Misclassification of heritable mortality undermines estimates of intrinsic life span heritability

This paper argues that the claim of a 55% heritability for human lifespan is fundamentally flawed because it incorrectly assumes susceptibility to "extrinsic" causes of death, such as infections and accidents, is non-genetic, thereby introducing selection bias that artificially inflates heritability estimates.

The Gist

The Big Picture: A Dispute Over "Nature vs. Nurture"

Imagine you are trying to figure out how much of a person's lifespan is determined by their genes (nature) versus their environment (nurture).

Recently, a group of scientists (Shenhar et al.) published a study claiming that if you remove all "accidental" or "external" causes of death (like car crashes or infections), the heritability of human lifespan jumps to a massive 55%. They argued that these external deaths are just "noise" that hides the true genetic signal.

Fergus Hamilton, the author of this paper, is saying: "Stop the press. That math is wrong."

He argues that by trying to remove "external" deaths, the researchers accidentally threw away the very genetic data they were looking for, creating a fake, inflated number.

---

The Core Analogy: The "Bad Weather" Filter

To understand Hamilton's argument, let's use an analogy of running a race in the rain.

1. The Race: Life is a long race.
2. The Runners: People have different natural running speeds (their genetics).
3. The Rain: This represents "extrinsic mortality" (accidents, infections, bad luck).

What Shenhar et al. did:
They looked at the race results and said, "The rain messed up the data! Some fast runners slipped and fell because of the rain, and some slow runners got lucky and didn't slip. Let's pretend the rain never happened. If we remove the rain, we can see who the true fastest runners are."

They calculated that without the rain, the difference between runners is huge (55% heritability).

What Hamilton says is wrong:
Hamilton argues that the rain isn't just random noise. Some runners are genetically built to be better at running in the rain (stronger immune systems, better balance).

• If a runner slips in the rain, it might be because they have "weak ankles" (genetics), not just because it was raining.
• By saying "Let's pretend the rain never happened," the researchers are effectively saying, "We will ignore the fact that some people are genetically better at handling the rain."

The Result:
When you remove the "rain" (infections/accidents) from the data, you aren't just removing noise; you are removing the part of the race where genetics actually mattered. This makes it look like only the "intrinsic" running speed matters, artificially inflating the importance of genes.

---

Key Arguments Explained Simply

1. Infections Are Genetic (The "Shield" Analogy)

The researchers claimed that dying from an infection is purely bad luck (environment). Hamilton says, "No way."

• The Analogy: Think of your immune system as a shield. Some people are born with a heavy, strong shield (good genes); others have a flimsy one.
• The Evidence: Hamilton points to studies of adopted children. If a biological parent died young from an infection, the adopted child is much more likely to die young from an infection too. But if the adoptive parent died of an infection, it didn't change the child's risk.
• The Takeaway: Susceptibility to infection is heavily genetic. You can't treat infection deaths as "random noise" because your genes determine how well you fight them.

2. The "Survivor Bias" Trap (The "Filter" Analogy)

This is the most technical part, but here is the simple version:

• The Scenario: Imagine a filter that only lets people who survived a dangerous storm pass through to the next stage of the race.
• The Problem: If your genes helped you survive the storm, you are now in the "survivor" group. If you look only at the survivors, you will think, "Wow, everyone here is a great runner!"
• The Reality: You are ignoring the people who were great runners but got washed away by the storm because they had a specific weakness.
• Hamilton's Point: By removing early deaths (infections/accidents), the researchers are looking only at the "survivors." This creates a selection bias. It makes it look like genetics controls everything about the later years, when in reality, genetics helped them survive the earlier years too.

3. The "Additive" Mistake

The researchers used a math model that assumes your "internal aging" and "external risks" are two separate things that just add up (like adding apples and oranges).

• Hamilton's Counter: In real life, they mix together. Your genes (apples) determine how your body reacts to the environment (oranges).
• The Analogy: A car's engine (genes) and the road conditions (environment) interact. A bad engine breaks down faster on a rocky road. You can't just say, "Let's ignore the rocky road" to see how good the engine is, because the engine's quality was tested by the road.

4. The "Historical" Excuse Doesn't Work

The researchers might say, "But we are studying old people from 100 years ago when infections were common. It doesn't matter for us today."

• Hamilton's Rebuttal:
  1. The data they used is from 100 years ago, so infections were a huge part of the picture.
  2. Infections still kill people today (pneumonia, flu, COVID).
  3. Even if we ignore infections, the same genes that fight infections also fight aging and cancer. So, throwing out infection data throws out vital genetic clues about aging.

---

The Conclusion: Why the 55% Number is Fake

Hamilton concludes that the 55% heritability number is a mathematical illusion.

• The Real Number: Other studies (looking at millions of family trees and DNA) suggest lifespan heritability is actually quite low, around 7%.
• The Flaw: Shenhar et al. didn't "correct" for bad luck; they erased the part of life where genetics plays a huge role (surviving infections).
• The Final Verdict: You cannot calculate the "true" heritability of a population by imagining a world where no one ever gets sick or has an accident. That world doesn't exist, and the math based on it doesn't apply to real humans.

In short: You can't measure how strong a person's legs are by only looking at people who never fell down. You have to look at how they handled the falls, too.

Technical Summary

Based on the preprint provided, here is a detailed technical summary of the paper "Misclassification of heritable mortality undermines estimates of intrinsic life span heritability" by Fergus Hamilton.

1. Problem Statement

The paper critiques a recent study published in Science by Shenhar et al., which claims that human lifespan heritability is approximately 55% after correcting for "extrinsic mortality" (deaths caused by external factors like accidents and infections).

• The Core Conflict: Shenhar et al. argue that extrinsic mortality acts as environmental noise that masks genetic signals, and removing it reveals a much higher intrinsic heritability than previously estimated by large pedigree studies (which suggest heritability is often <10%).
• The Author's Challenge: Hamilton argues that the premise that extrinsic mortality (specifically infectious disease) is purely environmental and independent of genetics is biologically untenable. By treating infection susceptibility as non-genetic, Shenhar et al. introduce a fundamental methodological error that inflates heritability estimates rather than correcting for confounding.

2. Methodology and Theoretical Framework

Hamilton deconstructs the methodology used by Shenhar et al., focusing on the Gompertz-Makeham model and statistical assumptions:

• The Model Flaw: Shenhar et al. decompose mortality ($m(t)$) into an extrinsic component ($m_{ex}$) and an intrinsic aging component ($\theta$). They assume $m_{ex}$ is a cohort-level constant shared by all individuals and that genetic variation only operates through $\theta$.
• The Counterfactual Error: The authors define "heritability of intrinsic lifespan" (HIL) by extrapolating to a counterfactual world where $m_{ex} = 0$. Hamilton argues that because the model assumes genetics do not influence $m_{ex}$, setting $m_{ex}$ to zero mathematically forces the remaining variance to be attributed to genetics, artificially inflating the heritability estimate.
• Collider Bias: Hamilton identifies that conditioning on survival (by removing early deaths from extrinsic causes) creates selection bias. If genetic factors influence both susceptibility to extrinsic death (e.g., infection) and intrinsic aging, survival acts as a "collider." Conditioning on survival induces a spurious correlation, making heritability appear higher even if the underlying genetic architecture hasn't changed.
• Additivity Assumption: The critique highlights that the Gompertz-Makeham framework assumes intrinsic and extrinsic mortality are additive and independent. This ignores gene-environment interactions (e.g., sickle cell trait vs. malaria), where genetic risk is only expressed through specific environmental insults.

3. Key Contributions and Evidence

Hamilton provides three lines of evidence to refute the claim that infection mortality is non-genetic:

• Immunogenetic Evidence:
  • The HLA/MHC region is highly polymorphic due to pathogen-mediated balancing selection.
  • Specific genetic variants (e.g., in red blood cells) are known drivers of susceptibility to malaria and other infections.
• Adoption and Twin Studies:
  • Citing Sørensen et al. (1988): Biological parents' death from infection before age 50 conferred a Relative Risk (RR) of 5.81 for infection death in adoptees, whereas adoptive parents' death conferred no increased risk (RR 0.73).
  • Danish twin studies confirm infection mortality heritability at ~40%, with monozygotic concordance significantly exceeding dizygotic concordance.
• Contradiction with Molecular Data:
  • If lifespan were truly 55% heritable, Genome-Wide Association Studies (GWAS) should explain a substantial fraction of variance. Instead, the largest longevity GWAS identify only a few loci (e.g., APOE, FOXO3) explaining a small fraction of variance, consistent with lower heritability estimates (<10%).

4. Results and Findings

• Simulation Confirmation: Hamilton's simulations (referenced as Supplement S1) demonstrate that when genetics influence both extrinsic death risk and intrinsic lifespan, removing extrinsic deaths causes Falconer's heritability ($h^2$) to increase artificially, replicating Shenhar et al.'s pattern without any actual change in genetic architecture.
• Inconsistency in Age Restriction: The paper notes that the "correction" is applied inconsistently. While extrinsic mortality is removed for young/middle-aged cohorts, non-infectious extrinsic mortality (accidents, falls) actually increases in older age groups (U-shaped curve), yet is not removed there.
• Historical Relevance: The critique emphasizes that the cohorts studied by Shenhar et al. (Danish/Swedish twins born 1870–1935) lived during eras of high infectious mortality. Furthermore, infection remains a top cause of death today (e.g., pneumonia, influenza, COVID-19), and GWAS for COVID-19 severity have identified specific genetic loci (OAS1, TYK2).

5. Significance and Conclusion

• Refutation of High Heritability: The paper concludes that the ~55% heritability estimate is not the true heritability of any actual human population (past or present). It describes a biologically implausible counterfactual population where genetic susceptibility to infection has been mathematically erased.
• Methodological Warning: The study warns that treating gene-environment interactions as "noise" to be removed leads to selection bias rather than confounding correction.
• Implication for Future Research: The findings suggest that the "missing heritability" in lifespan is not hidden by extrinsic noise but is likely much lower than 55%, aligning better with large-scale pedigree and GWAS data. The paper calls for a re-evaluation of models that assume independence between genetic susceptibility to environmental causes and intrinsic aging.

In summary: Hamilton argues that Shenhar et al. made a category error by classifying infection susceptibility as purely environmental. By mathematically removing a highly heritable cause of death, they inadvertently forced the statistical model to attribute all remaining variance to genetics, resulting in a gross overestimation of intrinsic lifespan heritability.