Manolis Kellis: Biology of Disease | Lex Fridman Podcast #133

Scientific Paradigm Shift

• Human genetics has overtaken traditional model organisms (mice, flies, yeast) as the primary driver for understanding basic biology and disease mechanisms.
• While traditional biology relied on "perturbation experiments" in animals to infer human function, modern science leverages the "natural experiments" of 6 million genetic variants segregating across 7 billion humans.
• Disease complexity is comparable to the human brain's circuitry, involving intertwined layers of genome, epigenome, immune system, and neural circuitry complexity.

Data Scale and Power Analysis

• Every individual carries approximately 6 million unique genetic variants (polymorphic sites), representing 6 million potential "perturbations" or experiments.
• Statistical power for detecting disease associations depends on effect size (difference between variant means) and natural phenotypic variance (standard deviation).
• Large, well-phenotyped cohorts (e.g., Electronic Health Records) enable the analysis of millions of variants against hundreds of phenotypes simultaneously, moving beyond single-trait family pedigrees.

Methodology: From Association to Mechanism

• The 6-Step Pipeline to Causality:
  • 1. Tissue Identification: Use epigenomic enrichment to determine which tissue/cell type drives the disease signal (e.g., Microglia for Alzheimer's, Pancreatic Islets for Type 2 Diabetes).
  • 2. Causal Variant Identification: Narrow down from blocks of co-inherited variants (Linkage Disequilibrium) to a single nucleotide responsible (e.g., RS1421085 for obesity).
  • 3. Motif Disruption: Identify the specific regulatory DNA word (motif) disrupted by the causal variant.
  • 4. Regulator Identification: Determine which protein (transcription factor) fails to bind due to the motif disruption (e.g., ARID5B in the FTO obesity locus).
  • 5. Target Gene Discovery: Link the regulatory element to its target gene using physical (Hi-C), genetic (eQTL), activity, or perturbation links.
  • 6. Pathway Function: Determine the biological process mediated by the target gene (e.g., IRX3/IRX5 controlling thermogenesis vs. lipogenesis).
• Genomic Circuitry: 93% of disease-associated genetic variants lie outside protein-coding regions, often influencing genes located millions of nucleotides away via long-range chromosomal interactions.

Key Disease Insights & Case Studies

• Alzheimer's Disease:
  • Surprisingly, genetic variants showed zero enrichment in standard neuronal brain samples.
  • Variants were strongly enriched in Microglia (resident brain immune cells), revealing an unexpected immune basis for the disease.
  • Disease progression models now track pathology from the entorhinal cortex to the neocortex, identifying early biomarkers for intervention.
• Obesity (FTO Locus):
  • The "FTO" gene was a red herring; the causal variant (RS1421085) controls IRX3 and IRX5 genes located 1.2 million nucleotides away.
  • IRX3/IRX5 act as master regulators deciding between white fat storage (lipogenesis) and beige fat energy burning (thermogenesis).
  • Manolis Kellis carries a homozygous risk variant that prevents thermogenesis, causing him to store energy rather than burn it as heat.
• Schizophrenia:
  • Variants converge on synaptic pruning mechanisms mediated by the Complement system (an immune pathway).
  • Disrupted pruning leads to abnormal neural connectivity; this immune-genetic link was previously unknown.

Technological Enablers

• CRISPR-Cas9 & Base Editing:
  • Utilized not just for "genome vandalism" (cutting), but for precise editing (prime editing) and regulatory modulation (dead Cas9) to turn enhancers on/off.
  • Allows for the systematic testing of causality by editing single nucleotides to flip phenotypes (e.g., obese to lean).
• Massively Parallel Reporter Assays (MPRA):
  • Technologies like Hydra and Sharper MPRA allow the testing of up to 7 million genetic variants and enhancer regions simultaneously.
  • Uses pooled assays where 10,000+ hypotheses are tested in a single experiment using barcoded DNA.
• Single-Cell Profiling:
  • Uses microfluidics or "bottleneck" randomization to barcode millions of cells, enabling resolution of cell-type-specific effects (e.g., distinguishing neuron subtypes).
  • Simultaneous single-cell RNA and ATAC-seq (epigenome) allows mapping of enhancer-gene connectivity across cell types.
  • Recent datasets include 10 million cells from human brains across 12 disorders.

Future of Medicine & Therapeutics

• Systems Medicine: The future lies in intervening at the network/pathway level rather than single genes to avoid pleiotropic side effects (where one gene affects multiple traits).
• Cell-Type Specificity: Therapeutics will target specific cell types (e.g., liver vs. brain) using synthetic biology constructs that sense environmental microRNAs.
• Speed of Discovery: The timeline from discovery to therapeutic intervention is shortening from 50 years to roughly 10 years due to computational and high-throughput tools.
• Somatic Mosaicism: Single-cell sequencing allows detection of mutations acquired after the zygote stage, revealing distinct genetic drivers for neuronal energetics and oligodendrocyte function in aging.

Philosophical & Structural Implications

• Biology is no longer siloed by organ or system; genetics unifies the field, forcing immunologists to study neurodegeneration and neurologists to study metabolism.
• The genome is viewed as a digital circuit; understanding disease requires a computer science approach to deconvolve massive, noisy matrices of genotype and phenotype.
• Human Longevity: Pushing human lifespan beyond evolutionary limits (120+ years) requires a complete redesign of biological systems to handle new challenges like late-onset cancer and Alzheimer's, similar to redesigning a car to go from 70 to 400 mph.