoddard, 2003). Our results, and those of Salazar and Stoddard (Salazar and Stoddard, 2008) for a pulse-type species, show that maintaining a large EOD signal is energetically costly, which should make it an honest indicator of physical condition. Further research combining studies of the electrical behaviour and the physiology of weakly electric fish will prove to be fruitful and stimulating for understanding the forces shaping the evolution of animal communication systems.

We thank Martin Cuddy, Jamie Dallaire, Gloria Booth-Morrison, Chris Wong and Emily Darbyson for their assistance with various aspects of this project; and to John Lewis and the anonymous reviewers for comments that improved the quality of this manuscript.