Study area

The Sonoran Desert has average high temperatures of over 38°C in summer and is one of the hottest and driest regions of North America (INRMP 2003). From June to October, temperatures can exceed 32°C for more than 100 consecutive days on CPNWR (USFWS 2002). Average annual precipitation on the CPNWR ranges from 20 cm in the east to 7.5 cm in the west falling in monsoons during July–August and December–January (USFWS 2002). April to June is the dry season.

In May and June of 2012–2014, we collected pronghorn fecal pellets in the captive pen and from wild individuals at 12 developed watering holes (hereafter drinkers) and nine sites not associated with drinkers on Organ Pipe National Monument (ORPI), CPNWR, and the adjacent Barry M. Goldwater Range (BMGR) in southwest Arizona, USA (Fig. 1). The 2.56 km² captive pen holds two separate populations which total ∼⃒100 free ranging individuals (USFWS 2015), and the wild population in the US was estimated at 202 individuals (95% CI: 171–334) in December 2014 (USFWS 2015). Arizona Game and Fish Department (AZGFD) and US Fish and Wildlife Service (USFWS) personnel provide alfalfa at feed stations both in the wild and the pen. Conditions in the pen are kept as natural as possible and portions of the pen are irrigated to encourage growth of natural forage. Fawning starts in the pen in mid-February and most fawns are born in March and April (Wilson et al. 2008). Wild pronghorn are fed alfalfa weekly during the dry months at six of the 12 drinkers outside the pen. Feeding and watering typically begin in April or May and continue through October or November depending on annual rainfall amounts (J. Atkinson, U.S. Fish and Wildlife Service, pers. comm.). In the wild, fawns are typically born between February and June with one birth reported as early as January in 2013 (Bright and Hervert 2005, USFWS 2015).

Figure 1.

Map depicting Sonoran pronghorn fecal sampling locations (drinkers and non-drinkers) in 2013 and 2014 on Barry M. Goldwater Range (BMGR), Cabeza Prieta National Wildlife Refuge (CPNWR), and Organ Pipe National Monument (ORPI), southern Arizona, USA. Each star represents general locations of multiple sampling areas.

Sample collection

Each December, annual capture operations are conducted in the captive pen by AZGFD and USFWS, during which individuals are captured, radio-collared, and a blood sample is collected. Fawns captured in the pen are ear-tagged, and radio-collared if recaptured in subsequent captures. Young of the year are easily identified (e.g. size, horn development) and classified as fawn (0–11 months), and individuals captured as fawns in the previous year are known yearlings (12–23 months). However, not all fawns are caught during annual capture operations and consequently, an individual may not be handled until it is > 1 year old and is potentially misclassified as to actual age. Thus, all captured animals of unknown age are classified as adults. Some captured (captive) individuals are subsequently released into the wild, at a ratio of approximately two males to one female (USFWS 2015). To obtain DNA, blood samples were collected from 58 captured individuals in December 2012 and 2013 when feasible (i.e. if health and safety of the animal was not at risk due to stress) (USFWS 2015). These samples provided a genotype of an individual of known age class for later matching to genotypes obtained from fecal pellets collected in the pen or in the wild.

In May 2012 in the captive pen, we collected five fecal pellets (per Morden et al. 2011) less than 24 h old from each of 185 fecal pellet piles in three pellet size groups defined visually as small, medium, and large. Size refers to the pellet, not size of pile, and all pellets from a pile were given the same size classification. While we recognize our size classification is subjective, we wanted to ensure collection of all age and sex classes, and this size classification was used only to structure collection and was not part of the analyses. To ensure collection of samples less than 24 h old, we cleared pellets from the area around feed stations in the captive pen on the day prior to collection. We collected fresh pellets as our pilot studies indicated low nuclear DNA (used for individual identification) PCR success rates (2–28%) by day 14 and 0% by day 60 (Woodruff et al. 2015, Woodruff unpubl.). We excluded piles that appeared to be from more than one individual based on pellet shape, color and size to minimize the potential for wasted effort in the laboratory, as these are likely mixed samples. We placed pellets in paper coin envelopes and stored them at room temperature in a plastic ziploc bag with ∼⃒ 250 ml of silica desiccant to minimize DNA degradation prior to analysis (Soto-Calderon et al. 2009, DeMay et al. 2013). To determine the age of the individual from which the fecal sample was collected, we matched 7–16 locus microsatellite genotypes of fecal samples to the blood samples.

Due to high rates of capture myopathy, only limited capture and radio-collaring of wild Sonoran pronghorn occurs, and the majority of marked animals in the wild are captive released individuals (USFWS 2015). To test our ability to assign age class, fawn or non-fawn (≥ 1 year old), in the wild population, we measured pellet samples collected from wild pronghorn in May and June of both 2013 and 2014 as part of a larger mark—recapture study (or detection—redetection since there is no physical capture; Woodruff et al. 2016). We visited drinkers when pronghorn had left the area and collected fecal pellets surrounding feed stations and drinkers. At least six pellets from each sample were collected, and stored as described above. We collected six pellets per sample to ensure we had a sufficient number of pellets to perform two DNA extractions (Woodruff et al. 2015). Samples were initially field-classified as fawn or non-fawn based on visual assessment of size and morphology. We had two sources of known age wild individuals: 1) genotypes of known age individuals released from the captive pen obtained from blood samples, which consisted of two known adults captive released in 2012 and sampled (i.e. through fecal pellets) in the wild in 2013, and 2) 15 individuals redetected in 2014 (thus, known non-fawn) from fecal pellet genetic analysis in 2013.

DNA extraction and genotyping

DNA extraction of blood samples was conducted using a Qiagen DNeasy blood and tissue kit (n = 10) or by overnight lysis with ProteinaseK (10 mg ml-1) at 55°C, followed by a modified protocol (n = 48) based upon the standard phenol/chloroform extraction and isopropanol/sodium acetate precipitation (Sambrook et al. 1989). We used Phase Lock gel tubes (5-Prime) to aid in the separation between organic and aqueous phases and resuspended the DNA in Low TE (10 mM Tris-pH 8.0, 0.01 mM EDTA). Fecal pellet DNA was extracted using the QIAamp DNA Stool Mini Kit following methods described in Adams et al. (2011) and Woodruff et al. (2015). For individual ID, ten nuclear DNA (nDNA) microsatellite loci ranging in size from 90–278 base pairs and one sex ID locus were amplified in a single multiplex reaction (Lou 1998, Carling et al. 2003, Munguia-Vega et al. 2013). The 7 μl PCR reaction contained 1 × Qiagen Master Mix, 0.5 × Qiagen Q-Solution, 1.71 μM Anam97, 0.04 μM Anam50, 0.07 μM Anam82, 0.01 μM Anam79, 0.86 μM Aam13, 0.43 μM Aam11, 0.14 μM ADCYC, 0.26 μM Aam10, 0.04 μM Aam1, 0.04 μM Aam2, 0.29 μM KY (sex ID), and 1.5 μl DNA extract. The PCR profile included an initial denaturation of 95°C for 15 min, followed by a touchdown of 20 cycles with a 30 s denaturation at 94°C, 90 s annealing step at 63°C decreasing 0.5°C each cycle, and a 60 s extension at 72°C, followed by 34 cycles of a 30 s denaturation at 94°C a 90 s annealing step at 53°C, and a 60 s extension at 72°C. The cycle finished with a 30 min final extension at 60°C.

We initially screened all samples with two PCR replicates to assess sample quality, and samples failing to amplify at ≥ 5 loci were dropped from additional genotyping to remove low quality, error-prone samples from the dataset. To obtain a consensus genotype, three to eight PCR replicates were performed per sample. Consensus genotypes were based on multiple runs of a sample as follows: 1) for homozygotes, the allele was present at least three times, and 2) for heterozygotes, we had to see each allele at least two times. We repeated this testing and evaluating process until we obtained a consensus genotype at a minimum of seven loci to meet the matching criteria of ≤ 0.01 probability of identity siblings (P(ID)sibs) (Waits et al. 2001). Consensus genotypes were determined in Microsoft Access (Skrbinsek 2010), and matching and P(ID)sibs analysis was conducted in GenAlEx 6.5 (Peakall and Smouse 2006). Within the captive pen, fecal pellet genotypes were first matched to other fecal pellet genotypes. Then, unique genotypes were matched to genotypes of blood samples for individuals of known age. In the wild population, fecal pellet genotypes were matched to other fecal pellet genotypes, and individuals redetected in year two were then known to be ≥ 1 year old (non-fawn).

Pellet dimension measurements

Using digital calipers we measured (mm) maximum length (L), maximum width (W), and calculated length-to-width ratio (L/W), and approximate volume (V: 4/3π (L/2)(W/2)2; volume of an ellipsoid — the approximate shape of pellets), where W is used for width and diameter, of each of the five pellets per sample from the captive pen within two hours of sample collection. We calculated the mean measurements of the five pellets to represent the sample. From wild individuals, we attempted to measure five pellets from at least one pellet group from each individual (confirmed by DNA analysis). However, we did not always have enough pellets postextraction for measuring, and thus not all individuals had a measured sample. Samples with fewer than two pellets were discarded and we excluded broken, split, or partial pellets as recommend by Zahratka and Buskirk (2007). To evaluate the effect of time and desiccation on the measurements, we re-measured 16 samples after seven days of exposure to local field conditions and reran the calculations. Mean daily high temperature was 37.3°C (range: 33.6–40.7°C), and there was no precipitation during this 7-day period.

Statistical analysis

We conducted all statistical analyses using R ver. 3.1.2 (<  www.r-project.org >). We used t-tests with an alpha of 0.05 to examine differences in pellet measurements between captive and wild individuals and between measured (fresh) and re-measured (7-day old) samples. We used logistic regression to distinguish between age classes. Using logistic regression (brglm function, R package brglm; Kosmidis 2013), we evaluated the predictive accuracy of models using each measurement (e.g. length, volume) as a single predictor variable, as well as models using all possible combinations of the predictor variables (e.g. length and volume). Predictive accuracy was estimated using randomized five-fold crossvalidation with the cvFit function from the R package cvTools (Alfons 2012) with 100 replications, and models were also ranked using Akaike's information criterion corrected for small sample size (AIC_c). To avoid infinite parameter estimates due to complete or quasi-complete separation, the model parameters were estimated using bias-reduced maximum likelihood (Firth 1993, Kosmidis and Firth 2009). Probability of correct classification was compared to a null model (i.e. no explanatory variables) where the predictive accuracy depends only on the base rate of classes in the population (e.g. if 60% of the population are adults, then prediction accuracy would be 60% if every sample were called adult).

For age class, cross validation analyses indicated that it was not possible to reliably distinguish between yearlings and adults based on pellet morphology significantly better than using the base rate (probability = 0.69–0.72; null model probability = 0.67), so our age class analysis focused on separate analyses distinguishing fawn from yearling and fawn from adult. We used only known age captive individuals and known age captive released wild individuals for age class analysis (i.e. excluded wild individuals of unknown age). Finally, we combined yearlings and adults into a single age class (non-fawn) and tested the accuracy of our visual fieldclassification of wild samples by age class, i.e. fawn or nonfawn. We used the fitted model developed from known age individuals (captive fawns and captive and wild non-fawns) to predict age class of wild samples. We used models with single explanatory variables of width and volume because these were two of the best AIC_c ranked models for predicting age class. In cases where the model-predicted age class differed in multiple samples from the same individual (i.e. one predicted adult, one predicted fawn), we ran an additional model (length + length—width ratio).

Pellet size and age class

For both captive and wild, length, width and volume of yearling and adult (or non-fawn) pellets were greater than fawns, but length—width ratio was larger for fawns, indicating fawn pellets were rounder than either yearling or adult pellets (Fig. 2, Table 3a). Probability of correct age class classification for captive fawn versus yearling using a single variable was 0.98 for width only and was the best AIC_c ranked model (Table 4a). Other single variable models and combinations of ≥ 2 explanatory variables performed similarly with predictive probability ranging from 0.91 to 0.98 and delta AIC_cs (difference in AIC_c compared to the lowest AIC_c) from 1.66 to 7.16. Length as a single explanatory variable was the worst ranked model with probability of correct classification of 0.77 and delta AIC_c value of 7.16. The null model (i.e. the model with no explanatory variables) had probability of correct classification of 0.32 (i.e. if we guess fawn, we will be right 32% of the time because 32% of our samples were fawn).

Figure 2.

Scatterplot of fecal pellet measurements of captive and wild Sonoran pronghorn collected May and June 2013 on the Cabeza Prieta National Wildlife Refuge, Arizona, USA. Captive individuals are known age and wild individuals are as predicted by models. The dotted lines represent contours of constant length—width ratio, with points in the upper left representing pellets that are relatively long and narrow, while lower right points represent pellets that are relatively short and wide.

Table 3.

(a) Descriptive statistics and associated standard errors (SE) of fecal pellets from captive and wild Sonoran pronghorn collected May and June 2012–2014, Arizona, USA. Wild non-fawn individuals include only redetected individuals, and wild fawns were determined using model predictions. (b) p-values from t-tests comparing differences in fecal pellet measurements between captive and wild Sonoran pronghorn.

The best AIC_c ranked model for captive fawn versus adult again included pellet width only with probability of correct age class classification of 0.98 (Table 4b). The volume only model and combinations of ≥ 2 explanatory variables performed similarly with predictive probability ranging from 0.93 to 0.97 and delta AIC_cs from 1.46 to 8.41. Length and length—width ratio as single explanatory variables were the worst ranked models with probability of correct classification of 0.85 and 0.77 and delta AIC_c values of 17.77 and 26.65, respectively. The null model (i.e. the model with no explanatory variables) had probability of correct classification of 0.50 (i.e. 50% of pellet samples were adult and 50% were fawn).

Captive versus wild measurements of pellets

Mean length, width and volume of wild non-fawn pellets (≥ 1 year as evidenced by redetection) were smaller than those of captive non-fawns (Fig. 3). Mean length—width pellet ratio, however, was consistent for non-fawn captive (1.59) and wild (1.68) individuals. Pellet length, width and volume were significantly different between captive and wild non-fawns (Table 3b). For fawns, only pellet width was significantly different between captive and wild individuals.

Table 4.

Results of the logistic regression models evaluating the influence of pellet measurements on the ability to distinguish (a) captive Sonoran pronghorn fawn versus yearling and (b) fawn versus adult. AIC_c values and standard errors are approximate due to some pseudoreplication from two or more pellet samples from the same animal. Only models with delta AIC_c ≥ 2 are shown. Pellets were collected in the captive pen on the Cabeza Prieta National Wildlife Refuge, Arizona, USA in May and June 2013 and 2014. We included single dimension measurements as well as all combinations of measurements.

Prediction of age class from wild pellet samples

In 2014, we redetected 15 individuals (n = 39 samples), thus known to be non-fawn (≥ 1 year old). All samples from redetected individuals were correctly model-predicted as non-fawn in either the width only or volume only model (Table 5, Supplementary material Appendix 1 Table A1).

Ten unique individuals (a total of 16 samples) had conflicting model-predicted (width only and volume only models) age class (i.e. one model predicted non-fawn, one predicted fawn). There was no consistency in which model (e.g. width only) predicted fawn or non-fawn in these cases. For eight of these individuals, we measured multiple samples, and we took the majority model-predicted classification (i.e. four model-predicted as non-fawn, two model-predicted as fawn, we called it non-fawn). The two other individuals were single detections (i.e. we had only one sample) for which we ran an additional fitted model (length + length—width ratio) and used the age class that was predicted twice (i.e. volume and width models both predicted fawn, and length + length—width ratio predicted adult, we called the individual fawn).

Figure 3.

Box plots showing mean measurements of Sonoran pronghorn fecal pellets collected in May and June 2012–2014 in Arizona, USA, from captive and wild fawn (F) and non-fawn (NF) for length, width, length—width ratio (LW), and volume. The box signifies the upper and lower quartiles and the median is represented by the thick black line. Black dots represent outliers.

Correct visual field-classification by age class varied considerably between years. In 2013, we measured 128 samples from 76% (n = 72 individuals) of the total wild individuals we detected (i.e. detected their genotype) (Table 2). Ninety-six percent of visual field-based classifications matched the model predictions (Supplementary material Appendix 1 Table A1). In 2014, we measured 130 samples from 64 individuals, and 65% of samples were correctly field-classified by visual observation (i.e. matched the model prediction).

Table 5.

Comparison of visual field classification and measurement prediction for pellet samples collected for wild pronghorn in 2013 and 2014, Arizona, USA.

In 2013, only one sample that was visually field-classified as non-fawn (n = 119) was model-predicted as fawn, whereas three samples visually field-classified as fawn were subsequently model-predicted as non-fawn (Table 5, Supplementary material Appendix 1 Table A1). Similarly, in 2014, 98% (n = 62) of samples visually field-classified as non-fawn were model-predicted as non-fawn, but only 33% (n = 22) of samples field-classified as fawn were model-predicted fawn. Nearly all samples (46 of 49 in the years combined) incorrectly visually field-classified as fawn were male.

Pellet size and age class

Our visual classification of pellets of captive pronghorn by size (small, medium, large) was a mechanism to facilitate collection of all age and sex classes and was not used in the analysis, but our results indicate this was an appropriate collection method for obtaining a range of age classes for our analyses. As expected, pellet size was larger for yearlings and adults than for fawns. The probability of correctly assigning age class models was similar for fawns and yearlings and fawns and adults and performed notably better than the null models (null probability 0.32 and 0.50, respectively). On the other hand, both Morden et al. (2011) and MacCracken and Van Ballenberghe (1987) found higher rates of correct model classification for adults compared to juveniles. Contrary to other studies (Delibes-Mattos et al. 2009, Morden et al. 2011), our cross-validation analysis showed that single measure models assigned age class with roughly the same accuracy as models with multiple variables. Our small sample size (n = 4) for adults is perhaps the reason we could not distinguish between adults and yearlings. Collecting and measuring fecal pellets from additional known adults would potentially allow for this distinction.

Some variables were correlated as would be expected given that two variables (length—width ratio and volume) are calculated using combinations of the other variables, and length and width also tend to be correlated. Thus, it is not surprising that models with additional variables generally did not substantially outperform the width-only model.

Captive versus wild

Similar to other research, our results suggest that the most accurate predictive models are built from samples collected from the target population (MacCracken and Van Ballenberghe 1987, Chapman 2004, Morden et al. 2011). Differences in diet (e.g. seasonal variation, captive versus wild) can affect size of fecal output (Campos-Arceiz et al. 2008, Morden et al. 2011) and defecation rate (Rogers 1987, Mayle et al. 1996, Chapman 2004, Ferretti et al. 2014) with both reduced (Irby 1981, Asa et al. 1985, Rogers 1987, Kitchen and Martin 1996, Chapman 2004) and increased defecation rates in captive individuals (Smith 1964, Dinerstein and Dublin 1982). Brashares and Arcese (1999a, b) suggest dominant male oribi Ourebia ourebi restrict defecation volume and occurrence in order to increase the frequency of territory marking events. Consequently, there could be differences in defecation rate, and potentially pellet size, between captive and wild individuals, as well as fed and unfed individuals in the wild. However, we believe that our approach to develop predictive models using captive individuals was justified and effective for multiple reasons. First, in our study system, any future fecal DNA sampling will be conducted at drinkers, nearly 50% of which also provide supplemental feed. Additionally, as conditions in the pen are semi-natural, captive pronghorn also feed on natural forage year-round due to irrigated forage plots. Second, we were able to test our methods on 15 wild animals known to be non-fawn. All were correctly identified as non-fawn by the model-prediction. Third, length—width ratios between captive and wild animals were not significantly different and indicated that pellet size changed consistently across measurements from captive to wild. Fourth, our results illustrate significant differences in pellet size between fawn and nonfawn in both the captive and wild populations, with wild fawn having the smallest measurements and thus having a lower chance of being misclassified as non-fawn compared to captive animals.

Accuracy of visual-based field classification

Our results suggest that age classification based on visual assessment is imprecise and likely influenced by individual differences in observers. In 2013, we had two people collecting samples who were trained simultaneously and extensively. In 2014, four new personnel collected samples and training was limited. Based on model predictions, visual-based field classification was more accurate in 2013 (96% correct) compared to 2014 (65% correct). Samples were more often field-classified as fawn and consequently model-predicted as, or known to be, non-fawn (n = 19) than vice versa (n = 2). While we detected twice as many males as females (Woodruff et al. 2016), 94% (46 of 49) of samples incorrectly visually field-classified as fawn were male suggesting fecal pellets of adult males are more difficult to classify than adult females, contrary to MacCracken and Van Ballenberghe (1987) where correct classification was similar for males and females. In 2014, two redetected individuals made up 24% (n = 11) of the samples incorrectly field-classified as fawn. One of these individuals was fieldclassified and model predicted as fawn in 2013 and would putatively be a yearling in 2014 perhaps providing explanation for the misclassification. The other individual was detected at a site with supplemental feed and field-classified and model predicted as non-fawn in 2013. However, in 2014 this individual was redetected at a site with no supplemental feed and was field-classified as fawn in all samples, but model-predicted as adult in 64% (n = 7) of samples. While based on only a single individual, this strengthens the idea that fed individuals produce larger pellets (Hummel et al. 2008). We recognize the weaknesses of our visual assessment method, including the skill level and subjectivity of observers, but we felt it was valuable to test the method. With in-depth training of observers (as seen in 2013), the accuracy of this method could be improved.

In turn, we recommend measuring pellets when trying to determine age class to increase accuracy. We also note that developing a model based on measurements from wild pronghorn would strengthen predictive power. However, we posit that when obtaining samples from wild individuals is not feasible, samples from captive animals provide a reasonable surrogate (Chapman 2004).

Management implications

In species that are highly sensitive to capture and disturbance, like Sonoran pronghorn, noninvasive genetic sampling methods are an appealing alternative approach for obtaining critical demographic data. Managers of Sonoran pronghorn do not have current sex and age-based survival estimates because these data cannot be obtained with the current aerial survey approach used for population estimation (USFWS 2015). Because we measured samples collected in May and June and relatively soon after fawning, we do not know at what age fawn samples are no longer distinguishable from adults. Thus, it would be beneficial to collect and measure samples several times throughout the year to determine at what age fawns can no longer be reliably distinguished by pellet size. Also, as recruitment for ungulates is typically estimated when young are <6–9 months of age (DeCesare et al. 2012, Christianson and Creel 2014), alternative timing of pellet collection might be preferable. However, fecal pellet DNA sampling for population monitoring (i.e. sufficient sample size) is likely feasible only during the months when pronghorn are congregating at watering holes (April–June) due to the hot, dry conditions (Woodruff et al. 2016). Accuracy of age classification could potentially be improved by developing separate models for sites with and without supplemental feed, as our limited data (not shown) indicates pellets from individuals at sites with supplemental feed are larger. We also note that while seven days of weathering did not alter the pellet measurements, results could differ if pellets are older than seven days. Additional pilot studies are needed to evaluate this. However, other research shows older Sonoran pronghorn pellets have lower DNA amplification success rates, and if the objective is to use these measurement methods in conjunction with individual identification from fecal DNA, pellets older than seven days will be of little use (Woodruff et al. 2015).

While we could not determine sex using morphological methods (data not shown), sex is easily ascertained using genetic methods, and our method shows strong potential for use in wild populations of pronghorn to distinguish fawns from yearlings or adults. The ability to distinguish age structure using fecal pellet measurements greatly improves the applicability of fecal pellet collection and has significant implications for management of the Sonoran pronghorn population. We recommend analyzing pellets using both single width and volume variable models and running the combined model (length + length—width ratio) if the width and volume models disagree. A long-term monitoring scheme combining pellet measurements with individual identification using DNA would allow the identification of fawns that would remain known age individuals for the remainder of the monitoring years. This would provide a powerful tool for documenting the population's age and sex structure, annual recruitment, and age-specific annual survival rates. As research has shown, distinguishing age classes is possible in other ungulate species, and as the use of noninvasive genetic methods increases in ungulates, these methods could potentially be applied to other species.