Book Review: Conceptual Breakthroughs in the Evolutionary Biology of Aging

The reductionist scientists have promised that we can cure aging by studying molecular pathways (nutrient sensing, hormonal signalling, IGF-1), searching for small molecules having large effects on metabolism (semaglutide, rapamycin, etc), and genes with large effects (daf-2 in doubles nematode lifespan), but all of these approaches have failed to deliver. Since I have a background in physics (BS) and practice engineering (CS), I found my reductionist mindset taking a heavy beating as the book persistently accumulated evidence that such an approach will not save us.

Setting the Stage

For context let’s compare this book to a couple others that I’ve read.

I found David Sinclair’s book, Lifespan: Why We Age―and Why We Don’t Have To, light on the science with a confusing description of epigenetics as the scratches on a CD. We know that each generation restores an epigenetic profile and we have Yamanka Factors to accomplish this, but not yet in a specifically controlled manner that would avoid cancer. It overpromises our ability to figure that out. The second half of that book reads like shills for various companies offering quick hacks of little utility. I won’t recommend reading it and Sinclair, as a famous scientist, should feel embarrassed about the quality of its content. It aims at the lay public and has far less science than Rose’s Conceptual Breakthroughs in the Evolutionary Biology of Aging.

I do recommend Peter Attia’s book Outlive: The Science and Art of Longevity for anyone looking to change their perspective about about the inevitability of aging. His concept of picking daily activities that you would like to perform as your personal Centenarian Decathlon, a way of preparing for your marginal decade, lets the reader take meaningful action today. It also aims at the lay public and has far more actionable advice than Rose’s Conceptual Breakthroughs in the Evolutionary Biology of Aging.

Finally, the most comparable. Aubrey De Grey’s book Ending Aging, while dated, has a large volume of citations to peer reviewed literature. It outlines a scientific reductionist approach to solving aging by attacking seven different areas of physiology that fail with increasing age. In particular, corresponding to De Grey’s own background, it highlights various problems that cells experience with mitochondrial dysfunction, and suggests that we can avoid them by genetic therapy that moves the remaining mitochondrial genes into the cell nucleus. A difficult problem, with clear scope and a specific outcome, designed for funding by a research grant. He promotes the idea that with a thousand similar fixes we can defeat death. Rose’s Conceptual Breakthroughs in the Evolutionary Biology of Aging contains several passages, backed by research of a different field, that strongly suggests De Grey has a case of mistaken optimism bordering on faith. From the evidence delivered, I can’t disagree, even as I still share that faith.

Outline of the Conceptual Breakthroughs

The book tells a story about the development of the field. It does so with a formality that I quite enjoyed. Each chapter being only a few pages (really no more than 5) of clear sections — The Standard Paradigm, The Conceptual Breakthrough, Impact, and References and Further Reading. I can only imagine that at times this format must have felt quite constrained. As a reader, I see that it forces a brief highlighting of each stepping stone as the field developed over time. I strongly appreciate the brevity, like I’ve been given sequence of executive summaries for each breakthrough in the field over the last 100-ish years. For this reason, I would strongly recommend the book to any undergraduate majoring in gerontology or experimental evolution.

Looking at the chapters list, we see extremely sparse development up until 1900s. Chapter 1: 384 — 322 B.C The first biologist on Aging. Chapter 2: 1645 A tale of two Bacons. Chapter 3: 1881 Natural selection is the ultimate determinate of aging. Then Chapter 4: 1922 Early laboratory experiments on demography. From there the pace quickens as the dates get shorter. In Chapter 20 we hit the 1980’s and it becomes even denser and much more exciting. Chapter 40 only brings us to 1993, and by this point it becomes quite obvious that research breakthroughs, each taking a few years of experimental work, have been occurring in parallel at multiple labs.

Rose makes it quite clear that the field began with personal observations recorded by (but likely not unique to) particular luminaries using verbal arguments and inferences. The field doesn’t really make any strong progress beyond vociferous opinions and emphatic gesturing at particular examples until it acquires a mathematical footing starting in 1928. That tooling forces a precision to the arguments that ground to a particular model. From here the scientists take experimental and field observations map them to the model’s predictions. Progress becomes both possible and clear as researchers discover gaps and make adjustments that further refine the models.

Rose did not shy away from including some mathematics! And I took a diversion into learning about the Leslie matrix, which field biologists use to estimate whether a wild population can be safely culled (because it grows to quickly) or needs protection (because it doesn’t have enough reproductive capacity). Because the book emphasizes the conceptual breakthrough aspect, most of the math references particular terms of the most important equations, highlighting the coefficients and what they mean as parameters in a model. For example, from Chapter 24:

Over three articles Rose (1982, 193, 1985) showed that antagonistic pleiotropy often leads to the maintenance of genetic polymorphisms. At selective equilibrium, these genetic polymorphisms have little additive genetic variance for fitness itself. But they nonetheless feature abundant selectable (i.e. additive genetic) variation for individual life-history characters.

In order to maintain genetic polymorphisms, the following expressions must be sufficiently large
e^{-r_{11} x_a}\ l(x_a)f_a - e^{-r_{11} x_b}\ l(x_b)h_b f_b
As well as
e^{-r_{22} x_b}\ l(x_b)f_b - e^{-r_{22} x_a}\ l(x_a)h_a f_a
Here l(x_a) represents the age-specific survivorship, as before the dominance parameters are h_a and h_b. When these dominance parameters are small, it is more likely that polymorphisms will be maintained.

That last sentence being the important message for those that can’t follow the mathematics or (like me) chose not to dive deep enough (read the referenced material). Importantly, the book does not contain the derivations. So, in a sense, no reader has enough equipment to follow the math. But that doesn’t matter, because we only need to look at a few terms. I get to walk away with the illusion of understanding, since I interpret the expression as “a difference in expression between alleles a and b” where each has some terms weighting for age, genetic dominance, and prevalence in the population. The book focuses on highlighting the meaningful terms and communicating their value as part of the narrative of scientists trying to figure out, through precise description and experimentation, how evolution affects aging profiles of different species. This decision makes the book a great reference for anyone choosing to major in the field, because it contextualizes the math you can expect to encounter.

Some Important Lessons

From the later chapters it became quite clear, that most of what we have learned about the evolution of aging comes from experiments in fruit flies. Due to their short lifespan, normally 40-50 days, with some experimental selection setups pushing that to 110 days, fruit flies make a great candidate organism that a graduate student, who has about 6 years of research time, can study for several generations. Occasionally lessons come from the field, as we get to see distinguishing selection pressures on symmetrically fissile organisms (those that reproduce clones by budding) vs semelparous species (like salmon that die soon after mating) vs iteroparous species (like humans and other animals that participate in several mating sessions).

Because of a quite early divergence between our species, humans, fruit flies, and nematodes all have quite different physiology. So we do not expect any lessons on manipulating metabolic pathways or particular genes, like daf-2 or the induction of dauer stage in nematode, to translate to humans. And this remains true in spite of a conserved IGF-1 pathway across all three species. However, because humans and fruit flies both outbreed, reproduce quickly and with preference for non-kin, neither of us escapes the pressure of age-specific selection processes. So lessons about the evolvability of age-related physiology in fruit flies can generalize to humans.

Lesson: Resiliency and Longevity Trade-offs

For example, measurements of aging characteristics of fruit flies tested their resiliency to particular stressors like starvation, alcohol, dehydration, and temperature. When we see a statistical correlation among several species that delayed reproduction implies longer lifespan, and then reproduce that observation in a lab though selection on fruit flies, we can expect it will generalize to humans.

When we see that applying a selection pressure for a particular stressor initially also increases lifespan (e.g. because resilience generalizes) we might expect that to also occur in humans. Yet experiments showed the presence of a Pareto frontier, once that hits, optimizing further for resilience to the stressor costs longevity. As a particular example from fruit fly research, more fat stores provide resilience to starvation and also increase longevity, up to a point. Beyond that point, obesity provides starvation resistance, but it has trade-offs that cost the fly some lifespan. We already see this in human data, where obesity does act as a metabolic buffer to particular acute stressors, yet simultaneously correlates to a shorter expected lifespan.

Lesson: Integrating Characteristics

We use similar metrics in humans: grip strength, VO2Max, endurance, glucose response, reaction time, etc. Importantly, these metrics tend to integrate various factors into one value. While I don’t believe that exercising my forearms will help me live longer, I do think that muscularity or its weakness, will predict my ability to endure a fall without breaking a hip. Surviving an event like that without injury does mean I can stay alive longer.

I want to emphasize a different metric to make the point. VO2Max can bottleneck in many ways. Can you get enough air into your lungs? Do they effectively transfer that oxygen to the blood? Does your heart pump efficiently? Can your vasculature deliver the oxygenated blood to muscle cells? Do they take up the oxygen? Can they transport it to the mitochondria? Do your muscles have a large or small number of mitochondria? Will they switch over to anaerobic activity too easily? Now you can see this one metric integrates quite a number physiological factors. It would be difficult to game this with a few hacks.

Longevity, as a trait of a highly complex system, needs similar non-gameable metrics.

Lesson: Complexity of Physiology

In the later parts of the book, we find that genome assays of fruit flies that have been selectively bred for late reproduction (i.e. those bred for longevity) have a large number of small changes across their genome. In contrast to nematodes, which had a particular life stage (dauer) that researchers could trigger via a single mutation (daf-2 mutants with the required daf-16 activity). In this research, fruit flys behave more like humans. They prefer to outbreed instead of self-fertilizing, and they have frequent reproductive events post maturity.

But these findings (i.e. large number of changes spread across the genome) indicate that longevity, as a physiological trait, occurs in a highly complex fashion. Our current science techniques of tweaking single genes together with an A/B difference comparison lends itself toward finding single genes of large effect and cannot help us to find a multitude of genes with small effects. In part because we don’t know which genes to target — not even after-the-fact, as several different long-lived fly populations each had a different mutation profiles, with little overlap. But also because we don’t have the mathematical equipment to identify the small effects from noise and performing any study in volumes enough to tease out that signal remains prohibitively expensive.

I expect that AI/ML will be the new tools that we need for tackling this level of complexity. Don’t trust anyone offering a miracle molecule for your longevity. It might save you from a particular problem (e.g. a cyclodextrin that cures arterial plaque), but living longer remains more complex than patching one big problem at a time.

Lesson: Mortality Risk Plateaus

Unsurprisingly, we can characterize aging as a time period increase in mortality risk. Year-over-Year for humans, Day-over-Day for a fruit fly. For both humans and fruit flies, we see that in late life the mortality risk plateaus. Unfortunately, for humans, this happens quite late, at age 105 (according to Barbi et. al. 2018). But at that age, the quite high YoY risk means you won’t make it very many years.

So we arrive at an important longevity goal: push the mortality plateau down to earlier ages with lower YoY risk!

Lesson: Age-Dependent Structure

The age-dependent structure of a population became an inescapably recurring theme throughout the book. It had to. All of the experiments examine physiological factors that have an age-dependent structure.

A finding that I did not at all expect: fruit flies have an age-dependent ability handle diet. Experiments showed Hamiltonian patterns of age-dependent adaptation to novel environments that lab populations of flies raised on banana and high-sugar syrups since 1981 derived from wild North American populations feeding on apples (for centuries, because N. America doesn’t cultivate banana), demonstrate improved health when switching to apple in late life. Note, that due to the ~500 generations of evolution on banana in the lab, the juveniles demonstrate lower health on banana.

I suspect this finding generalizes to humans, though I haven’t seen large-scale peer-reviewed studies confirming that. However, we do have the following indications:

  • Introduction of agriculture correlates to poor health, at the time of its introduction
  • Individuals from populations that do not have any evolution on the agricultural diet fare poorly when introduced to it
  • Nobody has had time to adapt to modern industrial diet
  • Anecdotally, switching over to paleo diet after ~45 improves health

Ultimately, this lesson reinforces our earlier lesson on the Complexity of Physiology. I didn’t expect diet to show age-dependent structure, but given that changing nutrition, digestion, etc also involve a large number of small mutations, perhaps nearly everything related to our wellbeing will show age-dependent structure?

Lesson: Experimental Difficulty

Finally, in studying evolution by evolving organisms in the lab, researchers must avoid gene-environment interactions that confound the experiment. For example, we data on aging of wild-caught rodents studies in the lab would not project to conclusions about aging rodents in the wild. In spite of both animals coming from the same source, the lab study remains confounded by novel conditions (lighting, cages, diet), any of which could affect aging traits.

But GxE is a more severe problem for experiments on aging than the previous paragraph suggests. We have found that it obscures, if not wholly undermines, experimental interpretation even when we, as experimenters, were determined to avoid it. Perhaps the best explanation we can give for the insidious difficulties of GxE in studies of life-history characters is that life-history characters numerically determine fitness. If there is any opportunity for a species to evolve patterns of environmental responsiveness of life-history characters that can increase fitness, then it is likely for a species to do so.

This raises the possibility of artificial effects from such novel environment, relative to populations that have reached an evolutionary equilibrium. The intuition is that there could be segregating genetic variation that affects general suitedness to the new environment, thereby enhancing life-history characters without any antagonism among these effects. Such segregating genetic variation would then bias genetic correlations among functional characters towards positive values, though not necessarily elevating them above zero.

Overall, despite the salience of this work on genotype-by-environment interaction as a source of artifacts, it has had relatively little impact on subsequent work across the subdisciplines of biology, including those parts of gerontology uninformed by evolutionary research.

So experimental design remains tricky for biologists, but especially so for evolutionary biologists, and emphatically so for those studying aging. To repeat the point, species survive by acquiring genetic programs that allow for adaptation to novel environments. Consequently, labs can accidentally engage these programs, confounding the results from the intended conclusion. Experiments designed to measure the genetic correlation bias towards positive values succeeded in doing so.

What made the book entertaining

Rose really doesn’t hold back. For your amusement, I repeat here some of the spicy passages. Get the book to read more!

From Ch 26, Direct demonstration of nonaging in fissile species

Unfortunately, it should be stated that cell-molecular theories of aging have not been adjusted to reflect these profound findings. Their aging experiments all involve species that have the same essential cell-molecular features: telomeres, lysosomes, free-radical damage, and so on. Yet some of these species have demographic aging, while others do not, in the patterns predicted by the evolutionary theory of aging. There is no physiological necessity to biological aging whatsoever. The process is entirely contingent on age-specific patterns of selection, a point that subsequent work on patterns of mortality plataues would make even more apparent in the 1990s.
Evidently, the findings of experiments of this kind are as abhorrent for cell-molecular biologists as telescopic studies of the orbits of the moons of Jupiter were for Aristotelian astrophysicists in the 17th Century. That may be why the generally ignore them.

From Ch 34, 1988: Evidence for senescence in the wild

In particular, since a major goal among many who study senescence is to gain better insights into human gerontology specifically, large mammals in the wild might be an important window into some aspects of aging in humans, because we ourselves are large mammals that do not live and breed in laboratories, leaving aside some desperate graduate students and postdocs.

From Ch 44, 1996–98: Physiological research on evolution of aging supports organismal mechanisms

The reductionist cell-molecular paradigm for aging assumes that the key controls of aging are to be found in such phenomena as DNA repair, tranlation-error catastrophe, and free-radical damage. … A major challenge these proposed mechanisms face is that it is difficult to demonstrate that cellular or molecular damage is the cause of aging directly. On the other hand, this gives them reduced falsifiability, making it very difficult to collect evidence eliminating such hypotheses from contention. … Fortunately for cell-molecular biologists, major aging research agencies like the U.S. National Institute of Aging no longer fund research of that kind [focused on falsifying hypotheses], leaving the standard paradigm of cell-molecular theories of aging as intact as the geocentric theory of astrophysics was during the lives of Giordano Bruno and Galileo.

From Ch 53, 2010: Genome-wide sequencing of evolved aging reveals many sites

Whether or not conventional gerontologists wish to continue to conduct research with genetic variants that have little to do with aging in outbred populations is, in large part, up to them. If their interest is in outbred human populations, then research based on “longevity genes” will be of little biomedical value. On the other hand, if they want to help decipher the vagaries of the inbred rodent strains that have been of such obsessive interest to the American National Institute of Health, then they should continue to work on the problem of irrreproducibility in mutant strains derived from inbred lines of Drosophila and the like (vid. Khazaeli et al. 2015)

Where I’ve Changed My Mind

I began my reading as a material reductionist, thinking that we could follow Aubrey de Grey’s program of solving a 1,000 problems of microbiology. That position takes some real heat throughout the book, with an ongoing comparison to the abandoned geogentric model of the solar system. Now, with the evidence presented by this book, I have to recognize that aging has (a) age-dependent structure and (b) occurs in an immensely complex system.

The first take-away surprised me. I had a model of steadily accumulated sources of dysregulation, such as loss of epi-genomic control leading, shortening of telomeres, growing plaque deposits in the arteries, junk in the extracellular matrix, decline of stem cell populations, etc. All of which contribute to a steady loss of resiliency. The data shows exponential mortality curves rather than linear ones, so it seems that this “damage” compounds. The book shows mortality plateaus, which gives us a completely different approach. We should now be searching for ways to plateau early in life, when the mortality risk remains low.

I find the second take-away discouraging. We simply lack good tools of scientific investigation for disentangling complex feedback loops. I hope that the massively large statistical inference engines that we call AI and ML can help us here. I do not expect that we can solve aging by finding a handful of large-effect genes, or single molecules. Although a single cell itself represents a complex system, the research on cell-molecular biology remains insufficient, our knowledge of metabolic pathways hopelessly inadequate, for suggesting therapies at the organismal level.

Nevertheless, I still can’t escape the reductionist observation: aging looks like physiological failure. From a background in physics, it looks like the system of feedback loops that preserves homeostasis fail, with redundancy and backup circuits also failing, until a small stressor finally overwhelms the system and it collapses with a final breath.

An Actionable Conclusion

For those who want to take action now, instead of reading science, the book’s Conclusion offers a few paragraphs:

In the same spirit [as physicians persuaded into washing their hands], evolutionary biologists with Eurasian ancestry can take into account results like those of Rutledge [1][2] … to avoid novel processed foods and, at later ages, endeavor to emulate preagricultural diets. For those with ancestors who did not consume agricultural foods prior to 1500, avoidance of agricultural foods at every age might be the best option.
In the same vein, it might be advisable to adopt preagricultural lifestyles beyond patterns of food consumption. This could include increased levels of chronic activity, or sustaining closer networks of friends and allies that might be analogous to those which prevailed before agricultural settlement.