[@DwarkeshPatel] Why evolution designed us to die fast, & how we can change that – Jacob Kimmel

Link: https://youtu.be/XCLODgdCmKA

Short Summary

Here's a summary of the key takeaway and an executive summary of the YouTube transcript:

• Most Important Action Item/Takeaway: Focus on problems evolution hasn't fully optimized for, as these areas offer the most promising opportunities for intervention and improvement.

• Executive Summary: Jacob Kimmel of NewLimit argues that evolution hasn't optimized for human lifespan, leaving room for significant improvements through epigenetic reprogramming. NewLimit aims to identify combinations of transcription factors to remodel the epigenome and revert cells to a younger, healthier state, offering potential for treating age-related diseases.

Key Quotes

Here are five quotes from the transcript that I consider particularly insightful:

1. "You always have to start by asking yourself, 'Did evolution spend a lot of time optimizing this? If yes, my job is going to be insanely hard. If not, potentially there are some low-hanging fruit.'" This provides a valuable framework for approaching biological engineering problems.

2. "So there is a notion by which a population being laden demographically with many aged individuals, even if they did have fecundity persisting out some period later in life, is actually net negative for the genome's proliferation and that really a genome should optimize for turnover and population size at max fitness." This highlights the counterintuitive idea that aging might be selected for due to population-level dynamics.

3. "It turns out that TF's binding DNA is a pretty darn big surface. Small molecules aren't great at disrupting that and certainly even worse at activating it. Small molecules can get all the way into the nucleus, but they can't do much once they're there. They're just too small." This succinctly explains the limitations of traditional small molecule drugs for targeting transcription factors.

4. "Ultimately, we're probably going to have to solve delivery the way that our own genome solved delivery... The problem was solved by the immune system." This suggests a potential bio-inspired approach to drug delivery by leveraging the inherent capabilities of the immune system.

5. "If you have more of these medicines for everyone, medicines that keep you healthier longer rather than medicines that only fix a problem once you're already very sick, I think you actually avoid a lot of the types of administration costs." This argues that preventative medicines could potentially reduce overall healthcare spending by shifting the focus from treating late-stage diseases to maintaining health.

Detailed Summary

Here's a detailed summary of the YouTube video transcript, focusing on the key topics, arguments, and information, excluding sponsor announcements and other advertisements:

• Introduction:

  • Jacob Kimmel, president and co-founder of NewLimit, is interviewed. NewLimit focuses on epigenetically reprogramming cells to their younger states.

• Evolutionary Perspective on Aging:

  • The Puzzle: Why doesn't evolution select for longer, healthier lifespans to allow for more offspring and better care for them?
  • Multiple Angles:
    • Selective Pressure: Is there sufficient evolutionary pressure to live longer?
    • Anti-Selective Pressure: Are there pressures working against longevity?
    • Optimizer Constraints: Are there limitations on how natural selection can optimize for longevity?
  • High Baseline Hazard Rate: Historically, the likelihood of death from non-aging-related causes (disease, accidents, predators) was very high. This reduced the selective pressure for longevity because individuals were unlikely to reach the point where aging was the primary limitation.
  • Analogy to Intelligence: Similar to arguments about why human-level intelligence took so long to evolve, lifespan extension might not have been a primary focus due to high hazard rates. The need to rapidly reach adulthood to contribute resources to the group would also compete with traits like intelligence that require a prolonged adolescence.
  • Engineering Perspective: If evolution hasn't strongly optimized for a trait, it might be easier to engineer it (lower-hanging fruit).

• Kin Selection and Aging:

  • Selfish Gene View: If evolution prioritizes genome propagation rather than individual well-being, optimizing for longevity becomes complicated.
  • Regularization Term: Extending lifespan without eliminating aging may lead to older, less fit individuals consuming resources that could be used to support more reproductively active younger individuals. Aging can therefore be seen as a "length regularizer," limiting calorie consumption per genome.
  • Population Turnover: The genome might optimize for turnover and population size at max fitness, rather than long individual lifespans.

• Optimization Constraints on Longevity:

  • ML Analogy: The genome is like a set of parameters, and evolution is like an optimization algorithm. Similar to the limitations of training neural networks, there are constraints on how the genome can be optimized.
  • Mutation Rate: A low mutation rate limits the step size for genetic updates. Too high a mutation rate leads to cancer.
  • Population Size: Limited population sizes constrain the number of variants that can be screened in parallel.
  • Infectious Disease: Much of evolutionary pressure is spent on fighting infectious diseases, diverting resources from longevity optimization.
  • Conclusion: A combination of absent positive selection, potential negative selection, and optimization constraints may explain why humans don't live forever. Longevity can be thought of as not having a high enough weight on the things the genome is optimizing for.

• Antibiotics and Evolution:

  • The Question: Why didn't humans evolve their own antibiotics?
  • Source of Antibiotics: They are metabolites largely of other bacteria or other fungi.
  • Evolutionary Arms Race: Pathogens and hosts (or competing microbes) engage in a rapid evolutionary arms race. Bacteria evolve resistance, and fungi evolve new antibiotics.
  • Population Size and Mutation Rate: Bacteria have vast population sizes and high mutation rates, allowing them to adapt quickly. Metazoans like humans have lower population sizes and lower mutation rates because of cancer risks.
  • Historical "Naive Antibiotics": There may have been historical antibiotics that pathogens have already evolved around. Evidence exists for co-evolution of pathogens and hosts with genes that have been lost.

• Gene Duplication and Evolution:

  • Explaining Adaptation: Gene duplication is a key mechanism for adapting to new environments and pathogens.
  • Copy-Paste Mechanism: When a new environmental challenge arises, a gene can be duplicated, allowing one copy to maintain its original function while the other copy mutates to address the new challenge.
  • Tolerated Edits: Through duplication you can create a scenario where the edits are tolerable because of a backup copy.
  • Homologous genes: Duplication and swapping of genes with similar homologous sequences, leads to specialized functions.

• NewLimit's Approach: Epigenetic Reprogramming:

  • The Epigenome: It's a layer of regulation that determines which genes are expressed in different cell types. It can degrade with age.
  • Transcription Factors: They are the "orchestra conductors" of the genome that bind to DNA and control gene expression.
  • Remodeling the Epigenome: NewLimit aims to remodel the epigenome back to a younger state by finding combinations of transcription factors that can shift chemical marks on DNA.
  • Challenges: Each transcription factor binds to many places in the genome. Aging might not involve moving perfectly along any of the vectors in the basis set.
  • Looks Like Assay: Reverting aged cells to younger looking ones at a gene level.
  • Functional Level: Can I actually make an aged cell perform it's roles the same way a young cell would?

• Addressing Potential Deleterious Effects:

  • Canonical Examples: They can reverse age of a cell, but might simultaneously be changing cell type.
  • Tumor Risk: Using Shinya Yamanaka's factors to induce pluripotency can cause tumors (teratomas).
  • Cell Type Verification: NewLimit checks whether reprogrammed cells still look like the correct cell type and screens for other pathologies.

• The Need for AI Models:

  • Yamanaka's Experiment: He found factors with high expression in embryonic cells, and then tested which sets made a somatic cell turn into a stem cell.
  • Complexity of Aging: Measuring success in reversing aging is much more complex than measuring stem cell conversion. It needs a really complex molecular measurement.
  • Single-Cell Genomics: Enabled by new tech, taking a cell, ripping it open, sequencing the mRNAs it's using.
  • Discriminatory Models: Models are trained to discriminate young and aged cells based on gene expression profiles.
  • Amplification Problem: Unlike Yamanaka's factors, a medicine needs to be efficient across many cells and has a much higher bar.
  • Combinatorial Search Space: Screening all possible combinations of transcription factors (10^16) is not tractable. Models can predict the effect of interventions from sparse sampling.
  • Generative Problem: Models can be used to generate and sample combinations most likely to take the cell to a target state.

• Transcription Factors as a Modality:

  • Argument For This: TFs have relatively modular, self-contained effects that evolution designed.
  • Evidenced by development: You and I were a single cell, then a bag of undifferentiated cells. Then somehow were are humans with hundreds of different cell types. All done by TF's.
  • Limited Edit Size: Small edits to TFs can lead to large phenotypic changes due to evolutionary constraints.
  • Analogies to Code: TFs are like the "queries" in an attention mechanism, with genome sequences as "keys" and genes as "values."

• Pathogens and Transcription Factors:

  • HIV Example: HIV uses a protein called Tat to activate NF-κB, a master transcription factor in immune cells, to drive its own transcription.

• Why Aren't There More Drugs Targeting Transcription Factors Directly?

  • Indirect Targeting: Many existing drugs affect transcription factor activity indirectly through receptors and signaling pathways.
  • Small Molecule Limitations: Small molecules are often too small to disrupt the large interface between transcription factors and DNA.
  • Protein Size Limitations: Proteins (recombinant/antibodies) are too big to get through the cell membrane.
  • Nucleic Acid Medicines: New technologies like lipid nanoparticles and viral vectors can deliver RNAs into cells to modulate transcription factor activity.

• Delivery Methods and Future Directions:

  • Current Modalities:
    • Lipid Nanoparticles (LNPs): Fat bubbles get taken up by tissues that take up fat like the liver, they are able to release RNA. Engineering the fats to go to different cell types.
    • Viral Vectors (AAVs): Viruses can get into our cells. Carry DNA genomes, have small delivery trucks. Can go to certain cell types.
  • NOT Gates: Can engineer sequences to add a NOT gate. Engineering the virus to go to new cell types.
  • Controversial Opinion: Current delivery methods (viral vectors and LNPs) will not be the primary methods in the long term.
  • Future: Immune System Delivery: The immune system is engineered to patrol the body, find arbitrary signals, and deliver cargo.
  • Engineering Cells: Will have cells engineered to perform these ornate function that will live within you.

• Transplant Experiments:

  • Old to Young: Reduced risk of several other diseases and overall better survival in patients that have a younger liver.
  • HSCs: Mostly from book from Frederick Appelbaum. Replace one special cell type, knock-on effects throughout body.
  • Gene Breaking: Mice can have dramatically shortened lifespans because of a single broken gene that encodes transcription factor in their mitochondria.

• Payload Size and Treatment Frequency:

  • Payload Size: TFs that they've found that have efficacy are somewhere between one and five. Encapsulation is current mRNA vaccines. The dose has shown to be not limiting either, with strong efficacy at low doses.
  • Chronic vs. One-Time: Can be in principle one time only.

• Non-Human Transcription Factors:

  • Unlikely to be Right Search Space: The state they're trying to access is encoded by some combination of TFs.
  • No Guarantees: There are no guarantees that aging progresses along the same basis.
  • Examples with Other Organisms: Scientists took yeast proteins and saw that they didn't just take human proteins to make aging cells.

• Elastin and Skin Sagging:

  • Not Cellular: Believed to be caused by a protein called elastin that doesn't polymerize and form new fibers after development.
  • Potential Solution: Programming cells to reinvigorate the polymerization process.

• Eroom's Law:

  • Decline in New Drugs: Consistent decrease in new medicines per billion dollars invested.
  • ML Scaling: This is the opposite of scaling laws you have in ML, where you have exponentially more investment and hype.

• Diminishing Outputs and General Purpose Models:

  • Returns to scale are expected to increase super exponentially: This is in contrast to scaled outputs in Eroom's Law, where products are not scaling in revenue.
  • Traditional investment in biotech: Does not necessarily engender you to be able to treat disease "Y" more readily. It means that the ability to make these molecules more rapidly isn't reducing the largest risk in the process. It isn't treating the biggest risk in the process.
  • General Model in Biology: Most of the risk is not in how to make an antibody to target my particular target, it is what is to target.

• Other Modalities of Drug Discovery:

  • What about a Goldilocks Problem? Where things have to be small enough but large enough? It does not have a formal way of explaining this. The amount of well-known targets could probably fit on a single page. You need to find a better model to combat this.

• What is General Purpose Thing?

  • Virtual Cell: In this model people are trying to do and measure as many molecules as possible and then measuring how the changes in the gene goes to.

• Perturb-seq and its Limitations:

  • Many technical improvements were needed: This meant that the reagents became cheaper. Before, almost all of the experiments were labeled incorrectly.
  • The model today: Has to learn a distribution of effects even just in this small region, it is going to be effective and we can make really amazing products. The NewLimit is a small subset of the virtual cell process.

• Future of Pharma Industry:

  • What will it take to be innovative? Gray market issues are a question of IP enforcement. There is a need for drugs with long-term durability. There needs to be a great mechanism to test them over time.
  • More direct to consumer model: The payer systems might be able to be abstracted. We need to focus on consumer demand. It might be a payment-over-time plan, like financing other large purchases. Healthcare is already 20% of GDP. We need more focus to treat all people. Not just a narrow population.

• Role of Traditional Pharma:

  • Many of them see it a bit like a venture capital firm: They tend to externalize their R&D, and collab to get things done. The smaller biotechs take on most of the smaller discovery.
  • The Market Structure: Has small biotechs (startups) then working with an oligopsony of pharmas. There are limited buyers of the therapeutics.

This bullet-point summary captures the core ideas discussed in the interview and highlights NewLimit's work within the broader context of aging research and drug development.