Epigenetic clocks may come out of rhythm—implications for the estimation of chronological age in forensic casework

The effects of adverse living conditions have been addressed by numerous publications. Cumulative lifetime stress results in an increased epigenetic age measured via Horvath clock [14]. This effect was especially pronounced in advanced age. Moreover, severe stress-related diseases (psychotraumata or post-traumatic stress disorder (PTSD)) may lead to increased epigenetic age estimation. Boks et al. [15] found increased epigenetic age estimates in 96 traumatized male soldiers based on the application of the Horvath clock. Wolf and colleagues [16‐18] investigated the effects of childhood or lifetime trauma: Regarding the Horvath clock, they did not find effects of PTSD and childhood trauma on epigenetic age. This finding was supported by Mehta et al. [19]. In contrast, a correlation between increased epigenetic age estimation and childhood trauma as well as lifetime PTSD severity was evident with the Hannum clock. The data presented in a meta-analysis [20] hint to a higher impact of traumata experienced during childhood compared with those experienced in adulthood.

A low socioeconomic status and/or education level may also be associated with increased epigenetic age estimations. Chen et al. [21] investigated epigenetic ages of 379 African American adolescents in Georgia by analysing DNA from leukocytes. The epigenetic age of those individuals was increased by 1.42 years per measure of economic adversity. Another group [22] reported increased epigenetic age estimates in black middle-aged women with low income. They also investigated “the extent to which various health-related behaviours such as diet, exercise, smoking, alcohol consumption, and having health insurance could explain the effect of income on aging”, but did not find “significant relationship between these variables and the speed to which aging occurred, and controlling for them had no impact on the association between income and biological aging.” Fiorito et al. [23] described increased epigenetic age estimation in individuals with low socioeconomic status. The authors pointed out that this effect could be ameliorated by increasing socioeconomic status during life course; however, epigenetic age estimates still remained higher in those individuals compared with those with higher socioeconomic status. Jovanovic et al. [24] investigated DNA isolated from saliva of children and found an association between increased epigenetic age estimates and low socioeconomic status. These results are supported by the findings of Hughes et al. [25], who investigated the epigenetic age of individuals whose parents had been in semiskilled or unskilled occupations when the investigated individuals were 14 years old. At an age of 26 years, these individuals exhibited an increased epigenetic age of 1.07 years. In individuals without a working parent at an age of 14 years, this effect was even more severe, resulting in an increase of epigenetic age estimates of 1.85 years at an age of 26 years. The authors state that these “differences were not explained by smoking, adiposity or alcohol consumption, suggesting mechanisms independent of health behaviours are involved.” These results are again supported by the data of Austin et al. [26], who reported an increase in epigenetic age in individuals with low early life socioeconomic status, as well as by Fiorito et al. [27], Tajuddin et al. [28], Thurston et al. [29] (regarding education) and Marini et al. [30] (regarding children), each of them with application of varying epigenetic clocks. Fiorito et al. [27] stated again that the increases of epigenetic age with low socioeconomic status were mainly independent of other lifestyle-related risk factors like smoking, obesity, alcohol intake, and low levels of physical activity. It is not clear yet, why a low socioeconomic status alters epigenetic age. Some groups assumed that this might be due to the stress of experiencing economic hardship, and that changes of DNA methylation are mediators of the association of low socioeconomic situation and a higher risk of diseases (e.g. cardiovascular diseases) [24, 26, 72, 73].

In contrast, high socioeconomic position or education levels have been shown to be associated with reduced epigenetic age estimates [31]. McCartney et al. [32] reported that each degree of increase in education level or socioeconomic status led to 0.05 years decrease of epigenetic age estimates. Additionally, Quach et al. [33] stated that higher education is associated with a decrease in epigenetic age estimates (4.14 years) and also described an inverse relationship between income and epigenetic age.

Exposure to violence was also suggested to impact epigenetic age. Based on the analysis of DNA isolated from saliva of children, Jovanovic et al. [24] reported an association between an increase of epigenetic age estimates and experiences of violence in a dose dependent manner. Similarly, Sumner et al. [34] found increased epigenetic age estimates and an advanced pubertal stage in children with threat-related early life adversity (e.g. violence). These results were supported by Marini at al. [30], who found “that exposure to abuse, financial hardship, or neighbourhood disadvantage during sensitive periods in early and middle childhood (…) (led to a) deviation of Hannum based epigenetic age from chronological age, even after considering the role of adversity accumulation and recency”. Regarding Horvath clock, they did not find differences between estimated ages of children with or without exposure to violence. Brody et al. [35] described the effects of exposure to higher levels of racial discrimination on epigenetic age estimates and reported increased epigenetic age estimates in discriminated adolescents with less supportive families (known to ameliorate the effects of exposure to racial discrimination).

In forensic case work, age estimation in young migrants without valid identity documents has become highly relevant in these times of migration and flight [74]. In this context, the possible impact of adverse living conditions on epigenetic age estimation is of major importance. If an accumulation of stressors like experiences of violence and low socioeconomic status may cause increased epigenetic age estimates and, consequently, a false high age estimate, this may have serious consequences for the individual (e.g. legal responsibility).

Spolnicka et al. [11] tested an epigenetic model composed of five markers from five genes (ELOVL2, C1orf132, KLF14, FHL2 and TRIM59) for the estimation of chronological age in three disease groups (late and early onset Alzheimer’s disease, Graves’ disease). They reported aberrant hypermethylation and decreased prediction accuracy of chronological age for TRIM59 and KLF14 markers in the group of early onset Alzheimer’s disease. In Graves’ disease patients, an aberrant hypermethylation was observed for TRIM59, an aberrant hypomethylation for FHL2. In contrast, ELOVL2 and C1orf132 showed unchanged prediction accuracy in all disease groups. Analysis of the identical five markers in 39 blood samples from patients with chronic lymphocytic leukaemia (CLL) and 92 healthy individuals (control group) resulted in highly statistically differences between patients and controls for all CpGs, indicating a strong influence of CLL on age-related methylation [12]. The authors concluded that “DNA methylation signature in blood does not predict calendar age in patients with chronic lymphocytic leukemia”.

Altered DNA methylation patterns were not only observed in cancer which affects blood cell lines [12], but also in other common types of cancer like colorectal, gastric, kidney, lung, prostate and urothelial cancer [2, 36‐41]. Dugue et al. [36] found that “epigenetic aging was associated with increased cancer risk, ranging from 4% to 9% per five-year age acceleration”. Several other publications reported associations between cancer/cancer risk and an increase of epigenetic age estimates in blood of up to 2.5 years [37‐41]. In fact, cancer is a monoclonal disease and hence the malignant cells capture the epigenetic makeup of the tumour initiating cell [75]. There is even evidence that for many types of cancer, the acceleration of epigenetic age is of prognostic value for disease development [76]. However, it is unknown so far if increased epigenetic age is a cause or consequence of cancer development. The aforementioned data are also of relevance for forensic casework: Blood samples of persons suffering from cancer may exhibit altered DNA-methylation levels that might result in false high age estimates. This may also be true if the cancer already exists, but is not diagnosed yet.

Based on the DNA-methylation pattern in blood, epigenetic age of individuals infected with HIV was estimated 5.2 years higher compared with healthy controls [45]. These results were confirmed by Gross and colleagues [46]. Similarly, H. pylori and Cytomegalovirus (CMV) infections were associated with increased epigenetic age estimates [47, 48]. In forensic casework, the impact of those infections on epigenetic age estimates has to be kept in mind, as they might result in deviations between estimated and chronological ages of up to 6 years. This point is of high importance regarding age estimation of unaccompanied young refugees without valid documents, as many refugees stem from African regions with high HIV-incidence, and additionally CMV infections are very common with up to 95% incidence in adults in many countries [77].

Obesity, high body mass index (BMI), and blood cholesterol have been shown to increase epigenetic age estimates. Nevalainen et al. [42] found increased epigenetic age estimates in middle-aged individuals with high BMI. These findings were supported by data of Simpkin et al. [43], McCartney et al. [32], Li et al. [44] and Thurston et al. [29]. Recently, Fiorito et al. [27] published data from a large multi-cohort study (n = 16,245), revealing increased epigenetic age estimates of up to 1.08 years (p < 0.001) for individuals with high BMI (BMI ≥ 30). Individuals with lower BMI may exhibit low epigenetic ages [33].

Moderate alcohol consumption (1–7 drinks/week) was suggested to decrease epigenetic age [33]. In contrast, habitual alcohol consumption [27] and alcohol dependence [49] were associated with an increase in epigenetic age. These data were recently supported by Luo et al. [50].

There is clear evidence that smoking evokes specific DNA methylation changes [78]. The data available for the impact of smoking on epigenetic age estimations are still inconsistent. Some groups did not report any effects [33, 40, 51, 52]. Studies comprising larger sample sizes report increased epigenetic age estimates in smokers: McCartney et al. [32] investigated DNA from 5100 blood samples and found increased epigenetic age estimates in smokers. Fiorito et al. [27] investigated 16,245 blood samples and reported that smoking was associated with an increased epigenetic estimate of up to 1.57 years. Additionally, Yang et al. [53] reported that cumulative smoking (pack-years) was significantly associated with epigenetic acceleration.

Even exposure to environmental toxins has been shown to affect epigenetic age. Ward-Caviness et al. [54] reported an increase of epigenetic age with increased exposure to fine particulate matter and nitrogen oxide. Similarly, Li et al. [55] as well as Nwanaji-Enwerem et al. [56, 57] found increased epigenetic ages after exposure to black carbon or fine particulate matter. Lind et al. [58] described increased epigenetic age estimates in individuals exposed to pesticides. In contrast to this, White et al. [59] found a deceleration of the epigenetic age after exposure to NO₂. However, the age acceleration effect of fine particulate matter varied from 2 to 6 years and depends on its composition.

In age estimation models comprising of high numbers of markers, males are estimated older than females of identical age. This finding was first described by Hannum et al. [3], who found that—on the epigenetic level—men age approximately 4% faster than women. These results were supported by the findings of other studies based on different models for age estimation [28, 32, 60, 61]. Notably, no difference in epigenetic age estimates between men and women were reported in one study based on a model comprising of only 8 CpGs [52].

Cho et al. [9] applied an epigenetic model for the estimation of chronological age (markers located in the ELOVL2, C1orf132, TRIM59, KLF14, and FHL2 genes) derived from a Polish population [63] to blood samples from 100 Koreans. They reported that the age predictive performance of the tested model was “relatively consistent across different population groups”. However, at certain loci (FHL2, C1orf132, KLF14) the extent of the age association in Koreans was not identical to that of the Polish, and retraining of the models produced better prediction accuracy. Several other publications describe differences in epigenetic age estimates between same aged individuals of different ethnic groups, too [28, 29, 60, 62]. As an example, Hispanics and Tsimane Amerindians have been shown to exhibit accelerated epigenetic ageing compared with Caucasians, while African-Americans were suggested to have decelerated epigenetic ageing compared with Caucasians and Hispanics [60].

Facing a growing list of factors influencing DNA methylation, the identification of relevant factors in the forensic context (estimation of chronological age) is of great importance. A factor is relevant, if it may cause significant and systematic deviations of epigenetic age estimates from chronological age and if it is likely to be present in forensic cases.

A first step towards the identification of relevant factors can be a pre-selection by analysis of published literature including clinical and basic science data, as done in this review. Taking into account the data in Tables 1 and 2, such pre-selected factors could be—for example—CMV infections (differences between epigenetic and chronological ages of ~ 5 years, high incidences especially of CMV infections).

In a second step, the influence of these factors on the quality of age estimation by models used for forensic age predictions can be tested by a targeted investigation of populations bearing possible confounding factors—a strategy that has already been chosen by Spolnicka et al. to test the relevance of physical activity and selected diseases [10‐12].

However, some confounding factors may be interrelated (e.g. BMI, nutrition, low fitness and low socioeconomic status) and the individual impact of each of these interrelated factors can possibly not be differentiated clearly, since an independent investigation of them appears difficult.

The identification of forensically relevant factors, however, does not solve the question of how one can recognize if such a factor has to be taken into account in the individual forensic case. This problem arises especially in postmortem cases (unidentified deceased) and the analysis of blood traces of unknown donors for crime scene investigations. Spolnicka et al. [12] proposed a classification model for the identification of patients with chronic lymphocytic leukaemia (which has a strong impact on DNA methylation), deduced from the DNA methylation data. In living individuals, age estimation should consider the detailed medical history; specific questions should be asked about possible factors of influence.

Models that are as robust against the influence of confounding biological factors as possible have to be developed to minimize the influences of the identified confounding factors on the quality of forensic age estimation.

Theoretically, models that include large numbers of CpGs could have the advantage that the influence of confounding biological factors on some markers may be “compensated” by an unaffected DNA methylation of other CpGs. However, even “large clocks” may be influenced by confounding biological factors (see “Horvath clock” (353 CpGs) and “Hannum clock” (71 CpGs) in Tables 1 and 2). The same is true for epigenetic models composed of a small number of CpGs, as shown for example by Spolnicka et al. [10] (5 CpGs). Within such models, some DNA methylation markers may be more sensitive towards confounding biological factors than others may, as shown for example for ELOVL2 and C1orf132 in three disease groups [11]. Focusing on a selected number of such “robust” CpGs in targeted assays may be a promising approach for the development of “robust” models for epigenetic age estimation; such “specialized” clocks [13, 80] may be much more powerful for forensic age estimation. This approach will result in models composed on only few CpGs, which will also be applicable in settings with low amounts of DNA of good quality. Another approach may be the consideration of identified confounding factors as parameters in prediction models.

The aim of optimization of epigenetic age estimation poses many challenges. Apart from aspects like the quality of DNA, applicability and suitability of methods, predictive markers, and integration of DNA methylation measurements into predictive models [80], it is important to elucidate how biological confounding factors affect epigenetic age-predictions. This interplay needs to be understood for reliable applications in forensic contexts. Collaborative research with coordinated research strategies is required to address this multitude of open research questions and to improve epigenetic methods for the estimation of chronological age.