First evidence for a latitudinal body mass effect in extant Crocodylia and the relationships of their reproductive characters

Relationships between distribution patterns and body size have been documented in many endothermic taxa. However, the evidence for these trends in ectotherms generally is equivocal, and there have been no studies of effects in crocodylians specifically. Here, we examine the relationship between latitudinal distribution and body mass in 20 extant species of crocodylians, as well as the relationships between seven important reproductive variables. Using phylogenetically independent contrasts to inform generalized linear models, we provide the first evidence of a latitudinal effect on adult female body mass in crocodylians. In addition, we explore the relationships between reproductive variables including egg mass, hatchling mass and clutch size. We report no correlation between egg mass and clutch size, upholding previously reported within-species trends. We also find no evidence of a correlation between measures of latitudinal range and incubation temperature, contrasting with the trends found in turtles.

Crocodylians are a vital component of their ecosystems, often acting as keystone species in tropical regions (Ashton, 2010). They represent the last surviving pseudosuchian archosaurs, a clade that once inhabited every continent and that has persisted for at least 230 Myr (Irmis et al., 2013;Turner et al., 2017). Having survived the devastating Cretaceous-Palaeogene (K-Pg) mass extinction (Bronzati et al., 2015;Puértolas-Pascual et al., 2016), as well as a number of other, smaller or more protracted major extinction events (Hutchison, 1982;Markwick, 1993;Toljagíc & Butler, 2013;Mannion et al., 2015), crocodylians have demonstrated remarkable resilience to cataclysmic climate change and habitat loss (Toljagíc & Butler, 2013). Despite this, half of all extant crocodylian species are currently threatened with extinction and, at the current rate, vertebrate species loss will soon equal or even exceed that of the K-Pg event (Barnosky et al., 2011;Pievani, 2014;Lécuyer, 2018). To understand possible future trends in crocodylian diversity more fully, it is important that their current diversity, ecology and distribution are fully characterized (Fig. 1). Despite previous work in this area (Martin, 2007;Pierce et al., 2008;Nicolaï & Matzke, 2019), the nature of the relationship between geographical distribution and the reproductive biology of living crocodylians is unclear.
The resilience of crocodylians to historical mass extinctions is also much debated. Previous work has implicated many different factors in their survival, including diet (Sheehan & Hansen, 1986), aquaticism (Robertson et al., 2013), famine resistance (Robertson et al., 2013), induction of torpor at low air or water temperatures (Almandarz, 1975;Brisbin et al., 1982;Sun et al., 2019), and their propensity to burrow to take refuge from harsh environmental conditions (Thorbjarnarson, 1989;Mobaraki et al., 2015). Another hypothesis concerns their unique reproductive biology (Charruau et al., 2017). Crocodylians have no sex chromosomes and hatchling sex is thought to be fully controlled by incubation temperature. Temperaturedependent sex determination (TSD) systems have a threshold that yields an approximately 1:1 ratio of males and females at equilibrium (Escobedo-Galván et al., 2016). Higher temperatures produce a greater proportion of males, whereas lower temperatures produce more females (Deeming & Ferguson, 1989;González et al., 2019). Counterintuitively, the highest tolerable incubation temperatures produce mostly females, although these frequently fail to thrive (Marco et al., 2017). TSD was proposed as a possible buffer to extinction by Woodward & Murray (1993). Harsh environmental conditions are usually associated with a fall in ambient temperatures, and in species that utilize TSD these can result in a female-biased primary sex ratio (Tomillo et al., 2014;Carter et al., 2018). While lower temperatures may be consistent with the initial 'nuclear winter' effect of the end-Cretaceous bolide impact (Vellekoop et al. 2014(Vellekoop et al. , 2016, the longer-term warming resulting from increasing atmospheric greenhouse gas concentrations (Harrington, 2001;Turner, 2018) would be expected to skew sex ratios in favour of males. The latter scenario has been hypothesized as the most likely outcome of current trends in global warming (Miller et al., 2004). Although TSD has been well documented within crocodylian species (Ferguson & Joanen, 1982;Deeming, 2004;Piña et al., 2007), no comparisons of the effects of threshold temperature on development have been attempted across species, and little is known concerning the interrelationships between crocodylian reproductive characters.
The reproductive characters of crocodylians were first surveyed by Thorbjarnarson (1996). This work identified allometric relationships between egg mass,   clutch size, clutch mass and female length: patterns that had previously been observed in other vertebrate groups (Rohwer, 1988;Sinervo & Licht, 1991;Shine, 1992). However, the correlations and causal interactions between these reproductive variables were not the focus of this study.
Here, we used generalized linear models (GLMs) to test the relationships between the latitudinal distribution of extant crocodilian taxa and a variety of their important reproductive variables. We also tested whether there are significant relationships between those reproductive variables. In the context of these findings, we highlight some key questions concerning the reproductive biology, behaviour and social systems of wild crocodylians that could be addressed by future empirical and modelling work.

Data collection
We compiled data from the literature for seven morphological, environmental and reproductive variables across 24 extant crocodilian species (Supporting Information Appendix S1). We used the species list in Grigg & Kirschner (2015) (Fig. 1) and included one recently described species (Mecistops leptorhynchus) (Shirley 2018). Recent taxonomic uncertaintly makes the attribution of some reproductive data unclear, and two species (Crocodylus suchus and Osteolaemus osborni) were therefore excluded (Shirley et al., 2018;Hallmann and Griebeller, 2018;Isberg et al., 2019). The data compiled were: 1. Maximum latitudinal range (°) 2. Midpoint of latitudinal range (°) 3. Mean mass of breeding-age female (kg) 4. Mean fresh egg weight (g) 5. Mean hatchling mass (g) 6. Threshold incubation temperature, or the temperature at which a clutch will produce ~50% males and 50% females (°C). The lower (female to male) threshold was selected in this case because females are often inviable at the higher (male to female) threshold. 7. Mean clutch size (number of eggs laid) 8. Mean duration of incubation (days) We reviewed the literature in Google Scholar by searching for each species individually, coupled with appropriate terms (e.g. 'Crocodylus porosus clutch size', 'Caiman yacare body mass'). Variations on these were used as necessary (e.g. 'Caiman yacare'/'Caiman jacare' + 'body mass'/'female mass'/'weight'/'kg'). These were searched until pages returned only irrelevant results. Data were collated into a single spreadsheet (Supporting Information Appendix S1). Continuous variables (female mass, egg mass, hatchling mass, clutch size and duration of incubation) were averaged to smooth over outlier results that were unrepresentative. The literature contains data from individual captive-bred and wild-caught specimens, as well as values that are themselves averages from various heterogeneous samples. This constrains the precision of our data, but we believe that there is a homogeneous distribution of error. In addition, data on female mass were often derived from adults of different ages, and these were then averaged across multiple individuals and studies (Appendix S1). Data on latitudinal range were taken from the IUCN Red List of threatened species (IUCN, 2019).
For one species, Tomistoma schlegelii, there was insufficient information in the literature, and we therefore used data from our own observations. Hatchling mass data (see Supporting Information Appendix S1) were averaged from a captive-born clutch laid at 'Crocodiles of the World', Oxfordshire, UK. This clutch consisted of 17 eggs, of which 14 successfully hatched (although one infant failed to thrive and died shortly after hatching). Eggs were laid on 16 April 2016 by an imported farm-raised female measuring 2.3 m (snout-vent length). Eggs were separated and artificially incubated at 32.6 °C (three eggs), 32.0 °C (five eggs), 31.5 °C (five eggs) and 29.8 °C (four eggs), and hatched after 92, 94, 95 and 103 days, respectively. Hatchlings were weighed using a Marsden Super-SS B-100 waterproof electronic scale, accurate to 0.10 g.
5. Threshold incubation temperature is positively correlated with latitudinal midpoint, maximum latitudinal range and/or hatchling mass, as in turtles (Ewert et al., 2005).
The phylogeny of O' (Brien et al. 2019) was used to produce phylogenetically independent contrast (pic) values for each variable. These pic values (rather than our raw data) were used in the subsequent GLM analyses. This tree was constructed using multiple specimens representing each species and the tree was first pruned to remove duplicate  , leaving 20 species in the final analyses. We used the Akaike information criterion (AIC) to identify the minimum adequate model by backward stepwise deletion of poorly fitting variables. All models used the 'Gaussian' error family and 'identity' link function, and were implemented in the statistical software environment R, using the GGally, ape and phytools packages (version 3.5.3) (R Core Team, 2013).
We c o n s t r u c t e d f i v e G L M s b a s e d o n t h e phylogenetically independent contrasts of all variables (Table 1). Models 3 and 5 were produced from a modified phylogeny of 17 species, excluding Crocodylus novaeguineae, Osteolaemus tetraspis and Tomistoma schlegelii. This was because threshold incubation temperature data were not available for these species, and also because threshold incubation temperature was shown through our Spearman's rank test (Fig. 2) to correlate with hatchling mass (model 3), incubation duration and latitudinal values (model 5).

Female maSS
Model 1 was constructed to account for the potential effects of latitudinal midpoint, maximum latitudinal range, egg mass, hatchling mass, clutch size and duration of incubation on female mass (Supporting Information Appendix S2 -initial and final models). The final model (Table 2) was the result of backward stepwise deletion based on AIC value as a measure of model efficiency. This model explained 85% of the deviance associated with adult female mass (D 2 ). Mean clutch size (P = 0.0017) was the most significant correlator to log mean female mass, followed by mean latitudinal range (P = 0.0026), log mean hatchling mass (0.0027) and maximum latitudinal range (P = 0.0133). Log mean egg mass and mean incubation duration both correlated poorly with log mean female mass and were dropped from the model.

egg maSS
Model 2 was constructed to account for the potential effects of latitudinal midpoint, maximum latitudinal range, female mass and clutch size on egg mass (Supporting Information Appendix S2). The final model (Table 3) explained 68% of the deviance associated with egg mass (D 2 ). Log mean female mass (P < 0.0001) was the only

HatcHling maSS
Model 3 was constructed to account for the potential effects of latitudinal midpoint, maximum latitudinal range, female mass, egg mass, clutch size, threshold incubation temperature and incubation duration on hatchling mass (Appendix S2). The final model (Table 4) explained 90% of the deviance associated with hatchling mass (D 2 ). Log mean egg mass (P < 0.0001) was the only significant correlator to hatchling mass. Log mean female mass, latitudinal midpoint, mean clutch size and mean incubation duration correlated poorly with hatchling mass and were dropped from the model.

clutcH Size
Model 4 was constructed to account for the potential effects of latitudinal midpoint, maximum latitudinal range, female mass, egg mass and hatchling mass (Supporting Information Appendix S2). The final model (Table 5) explained 87% of the deviance associated with clutch size (D 2 ). The most significant correlators to mean clutch size were log mean female mass (P < 0.0001) and maximum latitudinal range (P < 0.0001), followed by latitudinal midpoint (P = 0.0007) and mean duration of incubation (P = 0.1071). Egg mass and hatchling mass were found to correlate poorly with clutch size, and were dropped from the model.

tHreSHolD incubation temperature
Model 5 was constructed to account for the potential effects of latitudinal midpoint, maximum latitudinal range, female mass, egg mass, hatchling mass, clutch size and incubation duration on threshold incubation temperature (Supporting Information Appendix S2). The final model ( Table 6) explained 42% of the deviance associated with threshold incubation temperature (D 2 ). The only significant correlator to threshold   incubation temperature was mean incubation duration (P = 0.0065). Log mean female mass, log mean egg mass, log mean hatchling mass, mean clutch size, latitudinal midpoint and maximum latitudinal range all correlated poorly with threshold incubation temperature and were dropped from the model.

general overview
A robust, positive correlation between latitude and body size has been well documented in endotherms (Gillman et al., 2009;Torres-Romero et al., 2016), but is much less compelling among ectotherms (Ashton & Feldman, 2003;Cruz et al., 2005;Lindeman, 2008), and has never been tested for in crocodylians. Here, we demonstrate that, despite low overall correlation coefficients, both latitudinal midpoint and maximum latitudinal range correlate significantly with adult female mass once the effects of phylogeny and other potentially confounding reproductive variables are controlled for (P = 0.0026 and 0.0133 respectively, see Model 1). Furthermore, we describe significant relationships between some important morphological reproductive variables across crocodylians as a whole, a phenomenon that has previously been described only within species (Verdade, 2001;Campos et al., 2008;Murray et al., 2013). We demonstrate that these relationships are discrete with, for example, a significant effect of female mass on egg mass, and of egg mass on hatchling mass, but no direct effect of female mass on hatchling mass. Female mass is a significant predictor of clutch size in crocodylians, as well as both the latitudinal midpoint and maximum latitudinal range. We did not detect any significant relationship between threshold incubation temperature and either latitudinal midpoint or maximum latitudinal range (see Model 5). Some of the most biologically significant relationships described by our models are presented in Figure 3.

implicationS oF reSultS
The strongest associations in our study are between morphological reproductive variables (Models 1, 2 and 3). These include adult female mass, fresh egg weight, hatchling mass and clutch size (Fig. 3). Intraspecific relationships of this kind have previously been documented in the American alligator (Alligator mississippiensis) (Murray et al., 2013) and in the broad-snouted, spectacled and Yacare caimans (Caiman latirostris, Caiman crocodilus and Caiman yacare) (Verdade, 2001;Larriera et al., 2004;Campos et al., 2008). Here, we verify the same trends across extant crocodylians as a whole. We report a significant positive effect of female mass upon clutch size (Model 4), a result that contrasts with the findings of previous species-level studies (Verdade, 2001). We find no evidence for any correlation between egg mass and clutch size, which is a biological trade-off that has been described in many other taxa (Lack, 1967;Sinervo & Licht, 1991;Rowe, 1994). We found no significant relationship between threshold incubation temperature and hatchling mass (Models 3 and 5), suggesting that the temperature required to produce an even sex ratio is not significantly different in crocodylian species that produce large hatchlings compared with those that produce small hatchlings. This result is consistent with previous studies that examined trends within species (Joanen & McNease, 1989;Campos, 1993). The effects of contemporary climatic warming on the reproductive biology of crocodylians have yet to be examined (Zhang et al., 2009;Simoncini et al., 2014), but based on deeptime associations between crocodylian distributions   (Markwick, 1998;Quintero & Wiens, 2013;Mannion et al., 2015), it is plausible that all crocodylians are vulnerable to rapid warming, and that this should be considered a threat to local populations (Dudgeon, 2014). Our results suggest that the only significant correlator to threshold incubation temperature in crocodylians is incubation duration (Model 5). This factor may help to explain the diversity of threshold incubation temperatures as, within species, eggs incubated at higher temperatures tend to hatch sooner than those incubated at lower temperatures (see section on Tomistoma schlegelii, Methods).
We identified no significant relationships between the threshold incubation temperature and either the mean or the maximum latitudinal range of crocodylians (Model 5). This suggests that the nest temperature of crocodylians is affected by factors other than latitudinal distribution. This stands in contrast to the relationship observed in turtles, in which latitude has a significant effect on reproductive traits and nest temperature (Iverson et al., 1993;Ewert et al., 2005). The clutch size of turtles correlates with latitude (Iverson et al., 1993) and their sex ratio is affected by annual temperature fluctuations (Janzen, 1994). This supports observed differences in the nest functions and parental roles of turtles and crocodylians, and suggests that the presence of adult crocodylians at the nest plays a much greater role in the temperature regulation of their nests than in turtles. Our results suggest that other factors (possibly the position, construction, location, shape or size of the nest) could contribute to the maintenance of an acceptable nest temperature (see also Grigg, 1987;Brazaitis & Wantanabe, 2011). In turtles, nest temperature is determined by ambient temperature and climate to a much higher degree (Hays et al., 2010;Refsnider et al., 2013;. Future research on the behaviour of nesting crocodylians may be necessary to shed further light on nest temperature regulation.

limitationS anD Future work
We note that some sources in the literature measured female mass at first laying, while others measured older females. Because crocodylians grow consistently after sexual maturity (Seymour et al., 2013), this could produce variable results. While our data were means from all of the published studies we were able to identify, we recognize the limitations of literature-based data collection, and acknowledge that the standardized collection of primary data would be preferable. This use of species mean values for mass, latitudinal distribution and clutch size also masks both population-level and biogeographical variation (Simoncini et al., 2009). Gathering these additional data would offer greater power, and enable both within-and cross-species comparisons. shown are between latitudinal midpoint and log mean female mass (P = 0.0026); maximum latitudinal range and log mean female mass (P = 0.0133); log mean egg mass and log mean female mass (P < 0.0001); log mean egg mass and log mean hatchling mass (P < 0.0001); log mean female mass and mean clutch size (P < 0.0001); maximum latitudinal range and mean clutch size (P < 0.0001); latitudinal midpoint and mean clutch size (P = 0.0007); and mean incubation duration and threshold incubation temperature (P = 0.0065).
An extension of the present study could include additional parameters such as nest dimensions, nest humidity, geographical range and post-copulatory reproductive behaviours (nest defence, guarding of hatchlings, etc.) to test the influence of these on crocodylian distribution and behaviour.
Mortality rates among unguarded eggs and hatchlings have been suggested as factors that may promote selection for parental care (Klug & Bonsall, 2010;Klug et al., 2012). Because there seems to be a wide within-species range of parenting behaviours in crocodylians (Hunt & Watanabe, 1982;Platt et al., 2008), the group is a good model system for investigating the effects of different types of care. For example, do female crocodylians living in high-stress environments (at the edge of their range or in humandominated landscapes) exhibit the same rates and types of parenting behaviour as those living in highly protected habitats? This represents an interesting subject for future research, especially as the habitats of crocodylians are increasingly subject to human encroachment and climatic stress (Langley, 2005;Amarasinghe et al., 2015;Corvera et al., 2017).

concluSionS
We present the first evidence for a latitudinal effect on the body size of extant crocodylian species and make the novel observation that major reproductive characteristics of extant crocodylians follow a consistent pattern of effect across the entire order.
We report no significant relationship between either latitudinal midpoint or maximum latitudinal range and threshold incubation temperature across sampled crocodylians. This contrasts markedly with the patterns seen in turtles and raises additional questions about the parental roles of crocodylians. Further work will be needed to clarify the possible responses of crocodilian reproduction and its impact on their resiliance in relation to anthropogenic warming. This may have implications for future studies on the effects of climate, latitude and life history on ectothermic amniotes, and for conservationists and government departments responsible for legislating on wildlife protection and climate change mitigation strategies.