Discussion
To the best of our knowledge, this is the first study to establish a method for epigenetic age estimation in bears. We constructed a single regression model using one CpG, designated “SLC12A5-4”, which showed the strongest correlation between age and methylation level. Additionally, we constructed six elastic net regression models and six SVR models using various combinations of 13 CpGs. Evaluation based on the MAE of LOOCV suggested that the best SVR model is the model using four CpGs, designated “SLC12A5-1, -2, -3, and -4”. This model had the smallest median absolute error (MedAE) and the second smallest root mean square error (RMSE) for LOOCV. The four CpGs included in this model can be covered by a single PCR amplification, suggesting that this model provides the greatest benefits in terms of both accuracy and cost-effectiveness among all tested models. Although the best model in this study targeted only four CpG sites, the accuracy of the model was comparable with that of models developed previously using genome-wide approaches (targeting more than 30 CpG sites) in other carnivore species with shorter life spans than bears. For example, the MedAE values were approximately 0.8 years in studies of dogs, wolves (Thompson et al. 2017), and cats (Raj et al. 2021), whose life spans are approximately one-half to two-thirds of the brown bear lifespan. Furthermore, the DNA methylation levels in seven individuals that were sampled multiple times showed similar increasing trends to those of other samples included in the model. These repeated measurements were all obtained from wild individuals, and the data are valuable because changes in DNA methylation levels were tracked in the wild under highly variable environmental conditions. As shown in Table 1, the MAE, MedAE, and RMSE values were small even when age estimation was performed for wild samples that were not included in the model, indicating that the models we constructed were highly accurate.
The current method is superior to other commonly used age estimation methods in bear species in the following respects. First, this method provides accurate and precise age estimates. Traditionally, tooth-based age estimation has been commonly used for bears (Mundy & Fuller 1964; Marks & Erickson 1966; Stoneberg & Jonkel 1966). Ages estimated using this method are 80–90% consistent with the actual age of the animal. However, if the observer is inexperienced, the accuracy and precision of age estimation are reduced (Mclaughlin et al. 1990). In addition, the error is greater for older than younger bears (Mclaughlin et al. 1990; Harshyne et al. 1998; Costello et al. 2004), as the annuli of older bears are less distinct and interpretable. The current method based on pyrosequencing can avoid such human errors and overcome the technical difficulties facing age estimation. Second, this method is less costly in terms of time, money, and human resources compared with traditional tooth-based age estimation, which requires multiple steps, including decalcification, neutralization, section preparation, staining, and counting of cementum annuli (Tochigi et al. 2018). Depending on the protocol, these steps may take several days (Matson et al. 1993). Third, the current method requires only 100 µL blood, which is much less invasive to the animal than removal of a tooth. Additionally, blood sampling is much easier than tooth removal, as pulling the tooth without breaking the root requires skill (Costello et al. 2004).
In the best model identified in this study, sex had no effect on Δage (predicted age − chronological age) or |Δage| (absolute difference between predicted age and chronological age). This finding indicates that no sex difference existed in age-dependent DNA methylation changes or individual variability and further suggests that our method can be applied regardless of sex. Among humans, DNA methylation levels change faster in males than in females (Nussey et al. 2013). Adult males have shorter lifespans than those of adult females in many long-lived vertebrates (Clutton-Brock & Isvaran 2007), although it remains unclear whether the shorter lifespans of males are associated with faster epigenetic aging (Hägg & Jylhävä. 2021). Notably, we had small sample sizes for males, especially for wild bears. In contrast to females, which are philopatric, males born in our study region leave the area at the age of 2–3 years (i.e., dispersal behavior (Blanchard & Knight 1991; Shirane et al. 2019)), hindering sample collection from adult males of known ages. In addition, only three captive males over 10 years of age were included in the analysis. Therefore, further study is needed to determine the influence of sex on DNA methylation.
The error represented by |Δage| was larger for wild bears than for captive bears (Table 1), indicating that individual differences in DNA methylation are greater in wild than captive bears. One possible reason for this difference is that captive bears are fed the same type and quantity of food throughout the year in a stable environment, whereas wild bears consume a variety of foods in differing quantities depending on the season (Naves et al. 2006; Shirane et al. 2021). Additionally, foraging strategies differ among individuals even within a population (Servheen & Gunther 2022; Jimbo et al. 2022). Factors linked to lifestyle, including obesity, weight reduction, and overfeeding, have been suggested to affect DNA methylation (Samblas et al. 2019; Yamazaki et al. 2021). Wild brown bears show cyclical annual body mass patterns, with a continuous decrease from the beginning of winter hibernation to summer, and a rapid increase during autumnal hyperphagia (McLellan 2011; Schwartz et al. 2014). In addition, the annual fluctuation in food availability affects body condition (Shirane et al. 2021). The dietary diversity of wild bears and annual fluctuations in food availability may cause greater individual differences in DNA methylation levels compared with captive bears.
Interestingly, Δage tended to decrease with age in wild individuals. This result indicates that older wild individuals have lower age estimates and contradicts the fact that captive animals generally live longer than wild animals (Müller et al. 2010; Lemaître et al. 2013). Animals with short lifespans and high reproductive rates are reported to live longer in captivity than in the wild, but this trend is not always true for species with long lifespans and low reproductive rates (Tidière et al. 2016). Brown bears have a long lifespan and low reproductive rate, which may explain the current results. Another possible explanation is the presence or absence of hibernation periods. Wild brown bears spend 3–7 months hibernating in a reduced metabolic state (González-Bernardo et al. 2020), whereas captive bears in Noboribetsu Bear Park are fed throughout winter and do not hibernate. In general, hibernators live longer than similar-sized non-hibernators (Wilkinson & South 2002; Turbill et al. 2011; Wilkinson and Adams 2019). Recent studies of yellow-bellied marmots, Marmota flaviventris , (Pinho et al. 2022) and big brown bats, Eptesicus fuscu s, (Sullivan et al. 2022) suggest that this difference arises in part because hibernation slows epigenetic aging. Similarly, in bears, the low metabolic state characteristic of hibernation (Tøien et al. 2011) may reduce the rate of DNA methylation changes, lowering the epigenetic ages of aged individuals. However, this study included a limited number of wild bears, especially aged bears. Further study is needed to clarify this issue.
Among the four genes located adjacent to the CpGs whose methylation levels showed significant correlations with age in bears, SLC12A5and POU4F2 have been reported to show similar correlations in other carnivores (Ito et al. 2017; Raj et al. 2021). VGF andSCGN show changes in expression during aging at the protein level in T lymphocytes (Busse et al., 2014) and at the mRNA level in blood mononuclear cells (Tan et al. 2012) in humans, although whether those changes are due to changes in the DNA methylation level remains unknown. The gene SLC12A5 , located adjacent to the CpG site whose methylation levels were used for the best age estimation model, encodes an integral membrane KCl co-transporter that regulates cell volume, net trans-epithelial salt movement, and maintenance of a low intracellular Cl concentration in neurons (Payne et al. 1996). As far as we know, age-related changes in expression at the mRNA or protein levels have never been reported. In humans, the DNA methylation level of the CpG site proximal to this gene is a frequently used target for age estimation in a variety of samples, including blood, saliva, buccal swabs, and hair (Florath et al. 2014; Hong et al. 2017; Hao et al. 2021; Schwender et al. 2021). This finding suggests that age-dependent methylation changes at CpG sites adjacent to SLC12A5 occur in a tissue-independent manner. In addition, CpGs located in the promoter region of SLC12A5 showed strong age-related methylation changes in cats (Raj et al. 2021). Although no positive relationship between age and the methylation level of this site was found in dogs (Ito et al. 2017), the present findings suggest that CpG sites adjacent toSLC12A5 are a useful methylation marker for age estimation in other mammals.
In conclusion, we established an epigenetic age estimation model for brown bears using SVR models and obtained an MAE value of 1.3 years. This value is comparable with those from models established for other animals. The current method is more accurate, easier to perform, and less invasive than conventional tooth-based methods. Notably, our long-term field study enabled the establishment of age estimation models for bears, including both captive and wild bears. Furthermore, this age estimation model may be applicable to other bear species, such as Asian black bears (Ursus thibetanus ), American black bears (Ursus americanus ), and polar bears (Ursus maritimus ). The current study will contribute to ecological research, conservation, and management of bear species.