This study investigated how different age indicators derived from blood assays, MRI brain scans, other-ratings, and self-reports converge and diverge within older adults and how these constellations change with age. Overall, we only identified moderate correlations within specific domains of aging indicators, both cross-sectionally and longitudinally. In contrast, associations across domains were infrequent and modest in size. The pattern was similar regardless of whether cross-sectional or longitudinal data were used.
Given our study’s design, we can only draw inferences about sample-level associations. It is possible that a different pattern of findings emerges when one considers subgroups of people, particularly those at risk for functional decrements and impending death. We note that in follow-up analyses, we corroborated the general pattern of results among those who suffered from two and more chronic physical diseases at follow-up and excluded supposedly healthy participants. At the same time, the BASE-II sample is still relatively healthy, so results may not generalize to participants in poor and debilitating health conditions. One way to interpret our findings is thus that the diverse and loosely interconnected pattern of aging indicators may be a defining characteristic of normal if not healthy aging. It might only when these normal aging processes are influenced by secondary aging, involving pathological changes, and tertiary aging, culminating in mortality, that the intricate and heterogeneous profile transitions into a more uniform and homogeneous pattern characterized by generalized broad-based declines42,43.
Research on the associations between psychological and biological indicators of aging is extremely scarce. Even though a number of studies have highlighted the importance of physiological health for subjective age and vice versa 31–33, we are unaware of any study examining links between subjective and biological/epigenetic age indicators. Interestingly, stronger epigenetic age acceleration was associated with a narrower subjective health horizon, consistently at baseline and follow-up seven years later. It has been argued that an individual’s sense of aging can be shaped by aggregate cellular stress34. Individuals who have undergone challenging life experiences may perceive themselves as prematurely aged due to the cumulative impact of subclinical deterioration and allostatic load. Research has demonstrated that the allostatic load, which reflects the wear and tear on the body resulting from prolonged exposure to stressors, is linked to epigenetic age measures35. Conversely, psychological stress, a significant contributor to allostatic load, has been found to be associated with an older subjective age36–38. Also, it was surprising to see that skin age was not strongly linked to subjective age even though one might assume that visible signs of aging would be internalized and affect how old someone feels39–41. The validity of the subjective age measure used in this paper has been shown in a number of studies32,42, and so has the validity of the skin age measure. We note that skin age was estimated based on the number and distribution of lentigines at the back of the hands; possibly, the occurrence of facial lentigines may be more strongly associated with subjective age ratings.
We also note that there is currently no gold standard on how to measure biological age. Therefore, we analyzed a number of different aging indicators many of which represent the current state-of-the-art in quantifying the pace of aging. At the same time, we acknowledge that these measures do not represent an exhaustive list of biological changes that evolve with aging and also come with a number of technical and methodological limitations 26,64,65. For example, the variables analyzed often index processes that evolve on different (time) scales. For example, some markers like telomere length (DNAmTL) track the actual progress of age, whereas others are calculated by taking the difference between the measured (=„biological“) age and the chronological age (e.g. DNAm age acceleration). This means that some measures are tied to chronological age, whereas others are not associated with chronological age. We also note that the age markers compared here are different in the underlying conceptual rationale and were in part optimized using different criteria. For instance, some biomarkers were developed and optimized to predict future mortality and others were developed and optimized to estimate concurrent chronological age or biological age. Such conceptual and methodological differences may reduce the strength of associations, yet one would presumably still expect to find sizeable interrelations.
To our knowledge, no study so far has examined longitudinal associations between changes in multidomain indicators of aging. Our findings indicate that correlations between changes in age indicators were very small or absent. For example, changes in epigenetic age measures were not related in statistically significant ways to changes in any other indicators of aging. Such absence of across-domain associations was unexpected because it appears plausible that people who show an age acceleration in one domain of functioning (as proxied by one set of alternative age markers) should show similar trends in other domains of functioning (as proxied another set of alternative age markers). These findings raise intriguing questions about the underlying mechanisms and methodological considerations that may contribute to the lack of association.
One explanation could be the inherent complexity of aging processes themselves. Aging is a multifaceted phenomenon influenced by a variety of different biological, environmental, experiential, and behavioral factors, each contributing to different aspects of people’s aging trajectories. It is plausible that changes in epigenetic age measured from leukocyte DNA may capture distinct biological processes that are not fully reflected in other measures of aging, resulting in divergent trajectories over time. Another methodological consideration is the measurement sensitivity and specificity of the different aging indicators. Each indicator represents a different facet of aging, and their sensitivity to capture changes may vary. It is, thus, possible that the specific metrics used to assess epigenetic age might not fully align with the changes observed in other indicators, leading to the lack of association between them. Furthermore, the time frame and duration of the study may have influenced the observed associations. Aging encompasses dynamic processes that evolve over extended periods of time, and changes in aging indicators may not manifest in synchronous ways. Longitudinal studies with longer follow-up periods and more frequent assessments may provide a more comprehensive understanding of associations between changes in different aging indicators. For example, it remains unclear whether changes in aging indicators follow a linear trajectory or exhibit a non-linear patterns. Trajectories might not be perfectly aligned and thus the timing of examinations will impactwhether or not associations are found. It is also possible that the strength of correlations changes across the lifespan. Because aging involves cumulative processes, the impact of certain biomarkers may intensify and magnify in older ages and longitudinal associations between these biomarkers may be more likely to be observed towards the end of life and less so in midlife or earlier phases of old age.
Study Limitations and Outlook
In closing, we note the following limitations of our study design, measures, and samples. First, not all of the clocks determined at baseline had been assessed at the same point in time. For example, on average there were more than two years in between assessments of the epigenetic clocks and brain age (M = 2.28, SD = 1.19). We note, however, that we observed similar patterns of non-correlations for clocks that had been assessed within just a few days (i.e., participants filled out the questionnaire data in-between the two medical sessions during which blood samples were taken). It, thus, appears unlikely that the non-simultaneous assessment of some of the measures constitutes a major contributing factor to the results obtained here. We also note that our study design does not allow to examine trajectories of markers of aging and how they are associated with one another. Given that we only had two timepoints available over the course of on average seven years, we cannot draw further conclusions about the speed and shape of change. For example, it is possible that certain health events during the seven-year period have led to an accelerated decline in some individuals, whereas others might have experienced a steady change throughout the seven years. Future research needs to further examine the shape of aging trajectories.
Changes in some biomarkers may serve as antecedents or predictors of changes in other biomarkers. For instance, genetic and epigenetic biomarkers may precede or influence the development of clinical or psychological biomarkers of aging (e.g., older epigenetic and BioAge might precede older skin age). It will be interesting to assess such time-ordered associations in future work.
Similarly, consider the potential scenario where epigenetic biomarkers of aging, such as specific DNA methylation patterns, precede and influence the development of clinical or psychological biomarkers of aging. For instance, alterations in DNA methylation patterns associated with accelerated aging may be antecedents to cognitive decline or the manifestation of age-related diseases. While such associations were reported in the literature, no statistically significant associations were found between epigenetic clocks and a number of functional, geriatric and cognitive assessments or frailty in BASE-II before43,44. In this case, understanding the antecedent biomarkers can provide valuable insights into the underlying mechanisms of aging-related conditions and potentially guide targeted interventions for prevention or early intervention. Hence, identifying the specific biomarkers that have a stronger association with specific clinical outcomes becomes crucial for effective clinical management and personalized interventions.
We note that this example represents only a single potential scenario, and the relationships and temporal dynamics between different biomarkers of aging can be complex and context-dependent. The interplay among biomarkers and their time-ordered or even causal pathways can vary based on the specific aging process or disease under investigation. Thus, generalizing results from one scenario to other biomarkers or aging processes should be done with caution. To gain a comprehensive understanding of the intricate associations between biomarkers of aging, longitudinal studies with large and diverse populations are needed that elucidate the complex interdependencies and causal pathways among various biomarkers. Integrating multiple levels of data, including epigenetic, proteomic, and clinical measures, will likely provide a more comprehensive picture of the aging process43,45. Also, the aim of this paper was to describe interrelations among multi-domain age indicators. To better understand aging processes, one next step will be to examine their unique, conjoint, and interactive effects in predicting clinical outcomes. For example, studies have highlighted the predictive validity of alternative age indicators for substance use behaviors, atherosclerosis, cancer, and overall mortality46–50. It remains a largely open question though which of a multitude of biomarkers are most predictive of which outcome when examined together. We also acknowledge that while more comprehensive that most other studies our selection of biomarkers was limited by dataavailability. Importantly, the field of biomarkers of aging is still evolving, and ongoing research continues to uncover new biomarkers and further elucidate their role in aging processes.
Finally, we used data from the sample of the Berlin Aging Study II, which is restricted on several dimensions. First, the dataset was recruited within a geographically restricted area encompassing Berlin and its surroundings. Specifically, our sample lacked racial diversity, making it challenging to quantify the generalizability of our results. As a consequence, we can draw no inferences on other, more diverse populations. This remains an important limitation as previous work has suggested that there are disparities in aging patterns among different racial and ethnic groups. For instance, research indicates that non-Hispanic Blacks and Hispanics may exhibit accelerated aging, while non-Hispanic Whites may experience decelerated aging51. At the same time, it is important to acknowledge that these racial and ethnic differences are intricately linked to other factors, such as educational attainment (REF). Thus, it is important for future research to replicate and validate our findings in more geographically and racially/ethnically diverse samples.