Positive feedback promotes terrestrial emergence behaviour in an amphibious fish

Spatial and temporal environmental heterogeneity creates a fundamental phenotype–environment mismatching problem for animals, as phenotypes well suited to one environment are often poorly suited to different environments (Agrawal, 2020). To minimize these functional trade-offs, animals that inhabit variable environments either adopt generalist ‘jack-of-all-trades’ phenotypes or are responsive to environmental conditions (Kassen, 2002; West-Eberhard, 2003). One form of responsiveness is phenotypic plasticity, the ability of an animal to modify its phenotype to match prevailing conditions (Pfennig, 2021). Alternatively, animals may respond to variability by choosing to occupy the most favourable subset of available habitats (Edelaar et al., 2008). However, there is little empirical data about how plasticity and habitat choice interact to determine phenotypes in heterogeneous environments (Edelaar and Bolnick, 2019).

Phenotypic plasticity and habitat choice can sometimes negate or offset one another (Turko and Rossi, 2022). For example, in situations when organisms can always select optimal habitats, phenotypic plasticity is not needed to improve the environment–phenotype match (Edelaar et al., 2017; Schneider and Meyer, 2017). Similarly, phenotypically plastic animals are often habitat generalists, as phenotypes can easily be adjusted to suit various conditions (Asbury and Adolph, 2007; Manenti et al., 2013). However, plasticity and habitat choice can also reinforce one another via positive feedback, although empirical examples are rare (Dingemanse et al., 2010; Miner et al., 2005; West-Eberhard, 1989). In stream-living salamanders, individuals choose to inhabit fast-flowing riffles or slow-moving pools depending on morphological traits linked to swimming ability; phenotypic plasticity then reinforces these morphological differences between habitats (Lowe and Addis, 2019). Likewise, predatory ambush bugs choose to perch on flowers that are a similarly coloured to their own bodies, and then this match is improved via plastic changes to body colouration (Boyle and Start, 2020). These examples of habitat choice–plasticity feedbacks focus on spatially heterogeneous but temporally consistent environments. In many cases, however, environmental conditions vary in both space and time, but to date there has been little consideration of how plasticity and habitat choice interact in these scenarios.

In the face of temporal environmental change, animals must evaluate whether to remain and use plasticity to mitigate the phenotype–environment mismatch or instead move elsewhere (Hendry, 2016; Scheiner, 2016). If environmental changes are gradual, plasticity that produces a more ‘tolerant’ phenotype may be expected given that dispersal often has fitness costs (Edelaar et al., 2008). However, if the rate of environmental change is too rapid to be accommodated by plasticity or when there are limits to the expression of plasticity, animals instead may simply leave the changing environment. In these situations, movement to an alternative habitat may then cause plastic changes that enhance phenotype–environment matching in the new habitat. As a result, the longer an animal remains in the new habitat, the more it may become mismatched to the original habitat, making a return less likely. In this way, the existence of habitat choice–plasticity feedback may reduce the occurrence of habitat transitions but also increase their success.

One of the most dramatic environmental transitions experienced by any animal occurs when amphibious fishes move between water and land (Damsgaard et al., 2019; Ord and Cooke, 2016; Turko et al., 2021). The fundamental physical differences between aquatic and terrestrial habitats result in a suite of functional trade-offs (Dejours, 1988; Wright and Turko, 2016). For example, fishes in water benefit from a large gill surface area that enhances oxygen uptake and improves tolerance to hypoxia, but in terrestrial conditions, large gills may be a liability because of the potential for damage and evaporative water loss (Nilsson et al., 2012). This respiratory trade-off is especially problematic for amphibious fishes, as many species live in hypoxic aquatic habitats (Graham, 1997) and severe hypoxia is a common proximate reason why amphibious fishes leave water (e.g. Livingston et al., 2018; Mandic et al., 2009; Urbina et al., 2011; reviewed by Sayer, 2005). Once out of water, many amphibious fishes use alternative respiratory surfaces and gills may become nonfunctional (Randall, et al., 1981). Some amphibious fishes, such as the mangrove rivulus, Kryptolebias marmoratus (Ong et al., 2007), and bichir, Polypterus senegalus (Turko et al., 2019a), modulate this gill surface area trade-off by reversibly altering the size of the interlamellar cell mass (ILCM). In air, these fishes increase the ILCM, reducing overall functional gill surface area. However, upon the return to water, several days are required to decrease the ILCM and during this lag-time period, respiratory function in water is impaired (Turko et al., 2012). Thus, by moving to land to escape aquatic hypoxia, and then developing a slowly reversed terrestrial gill morphology that limits respiration in water, a feedback loop may emerge that promotes terrestrial habitat choice (Turko et al., 2018).

We experimentally tested the hypothesis that habitat choice (behavioural hypoxia avoidance) and morphological plasticity (gill remodelling) reinforce one another in a positive feedback loop using the amphibious mangrove rivulus as a model. These fish readily move between water and land and experience severely hypoxic aquatic habitats in the wild (Rossi et al., 2019; Taylor, 2012; Wright, 2012). We first tested the critical predictions that terrestrial acclimation should decrease aquatic hypoxia tolerance and increase the sensitivity of the hypoxic emergence response. Second, we tested the prediction that aquatic hypoxia exposure should increase the frequency of emergence behaviour, and in turn should decrease subsequent aquatic hypoxia tolerance owing to gill remodelling on land.

We used adult Kryptolebias marmoratus (Poey 1880) hermaphrodites (0.09±0.02 g; strain 50.91; Tatarenkov et al., 2010) bred in captivity at the University of Guelph. Fish were held individually in 120 ml plastic containers filled with 60 ml of brackish water (15‰, 25°C, 12 h:12 h light:dark cycle) and fed Artemia nauplii three times a week. To acclimate fish to terrestrial conditions (7 days), fish were placed on a filter paper substrate that was saturated with brackish water in standard housing containers as described previously (Ong et al., 2007). Control fish were held in identical containers filled with brackish water and were fasted for the 7-day acclimation period because mangrove rivulus cannot feed on land (Turko et al., 2019b; Wells et al., 2015). These acclimation conditions were used for each of the three experimental series described below. All experiments were approved by the University of Guelph Animal Care Committee (Animal Use Protocol 3891).

To test how the sensitivity of the hypoxic emergence response varied between air- and water-acclimated fish (each n=10), we measured the acute emergence response using established methods (Livingston et al., 2018; Regan et al., 2011). We placed individual fish into the water section of a container composed of a lower section filled with water surrounded by an air-exposed platform, and allowed them to adjust (30 min) until they displayed normal behaviour (e.g. routine swimming without freezing or darting, no sign of hyperventilation), indicating that they were minimally stressed. In two cases, fish jumped out of water during this acclimation period; these fish were excluded from all analyses. After the acclimation period, we began to decrease the oxygen concentration by bubbling nitrogen gas from 21 to 0 kPa over 20 min. We measured water oxygen saturation using a fibre-optic oxygen probe (Witrox-4, Loligo Systems, Copenhagen, Denmark) and recorded the oxygen level at emergence.

To assess hypoxia tolerance, we measured time until loss of equilibrium (LOE) under severe hypoxia using standard methods (Regan et al., 2017). Briefly, we placed water- or air-acclimated fish (each n=6) into mesh-sided chambers submerged in a 5-gallon (19 l) aquarium filled with brackish water to prevent the fish from reaching the air–water interface. The aquarium was covered with plastic film to minimize gas exchange with atmospheric air. Fish were allowed to recover for 30 min, after which we bubbled nitrogen gas to decrease the oxygen partial pressure (P_O2) from 21 to 0.5 kPa over 10 min, at which point it was held constant at 0.5 kPa for the remainder of the experiment (Regan and Richards, 2017; Regan et al., 2017). We recorded the time until fish lost equilibrium, defined as the point when the fish did not respond to a physical stimulus.

To test the prediction that hypoxia exposure should increase the frequency of emergence and subsequently decrease hypoxia tolerance, we exposed fish to one of three oxygen levels: normoxia (21 kPa oxygen; n=38), hypoxia (2.1 kPa oxygen, n =39) or extreme hypoxia (0.5 kPa; n=28) for 7 days. Within each group, fish were acclimated to either water or air for 7 days immediately before the experiment began to determine whether differences in gill morphology caused by air versus water acclimation influenced emergence behaviour in hypoxia. Fish were held individually in containers like the ones used to assess the hypoxic emergence response (described above), except the lower, water-filled portion of the containers were mesh-sided and submerged in a larger water bath, allowing for free circulation of water. Oxygen saturation of the water bath was maintained with an oxygen control system (OxyCTRL, Loligo Systems) that automatically bubbled nitrogen gas into a header tank when oxygen concentrations exceeded the experimental setpoint.

At the beginning of these trials, the oxygen level in the water was first adjusted to the desired setpoint (normoxic, hypoxic or extremely hypoxic) and allowed to stabilize before fish were added. Fish were then gently placed directly into the aquatic section of the chamber, and video recording for subsequent measurement of emergence behaviour began immediately. Emergence behaviour was recorded during the 12 h of light per day, as there is no difference in emergence behaviour between day and night (Turko et al., 2011). For the duration of this habitat choice experiment, fish were not fed and great care was taken to minimize all human disturbance so as to not influence their behaviour. After this 7 day experiment, we measured the acute emergence response (total n=58, n=9–10 per pre-acclimation/exposure combination) or time to LOE (total n=48, n=7–9 per pre-acclimation/exposure combination) on random subsets of fish as described above.

To determine whether there was a relationship between emergence behaviour and gill morphology, a subsample of fish (total n=30, n=4–8 per pre-acclimation/exposure combination) was euthanized with buffered tricaine methanesulfonate and processed for histology as described previously (Turko et al., 2011). We measured the proportion of ILCM coverage of 20 randomly selected lamellae spanning five filaments for each fish (Turko et al., 2014, 2018).

We noticed that some fish in the extreme hypoxia treatment never lost equilibrium despite remaining in water for lengths of time that exceeded the degree of tolerance measured in our LOE experiments, and hypothesized these fish were using aquatic surface respiration (Kramer and McClure, 1982). To test this hypothesis, we placed individual fish (air-acclimated n=4, water-acclimated n=5) in a 5-gallon glass aquarium filled halfway with water and covered on all sides with the exception of a small viewing window. Fish were allowed to adjust for 15 min, and then we slowly introduced nitrogen gas to reduce oxygen levels at a constant rate (−0.4 kPa O₂ min⁻¹). Each trial was video recorded, and we quantified the proportion of time each individual fish used aquatic surface respiration (defined as when the mouth of the fish is within 1 mm of the surface of the water; Kramer and McClure, 1982) as a function of oxygen saturation in the water. We quantified the use of aquatic surface respiration in 10 min intervals, during which time P_O2 declined by ∼4 kPa. The mean P_O2 of each interval was used for statistical analysis. Scoring of aquatic surface respiration was done by a single observer (L.T.) who was not blinded to the acclimation condition of each fish.

All statistics were calculated using R version 4.0.2. To compare oxygen saturation at emergence and time to LOE in water- versus air-acclimated fish, we used Student's t-tests (Series 1). To compare oxygen saturation at emergence, time to LOE and ILCM coverage among hypoxia exposure conditions (Series 2), we used two-way ANOVA, followed by Tukey's post hoc tests. To assess changes in emergence behaviour over time within each acclimation condition and hypoxia exposure condition, we used permutational two-way ANOVA (1,000,000 permutations, package permuco; https://CRAN.R-project.org/package=permuco) because our data did not meet assumptions of normality or homoscedasticity. Permutational ANOVA is similar to a parametric ANOVA, but rather than assuming a Gaussian distribution, the dataset is repeatedly randomized, and the proportion of randomly ordered datasets with greater treatment effects than the actual dataset is calculated. We used a series of two-way ANOVAs (one per oxygen tension) rather than a single three-way ANOVA because we found a marginal three-way interaction in our exploratory analysis (F=1.73, P=0.056) and because the key predictions made by the positive feedback hypothesis relate to patterns over time within each treatment rather than statistical comparison among treatments. Individual identity was included as a random factor in all models to account for repeated measurements over the 7-day trials. Post hoc comparisons were calculated using permutational t-tests (1,000,000 permutations) implemented using the pairwise.perm.t.test function (package RVAideMemoire; https://CRAN.R-project.org/package=RVAideMemoire), and P-values were adjusted for multiple comparisons using the Benjamini and Yekutieli correction, which controls the false discovery rate (Benjamini and Yekutieli, 2001). We used similar permutational ANOVA and post hoc permutation t-tests (with Benjamini and Yekutieli corrected P-values) to assess changes in the proportion of time fish used aquatic surface respiration during the induction of hypoxia (Series 3), as these data did not meet parametric assumptions of normality or homoscedasticity.

Fish that were acclimated to air emerged at oxygen levels that were 2-fold higher than the oxygen levels that caused water-acclimated fish to emerge (t-test P=0.004; Fig. 1A). Hypoxia tolerance (time to LOE) of air-acclimated fish was approximately half that of water-acclimated fish (t-test, P=0.0007; Fig. 1B).

In our habitat choice experiment, exposure to aquatic hypoxia dramatically increased the amount of time fish spent out of water (Fig. 2). Fish in normoxia rarely emerged from water over the course of the trial, and emergence frequency in previously air-acclimated fish slightly but significantly decreased over the 7 day trial (time×acclimation P=0.0069; Fig. 2A). In moderate hypoxia, emergence frequency similarly decreased over time, and previously air-acclimated fish initially spent significantly more time out of water than water-acclimated fish (time×acclimation P=0.0072; Fig. 2B). In extreme hypoxia, fish regularly emerged from water (∼30% of the time) but this did not depend on previous acclimation condition (P=0.12) or time (P=0.057, interaction P=0.11; Fig. 2C).

At the end of the habitat choice experiment, exposure to extreme aquatic hypoxia almost doubled the proportion of gill lamellae that were covered by an ILCM compared with normoxic or hypoxic conditions (two-way ANOVA, P=0.002; Fig. 2A–D). Fish exposed to extreme hypoxia also had a significantly more sensitive hypoxic emergence response (two-way ANOVA, P<0.0001; Fig. 3E). Finally, exposure to aquatic hypoxia decreased hypoxia tolerance (time to LOE) by approximately 50% (two-way ANOVA, P<0.00001; Fig. 3F). In all cases, there was no significant effect of previous acclimation to air versus water (all P>0.05).

Mangrove rivulus significantly increased their use of aquatic surface respiration when oxygen tension fell below 10 kPa (P<0.0005; Fig. 4). At the lowest P_O2 we tested (1.5 kPa), fish used aquatic surface respiration for 95% of the recording period. This pattern of aquatic surface respiration was not affected by prior acclimation to air or water (P>0.05; Fig. 4).

We found three lines of evidence consistent with the hypothesis that hypoxia-induced emergence behaviour can become self-reinforcing via positive feedback as gill remodelling on land decreases respiratory function in water and promotes emergence behaviour. First, air-acclimated fish emerged at higher oxygen levels relative to water-acclimated fish, indicating increased sensitivity to aquatic hypoxia. Second, air acclimation decreased aquatic hypoxia tolerance, increasing the reliance on avoidance behaviour for coping with harsh aquatic conditions. These are both necessary conditions for the establishment of a positive feedback loop. Third, after 7 days of exposure to hypoxic water with access to land, the sensitivity of the hypoxic emergence response increased while hypoxia tolerance decreased. Overall, our data indicate that emergence behaviour is an important strategy for coping with adverse aquatic conditions in this species, consistent with previous studies (Rossi et al., 2019; Turko et al., 2018). Importantly, our data also indicate that emergence behaviour makes it increasingly difficult for these fish to maintain homeostasis in hypoxic water. We suggest that this sort of positive feedback between plasticity and habitat choice has widespread potential to generate extreme phenotypes in animals. Furthermore, we speculate that if habitat choice–plasticity feedback results in genetic assimilation (Crispo, 2007; Pigliucci et al., 2006; Schneider and Meyer, 2017), this process may accelerate evolutionary habitat transitions such as the invasion of land by fishes.

There are two key elements necessary to generate positive feedback between habitat choice and phenotypic plasticity. First, expressions of plasticity that improve performance in one habitat must also decrease performance in alternative habitats (i.e. an antagonistic trade-off). Second, an animal must be capable of assessing its own phenotype and using this information to choose suitable habitats (i.e. matching habitat choice; Camacho et al., 2020). Our data indicate that both elements exist for mangrove rivulus moving between water and land, suggesting that the positive feedback hypothesis is plausible.

Mangrove rivulus face a suite of physiological trade-offs when moving between aquatic and terrestrial environments, the first requirement for positive feedback to occur. Many of these trade-offs involve the oxygen transport cascade. Gill remodelling on land may reduce evaporative water loss in air, but remodelling also decreases the capacity for oxygen uptake in water (Turko et al., 2012). Angiogenesis of cutaneous epidermal capillaries on land enhances aerial oxygen uptake (Blanchard et al., 2019), perhaps at the cost of oxygen loss to hypoxic water (Scott et al., 2017). The oxygen binding affinity of haemoglobin also increases in air, but in water this may impair oxygen offloading to metabolically active tissues (Tunnah et al., 2021; Turko et al., 2014). In the present study, terrestrial acclimation decreased overall aquatic hypoxia tolerance (i.e. time to LOE) by approximately 50%, indicating that the combined effects of these physiological trade-offs have a dramatic organism-level consequence.

We also found evidence that habitat choice depends on physiological state (i.e. self-assessment), fulfilling the second requirement for positive feedback. When acutely challenged with progressive aquatic hypoxia, air-acclimated fish emerged at oxygen concentrations that were more than double that of water-acclimated fish. This result implies that mangrove rivulus can assess their physiological tolerance of aquatic hypoxia and make anticipatory decisions, as the concentrations of oxygen that elicited emergence were ∼3- to 8-fold higher than the concentrations that these fish could tolerate for 20–40 min in our LOE experiments. The mechanism(s) used for self-assessment of hypoxia sensitivity are not known, but previous work has found that the hypoxic emergence response of this species is regulated by oxygen-sensitive neuroepithelial cells (Regan et al., 2011). One possibility is that self-assessment occurs via internally oriented neuroepithelial cells that sense blood P_O2 (Perry et al., 2009), as impaired respiratory function in air-acclimated fish would presumably cause a faster decrease in blood P_O2 compared with water-acclimated fish. Understanding the mechanism of self-assessment is an exciting area for future research.

Our habitat choice experiment provides further evidence for positive feedback between plasticity and habitat choice. We found that fish exposed to extreme aquatic hypoxia spent substantial amounts of time out of water (70-fold more than controls in normoxic water) and were more sensitive to aquatic hypoxia than fish exposed to normoxic water. Fish exposed to extreme hypoxia were also less hypoxia tolerant (i.e. time to LOE was reduced by ∼40%) and had reduced functional gill surface area. Notably, these plastic changes are opposite to those typically observed with hypoxia acclimation in fishes (Perry et al., 2009; Richards, 2009). For example, when mangrove rivulus were prevented from leaving water during hypoxia acclimation, gill surface area increased and respiratory function improved (Turko et al., 2012). In contrast, the ILCM of fish exposed to extreme hypoxia covered the lamellae to a similar extent as previously measured in forcibly air-acclimated fish (Turko et al., 2012, 2018). Overall, our results indicate that when mangrove rivulus are given the choice between living in extremely hypoxic water or on land, the respiratory phenotype is shaped by the terrestrial habitat. Thus, habitat choice may be a key factor that shapes physiological tolerance of extreme environments.

Mangrove rivulus exposed to extreme aquatic hypoxia spent much more time out of water than fish in the other treatments, as predicted by the positive feedback hypothesis. However, contrary to the feedback hypothesis, emergence rates did not increase over the duration of the extreme hypoxia experiment despite the plastic expression of a terrestrial respiratory phenotype in these fish. Emergence was also surprisingly infrequent in the moderate hypoxia treatment considering that acute exposure to this oxygen concentration elicited emergence in our first series of experiments. One explanation for both of these observations is that mangrove rivulus extensively used aquatic surface respiration (Chapman and Mckenzie, 2009; Kramer and McClure, 1982) in combination with emergence to cope with aquatic hypoxia. Aquatic surface respiration allows fishes to selectively breathe from the relatively well-oxygenated layer of water at the air–water interface, thus reducing the severity of aquatic hypoxia. Our results nonetheless illustrate the interactive effects of habitat choice and physiological plasticity. Even subtle habitat choice, i.e. a shift up in the water column to use aquatic surface respiration, may have minimized the extent of aquatic hypoxia experienced by the fish and thus blunted plastic responses typical of hypoxia acclimation when fish cannot access the air–water interface. Reliance on aquatic surface respiration in our experiment may be a laboratory artefact. In the wild, mangrove rivulus under similarly hypoxic aquatic conditions spend ∼90% of their time on land (Turko et al., 2018) and do not appear to frequently use aquatic surface respiration (A.J.T., personal observation), perhaps because this behaviour increases the risk of predation (Domenici et al., 2007; Kramer, 1983; Riesch et al., 2010). The terrestrial mangrove habitat provides many hiding opportunities amongst leaf litter and other debris, which we speculate may reduce the risk of predation compared with when fish use aquatic surface respiration.

A fundamental goal of physiological ecology is understanding the processes that promote phenotype–environment matching (Bolnick and Otto, 2013; Botero et al., 2015; Edelaar et al., 2008). Two important mechanisms are behavioural habitat choice and phenotypic plasticity, but these mechanisms have often been studied in isolation. However, our results indicate that plasticity and matching habitat choice interact and can feed back on one another. Understanding these interactions can be important for interpreting patterns of phenotypic variation and predicting how animals will respond to changing environments. For example, our data showing that hypoxia-exposed mangrove rivulus have relatively low gill surface area and hypoxia tolerance cannot be explained without considering both plasticity and habitat choice in tandem.

Positive feedback moves systems away from equilibrium and as a result has been hypothesized to facilitate major evolutionary transitions (Crespi, 2004; Hendry, 2016) such as the invasion of land. On evolutionary timescales, phenotypic change coupled with matching habitat choice is thought to promote local adaptation and, potentially, speciation (Camacho et al., 2020; Jacob et al., 2017; Muñoz and Losos, 2018; Nicolaus and Edelaar, 2018). Our results indicate that analogous feedback between plasticity and habitat choice can cause phenotypic divergence within the lifetime of individuals. Given that expressions of plasticity can cause evolutionary change through genetic assimilation (Crispo, 2007; Pigliucci et al., 2006; Schneider and Meyer, 2017), we speculate that positive feedback between habitat choice and plasticity may be an important first step in generating similar evolutionary feedback between these processes. For example, populations of mangrove rivulus that inhabit moderately versus severely hypoxic habitats should differ in emergence rates and thus respiratory phenotype. If these environmental differences persist, evolutionary processes (e.g. genetic assimilation and/or altered selective environment) may reinforce the phenotypic divergence between populations. Over time, this mechanism may therefore promote a major ecological transition by producing terrestrially adapted fishes that have poor aquatic performance owing to respiratory trade-offs.

We thank Matt Regan, Giulia Rossi, Louise Tunnah and Paige Cochrane for helpful discussions, and Matt Cornish, Mike Davies and Abiran Sritharan for animal care.