We analyzed time-depth recorder (TDR) datasets from three loggerhead, six leatherback, and four green turtles (Table 1). The TDRs deployed on loggerhead turtles were Satellite Relay Data Loggers (SRDL–Sea Mammal Research Unit, United Kingdom), while MK10 tags (Wildlife Computers, United States) were used for both leatherback and green turtles. The loggerhead TDRs were attached to the top of the carapace using epoxy (for details, see Patel et al., 2018). For leatherback and green turtles, the TDRs were tethered to the posterior position of the carapace (for details, see Robinson et al., 2016). The loggerhead turtles were sub-adults sampled at foraging areas in the Mid-Atlantic Bight, while the leatherback and green turtles were inter-nesting females sampled on the east coast of South Africa and the northwest coast of Costa Rica, respectively (Table 1). The TDRs recorded depth to within 1 m accuracy at time-intervals of either 4 or 10 s. We standardized the sampling interval for all TDRs by interpolating the data to 1 s intervals.

TDR datasets ranged in duration 187–476 days for loggerhead turtles, 10–34 days for leatherback turtles, and 11–54 days for green turtles. As loggerheads are known to modify their diving behavior in response to seasonal variation in water temperatures (Iverson et al., 2019) and conduct extended “hibernation” dives during winter (Hochscheid et al., 2005), we only analyzed loggerhead data from the first 46 to 139 days (Table 1). This ensured that all data were collected during the summer months (May to November) when tag-derived sea surface temperature values were between 19 and 26°C. In contrast, tracking duration was short enough for leatherback and green turtles to assume that there were no significant changes in diving behavior due to seasonal factors within the available datasets. Moreover, all the leatherback and green turtles’ data were collected from tropical or sub-tropical habitats, where the sea surface temperature values only varied between 24 and 29°C. The loggerhead TDR data have been previously published in Patel et al. (2018). The leatherback and green turtle data have been published in Blanco et al. (2013), Robinson et al. (2017), and Clyde-Brockway et al. (2019).

To estimate gas dynamics in diving sea turtles, we adapted a model that was developed for estimating blood and gas tissue N₂, O₂, and CO₂ tensions in marine mammals and penguins (Fahlman et al., 2006, 2007, 2009, 2018, 2021). In the model, gas exchange first occurs between the respiratory system and arterial blood, and then between the arterial blood and four compartments: (1) the brain; (2) fat and bone; (3) the central circulatory system, which included the heart, kidney, liver, and digestive tract but not blood; and (4) muscle, which included muscle, skin, connective tissue, and all other tissues and organs that were not included in the other compartments. Gas exchange next occurs between the four compartments and the mixed venous blood, and finally between the mixed venous blood and the respiratory system.

Gas exchange was driven by partial pressure/tension gradients as detailed in Fahlman et al. (2006), and a physiologic pulmonary shunt was allowed at increasing pressures due to alveolar compression (Bostrom et al., 2008; Fahlman et al., 2009). While reptilian lungs have faveoli or ediculi instead of the alveoli in mammals, the similarities in their structures mean that they likely respond similarly under pressure differentials. For ease of use, we hereafter refer to these differing features under the singular term faveoli. It should also be noted that sea turtles have the capacity to perform a right-to-left intracardiac shunt. As the role of this cardiac shunt in regulating N₂ is unknown in aquatic reptiles (Burggren et al., 2020), we therefore adjusted the parameters for the modeled pulmonary shunt, so that it mirrored the arterial PN₂ levels that previously reported for a sea turtles during forced submergence (Berkson, 1967).

The model assumed that breathing began the instant the animal reached the surface (i.e., a depth of 0 m). However, due to a combination of measurement error, the use of tethered tags on both leatherback and green turtles, and the original sampling frequency of the TDRs, not all breathing events were associated with a depth measurement of 0. While many studies address this issue by defining a dive as the period when a turtle descends beyond a pre-determined depth (e.g., 2 m), we chose to not use this dive definition as previous models have shown the risk of GE increases rapidly as a diver approaches the surface (Fahlman et al., 2018). Instead, we subtracted 2 m from all depth values. While this ensured that all breathing events were associated with a measurement of 0 depth, we acknowledge that this also slightly reduced the maximum depth of the dives.

When breathing, the arterial blood gas tensions of N₂, O₂, and CO₂ were set to 0.741 atmospheres absolute (ATA), 0.164 ATA, and 0.033 ATA, respectively, as have been previously reported for loggerhead turtles (Lutcavage and Lutz, 1991) with water vapor being 0.062 ATA (Lutcavage and Lutz, 1991; Fahlman et al., 2008). We also assumed that arterial blood gas tensions while breathing were equal to the partial pressures in the faveoli, and that all CO₂ that exchanged for O₂ remained in gas phase in the lungs and did not dissolve into the lung parenchyma (Fahlman et al., 2006).

We used the model published by Bostrom et al. (2008) to estimate faveolar volume at depth and compared the estimated faveolar collapse depth from this model with empirical data previously published in green turtles forcibly submerged inside hyperbaric chambers (Berkson, 1967). This model assumed that the total lung capacity (TLC) includes both the maximum faveolar volume (VfA_max) and the volume of dead space in the trachea and bronchi (VD; Fahlman et al., 2009, 2018). We obtained measurements of TLC for each species from the relevant literature (Table 2; Berkson, 1967; Table 3; Lapennas and Lutz, 1982; Lutz and Bentley, 1985; Lutcavage et al., 1990; Lutcavage and Lutz, 1991; Hochscheid et al., 2007). However, as we were only able to find measurements of VD for green turtles (Berkson, 1967; Gatz et al., 1987), we assumed that VD was equivalent (7% of TLC) in both loggerhead and leatherback turtles. We assumed that gas exchange only occurred in the faveoli, and that gas exchange stopped immediately upon faveolar collapse (i.e., when the faveolar volume was zero, VfA = 0).

To account for the relationship between pulmonary shunt and faveolar collapse, we used Eq. 4 in Fahlman et al. (2009). We also estimated the faveolar volume (Eq. 4 in Bostrom et al., 2008), using the parameters as previously defined for human alveoli (a = 1.04, b = 0.20, and c = 1.21). Finally, we estimated dead space in the upper airways (Eq. 5 in Bostrom et al., 2008) and applied the parameters as previously defined for a compliant human trachea (Kp = −6.44, n = 0.74). We used the same parameters for all three turtle species.

We generated estimates of the relative size of the four compartments in the model using computed tomography of five loggerhead turtles ranging in body mass from 3.2 to 48.7 kg (Appendix 1). To support these results, we also performed mass dissections of two loggerheads, one leatherback, and nine green turtles ranging in body mass from 1 to 355.9 kg and encompassing both sexes. Dissections were conducted by the Fundación Oceanogràfic de la Comunidad Valenciana which is registered as a research unit under the official ID number: ES460250001024 under the collaboration agreement with the Conselleria d’Agricultura, Medi Ambient, Canvi Climàtic i Desenvolupament Rural of the Valencian Regional Goverment, the National Oceanic and Atmospheric Administration (Florida Fish and Wildlife Conservation Commission Marine Turtle Permit 20–081), and North Carolina State University (North Carolina Endangered Species Permit 20ST42). Organs were removed intact and as much blood as possible drained from the tissues prior to weighing. To weigh the heart, the great vessels were severed at the pericardium. Due to the difficulties separating fat and muscle, we did not attempt to estimate the fat/bone and muscle compartments during dissections.

To estimate blood volumes, we used a value midway between those measured in leatherback and green turtles (Thorson, 1968; Lutcavage et al., 1992; Table 3). Blood was separated into arterial and venous, respectively, 33 and 67% of the total blood volume (Fahlman et al., 2006). After the arterial blood gases were exchanged with the tissue, the venous side from all compartments joined immediately to form mixed venous blood, as previously detailed (Fahlman et al., 2006, 2009).

As we observed higher variability between individuals than between species and proportions in compartment sizes among species were similar, we combined the results from all the species to characterize compartment sizes for a “generic” sea turtle (Table 3). This turtle was composed of 7.00% blood, 0.06% brain, 30.00% fat/bone, 9.00% central circulation, and 53.94% muscle.

We defined cardiac output (Qtot⋅) at the surface (i.e., depth = 0) based on measurements from green turtles (body mass = 1.2 kg; West et al., 1992). To account for allometric differences between species in body mass, we used the following equation:

where 8.4 was the Qtot⋅(l • min⁻¹) and sQtot⋅is the mass-specific Qtot⋅ (Davis and Kanatous, 1999). At the surface, the model transferred 60% of Qtot⋅to the central circulation compartment, 34% to the muscle compartment, 4% to the brain compartment, and 2% to the fat and bone compartment. As no empirical data on this exist for sea turtles, we chose these values based on comparable data from marine mammal studies (Fahlman et al., 2018). While diving, the distribution of blood flow was tested iteratively to optimize O₂ utilization and thus maximize the aerobic dive duration (Fahlman et al., 2006, 2009). This resulted in 92.8, 5.0, 2.0, and 0.2% of sQtot⋅ being directed to the central circulation, muscle, brain, and fat and bone, respectively, while diving.

When at the surface, we used a metabolic rate of 4.23 ml O₂ • min⁻¹ • kg⁻¹ for each compartment. This is comparable to those reported in the literature for exercising leatherback, loggerhead, and green turtles on land under air temperatures between 26 and 30°C (Jackson and Prange, 1979; Butler et al., 1984; Lutcavage et al., 1987, 1990, 1992; West et al., 1992). We used values for exercising MR instead of resting MR, as cardiac output is often higher at the surface to recover from previous dives (Southwood et al., 1999). When diving, we assumed that the metabolic rate was 5% of that at the surface, i.e., 0.21 ml O₂ • min⁻¹ • kg⁻¹ based on data from a leatherback turtle conducting a 34-min dive (Southwood et al., 1999). This value also lies within the range of metabolic rates reported by Jackson and Prange (1979), Butler et al. (1984), Lutcavage et al. (1987, 1990, 1992), and West et al. (1992), and it also ensured that the modeled turtles did not run out of O₂ during long dives.

We used published data on blood packed cell volume (PCV, %), hemoglobin ([Hb]), and muscle myoglobin ([Mb], Table 2) concentrations for each species (Lapennas and Lutz, 1982; Lutz and Bentley, 1985; Lutcavage et al., 1990). A P₅₀ of the O₂ dissociation curve for each species was gleaned from published literature (Table 3). We used a Hill coefficient of 2.7 for all species (Berkson, 1967; Wood et al., 1984; Lutcavage et al., 1990).

We used O₂, CO₂, and N₂ gas balance equations from Fahlman et al. (2006) to estimate gas transport in the lungs and tissues over short-time intervals and small blood volumes (package volume). For these equations, the partial pressure/tension of each gas was used as the driving force for gas exchange, and each compartment was modeled as a single homogenous tissue with a one-to-one perfusion rate and tissue gas solubility (Fahlman et al., 2006).

We ran several model variations on each of the turtle datasets to test the effects of varying: (1) body-condition and (2a) the surface sQtot⋅and (2b) diving sQtot⋅ on venous PN₂

(1) To assess how changes in body-fat affect venous PN₂, we varied the relative volume of the fat/bone compartment in the control model (30%) to either 20 or 40% to represent extremes of nutritional condition. When changing the relative size of the fat/bone compartment, the contribution of the other compartments was adjusted accordingly to ensure that they maintained the same overall mass (Table 3)

(2a and b) To assess the effect of varying the sQtot⋅, we adjusted the surface sQtot⋅from the control model (sQtot⋅of 7 ml • min⁻¹ • kg⁻¹) to either 2.5 ml • min⁻¹ • kg⁻¹ (Surface-Low) or 10 ml • min⁻¹ • kg⁻¹ (Surface-High). We also examined the effect of varying the sQtot⋅during diving from the control model (5.0% of the control sQtot⋅) to either 10.0% (dive-low) or 3.3% (dive-high).

Finally, to evaluate how changes in body-fat or cardiac output affected the risk of GE, we calculated the end dive PN₂ saturation of the mixed venous blood during the last 5 s of each dive. We then measured the percentage change between the mean (over all dives) and the maximum end dive PN₂ for the different models using the formula:

While there was no published data available on arterial PN₂ levels in free-diving sea turtles, Berkson (1967) measured arterial PN₂ in green turtles during forced submergence. Thus, to assess the accuracy of our model for estimated arterial PN2, we compared the modeled results to those observed in Berkson (1967) during comparable dives.

As these are modeling based on the physiological assumptions from a limited number of individuals, we opted to only perform limited statistical analyses on the results. As such, these results should be considered largely descriptive and are simply to provide a framework for developing testable hypotheses in future studies.

When comparing data on arterial PN₂ during forced submergence from Berkson (1967) to the model output, the modeled data were never more than 1.5 ATA from the measured data (Figure 1). Furthermore, the modeled data strongly reflected the observed changes in arterial PN₂ at different pressures.

The model output showed that the accumulation of N₂ into the central circulation occurred faster than it did in the fat (Figure 2). Because of this, PN₂ within the central circulation compartment tended to reflect the ambient pressure, while fat PN₂ was more influenced by the dive profiles over a series of hours to days (Figure 3). As such, central circulation only briefly contributes to GE risk during the ascent as N₂ was rapidly removed and the supersaturation decreased. The fat compartment, on the other hand, slowly varied with time and depended on the previous diving pattern. Furthermore, unlike the central circulation, supersaturation occured close to the surface and during the entire surface interval. Nevertheless, the overall fat PN₂ saturation seldom exceeded 1 ATA and therefore does not contribute appreciably to GE risk.

Using the control model and combining the results from all three species, mean end-dive PN₂ increased with depth for central circulation (χ² = 32, 2 df, p < 0.01), muscle (χ² = 22, 2 df, p < 0.01), brain (χ² = 20, 2 df, p < 0.01), and mixed venous (χ² = 34, 2 df, p < 0.01) but not for the fat compartment (χ² = 2.4, 2 df, p > 0.2). When comparing dives within the shallowest dive bin (<30 m) to the deepest dive bin (>90 m), mean end-dive PN₂ increased by 245% for central circulation, 11% for muscle, 189% for brain, and 221% for the mixed venous compartment. When the three species were considered separately, leatherback turtles had the lowest mean end-dive PN₂, with green turtles having intermediate values, and loggerhead turtles having the highest (χ² = 8.49, 2 df, p = 0.014; End-dive PN2 = 2.73 (±0.31)−0.56 (±0.28) × leatherback + 0.54 (±0.41) × loggerhead−0.63 (±0.30) medium−1.71 (±0.29) shallow).

With data from all three species combined, maximum end-dive PN₂ increased with depth for central circulation (χ² = 11.5, 2 df, p < 0.01), brain (χ² = 8.8, 2 df, p = 0.012), and mixed venous blood (χ² = 12.1, 2 df, p < 0.01), but decreased for fat (χ² = 7.9, 2 df, p = 0.019), and did not change for muscle (χ² = 5.1, 2 df, p = 0.079). Comparing dives in the shallow depth range to those in the deep depth bin, maximum end-dive PN₂ increased by 80% for central circulation, by 68% for brain, and by 81% for mixed venous blood. For brain, both depth range and species helped to explain the variation (χ² = 7.39, 2 df, p = 0.025). Overall, mean end-dive PN₂ for brain increased with depth and leatherback turtles had the lowest maximum end-dive PN₂, while green turtles had intermediate values, and loggerhead turtles had the highest (change in end-dive brain PN₂ (%) = 4.23 (±0.65) − 0.50 (±0.66) × leatherback + 1.95 (±0.92) × loggerhead – 1.03 (±0.54) medium −1.86 (±0.53)). The maximum end-dive PN2 for mixed venous blood was 9.40 ATA, 6.17 ATA, and 5.03 ATA for leatherback, green, and loggerhead turtles, respectively, representing supersaturation ratios [(mixed venous PN₂ – ambient pressure) • ambient pressure⁻¹, or M-ratios] of 12.7, 8.3, and 6.8.

When body fat was increased from 30 to 40% of body mass or decreased from 30 to 20% of body mass, there was no effect observed on mean or maximum end-dive mixed venous PN₂ for either shallow or intermediate depths when all three species were combined (Figure 5). However, for the deepest depth bin, increasing or decreasing body-fat reduced the mean end-dive mixed venous PN₂ (χ² > 10.0, 2 df, p < 0.01; Figures 4A,B) and the maximum end-dive mixed venous PN₂ (Figures 5A,B). Thus, a model that included depth, species, and body-fat was warranted; % change in maximal end-dive mixed venous PN₂ = −13.3 (±1.2) − 12.4 (±2.6) leatherback – 5.2 (±3.8) logger + 12.5 (±3.0) medium depth + 11.0 (±2.9) shallow depth + 8.6 (±2.3) emaciated (χ² > 14.4, 1 df, p < 0.01).

When sQtot⋅was either increased or decreased, there were no consistent differences in mean mixed venous end-dive PN₂ among species at the shallow depth bin (p > 0.3 for all, Figures 4C,D). For both medium and deep depths, when sQtot⋅was increased to 10 ml • min⁻¹ • kg⁻¹ (Surface-High) this increased mean end-dive PN₂ by 10.6% (Figure 4D), while a decrease in sQtot⋅to 2.5 ml • min⁻¹ • kg⁻¹ (Surface-Low; Figure 4C) caused a 4.5% decrease in mean end-dive PN₂ [χ² > 26.5, 1 df, p < 0.01, change in end-dive PN₂ (%) = 10.6 (±1.9) – 15.1 (±2.7) × (Surface-Low)].

For maximum end-dive mixed venous PN₂, the changes were similar, with 43% increase in surface sQtot⋅(Surface-High) causing an increase in maximum end-dive PN₂ by 13.4% (Figure 5D), and 64% decrease (Surface-Low) in sQtot⋅causing a 5.4% decrease [Figure 5C, χ² > 31.0, 1 df, p < 0.01, change in end-dive PN₂ (%) = 13.4(±2.1) – 18.7 (±3.0) × (Surface-Low)].

When altering diving sQtot⋅, there was no statistically significant effect of dive depth on mean end-dive mixed venous PN₂ when all three species were combined (p > 0.5 for all, Figures 4E,F). Considering all species and depths together, an increase in diving sQtot⋅ from 5 to 10% lead to 20.6% increase in mean end-dive mixed venous PN₂ (Figure 5E)_. A decrease in diving sQtot⋅from 5 to 3.3% (Figure 4F) lead to a decrease of 13.2% in the mean end-dive mixed venous PN₂ [χ² > 52.8, 1 df, p < 0.01, change in end-dive PN₂ (%) = 20.6(±2.7) – 33.9 (±3.8) × (1/30)].

When varying diving sQtot⋅, there was uniform effect on the maximal end-dive PN₂, however, there were differences between species. Specifically, the maximum end-dive mixed venous PN₂ was higher for leatherback turtles than either loggerhead or green turtles at all depths. In addition, the lower diving sQtot⋅significantly reduced the maximal end-dive PN₂ (Figures 5E,F); [χ² > 6.21, 1 df, p = 0.013, change in maximal end-dive PN₂ (%) = 32.2 (±3.1) – 44.9 (±3.8) × (1/30) – 9.6 (±3.8) loggerhead/green].

Our goal was to construct the initial framework for a model to understand gas dynamics in diving sea turtles. As no such models previously existed, a logical first step was to adapt a previously constructed and validated model for another air-breathing vertebrate. While we chose to adapt a model originally designed for marine mammals (Fahlman et al., 2006, 2009, 2021), there are several key anatomical and physiological differences between marine mammals and sea turtles that have not yet been incorporated into the new model. Firstly, and of considerable importance, is that sea turtles, unlike marine mammals, are at least partially ectotherms. Specifically, even though several studies have shown that sea turtles can maintain body-temperature elevated above ambient conditions (Paladino et al., 1990; Sato et al., 1995), their body temperatures still fluctuate over wider ranges than marine mammals. While we attempted to control for this by only analyzing diving data for sea turtles within a restricted range of sea surface temperatures (19–29°C), the effect of temperature on both the metabolic rate and the gas tensions/solubilities cannot be ignored (Lutz et al., 1989).

Another important difference between marine mammals and sea turtles is that sea turtles have muscular sphincters within their pulmonary arteries and a partially compartmentalized ventricle that allow for central intracardiac shunting (García-Párraga et al., 2018a; Burggren et al., 2020). These features could allow a complete shunt and cessation of gas exchange as the animal begins the dive. In fact, this could be the mechanism by which sea turtles avoid GE (García-Párraga et al., 2018b; Fahlman et al., 2021). In this model, we assumed that the pulmonary shunt develops due to the passive compression of the terminal air spaces and the structural properties of the conducting airways and faveoli (Bostrom et al., 2008; Fahlman et al., 2009). While this is not necessarily the case in reptiles, the exact factors influencing the cardiac shunt in sea turtles are not well-established (Burggren et al., 2020). Without having a solid understanding of the functioning of the cardiac shunt, we adjusted the model so that the arterial PN₂ reflected those of forced-diving sea turtles in Berkson (1967). This meant that the physiological shunt and the potential anatomical shunt were expressed together as a pulmonary physiological shunt. In addition, we acknowledge that forced-diving experiments in Berkson (1967) would likely elevate blood and tissue N₂ levels beyond normal values, and our model likely thus provides a conservative estimate of blood and tissue N₂ levels. However, if we consider exercise as one of the main driving forces in the cardiac output adjustment (Okuyama et al., 2020), physically restrained animals could minimize N₂ uptake reflecting lower N₂ values compared to free swimming or entrapped individuals with restricted movement.

Other significant assumptions with the model resulted from differences between the datasets available for each species. For example, we utilized diving data from subadult loggerhead in foraging habitats, while the data for leatherback and green turtles were from adult females in inter-nesting habitats. Indeed, it is known that foraging and inter-nesting turtles exhibit differing dive behaviors (Robinson and Paladino, 2015). Other factors including life-stage and temperature are also known to influence diving behavior but are once again not standardized between our turtle diving datasets.

As is the case for all theoretical physiological models, the complexity of the entire system cannot be incorporated holistically into a mathematical model. While such arguments are often used to disparage the value of such exercises, we stress that we aimed to build the foundation upon which more detailed assessments of how factors, such as behavior, life-stage, or life-stage, can influence the risk of GE in sea turtles. As such, the goal of this exercise was to try and to identify which factors and assumptions need to be prioritized in future models to gain a more accurate understanding of gas dynamics in sea turtles.

In the basic model, the mean end-dive PN₂ differed between species and maximum depth of the previous dive. For shallow and deep dives, the end-dive PN₂ values were, respectively, between 38–111% and 193–342% higher as compared with the surface equilibrium PN₂. Interestingly, maximum PN₂ increased with dive depth in all turtle species. This suggests that the risk of GE increases with depth as observed in previous studies (Fahlman et al., 2017).

While we observed no effect of body-fat on end-dive mixed venous PN₂ at depths less than 90 m, significant differences were observed beyond 90 m. Interestingly, end-dive mixed venous PN₂ was highest at intermediate body-fat levels, and it dropped if body-fat was either increased to 40% or dropped to 20%. Moreover, these differences were up to 25% lower in leatherback turtles relative to loggerhead or green turtles. These findings suggest that the higher levels of fat in leatherback turtles could perhaps reduce the effects of GE and play a role in the ability of leatherbacks to dive several times deeper than any other sea turtle species (Robinson and Paladino, 2015). That said it is unclear why lower body-fat values also result in reduced end dive PN₂.

Past studies have shown that variation in blood flow, through changes in cardiac output, and the level of gas exchange are the physiological variables that have the greatest effect on N₂ uptake and removal (Fahlman et al., 2006, 2009; Hooker et al., 2009). Our model provided evidence that this may also be the case for sea turtles as a reduction of the diving sQtot⋅ to 3.3% caused a mean 13.2% decrease in end-dive PN₂, while increasing the diving sQtot⋅ to 10% caused a mean 20.6% increase in end-dive PN₂. Similar patterns were also observed for maximum end-dive PN₂. Nevertheless, these patterns differed by species and leatherback turtles exhibited lower mean end-dive PN₂ at both higher and lower diving sQtot⋅.

Our model suggests that during “natural” diving behavior sea turtles experience blood and tissue N₂ levels that would cause decompression sickness in land mammals of similar size. Indeed, the maximal end-dive PN₂ values between species ranged from 5.03 ATA to 9.40 ATA, which is considerably higher than the level of end-dive PN₂ that would result in severe DCS in 50% of similarly sized humans (Fahlman et al., 2020). As it is highly unlikely that sea turtles are perpetually suffering from GE during routine diving behavior, we propose that turtles, much like marine mammals, must have additional behavioral, anatomical, and/or physiological mechanisms to reduce N₂ uptake that are not currently considered in this model (García-Párraga et al., 2018a; Burggren et al., 2020; Fahlman et al., 2021).

One mechanism that sea turtles may employ to lower the risk of GE is the Selective Gas Exchange hypothesis (García-Párraga et al., 2018b; Fahlman et al., 2021). This hypothesis, which is supported by theoretical, anatomical, and physiological studies (Fahlman et al., 2009, 2018, 2020; Olson et al., 2010; Hodanbosi et al., 2016; García-Párraga et al., 2018a), suggests that if ventilation and perfusion in the lung can be managed, it would allow exchange of O₂ and CO₂, with little or no exchange of N₂ (García-Párraga et al., 2018b; Fahlman et al., 2021). In support of this, it has been shown that the pulmonary arterial sphincters contract and relax when exposed to parasympathetic and sympathetic neurotransmitters, respectively (García-Párraga et al., 2018a). If sea turtles exhibit Selective Gas Exchange, this would help to significantly reduce our modeled blood and tissue N₂ levels and help to explain how turtles avoid GE during natural dives. It would also explain how the sympathetic stress response could result in elevated pulmonary blood flow to the still ventilated lung, yielding excessive N₂ uptake and GE (Fahlman et al., 2021).

It is also possible that turtles couple physiological and behavioral mechanisms to further reduce the risk of GE. Indeed, sea turtles increase their heart rate, and likely cardiac output, during the ascent phase (Okuyama et al., 2020). In addition, several breath-hold diving vertebrates, including sea turtles, the beluga (Delphinapterus leucas; Martin and Smith, 1992), beaked whales (Hyperoodon ampullatus; Hooker and Baird, 1999), macaroni penguin (Eudyptes chrysolophus; Sato et al., 2004), and king penguins (Aptenodytes patagonicus; Sato et al., 2004) have been shown to reduce the ascent rate while approaching the surface. It has been suggested that this reduction may be an adaptation to reduce end-dive PN₂ before reaching the surface (Fahlman et al., 2006; Fossette et al., 2010). When the reduction in ascent rate is coupled with increases in heart rate during the ascent, it may reduce end-dive PN₂ by as much as 45% in marine mammals (Fahlman et al., 2006). A similar reduction in the supersaturation was seen in all the turtles in this study (Figure 4). It also appears that the tendency of sea turtles to conduct a short “pause” in their accent around 10–20 m, which we observed in all three species, may also help to reduce the supersaturation as the turtle returned to the surface (Fossette et al., 2010).

These adaptations to reduce PN₂ during normal dives could also explain why sea turtles are predisposed to GE when caught by fisheries. For example, rapid accent to the surface in a fishing net coupled with disruption of the normal dive profile and physiological response to underwater entrapment may alter cardiac output and shunting, leading to rapid GE. It also suggests that slower retrieval of the fishing nets or trawls may help to significantly reduce blood and tissue N₂ levels in bycatch sea turtles. If operationally possible, this may provide a simple and cost-effective strategy to minimize GE in turtle by-catch and could have the added benefit of reducing fish by-catch as well (Grothues et al., 2017). However, a significant concern with longer ascent times is increased risk of drowning, which may negate any benefits.

Our model provides a framework for future studies on the analysis of GE in sea turtles. While we adapted the model to test the effects of body-fat and cardiac output on GE, this model could be adapted in other beneficial ways. Indeed, incorporating a simulated capture event, whereby the speed of the ascent to the surface or the total time spent at depth is varied. The model could also be adapted to assess how variation in the cardiac shunt plays a role in GE during both routine dives and during incidental capture. The strong effect of cardiac output on GE also highlights that more information is needed on cardiac dynamics in diving sea turtles.