To know the body measure
How is the biological age of a person determined today and what does machine learning have to do with it
Polina Loseva, "The Attic"
The idea of a "life line", which is hidden somewhere inside a person and measures the time allotted to him, seems attractive to many. Otherwise, why does someone turn gray already at 30, and someone rides a bike at 90? While some hold out their palms to fortune-tellers, while others go to the cuckoo with a question, others rely on the achievements of science and take tests for biomarkers of age. And although the latter method of determining the "true" biological age seems much more reliable than the first two, the interpretation of its results should be approached no less cautiously. "The Attic" tells by what signs scientists determine the age of people today and how these signs are searched for.
There is no clear definition of old age. There is also no certain age after which they become old. No one can know for sure whether he is old or not, but everyone knows that old age is followed by illness and death. The age recorded in the documents does not say much about the chances of delaying its approach. True, the Gompertz curve has long been known, describing the risk of death depending on age, but it gives only an average value. And for a particular person, this risk may be significantly higher than average – for example, if he is ill with progeroid diseases that accelerate aging, or, conversely, noticeably lower if, for example, he is an over-long-lived person. Therefore, doctors need a more reliable tool – something like a portrait of Dorian Gray, looking at which they could determine the real, biological age of a person.
It is not difficult to find it if you have a large enough sample of elderly people of the same age. It is enough to regularly measure their different physiological indicators and calculate the annual mortality. And then build models that predict the risk of dying based on previous measurements, and look for the most convincing prophet among them.
But what works well with the elderly doesn't apply well to the young. Although they are usually far from death, the distance of this "far" can vary greatly depending on their health. Today, there are many diseases that appear with age and accelerate aging: cancer, cardiovascular diseases, dementia, osteoporosis. The list is constantly updated. For example, there are still debates about obesity – is it considered a sign of accelerated aging or not? It is known that a lot of old cells accumulate in adipose tissue, which then infect their neighbors with old age. This means that a young person with obesity may be closer to death than his healthy peers.
Nevertheless, a rare soothsayer is able to look so far into the future. In order to speak with more confidence about the risks of death of young people, the study would have to be stretched for decades, and the answer is needed now. And it is needed not only for patients with suspected accelerated aging, but also for researchers who are struggling to create a pill for old age, because this is their only chance to check whether their medicine succeeds in rejuvenating the participants of the experiment or not.
Unlike older people, for whom the proximity of death is the measure of age, statistics come to the aid of young people. Scientists measure different parameters in hundreds of thousands of people of different ages, look for characteristic age-related changes and make an average graph for each parameter or their combination. Further on this graph, you can find the result for each subject and estimate his biological age relative to the average peer.
But it is important to remember that such an estimate depends very much on the original sample. The search for biological markers is caught in the same trap as genetic tests: predictions that come true for privileged white men may be meaningless for immigrant women, but the former are participants in research much more often than the latter.
In addition, differences between generations may be the consequences of social, not age-related changes. For example, some researchers believe that the special microbial composition in the intestines of elderly people is not related to age at all, but to the fact that they are used to eating other foods and the so-called Western diet saturated with fats and carbohydrates has not affected them. But, despite all these considerations, we do not yet have a more accurate way to determine the biological age of young people than a comparison with the average value.
What are we guessing at
If desired, secret knowledge about the future can be gleaned from anything: from coffee grounds to the flight of sacred birds. However, not every bird with its flight predicts the fate of the empire and not every dimension can serve as a reliable support for prophecies. In order to avoid divination on unverified sources, in 2004 the American Federation for Aging Research formulated the requirements that a marker of biological age should meet. Looking ahead, let's say that it turned out to be quite difficult to match them.
1. It should be easy to measure it, and the measurement process should not harm human health and accelerate aging. A logical limitation, which, however, causes a lot of inconvenience to scientists. For example, it restricts the set of cell types that can be used. Most of the tests have to be carried out on leukocytes (the only full-fledged blood cells), and it is not always clear whether the same age-related changes occur in cells of other tissues.
2. It should predict the risk of death. Having learned to predict the chronological age of people, it is necessary to check that you also recognize deviations from the average – those who age too quickly or too slowly. But we will be able to make sure of this only when these people start dying sooner or later. This means that all calculations carried out on young people have no real significance unless confirmed by similar measurements on the elderly and the link with the risk of dying.
3. It should be based on the biological processes of aging. This criterion is designed to weed out everyone who wants to guess at the coffee grounds and reflection in the water. Gray hair, for example, is most often associated with age, so it could serve as its marker. However, it has nothing to do with the wear and tear of the body, and people who turn gray at the age of 30 often die no earlier than their peers.
4. It should work not only for humans, but also for animals. And this requirement is related to the standards of clinical trials. If scientists one day find a long-awaited pill for old age and decide to treat people with it, then they will be required to conduct preliminary animal studies. And for the purity of the experiment, it would be good if the rejuvenation of mice was evaluated on the same grounds as the rejuvenation of humans.
Today, for predictions about the fate of people, scientists turn to three familiar moirs, three goddesses of fate, spinning the thread of human life – the three main markers of biological age.
Atropos
The most terrible of the moirs, Atropos, breaks the thread and chooses the method of death of a person. In medical works, fragility indices are used to predict its solution. First, researchers compile a list of symptoms that are quite common in the population and have a negative impact on health. It can be an age-related disease like osteoporosis, tumors or atherosclerosis, or a malfunction of individual organs: poor eyesight, weak muscle grip or inability to move without support. Each patient is given a "score" of 0 or 1 for each item and the scores are summed up throughout the list. The more problems a person has collected in his body, the higher the value of his fragility.
These indexes turned out to be very convenient. They are easy to assemble, just a medical examination is enough for this, and they reliably predict the immediate consequences, for example, the need for daily care or the very "risk of death from all causes". But they are very difficult to apply to young people – except for those who have got themselves some serious illness ahead of time.
How does the probability of survival change over time in people with different fragility indices (FI). Williams et al., The Journals of Gerontology: Series A.
Therefore, to assess the rate of aging here and now, an index of many parameters remotely related to age-related diseases is used. For example, in the American CALERIE study, which studies the effect of calorie restriction on health, scientists measure 18 different signs at once: body mass index, the amount of hemoglobin, cholesterol and urea in the blood, the condition of the mucous membranes, and so on. And it turned out that the biological age of 38-year-old study participants, measured using this composite parameter, ranges from 30 to 50. There is another nuance: none of the fragility indices tells us anything about the causes of aging. He measures only the consequences, anticipating the swing of the scissors Atropos.
Lachesis
The second moira, Lachesis, measures the length of the thread at the birth of a child. The biological analogue to this, of course, is the length of telomeres. Telomeres, the end sections of DNA, shorten with each cell division and are designed for an average of 50 divisions in humans. When they reach a critically short length, the cell loses the ability to reproduce and from that moment can be considered old.
For a long time it was believed that the length of telomeres determines the shelf life of the human body as a whole. It is also known that telomeres can be lost under the influence of oxidative stress or inflammation. Even psychological stress, as follows from some works, reduces the life measured to cells. And the average length of telomeres in humans, as it turned out, correlates with mortality, although it is not associated with the development of specific diseases.
However, upon closer examination it turns out that Lachesis is not so simple. She measures out her term to each person, guided by her own considerations. For example, women have longer telomeres than men, and Africans have shorter telomeres than Europeans. In addition, telomeres are longer the older the child's father was at the time of conception, and the shorter the older his mother was.
Moreover, the measured thread does not always shorten over the years, and sometimes, on the contrary, it grows! In stem cells that need to be divided, telomerase works – an enzyme that completes DNA from the ends. It can build up the thread faster or slower depending on the type of cell or living conditions. There are even cases when telomeres became longer over time – for example, in Costa Ricans they grew during the dry season and decreased during the rainy season.
How the telomere length changed in experiments with 11 sheep. LTL is the relative length of telomeres, the age is given in weeks by Dugdale and Richardson, Philosophical Transactions B.
In some older people, telomeres become longer after the age of 75. Finally, in a recent NASA twin experiment, it turned out that telomeres can grow in just a year of life in orbit. At least that's what happened to Scott Kelly.
Perhaps these stories are related to the fact that we do not know how to measure telomere length very accurately. In the vast majority of studies, researchers estimate the average length of these sequences, not taking into account that it may differ in different cells and even on different chromosomes within the same cell. Therefore, when we read that someone's telomeres have become longer, for example, as a result of meditation, this may mean, among other things, that their ratio of cells in the blood has changed. There are fewer old cells with short telomeres, and more young cells with long telomeres. And if this is so, then it turns out that it is not so easy to deceive Lachesis, and even meditation will not help here.
Telomeres, at first glance, seem like a convenient oracle: their length is easy to measure, they are associated with both the risk of death and the deep processes of cell aging. However, we do not fully understand how their length actually changes over the course of life. And we are not quite sure that it changes in the same way for all cells of the body. And besides, telomeres do not meet the fourth criterion of the biomarker: it is difficult to compare a person with other animals on this basis. In mice, without which no clinical trial is complete, telomerase works throughout life, and the telomeres themselves are much longer. Nevertheless, it does not help them to live longer.
Clotho
The third sister, Clotho, spins the thread of human destiny, winding it on her spindle, and the same thing happens with DNA in every cell of the body. Over the course of life, the DNA strands in the cell nucleus are repackaged: many sections are folded, thereby hiding the information recorded on them, and others, on the contrary, are revealed. For this, methyl groups are responsible, which enzymes hang in certain places on DNA. The more methyl groups, the denser the folding, the smaller – the weaker.
Having collected data on where DNA is methylated with age, American Steve Horvath came up with the first epigenetic clock, or methylation clock, in 2013. They represent a set of 353 sites, among which 193 acquire a methyl group over time, and 160 lose. Later, a second type of watch appeared – the Hannam watch, shorter, with a total of 71 sections, and new options continue to arise.
The methylation clock, like telomeres, predicts the life span well, but it does not depend on either gender or race. They allow us to estimate the rate of aging even of individual cells. For example, with their help, it was possible to show that in patients with progeria (premature aging), intracellular time does not flow as in older people, but many tumor cells age faster than their "healthy" neighbors. Clotho spares no one, and epigenetic clocks can be built on the same principle for other organisms. However, for each species you will have to look for your own set of key sites.
The problem with the methylation clock is that we still don't understand why these particular regions of DNA turned out to be key to aging.
Despite the fact that with their help we can calculate quite accurately to what extent the thread of human life has already been spun, the set of plots itself is only a product of statistical data processing, and we do not know to what extent they can be trusted.
To avoid this misunderstanding, Horvath proposed combining the methylation clock with the fragility index into one system and called it PhenoAge. Together with his colleagues, he took 88 proteins, the number of which in blood plasma changes with age, and calibrated his watch on them, that is, he compiled a list of DNA sites on which methylation changes in accordance with the concentration of a particular protein in the blood. This was done for 12 proteins, and the unified model assembled from them was able to predict not only the actions of Clotho, but also the decisions of Atropos, that is, not just the life time, but also the time before the development of cardiovascular diseases or tumors.
Moira of Deep Learning
Despite the abundance of predictors that science has acquired over the past decades, it is still wary of unambiguous answers. The fact is that, no matter how confident individual seers may be, their prophecies do not fit together well.
From time to time, Hannam's watch may turn out to be more accurate than Horvath's watch, both of them can be surpassed in accuracy by the fragility index, and in some works none of the markers – neither the fragility index, nor the telomere length, nor the epigenetic clock – has justified the hopes placed on it.
This paradox can be solved only by assuming that each of the biomarkers measures only one side of aging, evaluating the process by which it was calibrated. One seer looks after the scissors of Atropos, another – for the movement of the hands of Lachesis, the third – for the rotation of the spindle of Clotho, but none of them is able to deduce from their observations the fate of the thread, the very biological age with which we ask them.
The true life span, if it is really measured out to us, still eludes the traps placed on it by the scientific method. Each diviner works within the scope of his specialization, depending on the context and circumstances. And the more specifically we formulate a question for him, the higher the chance that he will guess the answer.
However, in addition to trying to understand the "moirs" known to us, you can look behind their backs for something more grandiose, and so much so that the "look" of human intelligence is not visible. The main thing is to know the methodology of data processing. This is what attempts to determine biological age using machine learning look like from the outside.
Founder of Insilico Medicine Alexander Zhavoronkov and colleagues taught neural networks to predict age based on a variety of data, be it a blood test, gene expression profile, intestinal microflora or just a photo of the eye. And the accuracy of their prophecies turned out to be quite high: they determined the chronological age of randomly selected people with an accuracy of 2-6 years.
Eye scans that were used to train a neural network to predict a person's biological age. Bobrov et al., Aging.
Models built using machine learning so far satisfy only the first criterion: they are really easy to create without injuring the patient. From a biological point of view, they still resemble a bowl of water, along which, following unknown patterns, meaningful circles run. In order to trust the oracle ex machina, we will have to disassemble each model into its component parts and find out what is special about those genes and those intestinal microbes that the neural network has selected.
Until then, her predictions will have to be treated the same way as the ancients did – simply to believe (or not to believe) in the mysterious power of prophecy, based on something beyond human understanding.
Portal "Eternal youth" http://vechnayamolodost.ru