Why Do We Age? Evolutionary Theories of Senescence Tested

Why do we age, when natural selection is supposed to reward whatever keeps an organism alive and breeding? A body that can build itself from a single cell ought to be able to keep itself in repair, yet in most animals the risk of dying climbs year after year. Evolutionary biologists have had a broad answer since the mid-twentieth century: selection loses its grip on traits that act late in life. The broad answer is accepted. What remains unsettled is which mechanism carries most of the weight, how strongly the outside world’s dangers shape the pace of decline, and why some species barely seem to decline at all.

Below, the settled part is separated from the open part, and a small simulation built for this article reproduces a result that overturned a textbook prediction: whether a dangerous environment speeds the evolution of aging depends on how the population is held in check.

The question, stated precisely

“Why do we age” can mean two different things. The proximate question asks which cellular and molecular processes, such as DNA damage, protein misfolding or stem-cell exhaustion, produce the decline. The ultimate question asks why evolution allowed those processes to win. This piece is about the second one. Put precisely: given that organisms could in principle invest in indefinite repair, what evolutionary forces produced age-specific increases in mortality and decreases in fertility, and which of those forces dominates in a given lineage?

The answer matters beyond curiosity. If aging is mainly the residue of many weakly harmful late-acting mutations, there is no master switch, and interventions will have to address many separate defects. If it is mainly the price of genes that help early reproduction, some anti-aging interventions should carry costs in youth or fertility. Popular interest is real: a 2023 study linking fertility-boosting variants to shorter human life drew wide press coverage, including a ScienceDaily release.

What is established: selection weakens with age

The core idea traces to Fisher and Haldane, was given its classic verbal form by Peter Medawar in 1952, and was formalised by W. D. Hamilton in 1966. Flatt and Partridge’s review in BMC Biology summarises the lineage and the logic (Flatt & Partridge 2018). Every organism faces some risk of death from predators, disease, accidents and starvation. Even with no internal decline, fewer individuals remain alive at age 40 than at age 4. A mutation that harms only 40-year-olds is therefore expressed in fewer carriers, and those carriers have already done much of their breeding. Selection against it is weak.

Hamilton turned this into quantities now called the forces of selection. For a mutation that changes mortality at age a, the strength of selection is proportional to the expected reproduction still to come after age a, discounted by population growth and divided by generation time. Before maturity that force is flat and at its maximum. After the onset of reproduction it can only fall, and it approaches zero once almost no reproduction remains. Biologists sometimes call the region where the force is near zero the selection shadow.

Three points are not in serious dispute. The declining force is a mathematical consequence of age structure, not a hypothesis. It explains how intrinsic decline can evolve without aging being good for the species. And because it depends on the whole schedule of survival and reproduction, species with different life histories should age differently. Beyond this, agreement thins.

Three classic hypotheses, steelmanned

Mutation accumulation

Medawar’s version says that deleterious mutations with late effects drift and accumulate because selection is too weak to purge them. The decline is a pile of small, unrelated defects, each tolerated because it bites in the shadow. Evidence for it includes quantitative-genetic studies in which additive genetic variance for mortality or fertility rises with age, as the hypothesis predicts. Flatt and Partridge describe mutation accumulation as reasonably well supported by such work, while also noting published critiques. Against it: the predicted rise in genetic variance is not found in every study, and the hypothesis says little about why aging should be coordinated across organs.

Antagonistic pleiotropy

George C. Williams proposed in 1957 that some alleles help early survival or reproduction and harm the same body later. Because the early benefit is weighted more heavily by selection, such alleles can spread even though they cause decline. This turns aging from a passive failure into an evolved trade-off. It predicts negative genetic correlations between early and late fitness traits, and selection experiments on fruit flies have repeatedly shown that selecting for late reproduction lengthens life and can reduce early fecundity. The weakness is concreteness. Few specific alleles have been shown to work this way in wild populations, and Flatt and Partridge point out that the textbook human example, BRCA1/2 mutations, is hard to square with the hypothesis because those variants are rare.

Disposable soma

Thomas Kirkwood’s 1977 paper in Nature argued that somatic cells might economise on error correction to save energy, a view later developed into the disposable soma theory (Kirkwood 1977). The logic is an allocation argument: energy spent on maintaining a body that will probably be killed anyway is energy not spent on reproduction. It can be read as a physiological special case of antagonistic pleiotropy. Its appeal is that it links aging to metabolism and resource budgets. Its difficulty is that dietary-restriction and reproduction-ablation experiments do not always show the simple energy trade-off it implies, and the hypothesis is hard to falsify without measuring allocation directly.

A minority position holds that aging is itself an adaptation, programmed to benefit the group. Most evolutionary biologists reject it because group-level selection is usually weaker than selection on individuals.

The extrinsic mortality prediction and why it broke

Williams also made a bold comparative prediction: populations exposed to higher adult mortality from external causes should evolve faster aging. It is the most intuitive consequence of the selection shadow, and it was taught for decades as a direct corollary. The trouble is that Hamilton’s own mathematics does not deliver it. If an extra source of death strikes every age equally and the population is free to grow or shrink, the extra death lowers survivorship and lowers the growth rate by exactly compensating amounts. The forces of selection at each age are unchanged.

Moorad, Promislow and Silvertown reviewed this problem in Trends in Ecology & Evolution (Moorad et al. 2019). Building on earlier work by Peter Abrams, they concluded that population regulation is not enough on its own:

“density-dependent population regulation cannot, by itself, cause changes in selection.”

Moorad, Promislow & Silvertown, 2019

What matters is age specificity. If crowding mainly suppresses juvenile production or juvenile survival, then extra adult mortality frees up resources for the young and changes how selection is weighted across ages. Williams’ prediction then holds. If crowding kills all ages alike, it does not. De Vries, Galipaud and Kokko later wrote an explicitly pedagogical treatment, titled “a guide for the perplexed,” which reaches the same structure: high extrinsic mortality favours fast life histories when increasing density harms the production of juveniles (de Vries et al. 2023). The question has moved from “does danger speed aging?” to “how is each population regulated, and at which ages?”

What field, laboratory and human data show

The evidence is mixed in ways the theory now partly anticipates. Reznick and colleagues compared guppies from streams with and without major predators (Reznick et al. 2004). High-predation fish had already been shown to mature earlier and invest more in reproduction, but they did not age faster in mortality or reproduction. They did decline earlier in swimming performance, a pattern the authors called mosaic. A fruit-fly experiment in which adult mortality was imposed at high or low rates, reviewed by Flatt and Partridge, did produce the classic direction. Both results can be true at once if guppy and fly populations are regulated differently, which is exactly what the newer theory says to check.

At the widest scale, Jones and co-authors assembled mortality and fertility trajectories for 46 species, from mammals to plants and a green alga (Jones et al. 2014). Mortality rose steeply with age in some, stayed flat in others, and fell in a few, with no tight link between how long a species lives and how sharply it ages. Their verdict was blunt:

“Such variability is not predicted by the standard evolutionary theories of ageing.”

Jones et al., 2014, Nature

Defenders of the classic theory reply that negligible or negative senescence is allowed when organisms keep growing and fecundity rises with size, because then the force of selection need not fall much with age. That reply is reasonable, but it concedes that the standard models were too narrow.

Human genomic data have recently been brought to bear on the mechanism question. Rodríguez and colleagues used genome-wide association results for diseases with different ages of onset (Rodríguez et al. 2017). Risk alleles for late-onset diseases were more common and had larger effects than those for early-onset diseases, consistent with weak selection late in life, and they also found an excess of variants with opposite early and late effects. Long and Zhang went further with about 276,000 UK Biobank participants (Long & Zhang 2023):

“We observe a strong, negative genetic correlation between reproductive traits and life span.”

Long & Zhang, 2023, Science Advances

Of 583 reproduction-associated variants they tested, 123 affected survival to age 76, and 98 of those acted antagonistically. That is direct support for antagonistic pleiotropy in a living population, though it relies on association data, parental lifespans and a single ancestry group. Neither study shows that pleiotropy outweighs mutation accumulation overall; both suggest the two operate together. For how such variants relate to ongoing selection, see the earlier piece on whether humans are still evolving.

A small simulation built for this article

To make the regulation argument concrete, I wrote two short Python programs and ran them for this article. They are toy models. They are not fitted to any species and they demonstrate logic, not biology. If the vocabulary of selection, mutation and drift is unfamiliar, the explainer on whether evolution is random covers it.

The first program computes Hamilton’s force of selection on mortality for a simple life history: maturity at age 2, constant fecundity, a small intrinsic death rate of 0.01 per year, and an added extrinsic death rate m of 0.02, 0.10 or 0.30. It compares three cases: (A) the population grows or shrinks freely; (B) it is held constant by scaling fecundity, which stands in for crowding that suppresses juvenile production; and (C) it is held constant by an extra death rate applied equally to all ages. For each case I report the age at which the force has fallen to 50% and 10% of its pre-maturity value.

In cases A and C the selection shadow begins at the same age whatever the extrinsic death rate, exactly as Hamilton’s mathematics requires. Only in case B does more danger pull the shadow earlier, from 22 years down to 4 for the 50% mark. The same program asked when an antagonistically pleiotropic allele can spread: one giving 5% more fecundity at ages 2 to 4, at the cost of 0.05 extra hazard from some later age onward. With free growth, the allele invaded whenever its cost began at age 18 or later, for every value of m. Under fecundity regulation, the earliest invading onset age was 73, 30 and 7 for m = 0.02, 0.10 and 0.30.

The second program is an individual-based simulation of mutation accumulation. Up to 2,000 sexual individuals live for at most 40 years. Each of 40 loci can carry a mutation that adds 0.1 to the death rate at one specific age only, with a mutation rate of 0.002 per locus per birth and free recombination. Two forms of regulation were run: newborns fill only the space left by deaths (recruitment regulation), or every adult breeds and random culling across all ages trims the population back to 2,000. Each setting was run for 8,000 years with three random seeds; the code is below.

import numpy as np
K, A, MAT, MU0, U, D = 2000, 40, 2, 0.01, 0.002, 0.10   # pop cap, max age, maturity, base hazard, mutation rate/locus, effect

def run(m, seed, mode='recruit', steps=8000, B=0.6):
    rng = np.random.default_rng(seed)
    age = rng.integers(0, 10, K)
    G = np.zeros((K, A), dtype=bool)          # locus j raises hazard only at age j
    acc = np.zeros(A); n = 0
    for t in range(steps):
        haz = MU0 + m + D * G[np.arange(age.size), age]
        alive = (rng.random(age.size) < np.exp(-haz)) & (age + 1 < A)
        age, G = age[alive] + 1, G[alive]
        adults = np.flatnonzero(age >= MAT)
        nb = K - age.size if mode == 'recruit' else rng.poisson(B * adults.size)
        if nb > 0 and adults.size > 1:
            p1, p2 = rng.choice(adults, nb), rng.choice(adults, nb)
            pick = rng.random((nb, A)) < 0.5   # free recombination
            kids = np.where(pick, G[p1], G[p2]) | (rng.random((nb, A)) < U)
            G = np.vstack([G, kids]); age = np.concatenate([age, np.zeros(nb, int)])
        if mode == 'cull' and age.size > K:     # regulation: random death at all ages
            keep = rng.choice(age.size, K, replace=False); age, G = age[keep], G[keep]
        if t >= steps - 2000:
            acc += G.mean(0); n += 1
    return D * acc / n                         # evolved excess intrinsic hazard by age

for mode in ['recruit', 'cull']:
  for m in [0.05, 0.30]:
    prof = np.mean([run(m, s, mode) for s in (1, 2, 3)], axis=0)
    onset = int(np.argmax(prof > D / 2))
    print(f"{mode} m={m}: excess hazard ages 2,5,10,15,20,30 =",
          np.round(prof[[2, 5, 10, 15, 20, 30]], 3), " first age >0.05:", onset)

Under recruitment regulation, mutations piled up from about age 11 when extrinsic death was high but only from about age 34 when it was low. Under all-age culling the two danger levels gave nearly the same onset, 10 versus 9. In plain terms, the simulation reproduces the Abrams and Moorad point: danger speeds the evolution of aging only when crowding acts unevenly across ages.

What it does not show matters as much. The model has no growth with age, no trade-offs other than the one tested, no environmental variation, and mutational effects confined to single ages, which real mutations rarely are. It cannot say which hypothesis dominates in any real species. It shows only that the direction of the extrinsic-mortality effect is a property of the ecology, not a law.

The author’s reading, and what would settle it

What follows is my opinion, not a consensus. The declining force of selection is as secure as anything in evolutionary biology. Mutation accumulation and antagonistic pleiotropy are both real and act together; the human genomic results make it hard to argue that either is negligible. My best guess is that pleiotropy explains more of the coordinated, species-typical shape of aging, while mutation accumulation adds individual variation around it. Williams’ extrinsic-mortality prediction should no longer be taught as a corollary. It is a conditional claim, and its condition, age-specific density dependence, is rarely measured. Studies of interventions such as rapamycin, discussed in the piece on senolytics and rapamycin, sit downstream of this question: if pleiotropy dominates, lasting benefits late in life should often come with early-life costs.

Several studies would move the debate:

• Comparative work that measures how each population is regulated, and at which ages, before testing whether extrinsic mortality predicts the rate of aging. Prediction: the correlation should appear only where crowding mainly hits juveniles.
• Experimental evolution that manipulates regulation directly, for example removing juveniles versus culling all ages at matched total mortality. Prediction: only the first treatment accelerates aging.
• Large multi-ancestry biobank studies repeating the Long and Zhang test with direct rather than parental lifespans. Prediction under strong pleiotropy: the excess of antagonistic variants persists across ancestries.
• Mutation-accumulation lines in long-lived, indeterminately growing species. If mutational variance for mortality does not rise with age there, the classic shadow is genuinely weaker in those lineages.

Measurement is the bottleneck for several of these. Tools such as the epigenetic clocks used in longevity trials could help track the pace of decline in wild or experimental populations, provided their limits are respected.

Common questions

Is aging an adaptation?

Mainstream evolutionary theory says no, at least not directly. Aging is better understood as a by-product of weak selection late in life, either through tolerated harmful mutations or through genes favoured for their early benefits. A minority of researchers argue for programmed aging, but that view lacks broad support.

Do animals in dangerous environments always age faster?

No. That was Williams’ prediction, but theory now shows it depends on how populations are regulated. Guppies from high-predation streams, for example, did not age faster in mortality or reproduction, though their swimming performance declined earlier.

Why do some species barely age?

Jones and colleagues documented species whose mortality is flat or even falls with age. One common explanation is that when organisms keep growing and fecundity rises with size, the force of selection does not decline much with age. That explanation fits some cases but has not been tested systematically.

Does this mean fertility shortens human life?

Not for any individual. The 2023 UK Biobank study found that genetic variants associated with reproductive traits tended to have opposite effects on survival, which is a population-level pattern in allele effects. It says nothing about whether having children shortens a particular person’s life, and this article offers no medical advice.

Further reading

• Why We Age: What Science Is Discovering about the Body's Journey Through Life by Steven N. Austad. An accessible book-length account of evolutionary aging theory by a researcher who tested it in the field. See it on Amazon (affiliate link)
• Evolutionary Biology of Aging by Michael R. Rose. The classic monograph on the evolutionary theory of senescence and the Drosophila selection experiments. See it on Amazon (affiliate link)

As an Amazon Associate I earn from qualifying purchases. A commission may be earned if you buy through this link, at no extra cost to you.

References

1. Flatt, T., & Partridge, L. (2018). Horizons in the evolution of aging. BMC Biology, 16, 93. doi:10.1186/s12915-018-0562-z
2. Hamilton, W. D. (1966). The moulding of senescence by natural selection. Journal of Theoretical Biology, 12(1), 12-45. doi:10.1016/0022-5193(66)90184-6
3. Medawar, P. B. (1952). An Unsolved Problem of Biology. London: H. K. Lewis.
4. Williams, G. C. (1957). Pleiotropy, natural selection, and the evolution of senescence. Evolution, 11(4), 398-411.
5. Kirkwood, T. B. L. (1977). Evolution of ageing. Nature, 270, 301-304. doi:10.1038/270301a0
6. Moorad, J., Promislow, D., & Silvertown, J. (2019). Evolutionary ecology of senescence and a reassessment of Williams’ ‘extrinsic mortality’ hypothesis. Trends in Ecology & Evolution, 34(6), 519-530. doi:10.1016/j.tree.2019.02.006
7. de Vries, C., Galipaud, M., & Kokko, H. (2023). Extrinsic mortality and senescence: a guide for the perplexed. Peer Community Journal, 3, e29. doi:10.24072/pcjournal.253
8. Reznick, D. N., Bryant, M. J., Roff, D., Ghalambor, C. K., & Ghalambor, D. E. (2004). Effect of extrinsic mortality on the evolution of senescence in guppies. Nature, 431, 1095-1099. doi:10.1038/nature02936
9. Jones, O. R., Scheuerlein, A., Salguero-Gómez, R., et al. (2014). Diversity of ageing across the tree of life. Nature, 505, 169-173. doi:10.1038/nature12789
10. Rodríguez, J. A., Marigorta, U. M., Hughes, D. A., Spataro, N., Bosch, E., & Navarro, A. (2017). Antagonistic pleiotropy and mutation accumulation influence human senescence and disease. Nature Ecology & Evolution, 1, 55. doi:10.1038/s41559-016-0055
11. Long, E., & Zhang, J. (2023). Evidence for the role of selection for reproductively advantageous alleles in human aging. Science Advances, 9(49), eadh4990. doi:10.1126/sciadv.adh4990
12. ScienceDaily (2023). Press release on Long & Zhang (2023). sciencedaily.com