4 | Discussion
In this study, we found a significant correlation between age and
methylation rate in the gene regions, GRIA2 and CDKN2Ausing a DNA extracted from a non-invasive fecal sample. Although sex was
not affected by the correlation between age and methylation rate, the
methylation rate decreased for the females in nursing states. We also
succeeded in constructing an age estimation model using the methylation
rates of both genes. This study is the first to report the use of
multiple genes and DNA extracted from fecal samples to develop an age
estimation model. We estimated the ages of 19 unknown-age individuals
using our model (Table. S2). The error between the estimated age from
our model and the assigned age based on the year of first birth or
weaning exceeded the MAE after LOOCV was performed. On the other hand,
only one of the five individuals (individual number: #068FA), had a
high residual error, while the average error among the other four
individuals was 6.37 years (range: 4.58–8.53), which is closer to the
MAE after LOOCV. The reasons for the high estimation errors in the
individual is unknown. However, considering that fecal sample collection
was conducted underwater, it is plausible that feces from a different
individual, outside the camera’s field of view might have been
mistakenly collected. Indo-Pacific bottlenose dolphins are known to swim
in groups of 20-50 individuals (Wang, 2018), which increases the
likelihood of accidentally collecting fecal samples from a different
individual. To enhance the accuracy of age estimation using this method,
it is preferable to collect fecal samples multiple times from the same
individual and estimate the age based on those replicated samples.
The age estimation model which exhibited the highest accuracy and
precision were from the methylation rates of GRIA2 andCDKN2A (model 1), with an MAE of 5.08 years. This is 10–13%
(percentage error) of the life span of a Indo-Pacific bottlenose dolphin
which is 40–50 years (Wang, 2018) and it provides a sufficient level of
accuracy for ecological and conservation studies. In a study focusing on
Bechstein’s bats, a similar approach using multiple genes, includingGRIA2 from wing tissues, achieved an age estimation with a
standard deviation of 1.52 years (Wright et al ., 2018). Assuming
a lifespan of 20 years for this species, the error corresponds to 7.6%
of the lifespan. In the case of blood samples of domestic cats, an age
estimation error of 3.83 years has been reported (Qi et al .,
2021). Assuming a lifespan of 12 years for domestic cats, the error
accounts for approximately 31.9% of its total lifespan. Notably,
despite the use of fecal samples, which are associated with lower
precision, the range of error to lifespan was comparable to studies
conducted on other species. However, our model shows lower accuracy
compared to other cetaceans using skin samples (humpback whale: MAE =
3.575, Polanowski et al ., 2014; common bottlenose dolphin: RMSE =
5.14, Beal et al ., 2019; fin whale: MAE = 4.264, García‐Vernetet al ., 2021). In contrast to previous studies that used
pyrosequencing to measure methylation rates with DNA extracted from skin
samples, we used MS-HRM with DNA extracted from fecal samples. The
accuracy of MS-HRM was reported to be similar to that of pyrosequencing
(Migheli et al ., 2013), thus, the differences in detection limits
between the methods is not considered to be the cause of lower accuracy.
However, pyrosequencing allows for the estimation of methylation rate at
individual CpG sites, and only the most strongly correlated sites can be
selected for the estimation model. MS-HRM quantifies the methylation
rate within a certain range that encompasses multiple CpG sites.
Depending on the target regions, it may include several CpG sites with
varying degrees of correlation to age which can potentially lead to a
decrease in accuracy. Previous studies have also reported variations in
the correlation between methylation rates at individual CpG sites with
age (e.g. GRIA2 : 0.48–0.75, CDKN2A : 0–0.44, Bealet al ., 2019). In addition, the fecal samples used in this study
may also include not only the target dolphin’s DNA (Kita et al.,2017), but could also include DNA of intestinal bacteria (Suzukiet al., 2021), and prey species (Kita et al., 2018). The
low concentration of the target DNA may have caused low estimation
accuracy. As mentioned above, as samples were collected underwater,
there is a risk of mixing feces from other individuals that have
defecated nearby. In the future, when similar analyses are conducted
using DNA extracted from fecal samples, it may be possible to solve this
problem by using multiple replicated samples obtained from the same
individual to then detect and discard samples with relatively high
concentrations of target DNA or contamination from other individuals’
feces.
No significant sex differences were reported in the results of the age
estimation model that included GRIA2 for the common bottlenose
dolphins, which is closely related to our target species (Beal et
al ., 2019). Similarly, we did not find significant sex differences
between methylation rate and age in GRIA2 and CDKN2A(Table 3). Comparisons between the models also showed that the model
which excluded sex as an explanatory variable, demonstrated the highest
precision and accuracy. The results suggest that sex may not be
necessary for age estimation. It has been reported that on average,
females tended to have lower epigenetic age than males, and this
association strengthened over the course of human life (Simpkin et
al ., 2016). The sex differences in the rate of aging are consistent
with higher mortality rates and shorter average life expectancy in males
(Crimmins et al . 2019). The tendency for males to have a shorter
lifespan is a common phenomenon observed in mammals (Lemaítre et
al., 2020). In non-human mammals, a small number of species have
reported sex differences between DNA methylation rate and age in the
epigenetic clock (common bottlenose dolphin: Beal et al ., 2019;
domestic cat: Qi et al ., 2021). While many other mammalian
species have reported that there were no significant sex differences
found (humpback whale: Polanowski et al ., 2014; fin whale:
García-Varnet et al., 2020; beluga: Bors et al., 2021;
brown bear: Nakamura et al ., 2023). Most of these studies often
do not investigate sex differences in DNA methylation rates for each
gene but rather examine the significance of sex as one of the
explanatory variables in age estimation formulas that utilize multiple
genes. As a result of formulating estimation models with multiple genes,
the impact of sex may be concealed and overlooked. To address this
issue, further research focusing on the relationship between sex
differences and methylation rate for each gene is needed.
This is the first study to suggest the effect of female nursing states
on methylation rate and age in GRIA2 , and CDKN2A .GRIA2 are family of receptors (Henley & Wilkinson, 2013), whileCDKN2A encodes for several tumor suppressor proteins (Foulkeset al ., 1997; Zhao et al ., 2016). From a functional
perspective, it is difficult to consider the exact factors that could
lead to the decrease in methylation rates during the calving period.
However, a decrease in methylation could be associated with specific
physiological changes that occur during suckling and other parenting- or
calf-nursing-related activities. Although age-related changes in
methylation rate decreased in females specifically, no significant sex
differences were found overall. Therefore, it is possible that female
methylation rates are adjusted after calving by unknown factors, and
further studies examining the longitudinal variation in methylation
rates of individuals are needed in the future. The female nursing states
did not contribute to the precision and accuracy of the age estimation
model. This may be due to the different regression methods employed. The
ANCOVA was used to examine the sex differences in methylation rate
changes with age for each gene based on the least squares method. This
calculates the regression line by minimizing the distance from all data
plots on the scatterplot. On the other hand, the SVR used in building
the age estimation model in this study, is based on the maximum-margin
principle, where the regression line is determined by maximizing the
distance from the outer plots on the scatterplot. These conceptual
differences between the regression methods suggest that an effect
observed in one method may not be detectable in the other. It is
recommended that future analyses of similar studies should take this
effect into account.
Conventional age estimation has been based on counting the growth layers
formed on the tooth cross section in odontocetes (see Perrin & Myrick,
1980). This method requires capturing of individuals to collect their
teeth. In recent years, there have been attempts to develop an age
estimation method using epigenetic clock analysis in various taxa (e.g.
humans: Horvath, 2013; Bechstein’s bats: Wright et al ., 2018;
Asian elephants (Elephas maximus ) & African elephants
(Loxodonta africana ): Prado et al., 2020). All previous
studies of epigenetic clock analysis on cetaceans have used skin
samples, which require biopsy surveys or capture procedures to be
conducted in the wild (Polanowski et al ., 2014; Beal et
al ., 2019; Tanabe et al ., 2020; Bors et al ., 2021; Peterset al ., 2023). Biopsy surveys are generally considered less
invasive compared to capture methods; however, it is worth noting that
there have been instances of mortality, especially in small cetaceans
(Bearzi, 2000; Noren & Mocklin, 2012). Our method allows for
non-invasive age estimation as fecal samples were collected underwater,
without touching and disrupting the dolphins. As stated by Qi et
al., (2021), pyrosequencing is the gold standard method for quantifying
DNA methylation rate. However, each analysis requires 3–4 hours for
completion and costs $14, while MS-HRM offers a more cost-effective
alternative, with each analysis requiring only two hours and costing
$7. This cost-effectiveness, combined with the shorter turnaround time,
makes the MS-HRM method highly suitable for implementation in various
research sites.
The framework of this study can be extended to other cetacean species
and populations where fecal samples can be collected. For instance, the
Atlantic spotted dolphins (Stenella frontalis ) around the Bahamas
and the dwarf minke whales (Balaenoptera acutorostratasubsp.) in Australia have conducted underwater identification surveys
(see Herzing, 1997; Birtles et al., 2002). These populations have
favorable conditions for collecting fecal samples. In addition, fecal
samples of large cetaceans can be collected whilst being on board (see
Smith & Whitehead, 2000; Reidy et al., 2022). Identification
surveys on gray whales (Eschrichius robustus ), North Atlantic
right whales (Eubalaena glacialis ), Southern right whales
(E. australis ), sperm whales (Physeter macrocephalus ),
blue whales (B. musculus ), and humpback whales have been
conducted using natural marks such as color patterns and the shape of
flukes (Hammond et al., 1990). Thus, it may be possible to
introduce age estimation using fecal samples as well. Even in areas
where long-term individual identification surveys have not been
conducted by researchers, activities such as swim-with-dolphin programs,
scuba diving, and wading in close proximity to observation targets have
been developed in more than 11 genera of cetaceans, over 54 areas in 32
counties (Carzon et al., 2023). Therefore, there is potential for
collaboration with the tourism industry in these areas, where fecal
samples could be collected in conjunction with tourism activities. This
synergy between research and tourism allows for the collection of fecal
samples while visitors engage in educational and conservation efforts.
Terrestrial mammals may get better results because hydrolysis is less
likely to occur on land. If the fecal samples-based age estimation can
be applicated to terrestrial animals, it may lead to benefits for both
study species and researchers because fecal samples may be collected
non-invasively even in species that are difficult and/or dangerous to
encounter.
The successful quantification of methylation rates using fecal samples
of Indo-Pacific bottlenose dolphins in this study, suggests the
potential applications of age estimation using the same genes in other
cetacean species. This study serves as a steppingstone towards the
widespread application of non-invasive age estimation methods in various
mammal species, offering valuable contributions towards the
understanding of their ecology through age-related information.