Recently, several news outlets latched onto some interesting data reported this summer in Nature Aging: aging isn’t linear, but instead occurs in two sharp peaks over the course of human adulthood, corresponding approximately to ages 44 and 60.1 Great, another reason to dread the big “6-0.”
We know that aging is caused by changes at the molecular level that lead to downstream effects like disease and degeneration. Fully characterizing these changes is therefore an important step in understanding the aging process and potentially developing interventions to slow its progression. Thanks to the advent of methodologies to collect and analyze “big data” in biology, scientists are more empowered than ever to investigate these molecular underpinnings of aging, and thus, researchers Shen et al. undertook an ambitious study of millions of metrics in order to understand how they change over a human lifetime. Their findings – as reported in mainstream media – point to two distinct bursts of aging-related changes, but how should we interpret these data?
What they did
Researchers collected multiple samples from 108 U.S. subjects between the ages of 25-75 (51.9% female) for a median period of 1.7 years. Samples of blood, stool, and swabs of skin, oral mucosa, and nasal mucosa were collected every 3-6 months from all participants for as long as they were healthy, and Shen et al. used these samples to collect various “-omics” data (e.g., transcriptomics, metabolomics, proteomics) as well as microbiome data for respective tissues and typical laboratory tests such as hemoglobin A1C (HbA1C), a lipid panel, a complete blood count, and others. Defining baseline as data from the youngest participants (those between the ages of 25-40 years), the authors compared dysregulated molecules in each age group to the baseline and across other age groups.
What they found
The researchers found that a large majority (81.0%) of molecules showed changes in at least one age category compared to the baseline, but only a small portion of all the molecules and microbes (6.6%) showed a linear aging pattern. Indeed, in subsequent analyses, they found that thousands of the molecules they tested exhibited two major wave crests of change. Using sliding 20-year windows to compare molecule levels across successive 10-year age ranges (for example, molecule levels from participants aged 40-50 were compared to those of participants aged 50-60, and molecule levels from participants aged 41-51 were compared to those of participants aged 51-61, etc.), they observed peak changes at approximately 44 and 60 years of age (see Figure below). This means, for instance, that individuals at age 44 are more drastically different on the molecular level from individuals at age 43 than those at age 43 are from those at age 42, and so on, whereas individuals at approximately age 53 (a local minimum) show only relatively modest molecular differences from those at age 52 or 54. In other words, these data indicate that humans age non-linearly – experiencing the most rapid aging bursts around ages 44 and 60, while rates of aging appear to slow between ages 44-53 and after age 60.
Of note, these crests were apparent across multiple different types of -omics data, though some displayed only one of the two peaks (e.g., lipidomics data, which only demonstrated a peak around age 45, followed by fairly consistent decline in the number of significant differences between subsequent age groups). However, the authors also note that most of the dysregulated molecules were the same between the two peaks. Molecules present in both peaks included several associated with cardiovascular disease as well as many involved in skin and muscle instability, indicating that these crests might correspond to periods of increased cardiovascular risk or more rapid skin/muscle deterioration.
By applying principal component analysis to their multi-omics data (a method that reduces large, high-dimensionality datasets into a few essential factors – or “principal components” – that collectively account for most of the variability in the original dataset2), the authors found that the first “principal component” for many of the -omics datasets demonstrated significant correlations with participants’ ages. This means the variation in the levels of molecules and microbes observed across these datasets served as an effective predictor of participants’ ages – particularly with respect to metabolomics, cytokines, and oral microbiome data. Through additional analyses, the investigators identified eleven different clusters of molecules that changed during aging, within which they pointed out plausible connections to age-related processes and diseases such as metabolic dysregulation and reductions in DNA repair function.
Based on their finding of a nonlinear pattern of molecular dysregulation, Shen et al. surmised that there is a systemic, coordinated change across molecules and in many different locations in the body. They optimistically conclude that the identification of peaks in age-related changes around age 44 and 60 might inform the development of more effective strategies for prevention and early diagnosis of age-related diseases. Yet if we look past the complicated algorithms, colorful figures, and the enthusiastic conclusions, we quickly find that all is not as it seems.
A cross-sectional study disguised as longitudinal
Though the authors describe their data as “longitudinal,” this term is a stretch. While it’s true that participants provided multiple samples over time, the median follow-up was only 1.7 years (corresponding to 4-8 time points). That is not nearly enough time to collect any meaningful results about biological markers that might change with aging. Even the maximum follow-up period was just 6.8 years – hardly representative of the entirety of adulthood.
Thus, rather than following the same individuals as they progressed through life and determining whether common patterns existed among multiple people, Shen et al. assessed patterns of aging by comparing largely cross-sectional data from people of different ages. But cross-sectional comparisons are less reliable for evaluating changes over time than longitudinal data. It’s unclear if comparing someone who is 70 years old to someone who is 40 years old reveals anything useful about the 70-year-old at 40 or the 40-year-old at 70. Likewise, one cannot assume that a change seen in a handful of 60-year-olds at a specific point in time will necessarily occur in ten years to people who are 50 years old at that same point, as individuals who differ in age at one particular cross-sectional point also are likely to differ in other variables relevant to aging – as trends in diet, education, health awareness, environmental exposures, and countless other factors change with successive decades and generations.
Indeed, Shen et al.’s own findings reveal how little utility these cross-sectional patterns have on the level of any given individual. Examining data from the longest-participating subject (who began the study at age 59.5 and continued until age 66.3 – a range that partially overlaps with one of the alleged peak aging phases), the authors report, “it was not possible to identify obvious patterns in this short time window.”
The sample was very small and non-representative
Across all the samples taken and tests conducted, the investigators collected over 240 billion data points on over 130 million biological features. This may sound like a large amount, but all of these points were from only 108 individuals – very small for a study intended to draw conclusions about aging patterns in all of humanity. (If you collect data from only one person, even an infinite amount of data can tell you only about that one person.)
To make matters worse, the tiny sample is also not representative of a broad population. Every study participant was from the same area in California. Two-thirds were Caucasian, while most of the remainder were Asian. Black and Hispanic backgrounds were grossly underrepresented relative to their true proportions in the American population. Even the age range (a critical parameter for a study evaluating aging patterns) was non-representative, as the oldest participant was only 75 – a limit which would have entirely missed any aging changes that might take place beyond this age.
What’s behind these aging patterns?
Even if this had been a longitudinal study and had included a far greater number of participants representing far more diverse backgrounds, we still couldn’t extrapolate these data to draw any overarching insights on the aging process because we have no clear evidence that the patterns observed in this study reflect inherent biological phenomena as opposed to changes induced by evolving environments and life circumstances. Indeed, the discovery of two distinct periods of accelerated aging should raise questions about whether anything else might be going on at those stages of life.
External circumstances often change in predictable ways across a human lifetime. For instance, adults in their 40s are often reaching higher (and more stressful) levels of responsibility in their careers, caring for teenagers and elderly parents at the same time, and facing the financial pressure of their own mortgages and their children’s approaching college tuition bills. Around age 60, many experience decreases in their physical abilities, approach retirement and loss of their career as part of their identity, face the death of parents, spouses, or friends, and generally begin to face their own mortality. The molecular changes observed in this study could be linked to these kinds of external stressors (and/or any behavioral changes that might accompany them – such as reduced sleep or physical activity) rather than to any programmed chronological aging process, as this study assumes. And because the authors adjusted for a very small number of covariates across study participants, we have all the more reason to suspect that results reflect variables other than chronological age.
The probability of contributions from external factors also increases when we take into account the relative homogeneity of the study cohort and small sample size. White Californians from a similar socioeconomic class are likely to share cultural as well as biological similarities, which reinforces the point that some – or even all – of the noted changes could be due to sociocultural stages rather than biological ones. (To illustrate this point, let’s consider the so-called “freshman 15.” If you examine metabolic markers in a set of American prep school students from age 15 until age 22, you’d probably see changes peaking around age 18-19 due to the start of college and the new food environments related to that life change. If you instead look at this span of time in a population for which military service is mandated at age 18, I’d bet the patterns would look very different…)
What can we learn from this study?
This paper makes a nice story with its large volume of data, complicated analyses, and attractive charts and figures, but it’s easy to get lost in the trees and never see the forest. If we truly wanted to generate useful insights into patterns of aging based on multi-omics data, we would need a much larger, longer study including subjects of more diverse ethnic, geographic, and socioeconomic backgrounds. Such a study is unlikely to be conducted due to the high cost involved for a comprehensive analysis over several decades of time points.
Unfortunately, cutting corners in the case of the present study leaves us with little in the way of reliable knowledge gained. Shen et al. do indeed provide compelling evidence that humans (or at least, White Californians) undergo relatively dramatic biological changes at a molecular level around their mid-40s and their late 50s to early 60s. But their data reveal nothing about the relationship between these changes and any programmed biology of aging. The combination of the cross-sectional nature of the study and the small, non-representative cohort make it highly likely that the observed patterns reflect environmental, behavioral, or sociocultural variables that may correlate with particular ages but are not inherently linked to age itself.
Ultimately, this means that these patterns have little relevance on an individual scale – as any given individual may differ in the points at which these other variables come into play (e.g., some may experience the death of a parent at a relatively young age, while others might experience stresses related to childcare at relatively advanced ages). So while these results make for great headlines, they don’t need to be taken as a reason to fear the big “6-0” (or the “4-4”…) any more than any other birthday.
For a list of all previous weekly emails, click here.
References
- Shen X, Wang C, Zhou X, et al. Nonlinear dynamics of multi-omics profiles during human aging. Nat Aging. Published online August 14, 2024:1-16. doi:10.1038/s43587-024-00692-2
- Greenacre M, Groenen PJF, Hastie T, D’Enza AI, Markos A, Tuzhilina E. Principal component analysis. Nat Rev Methods Primers. 2022;2(1):1-21. doi:10.1038/s43586-022-00184-w
