Blood samples (n = 654: 576 samples from pregnancy or false pregnancy; 4 placental samples obtained from cord blood; and 74 samples from physiologic events) were collected from 57 animals during the period from 1983 through 2017 (Robeck et al., 2021; Table 1). These samples were collected as part of routine monitoring of animals for detection of reproductive events (n = 92). Reproductive events were defined as periods when serum P4 concentrations were increased above 1 ng/ml for longer than the normal luteal phase length of 21 days (O’Brien and Robeck, 2012). Physiologic events (n = 59 events) were comprised of samples collected during the following periods: FOLLICULAR (n = 16 samples from 12 animals and 12 events); OVULATION (n = 24 samples from 18 animals and 24 events); and LUTEAL (n = 34 samples from 15 animals and 23 events). Animals were housed at SeaWorld Parks in Orlando, San Antonio and San Diego. Animals were housed in enclosures containing ≥ 850 m³ of either natural processed (San Diego) or manufactured salt-water (Orlando, San Antonio).

Blood samples (n = 654 samples) were collected voluntarily (n = 509) or with manual restraint (n = 145) from animals ranging from weekly for peri-post ovulatory monitoring or monthly as part of routine husbandry management. Of these, there were 4 placental samples during normal pregnancy that were opportunistically collected from cord blood. Although most samples were collected using behavioral procedures without restraint, some of the historical samples collected prior to Jan 1, 2000, were collected using manual restraint. The exact number of these banked samples within this time period that were collected using restraint is unknown, but because the restraint method may have influenced GC sample concentration, the samples were further categorized as being either pre or post January 1, 2000, for the statistical analysis. Additionally, all ovulation samples (n = 24 samples/events from 18 animals) were collected as part of artificial insemination procedures and under restraint, thus these samples were also categorized accordingly (Robeck et al., 2013). Samples were collected from the ventral tail fluke using a 21-gauge winged blood collection set. Blood was collected by either the veterinary technician or attending veterinarian on staff and into BD Vacutainers (Becton Dickinson, Franklin Lakes, NJ, United States) containing activated thrombin. The thrombin-coagulated blood was centrifuged at 1500 rpm for 10 min, and the serum was decanted and frozen at −80°C for further testing. Although sampling time was not recorded for every sample, routine blood samples were typically collected in the mornings (before 12:00 h) as per standard husbandry procedure.

Bottlenose dolphin sera was extracted for use in the T and A4 hormone assays. Sample extractions were conducted identically to past hormone studies in killer whales (O’Brien et al., 2017; Robeck et al., 2017). Briefly, 3 ml of diethyl ether (Acros Organics, ThermoFisher, Waltham, MA, United States) was added to 0.15 ml of serum and vortexed at 1800 rpm for 5 min. The samples were then placed into an ultra-low (−80°C) freezer for 20 min to allow the aqueous layer to freeze. After, the solvent layer was poured off into a borosilicate glass tube and evaporated under compressed nitrogen gas. Samples were reconstituted with 0.3 ml of extraction buffer (0.2 M phosphate buffered saline, pH 7.5) and stored frozen at −20°C until assay. Mean ± sem extraction efficiency for this process (see aforementioned publications for description) was 93.7 ± 1.9% (n = 56).

All hormone concentrations were expressed as ng hormone per ml serum.

Within each animal and samples collected during a reproductive event, the date of ovulation was determined to align samples for analysis (Robeck et al., 2021). For this process, ovulation was determined by either knowing (n = 82) or estimating the date (n = 10). An ovulation date was considered “known” when: ovulation was determined by taking the midpoint between when samples were baseline and when they first became elevated post-ovulation and only relying on this estimation when the maximum time between these two samples was four weeks or less; based on observed estrus followed by elevated P4; ultrasonographic detection of ovulation; or daily urinary hormone analysis. Estimated ovulation dates were only done for normal births and were determined by subtracting the mean gestation length (376 days) for bottlenose dolphins from the date of parturition (O’Brien and Robeck, 2012). For abnormal pregnancies, only samples with known ovulation dates were included in the study (Table 1). Once the ovulation date was determined, the reproductive event period was divided based on stage or month post-ovulation (MPO). For non-pregnant animals, the follicular phase (FOLLICULAR) included any samples collected from one through five days before the day of ovulation (OVULATION) and luteal phase (LUTEAL) samples were samples collected during peak P4 which occurs between ten and 18 days post ovulation (O’Brien and Robeck, 2012).

Stage time periods were based on “trimesters” of a normal pregnancy and divided as follows: EARLY (days 1 to day 120 post-ovulation), MID (days 121 to 240), or LATE (days 241 until parturition). Reproductive event categories were defined as follows: Normal pregnancy (NORM) – a live calf that lived longer than 30 days (n = 217 samples within 27 pregnancies); Failure to thrive (FTT) – a live calf that lived from two to 30 days (n = 108 samples within 17 pregnancies); Perinatal loss (PNL) – a calf born either dead or alive but died prior to 24 h post-partum and had gestated for at least 352 days (the minimum gestation in a bottlenose dolphin that has resulted in the birth of a normal calf, E. Jensen, unpublished data, see Robeck et al., 2021, n = 93 samples within 20 pregnancies); Early loss (EL) – an ultrasound diagnosed, pregnant female that either reabsorbed or passed the conceptus or fetal tissue prior to day 121 (n = 58 samples within 12 pregnancies); and False pregnancy (FP) – a female with elevated P4 longer than a normal luteal phase (> 21 days; O’Brien and Robeck, 2012) without any ultrasonographic evidence of pregnancy (uterine membranes, fluid or conceptus) or was not in the presence of a breeding age male (n = 100 samples within 16 FPs).

Unless stated otherwise, statistical analyses were performed using Stata statistical software (version 16; StataCorp LP, College Station, TX, United States). We initially compared gestation length, age and parity across status categories. These comparisons were made using a two-level restricted maximum likelihood (REML) linear mixed model (LLM) with the dependent variables for each analysis being gestation length, age and parity, and with status as a categorical fixed variable and animal id as a random effects (level 2) variable. A post hoc marginal mean comparison was then made across status categories using a Šidák correction. Additionally, because earlier work indicated that age and parity were highly correlated (O’Brien and Robeck, 2012), we analyzed our data set to determine if this trend continued by performing a pairwise comparison across all data between age and parity using a REML LMM with id set as the random variable. Degrees of freedom adjustments for the small sample size were performed using the Kenward–Roger approximation (Kenward and Roger, 1997).

Hormone concentration comparisons were only made between NORM and one of the other abnormal reproductive events (FTT, PNL, EL, AB, or FP) and the analysis repeated until each potential paired comparison was completed. To compare hormone concentrations during either different stages or months post-conception between NORM and one of the other reproductive events, we used a two (animal id) or three level random effects (pregnancy id) LMM REML regression model (Cnaan et al., 1997; West et al., 2015; Robeck et al., 2017). Two or three level random effects models were compared using the likelihood ratio test and three level models (pregnancies within each animal) were used only if they provided significant improvement over 2 level models using animal ID only (West et al., 2015). Degrees of freedom adjustments for the small sample size were performed using the Kenward–Roger approximation (Kenward and Roger, 1997). For the REML regression models, the dependent variable was hormone and the fixed effect variables (level 1) were status (NORM and one of the other reproductive events) and pregnancy time period (either stage or month), animal age, season and method. Season was defined as samples collected during winter (December through February), spring (March through May), summer (June through August) or fall (September through November). Animal age and season were included as covariates to control for the effect that these variables may have on the different hormone concentrations being evaluated (Steinman et al., 2016; O’Brien et al., 2017; Robeck et al., 2017). In addition to age being previously identified as influencing hormone concentrations in normal pregnancies (O’Brien and Robeck, 2012; Steinman et al., 2016), and despite no apparent differences between animal age within each status group (Table 1), previous research indicated that animal age and not parity was a significant variable associated with pregnancy loss (O’Brien and Robeck, 2012). Although parity could have been added as covariate, its collinearity with age made it inappropriate to include in the model with age and, again, based on previously published results, we decided that age was the more appropriate variable to include in our analysis. We also included a categorical variable “method” which divided sample collection based on date of collection between pre and post Jan 1, 2000. This was an attempt to control for potential differences that may have occurred between collection methods (restraint versus behavioral) and the influence these methods may have had on hormone concentrations. All final mixed effects models were checked for normality using quantile plots of the standard residuals. If quantile-quantile (qnorm) plots of standardized residuals exhibited non-normal distribution then data were log transformed or square root transformed as predicted by the Shapiro–Wilk test (Ladder command, STATA) until residuals were normalized. Paired comparisons of the dependent variable marginal (predicted) means within each reproductive event against normal for each time category were made at a significance of P < 0.05. If appropriate, multiple comparisons of marginal means were performed using Bonferroni corrections at P < 0.05. For text, tables and graphs any transformed data were first back-transformed, and then, all data were presented as marginal means with 95% confidence intervals (CI) unless noted otherwise.

A total of 92 reproductive events and 59 physiologic events (FOLLICULAR, OVULATION or LUTEAL) were identified within 57 females (Tables 1, 2). Some animals had more than one type of reproductive event and, as a result, the total number of animals in Table 1 is greater than the actual number of animals in the study. No significant differences were detected between the mean age or parity of the females within each category of reproductive events and the length of time for EL and FP were significantly reduced compared to each of the other reproductive events (Table 1, Robeck et al., 2021). As would be expected, animal age was significantly (F₁,₅₂ = 124, P < 0.0001) associated with animal parity with age increasing at 2.63 ± 0.24 years for each successive reproductive event.

All final mixed model statistics for each hormone by time group (stage or month post-conception [MPC]) are presented in Supplementary Tables 1, 2 and statistics for significant variables for analysis by either stage or MPC are listed in Supplementary Tables 3, 4. Significant post hoc marginal mean comparisons are presented within text. Based on standardized residual analysis, all final models used the square-root of T and the logs of A4, cortisol, E2 and EC to provide the best residual fit.

Longitudinal patterns of circulating T, cortisol and EC during normal pregnancies were similar to our prior results (Steinman et al., 2016). Our results demonstrated that A4 had a similar trend to T during normal pregnancies with MID and LATE stages higher compared to EARLY, peaked during MPCs 7 and 8 then decreased until parturition. Our results support past research using LCMS-MS that has shown that A4 increases in a quadratic fashion with concentrations similar in MID and LATE stage during normal pregnancy and concentrations for both stages elevated compared to EARLY (Legacki et al., 2020). Increases of A4 during pregnancy have also been observed in the killer whale (Robeck et al., 2017; Legacki et al., 2020) and the beluga (Legacki et al., 2020), as well as humans (see review in Kuijper et al., 2013).

The longitudinal gestational profile of A4 was visually similar to T. In the killer whale, the peak in A4 concentrations precedes the T peak (MPCs 13 and 14, respectively) and is suggested to be related to their metabolic relationship where A4 is converted to T and estrogens via aromatization in the ovary or placenta (Robeck et al., 2017). Similarly, in our study, peak A4 (MPCs 7 and 8) preceded peak T (MPC 8 and 9). Concentrations of A4 were higher than T during all stages of normal pregnancy despite the larger cross-reactivity with other androgens (in particular dihydrotestosterone) for our T EIA. The same was observed when analyzed by LCMS-MS (Legacki et al., 2020). Because estrogens increase in late gestation in the bottlenose dolphin (Steinman et al., 2016), it is possible that more A4 would need to be biologically available for conversion to these other reproductive hormones and could explain the higher concentrations of A4 compared to T (Steinman et al., 2016).

Elevated androgen concentrations during pregnancy have been reported in the bottlenose dolphin (Boggs et al., 2019; Galligan et al., 2020) as well as other cetacean species (Robeck et al., 2017 and Wasser et al., 2017 [killer whale]; Richard et al., 2017 and Legacki et al., 2020 [beluga]; Rolland et al., 2005 and Corkeron et al., 2017 [North Atlantic right whale]; Hunt et al., 2019 [humpback whale]). Whether the source of androgens during bottlenose dolphin pregnancy is maternal or fetal-placental derived is unknown. However, the lower androgen concentrations we observed during FP compared to normal pregnancy indicates that these increases in androgens are pregnancy specific. In the killer whale, comparison of maternal serum and cord blood (representative of the placenta) has demonstrated that T, but not A4, is significantly higher in placental versus maternal sources (Robeck et al., 2017). This alone could imply that the major source of T is the fetus or placenta, while A4 is maternally derived. However, hormones may accumulate on one side of the fetal-maternal barrier due to preferential binding with proteins (Silberzahn et al., 1984). For the bottlenose dolphin, we found similar concentrations of both placental T and A4 as compared to maternal sources during LATE pregnancy. Despite this, we do not know for certain if androgens concentrations are higher in maternal versus placental sources or whether they concentrate in certain areas of the fetal-maternal unit due to other reasons.

The present study confirmed our previous findings of increased cortisol during late pregnancy in the bottlenose dolphin during NORM (Steinman et al., 2016). Increases in circulating GC or excretory metabolite concentrations during pregnancy have been documented in killer whales (Robeck et al., 2017), North Atlantic right whales (Hunt et al., 2006; Corkeron et al., 2017), humpback whales (Hunt et al., 2019) and blue whales (Valenzuela-Molina et al., 2018). Increases in circulating or excreted GC measures during pregnancy have been reported in several other wildlife species, including New and Old World primates, spotted hyenas (Crocuta crocuta), African elephants (Loxodonta africana) and dugongs (see review in Edwards and Boonstra, 2018). Glucocorticoid increases during late gestation are essential for maturation of the fetus in preparation for survival outside the uterine environment (Fisher, 1986), especially respiratory system development and maturation. Additionally, GC increases are also expected as part of the cascade of events required for induction of parturition in the cow (Adams and Wagner, 1970) and elephants (Meyer et al., 2004). Thus, it is unsurprising to see elevations of circulating GCs during LATE pregnancy in the bottlenose dolphin.

Concentrations of E2 increased throughout pregnancy, and placental E2 and EC measurements were the highest compared to all other maternal estrogen sources and reproductive stages (Table 2). Estrogens play an important role in the initiation and regulation of labor (Gibb et al., 2006). Thus, the late stage increases in estrogens that we observed may also be important for these processes. However, we did not have many samples within the immediate peri-parturient window that may have revealed more information regarding estrogen dynamics and its relationship with parturition. Our results in the present study with respect to EC also corroborated our previous findings (Steinman et al., 2016). We further were able to determine that the placenta is a major source of estrogens in bottlenose dolphin pregnancy. In primates, estrogens are necessary for stimulation of vascular endometrial growth factor and blood vessel growth in the placenta (Albrecht and Pepe, 2010). Hence, it is possible that the placental estrogens play a similar role in cetaceans.

Testosterone concentrations and patterns were similar between FTT and normal pregnancy (Tables 2, 3 and Figures 1, 2), but within A4 measurements, concentrations during EARLY and LATE were higher during FTT pregnancies. In humans, the influence of androgens on birth weight is more pronounced during the second trimester, where observations of elevated T during week 17 of gestation are associated with intrauterine growth restriction (Carlsen et al., 2006) and elevated A4 during the first trimester of pregnancy is associated with pre-eclampsia (pregnancy induced hypertension, see review in Kuijper et al., 2013). However, spontaneous pre-eclampsia is believed to be limited to human and non-human primate pregnancies and does not occur in other mammals (Berkane et al., 2017). Whether the differences in A4 concentrations we observed were an indicator of a potential problem or a result of chance remains to be determined. Evidence suggests that the majority (71%) of calf deaths after the first 24 h are related to failure of passive transfer and malnutrition, both of which are typically associated with poor or no nursing (Sweeney et al., 2010). Although the inability of a calf to nurse is multifactorial and can be influenced by dam abnormalities (e.g., due to illness or inability to aid the calf in learning to nurse), physically weak or immature calves due to gestational growth restriction (Osborn et al., 2012; Robeck et al., 2012) could potentially fall into this abnormal reproductive outcome group.

Cortisol was increased during all stages of FTT pregnancies, with the marginal mean concentrations two to three-fold higher than normal pregnancy. In wild killer whales, fecal GC metabolites are increased in unsuccessful pregnancies compared to successful ones (Wasser et al., 2017). Cortisol increases during the LATE stage of pregnancy play a role in maturation of the fetus for its survival outside the uterine environment (Fisher, 1986). However, excessive cortisol may be detrimental. In humans, increased cortisol during pregnancy is associated with early onset of labor and low birth weight (Austin and Leader, 2000). In the present study, there was no difference in the gestation lengths of FTT pregnancies compared to normal pregnancies (Table 1) suggesting that premature labor was likely not, or only minimally, a contributing factor to neonatal survival rates in the bottlenose dolphin. Birth weight and length are typically not determined in bottlenose dolphin calves so post-natal weight and length are only estimates, thus, it is unknown if there is an association with low birth weight and FTT pregnancies. Increased maternal cortisol in human pregnancy has also been shown to affect infant cognitive development suggesting that excess cortisol may have a programing influence on fetal development (Davis and Sandman, 2010). The placental enzyme 11β-hydroxysteroid dehydrogenase type 2 (11β-HSD2), regulates cortisol influence on the fetus by converting it to its inactive form, cortisone (Beitens et al., 1973). This enzyme increases during pregnancy then decreases toward the end of gestation to permit passage of more cortisol to the fetus to aid in late term organ development (Austin and Leader, 2000). However, 11β-HSD2 is only a partial barrier, and fetal cortisol measures are closely correlated with maternal concentrations (Gitau et al., 1998, 2001); consequently, excess maternal cortisol could have detrimental effects on the developing fetus. Investigations of 11β-HSD2 during pregnancy is limited to humans and rats (see review in Edwards and Boonstra, 2018), and any gestational effects of 11β-HSD2 regulation on other mammalian and wildlife species is unknown. Nonetheless, it is possible that the excess maternal cortisol we observed during FTT pregnancies had a negative influence on the developing fetus that did not affect the length of gestation but instead manifested negatively in neonate health. In laboratory rodent models, maternal stress negatively impacts both offspring growth and behavior (Meaney et al., 2007) while in snowshoe hares (Lepus americanus), elevated maternal fecal GC metabolites due to heightened predation risk results in lower reproductive output (less offspring) as well as lower quality offspring (Sheriff et al., 2009). In meerkats who have a cooperative breeding-social hierarchy structure, subordinate females who have been evicted from their social group while pregnant demonstrate elevated fecal GC metabolites. These subordinate females have lower reproductive function and are more likely to abort pregnancies (Young et al., 2006). As mentioned before with androgens, it is difficult to determine if poor nursing is due to maternal or calf causes, or a combination thereof. However, possible detrimental cognitive developmental effects on fetal programing associated with elevated maternal cortisol (Davis and Sandman, 2010) could explain poor nursing behavior on part of the calf.

For this study, we had a sufficient number of pregnancies and reproductive outcomes to be able to separate FTT pregnancies into its own category. Although there could be many different causes for neonatal mortality in the bottlenose dolphin, cortisol and possibly A4 analysis may be able to identify pregnancies with calves that are at risk. This information could improve calf survival by increasing post-natal monitoring of at risk calves to allow for rapid medical intervention if necessary. Furthermore, in the larger context of analyzing causes of reproductive failure in the bottlenose dolphin on a population level (e.g., Deepwater Horizon oil spill, Kellar et al., 2017), recognition of possible FTT pregnancies via hormone analysis may provide more insight into the possibility of differing effects of the environment or anthropogenic influences, both short and long term, on bottlenose dolphin pregnancy. The ability to identify at risk or compromised pregnancies over the short term can possibly be linked to long term reproductive failure and the overall health status of a population. Hence, when possible, reproductive studies should include FTT as a separate analytical category.

Perinatal loss included stillborn calves (calves born dead) as well as calves born live but died within 24 h. For bottlenose dolphins under human care, PNL rates of 5.2% and 11.5% have been reported (Sweeney et al., 2010; Robeck et al., 2021). During PNL in bottlenose dolphins, reduced concentrations of hormones have been reported for P4 (Bergfelt et al., 2011), relaxin (Bergfelt et al., 2017), total and free thyroxine (West et al., 2014) and PGs (Robeck et al., 2021). Across pregnancy, we found increased concentrations of cortisol and estrogens for PNL. When analyzed by stage, cortisol was four and three-fold higher during EARLY and LATE stages, respectively, but estrogens were not statistically different at the stage or MPC level. In pregnant ewes administered corticosteroids, a trend of increased circulating estrone has been demonstrated and indicates increased cortisol may stimulate placental production of estrogens (Keller-Wood et al., 2014). This may explain our observations of significant (P = 0.044 and 0.049 for E2 and EC, respectively) increased estrogens across PNL pregnancies.

In contrast with our findings, Kellar et al. (2017) has shown that blubber cortisol had no influence on the reproductive success rate on a population of bottlenose dolphins following the Deepwater Horizon event. However, that study only relied on single sample analysis and the timing of sample collection could have affected their results. Furthermore, cortisol concentrations may have been elevated for the population as a whole because of the effects of the oil spill, and discrete changes in cortisol concentrations may have been missed or dampened as a result. It has been reported that in pregnant ewes administered hydrocortisone to increase circulating maternal cortisol concentrations, there is a higher incidence of stillbirth and fetal death but no evidence of changes to uterine blood flow, placental hormone concentrations or birth weight (Keller-Wood et al., 2014). They suggest the relationship between poor reproductive outcome and increased maternal cortisol may be a result of cortisol metabolism by the dam and/or fetus (Keller-Wood et al., 2014). Additionally, in humans, late term fetal loss associated with elevated cortisol may result in changes to fetal cardiac function or size (Trainer, 2002). Our findings indicate there are potential negative influences of elevated cortisol outside of the normal pregnancy reference range on the feto-placental unit during pregnancy.

Like FTT, there was no difference in mean gestation length for PNL compared to normal pregnancy, so it is unlikely that increased cortisol resulted in premature labor for PNL in our study. We also assume that the major contributing factor in bottlenose dolphin stillbirth is dystocia or a prolonged labor (Robeck et al., 2021), and that dystocia may be associated with calf size (Robeck et al., 2012). Although dystocia is associated with increased cortisol at parturition in buffaloes (Sathya et al., 2007), we did not have many samples collected immediately prior to parturition (n = 3 and 5 for normal and PNL, respectively), so the elevated cortisol concentrations we observed during the late stage with respect to PNL were most likely not associated with any possible dystocia related influences. Continued efforts to collect and analyze samples in close temporal relationship to parturition would help answer some of these questions.

Like what has been reported with P4, where concentrations are increased above normal pregnancy concentrations during the EARLY stage in EL (Robeck et al., 2021), we also observed an increase above normal pregnancy for T. Hyperandrogenemia, especially during early pregnancy, is a contributor to recurrent pregnancy loss in humans (Okon et al., 1998), and elevated T and A4 are associated with preeclampsia, a condition in which the placenta may be small or damaged most likely due to abnormal blood vessel development in the placenta during early pregnancy (Troisi et al., 2003). Elevated androgens may be responsible for miscarriages due to implantational defects, possibly, because T can prevent endometrial estrogen-related gene expression (Kowalski et al., 2004). This may explain our observations of reduced estradiol in EL. Because androgens may be placental or fetal-derived, perhaps, the earlier than expected increase in T during EL, in combination with or separate from increased P4 during the same time, may be a signal of compromised placental or fetal health, e.g., chromosomal defects that can be maternally recognized. This maternal recognition may then lead to termination of a pregnancy with a high likelihood of a poor outcome. Causes of EL are largely unknown in the bottlenose dolphin, and in most instances, no placental or fetal tissue that can be examined for defects is expelled. Consequently, there is little that can be done to prevent this occurrence or examine it in detail.

The phenomenon of FP in cetaceans has not been well documented. Diagnosis of FP has relied on consistent (monthly or semi-monthly) monitoring of serum P4/PGs in combination with ovarian and uterine ultrasonography once elevations in these concentrations have exceeded the known luteal phase length for the species. To our knowledge, incidences of FP have been reported in three cetacean species: bottlenose dolphin (Sawyer-Steffan et al., 1983; Kirby and Ridgway, 1984; Yoshioka et al., 1986; Kirby, 1990; Robeck et al., 2021); killer whale (Robeck et al., 2018); and the false killer whale (Pseudorca crassidens, Robeck et al., 1994; Atkinson et al., 1999). Although FP has been previously reported in the bottlenose dolphin, the frequency of this occurrence is unknown. This highlights the importance of routine sampling and hormone monitoring with follow up ultrasonography in cases where the duration of P4/PG elevation exceeds the known luteal phase interval to provide more data and, hence, insight into this reproductive phenomenon.

Androgen and estrogen concentrations were remarkably different for FP compared to normal pregnancy with concentrations lower during MID and LATE stages and some corresponding MPCs for both T and A4. For the reproductive outcomes analyzed, hormone concentration differences from normal pregnancy were most evident for FP. The mean length of FP in the present study was 178 days, approximately 6 months. This is in agreement with past studies in bottlenose dolphins that have reported extended luteal phases/FPs lasting 5 to 6 months without evidence of pregnancy (Sawyer-Steffan et al., 1983; Kirby and Ridgway, 1984; Yoshioka et al., 1986; Kirby, 1990). As a result of this shortened length of FP compared to normal pregnancy, any influence of FP on androgen and estrogen concentrations would be expected to be minimal beyond this period, i.e., MID stage, after termination of the FP. Fecal androgen metabolite analysis in combination with pregnanediol glucuronide measures have been able to differentiate between pregnancy and FP in polar bears (Ursus maritimus, Stoops et al., 2012). In dogs, elevated E2 and reduced EC concentrations compared to pregnancy have been observed for FP (Chakraborty, 1987), but androgens were not distinguishable between the two states (Concannon and Castracane, 1985). However, unlike the bottlenose dolphin, in dogs, androgens peak in early pregnancy (within the first 20 days of a 60–65 days of gestation), so discrete changes between the two reproductive states may be missed during such a small window of time (Concannon and Castracane, 1985). The differences in androgens and estrogens we observed between normal pregnancy and FP indicates testing for these reproductive hormone classes may help distinguish between the two reproductive outcomes. Furthermore, because of the absence of a placenta and fetus in a FP, any androgen measurements during the FP would mainly be of CL or ovarian origin and supports the theory that the source of increased androgens we measured during actual pregnancies are most likely placental or fetal, and not maternally derived. When compared to FP, the estrogen increase observed during MID and LATE stages in normal pregnancy as well as placental concentrations suggests these increases are largely due to pregnancy and, as mentioned previously, are a good measure for distinguishing between the reproductive conditions.

Past work by our group has demonstrated that PG concentrations were better at differentiating between the two conditions compared to P4; during FP, PG concentrations were reduced compared to normal pregnancy by three months post-ovulation, whereas P4 was not until month nine (Robeck et al., 2021). Based on our past and present results, for pregnancy/false pregnancy confirmation, a hormone panel should include P4, PG, androgens and estrogens. The ability to discriminate between pregnancy and false pregnancy using a suite of hormone tests would be extremely useful for health assessments in wild populations where only a single sample may be collected and would decrease incorrect diagnoses of animals as pregnant. Pregnancy diagnosis based solely on P4 analysis is likely not able to detect differences between the two reproductive states.

Little to no seasonal effect was noted across all hormones and reproductive states. The exceptions within these hormones and reproductive states were found with testosterone (normal and FTT), estradiol (normal and false pregnancy) and marginally for cortisol (FTT). This supports previous studies that have also found little to no influence of season on delphinid hormone concentrations during and outside of pregnancy (St. Aubin et al., 1996; Steinman et al., 2016; Biancani et al., 2017; Robeck et al., 2017). We found no influence of age on reproductive outcome (Table 1). This was unexpected because earlier work in bottlenose dolphins demonstrated that animals older than 25 had higher incidences of failed pregnancies (AB) (O’Brien and Robeck, 2012). However, the previous study focused primarily on early loss (EL, < 120 days of pregnancy) and, therefore, the results and distribution of animals is not directly comparable to this study. Nonetheless, age did have some effect on hormone concentrations in the present study, including E2 and cortisol, but only during abnormal pregnancies. For cortisol, only concentrations during EL decreased with age. Additionally, and similar to our previous work (Steinman et al., 2016), we also found an association of age and estrogen concentrations whereby both E2 and EC decreased with age during PNL. However, in the killer whale, a closely related delphinid, no association of age and estrogen measurement during normal pregnancies has been found (Robeck et al., 2016). Whether the influence of age on estrogens is exclusively limited to PNL in the bottlenose dolphin or was an anomaly within our subjects remains to be determined and should be studied further. Apart from studies by our group in the bottlenose dolphin (O’Brien and Robeck, 2012; Steinman et al., 2016) and the killer whale (Robeck et al., 2016, 2017, 2018), the influence of age within pregnancy has not been reported. Accurate ages may be difficult to obtain in in situ settings but should be included in data analyses when available. Although bottlenose dolphins can give birth into their 40s (Dudley, 2008), most do not, and increased EL may be evidence of declining fertility with age or reproductive senescence. The various mechanisms for increased EL in the bottlenose dolphin are unknown but have been hypothesized to be related to a decrease in oocyte quality and numbers as has been observed in other mammalian species (O’Brien and Robeck, 2012; Agenor and Battacharya, 2015; Satué and Gardon, 2016; Cimadomo et al., 2018). Nonetheless, because population recruitment relies largely on the fitness of the dam, the influence of age on reproductive success needs to be evaluated. Because age and parity were correlated but parity was not associated with early loss in a previous study (O’Brien and Robeck, 2012) and because parity’s collinearity with age was not well-suited to our model, we did not include parity as covariate. As a result, although unlikely, parity may still have had an influence on hormone concentrations that went undetected.

Sample collection method (under restraint versus behavioral) had limited influence on hormone concentrations. Because it was impossible to know if samples collected prior to Jan 1, 2000, were collected with manual restraint or behavioral conditioning, despite our best efforts to uncover this information, we do not know for certain if all samples designated as manual restraint were actually collected under those circumstances. As a result, it is likely that some samples designated as restraint were collected behaviorally. Our intention with examining collection method as a variable was to determine if this could have influenced hormone concentrations. Our results indicated that cortisol concentrations increased under restraint during EL only, while A4 measurements were influenced during FP. The increase in A4 could be a result of androgen production by the adrenal glands, as has been observed in non-human primates (Möhle et al., 2002). Regardless, it appears that collection method had minimal effect on our results, but sample collection methods, if potentially stressful, should be included as an analytical variable, especially for in situ studies where animals are not habituated to restraint, but restraint is likely the only option for blood collection.