Hibernal Phenology of the Eastern Box Turtle, Terrapene carolina carolina

The eastern box turtle, Terrapene carolina carolina, is a terrestrial ectotherm that is vulnerable due to sustained population declines across its range in the Eastern United States. Where this species uses managed fields, conservation measures could be implemented to restrict field mowing until the local population enters into hibernation, and such mowing restrictions could be adjusted each year only if the precise timing of the entry into and emergence from hibernation could be predicted based on proximate environmental conditions. I monitored twenty turtles each winter over two years to determine movement and activity patterns at the William Floyd Estate, a management unit of Fire Island National Seashore in New York, USA. My objectives were to (1) identify environmental variables correlated with the timing of entry into and emergence from hibernation in order to inform local conservation measures relating to the timing of mowing and brush clearing activities, and (2) investigate whether physiological condition explained the broad variation in the timing of individual animals’ responses to the same environmental conditions. I determined the timing of entry into and emergence from hibernation using a combination of light sensors and temperature dataloggers and sampled blood from turtles in the second year before and after hibernation in order to measure plasma biochemical profiles. Cooler air temperatures were correlated with increased probability of entry into hibernation for any given week in the fall, while warmer air temperatures increased the probability of emergence from hibernation for any given week in the spring. Physiological condition was correlated with the number of days until immergence into hibernation in the fall. These findings indicate that environmental conditions are proximate cues that trigger the timing of hibernation and emergence in turtles, and that physiological condition may mediate these triggers by limiting or forcing responses. Based on these results, managers would benefit from using environmental monitoring to adjust the timing of mowing and brush clearing activities in order to take advantage of longer hibernation times during more severe winters or to limit risk to turtles during warmer falls or earlier onset of warming temperatures in the spring.

Claussen 1990), lethal cold exposure remains a significant cause of episodic mortality for hibernating adults Schwartz and Schwartz 1974). In addition, the warmer winters predicted under various climate change scenarios may decrease growth in hatchlings ) and result in increased metabolism and depletion of energy reserves prior to spring emergence, resulting in starvation and death .
The IUCN currently lists T. carolina as "vulnerable" due to population declines across its range exceeding 30% over the last 50 years (van Dijk 2013). These declines are thought to have been caused primarily by habitat destruction and fragmentation Budischak et al. 2006;, as well as other anthropogenic impacts, such as recreation (Belzer 1997) and changes in hydrology resulting from upstream water control (Hall et al. 1999). Because T.
carolina is a long-lived species exhibiting negligible senescence (Henry 2003) and low recruitment, adult-stage mortality significantly impacts population viability (Heppell 1998;. Anthropogenic causes of high levels of adult mortality include vehicular collisions ) and mowing and field clearing .
Mowing activities impact box turtle populations from direct mortality from the mower's blades and tires . For this reason, burrowing beneath the soil during winter may protect T. c. carolina from risk of mortality during mowing, and thus the timing of hibernation could be important for land managers attempting to conserve vulnerable populations.
The timing of hibernation in adult T. carolina has been shown to be correlated with environmental variables such as soil temperature and surface temperature, although there is substantial variation in turtle activity patterns between subspecies and even between populations, and several conflicting trends have been reported. For example, Grobman (1990) found that T. c. triunguis held in outdoor enclosures in Missouri emerged after five consecutive days of subsurface temperatures of at least 7 ˚C. In contrast,  reported no correlation between body temperatures and entry into or exit from hibernation, although while entry into hibernation was not related to the first killing frost, exit from hibernation always preceded the last killing frost. Interestingly,  found no significant interannual variation in duration of hibernation.  reported entry dates for T. c. carolina of Oct. 14 -29 for 89% of turtles in Indiana, with decreasing depths of hibernacula and emergence occurring after the point of inversion between surface and deep soil temperatures at the end of February. In addition to these differing relationships between environmental variables and hibernal phenology, it has been hypothesized that soil moisture may affect timing of emergence , but this hypothesis has not been investigated in wild populations.
More precise information about the timing of the adult turtles' entry into and exit from hibernation will be beneficial to managers attempting to reconcile competing advantages and disadvantages of the timing and frequency of mowing activities, as well as contribute to our understanding of this significant yet poorly understood aspect of this species' life history. The objectives of this study were to (1) identify whether T.
c. carolina behavior during winter months placed them at risk of injury or mortality from mowing activities, and (2)  February.

Determination of timing of immergence
We chose to quantify the proportion of days per week that turtles remained in the soil in the fall because turtles did not cease above-ground activity immediately after first burrowing was observed. Instead, turtles seemed to gradually spend a greater number of consecutive days in soil refugia over a period of several weeks, with occasional forays above ground, until seeming to cease above ground movements several weeks after initial burrowing. Therefore, in order to capture this decreased activity, we quantified the proportion of days per week that turtles remained in the soil in the fall by first validating maximum daily light levels recorded by the geolocators in the second study period (fall 2015spring 2016). A total of 264 direct observations made over the course of this period were used to identify the maximum daily light level recorded when a turtle was confirmed to be in its refugium and not visible above the leaf litter to an observer. This maximum daily light level of a turtle in its refugium (8.8 lux) was substantially lower than the minimum daily light level recorded when a turtle was observed on the surface (61.6 lux). The conservative definition of the maximum daily light level observed during confirmed refugium occupancy was used as the classification threshold to ensure that no surface activity would be incorrectly classified as refugium use.
We used classification trees (randomForest library,  to build a supervised classification of shell temperatures (minimum, maximum, and mean day versus night temperatures and day length were used as variables; a total of 5000 trees were grown, and five variables were sampled at each node split) for each day for all turtles in the second study period for which we had geolocator data (N=19).
The resulting classification (3.84% error) was used to predict daily hibernation status using shell temperatures and day lengths for turtles in the first study period (fall 2014 spring 2015; n=17) and the one turtle in the second study period for which geolocator data were not available.
Full data for all turtles were available beginning 17 days after the fall equinox in 2015; thus, the proportion of days for which turtles were considered to remain below-ground during the week were calculated beginning on the 17 th day after the fall

Determination of timing of emergence
Date of exit from hibernation was determined as the earliest of either of the following: (1) basking temperatures recorded by the shell data logger (shell temperatures several degrees above surface soil temperatures; , or (2) direct observation of the turtle above the leaf litter and active. In contrast to the gradual decrease in activity observed in the fall, once turtles came to the surface in the spring, they usually dispersed from the overwintering burrow within three days and rarely returned to a soil burrow after the emergence. Therefore, for each turtle, we classified weeks prior to emergence as "hibernating" and weeks including and after the date of exit as "emerged". We considered the window of emergence to begin with the first week any turtle emerged and to end with the last week any turtle emerged in either year.

Statistical analyses
All analyses were performed using R Statistical Software v. 3.3.2 (www.r-project.org, accessed 9 April 2017). We used t-tests to compare mean burrowing depth beneath the soil and duration of hibernation between years. Turtles found deceased after entry into hibernation (n=2 in 2014 -2015) were excluded from both comparisons, and turtles for which iButton data were not available (n=3 in 2014 -2015) were excluded from the latter comparison.
Habitat in which hibernacula were located was characterized as field if the turtle was outside of the tree line, otherwise it was considered forest. Available habitat in each category was quantified using ArcGIS Desktop v. 10.2 (Environmental Systems Research Institute, Inc., Redlands, CA, USA), and a test of proportions was performed to determine whether use differed from availability.
To identify candidate environmental variables that may influence the probability of hibernating for 7 consecutive days in any given week in the fall, a principal components analysis (PCA) was performed on the correlation matrix of 23 environmental variables. The first four components explained 93% of the variance ( We used a similar approach to determine the influence of the environment on the probability of emergence from hibernation in any particular week in the spring. The first four components from the PCA of 23 environmental variables explained 86.4% of the variance (Table 1.3). Because these loadings were similar to the principal components analysis performed for the fall dataset, and to ensure consistency between seasonal predictions, we chose the same four environmental variables to test in our model. These four variables and the candidate variables of year, MCL, sex, week, and body condition were tested singly in the generalized marginal model with turtle as the clustering factor, and variables with P > 0.2 were excluded from model selection. All possible additive combinations of the remaining candidate variables were then tested.
After comparing the candidate models using the independence model criterion, several models were within 2 QIC values of the lowest QIC, so the most parsimonious model of that set was selected as the final model.

RESULTS
In 2014, all turtles had moved out of fields and into the forest, with no straightlinecrossing of fields or roads as determined by weekly GPS locations, as of 2 November Only weekly mean daily temperature was included in the final model for the probability of hibernation during any one week in the fall (Table 1.2). Lower temperatures resulted in higher probability of hibernation in a given week (Figure 1.1).
The inflection point in the predicted response curve is at 13.9 ˚C, meaning that at weekly mean daily temperatures of less than 11.3 -17.1 ˚C, the probability of hibernation in any given week is predicted to be at least 50%. At weekly mean daily temperatures of less than 6.3 -10.8 ˚C, the probability of hibernation in any given week is predicted to be higher than 90%.
Weekly mean daily temperature and year were included in the final model for the probability of emerging from hibernation in the spring in any given week (Table   1.4). There was a slightly lower intercept for year 2015 than year 2016, although the parameter estimate was not significantly different from zero in the model. (Figure 1.2).
The inflection point in the predicted response curve for 2015 is at 11.8˚C and for 2016 is at 10.4 ˚C, meaning that at weekly mean daily temperatures of greater than 7.0 -15.4 ˚C, depending on yearly conditions, the probability of emergence from hibernation in any given week is predicted to be at least 50%. The probability of emergence in any given week is predicted to be higher than 10% when weekly mean daily temperature is warmer than 3.0 ˚C -10.3 ˚C, depending on yearly conditions. The scaling exponent estimated by SMA regression of body mass on MCL was 1.706, and the arithmetic mean value for MCL for the study population was 134.5 mm.

DISCUSSION
Turtles preferentially selected forest habitat for hibernacula locations. This avoidance of fields, similar to the avoidance of clearcuts in Indiana , is likely due to soil compaction and more variable thermal profiles of the different soil types typical in fields or areas without canopy cover; however, such avoidance also translates to lower risk to turtles from late fall, winter, and early spring mowing activities because turtles are unlikely to be present in fields at all. Furthermore, the radio-tracked individuals in this study were observed to decrease weekly distances moved and to cease using field and edge habitat in advance of entry into hibernation.
Movement out of fields and into forests, in conjunction with a reduction in overall movements, provides an additional buffer against the risk of mowing-caused mortalities even for turtles that have not yet entered into hibernation after mowing activities resume in the fall.
We determined that weekly mean air temperatures affected the probability of entry into and emergence from hibernation in any given week in the fall or spring, respectively. In the fall, cooler temperatures increased the probability that individuals would either begin burrowing into the soil or remain underground during the week.
Warmer temperatures in the spring increased the probability that individuals would emerge from hibernation during the week, supporting findings from previous studies on T. carolina (Grobman 1990;    to immergence into hibernation in the fall. We failed to find support for our hypothesis that plasma biochemical profiles would be related to the timing of emergence from hibernation in the spring; however, it may be that relative changes in physiological condition over the duration of hibernation, rather than absolute concentration of solutes measured after emergence, may be more biologically relevant to the timing of emergence. Physiological condition appears to act as a proximate mechanism for determining some aspects of an individual's response to changing environmental conditions in relation to hibernal phenology. This proximate mechanism for behavioral plasticity may enable persistence as climate change accelerates.

Introduction
Hibernation is a critical aspect of the life history of terrestrial ectotherms, and the timing of hibernation can influence access to mates, timing of reproduction, depletion of energy reserves, protection from desiccation, exposure to predation, and risk of experiencing lethal cold temperatures  Body condition is a commonly used index that measures muscle and fat stores after correcting for body size. However, the effects of changes in body condition of turtles on individual behavior may not be detectable because the amount of water stored in the bladder can overwhelm subtle changes in tissue mass ). In addition, the accumulation of fat stores in anticipation of hibernation has not been observed in many turtle species (Ultsch 1989), and  did not observe captive T. c. carolina utilizing stored fat reserves over the winter. Instead of relying on brown adipose tissue to fuel metabolism during periodic arousals while hibernating, as heterothermic mammals and birds have been observed to do , turtles generally rely on the storage of carbohydrates in the form of glycogen in order to fuel energy needs over the winter Ultsch 1989).
Glycogen is converted to glucose by enzymes in the liver even during winter, and because muscle tissue and other organs experience lower metabolic rates during this time, glucose accumulates in blood plasma, thus providing a source of immediately available energy upon emergence  as well as acting as a cryoprotectant (Costanzo, Lee Jr. & Wright 1993). Thus, the accumulation of glucose in blood plasma measured at emergence from hibernation may be a more accurate indicator of endogenous energy stores and hence physiological condition than body mass.
In addition to plasma glucose, there are many plasma biochemical variables that serve as indicators of various aspects of physiological condition in turtles and which show seasonal patterns of changes in concentration; these candidate variables may affect timing of immergence or emergence. We identified eight variables, including glucose, for which normal reference ranges in T. c. carolina have been reported in the literature and that collectively capture major physiological attributes, specifically metabolism, ion balance, and tissue and cell damage. Protein catabolism provides a source of energy for turtles during hibernation in addition to glycolysis . The accumulation of uric acid in blood plasma over the winter indicates protein catabolism, and uric acid can also serve as a cryoprotectant . Calcium is necessary for egg calcification , and is elevated in breeding females . The plasma electrolytes sodium and potassium are routinely measured as part of health assessments, and concentrations of both are depressed in unhealthy animals . Elevated levels of aspartate aminotransferase (AST) in plasma indicate tissue damage in liver or muscle, while elevated levels of lactate dehydrogenase (LDH) can indicate cryoinjury in overwintering turtles, specifically freeze/thaw damage to plasma membranes . Plasma osmolality provides a measure of hydration status and indicates overall concentration of dissolved solutes in the blood (Ultsch et al. 1999).
We hypothesized that plasma biochemical profiles of T. c. carolina consisting of these eight variables, but not body condition, would be correlated with immergence and emergence dates of hibernation. Specifically, we predicted that (1) turtles with better physiological condition in the fall would enter into hibernation earlier, and (2) turtles with better physiological condition in the spring would emerge from hibernation later.

Ethical procedures
All

Study area
Research was conducted at the 250 ha. William Floyd Estate, a management unit of Fire Island National Seashore, located on the southern coast of Long Island, New York, USA (40° 45'57" N, 72° 49' 26" W). This site occurs in the Long Island Coastal Lowlands section of the Atlantic Coastal Plain physiographic province, and elevation ranges from sea level to 5 m. Forested areas consisted mostly of a coastal-oak heath community, although some areas were dominated by eastern red-cedar, while upland fields were classified as Dactylis glomerata -Rumex acetosella cultivated herbaceous alliance .

Turtle monitoring
In August -September 2015, we performed visual encounter surveys throughout the study area to locate 20 adult box turtles > 350 g mass (10M:10F, sex determination followed . Radio transmitters (RI-2B 10 -15 g, Holohil Systems Ltd., Ontario, CAN) were attached to the carapace using marine epoxy so that turtles could subsequently be relocated over the winter until spring emergence. To determine hibernation exit dates, shell temperatures were monitored by attaching a miniature temperature datalogger (iButton Thermochron DS1922L, Embedded Data Systems LLC, KY, USA) to the carapace. Dataloggers were coated in plastic tool dip (Plasti-Dip International, MN, USA) to prevent water damage with minimal effect on accuracy , and were programmed to record at 0.5 °C resolution every 40 minutes. In order to resolve ambiguity around the determination of entry dates using temperature data alone, we additionally attached an archival light logger ("geolocator" Intigeo F100, Migrate Technology Ltd., Cambridge, UK) to the carapace, and programmed it to record maximum ambient light levels (0.28 -18604 lux) every 5 min. The combined weight of attachments (radio transmitter, iButton, geolocator, and all epoxy) averaged 4.58% of body mass (range 2.86 -7.06 %).
Turtles were located via radio telemetry at least weekly in the fall until they began to burrow into the soil or remained in the same location beneath the leaf litter for 7 consecutive days. Turtles were operationally defined as having entered into hibernation by remaining in the same location for 14 consecutive days, at which point monitoring via radio telemetry was reduced to monthly until March. We considered 14 days to be a conservative determination, as  defined entry into hibernation as the first day after which the turtle remained underground for at least one week. Weekly monitoring resumed in April to confirm spring emergence. Equipment was removed in May after all turtles had emerged and dispersed from hibernacula.

Quantifying body condition
Midline straight carapace length (MCL) was measured at first capture. Turtles were weighed to the nearest 5 g using a Pesola hanging scale at first capture, before and after equipment attachment or removal, at blood sampling, and upon first emergence in the spring. Mass measurements on each sampling occasion were corrected by removing the mass of attached equipment from total mass.

Blood sampling procedure
In October 2015, 0.5 ml of whole blood was collected via venipuncture of the dorsal subcarapacial cervical plexus (Hernandez-Divers, Hernandez-Divers & Wyneken 2002) in the first week, and subsequently every two weeks until the turtle was operationally determined to have entered into hibernation. A total of between one to three samples were collected from each individual, as monitoring visits occurred weekly and no individuals were observed aboveground during monitoring visits after October 30, 2015. In spring 2016, a single 0.5 ml blood sample was collected from each individual once emergence had been observed. Immediately following sample collection using 0.5 in 25 g needles with a 1 mL syringe, samples were transferred to heparinized vials and placed on ice until sample processing within five hours of collection.
Samples were transported to a processing area, where sample vials were manually inverted 10 times, following which approximately 50 µl whole blood was transferred into a heperanized capillary tube and centrifuged at 11,500 rpm for 5 min for the determination of packed RBC volume. The remaining whole blood sample was centrifuged for 8 min at 5000 rpm, and 50 µl separated plasma were pipetted immediately into a disposable test tube and analyzed within 2 hr after separation, with the exception of one sampling occasion when this timeline was not possible. On this occasion, samples were centrifuged and plasma was separated and refrigerated at 0 -4 ºC until analysis seven days later. Plasma osmolality was determined using a freezing point osmometer (µOsmette™ Model 5004, Precision Systems Inc., MA, USA) after calibration with control standards. The remaining volume of separated plasma was transferred to cryovials and placed on ice until transported from the field to -80 ºC storage. Plasma concentrations of AST, calcium, glucose, LDH, potassium, sodium, and uric acid were analyzed by IDEXX BioResearch Pathology Services (IDEXX Laboratories, Inc., ME, USA) within one year of collection.
We observed highly variable hematocrit measurements that we interpreted as indicating lymphatic fluid dilution of samples, a known complication from the chosen subcarapacial venipuncture location . Therefore, hematocrit levels were included in analyses based on the assumption that the concentrations of plasma biochemical variables would change linearly with hemodilution from lymph.

Soil temperature monitoring
In order to match the shell temperature to soil temperatures to determine the timing of spring emergence, once horizontal above-ground movements had apparently ceased, a ground stake was installed in the soil at 50 -200 cm from the turtle's hibernaculum.
Each ground stake had three temperature dataloggers (iButton Thermochron DS1922L or DS1921G, Embedded Data Systems LLC, KY, USA) set at the soil surface, 7.5 cm beneath the surface, and 15 cm beneath the surface to capture the probable range of burrowing depths of overwintering T. c. carolina in this region . Each datalogger was held in a waterproof plastic case containing desiccant that was glued to the wooden stake at the appropriate depth. Once the stake was installed in the soil, it was surrounded by a wire mesh cage to deter animal disturbance. Dataloggers were programmed to record at 0.5 °C resolution every 40 minutes, and were subsequently downloaded approximately every 50 days.
On occasions in which turtles shifted position, ground stakes were relocated so that the soil temperatures were monitored in close proximity to the hibernacula.

Determination of overwintering status
Date of entry into hibernation was determined by a classification system based on light levels recorded by the shell archival data loggers. A total of 264 direct observations made over the course of the study were used to validate geolocator readings. The maximum light level recorded when a turtle was confirmed to be in its refugia and not visible above the leaf litter by the observer was substantially lower than the minimum light level recorded when a turtle was observed on the surface (Figure 1). The restrictive definition of the maximum light level observed during refugia occupancy was used as the classification threshold to ensure that no surface activity would be incorrectly classified as refugia use. Weekly tracking observations were then used to augment this classification to ensure that no above-ground surface movements were incorrectly classified as refugia occupancy. The earliest date of all-day refugia use after which no horizontal surface movements were recorded between September -December 2015 was chosen as the date of entry into hibernation for analyses.
Date of exit from hibernation was determined as the earliest of the following: (1) basking temperatures recorded by the shell data logger (shell temperatures > 5 ºC above surface soil temperatures; ), (2) direct observation by observer of the turtle above the leaf litter and active, or (3) surface-level light levels recorded by the archival light logger. We chose to identify date of exit from hibernation using any of these three indicators because the light logger was placed on the rear of the shell, and in some cases, because turtles only partially emerged headfirst from hibernacula on the first day of emergence, the light logger did not always detect such emergence events.

Statistical analysis
All analyses were performed using R Statistical Software v. 3.3.2 (R Core Team 2014). One male turtle was excluded from all analyses because of geolocator failure. Body condition was calculated using the scaled mass index ), which standardizes the prediction for an individual's body condition by accounting for the scaling relationship between body mass and body size, whereas a traditional body condition index tends to be biased towards larger individuals because larger individuals will have larger absolute body components ).
The scaling exponent and coefficient for the index were determined using fall MCL and the first fall body mass measured for each individual. Males and females were combined because there were no significant differences (χ 2 (1) =0.540, p-value=0.462) in the slopes of the relationship between MCL and mass by sex. The scaling exponent was estimated to be 1.706, and the arithmetic mean value for MCL for the study population was 134.5 mm. This equation was then used to calculate an encounterspecific body condition index value for each sampling occasion for each individual.
In order to reduce dimensionality of the independent variables due to the small sample size, principal components analyses were performed on the correlation matrix of plasma biochemistry values and hematocrit levels for immergence and emergence separately. The principal components analysis for immergence included every sampling encounter. While this resulted in an uneven number of replicates per individual being included in the principal components analysis, we chose to use this approach in order to capture as much variation among the biochemical variables as possible during fall sampling. For two individuals with missing sodium and potassium concentrations due to small sample size, we substituted the values from the other sampling occasion for that same individual. One outlier was excluded from the spring principal components analysis; this turtle exited hibernation in mid-March, but was not sampled until 27 days later and thus the plasma biochemical profile was unlikely to accurately reflect physiological condition at time of emergence. Components explaining at least 75% of the variance were then included as candidates in model selection.
For investigating the effect of candidate components on dates of immergence, we included all individuals with at least two sampling events (n = 15) as the random effect in generalized linear mixed models using the "lme4" package . The dependent variable was the number of days from the sampling occasion until immergence date, i.e., time to immergence, which was modeled as a Poisson distribution as appropriate for non-zero count data. Each candidate component score, sex, and body condition (scaled to a mean of zero and a range of two standard deviations from the mean) were run in separate models, and variables with P > 0.2 were excluded from model comparisons. We next ran generalized linear mixed models, without interaction among the fixed effects to avoid overfitting because of the small sample size, using all possible additive combinations of candidate variables (Table 2.3), and the model with the lowest Akaike's information criterion corrected for small sample size (AICc) was selected as the final model.
For investigating the effect of candidate components on dates of emergence, we ran generalized linear models using the "lme4" package . The dependent variable was the Julian date of emergence, which was modeled as a Poisson distribution as appropriate for non-zero count data. Each candidate component score, sex, body condition (scaled to a mean of zero and a range of two standard deviations from the mean), and duration of hibernation (scaled to a mean of zero and a range of two standard deviations from the mean) were run in separate models, and variables with P > 0.2 were excluded from model comparisons. We next ran generalized linear models, without interaction among the fixed effects to avoid overfitting because of the small sample size, using all possible additive combinations of candidate variables (

Discussion
We found evidence that physiological condition may influence certain aspects of phenological asynchrony associated with hibernation in T. c. carolina at our study site.
We confirmed that lower concentrations of plasma solutes, indicative of good hydration, and higher levels of calcium are associated with shorter time to immergence into hibernation. It is likely that higher levels of calcium in the fall are indicative of turtles in better condition, because females will use body reserves of calcium to shell eggs  after ovulation occurs in May or June . This result corroborates the idea that behavioral plasticity enables risk-avoidance; in this case, turtles may be more likely to enter into hibernation earlier if in better physiological condition in order to decrease the risk of exposure to lethal cold temperatures, which can be a major source of mortality for T. c. carolina populations .
We failed to find support for our prediction that turtles in better physiological condition emerge from hibernation later, but we cannot be certain that this result was not related to our methods. It may be that the relative change in plasma biochemical variables from entry into hibernation until exit from hibernation is more biologically significant than absolute concentrations in plasma measured after emergence.
However, because our weekly tracking in the fall was not frequent enough to observe turtles when they were above the surface once activity levels had decreased in later October and November, we were not able to obtain blood samples for most individuals within less than five days of entry into hibernation. Turtles were not captured after they had entered into the soil to avoid disrupting the turtles' natural behavior. As a result, blood samples were likely collected too early in the season to accurately quantify physiological condition immediately prior to hibernation, and thus we cannot reliably compare immergence and emergence concentrations to obtain relative differences.
The complex relationships between various indicators of physiological condition and the timing of hibernation provide intriguing avenues for exploration into behavioral plasticity and constraints on seasonality within ectotherm populations. If physiological condition mediates the timing of immergence into hibernation, as these results indicate, then the environmental conditions in the preceding spring and summer that affect production of forage or abundance of invertebrates may be more relevant to the timing of seasonality in the fall, or even in the following spring, than previously realized. Since earlier immergence into hibernation seems to be in part a result of better physiological condition, and if this physiological condition has a carry-over effect on activity in the following spring, then the timing or size of clutches could be    Gray shading indicates 95% confidence region.