You have 4 million variants
in one patient's genome.
Which one is the culprit?

Forward vs reverse genetics

Today: forward. Next chapter: reverse.

Roadmap for today

1. Evolution of gene hunting · 1860s to today
2. GWAS · finding common variants
3. What GWAS tells you · and what it doesn't
4. Burden tests · finding rare variants
5. GWAS vs burden tests · two sides of one coin
6. Case study · schizophrenia architecture

1860s – 1900s · classical genetics

• Mendel: pedigrees and breeding
• Morgan: chromosome mapping in flies
• Tools: cross-breeding · pedigree analysis
• Output: inheritance patterns

1980s – 1990s · linkage analysis

• Track chromosomal regions through families
• Microsatellite markers · cM resolution
• Major successes: CFTR · HTT · DMD
• Limited to rare, high-penetrance Mendelian disease

2000s – 2010s · the GWAS era

• Microarrays: genotype millions of SNPs at scale
• 10,000 to 1,000,000 individuals
• Common variants only · MAF > 1%
• 100+ loci per complex trait

2010s – present · sequencing era

• Whole-exome and whole-genome sequencing
• Burden tests: aggregate rare variants per gene
• Reveals rare-variant architecture
• 10,000 to 100,000 individuals

The simple question

Across many people, are certain genetic variants more common in individuals with a disease than in those without?

The procedure

1. Recruit cases (e.g. 50,000 with diabetes) and matched controls
2. Genotype ~1 million SNPs per person
3. For every SNP: test frequency in cases vs controls
4. Apply genome-wide significance threshold: p < 5 × 10⁻⁸
5. Report list of associated loci

Real example · type 2 diabetes

• Suzuki et al. 2024, Nature
• 1.4 million participants
• 600+ risk loci identified
• Each individual locus shifts risk by ~5–10%
• Together: pathways for insulin secretion, beta-cell function

The summary statistics

Five applications of summary statistics

• Polygenic risk scores — combine thousands of variants into one number
• Genetic correlation — find shared architecture between traits
• Mendelian randomisation — natural experiments for causation
• Fine-mapping — narrow loci to credible causal variants
• Gene prioritisation — combine with eQTL data

Limitation 1 · most hits are not in genes

• Variant might be 500 kb away from the nearest gene
• Which gene does it regulate? Need extra experiments

Limitation 2 · linkage disequilibrium

• Multiple variants are correlated at any locus
• Statistical hit ≠ causal variant
• Need fine-mapping to narrow down

Limitation 3 · common variants only

• GWAS tests one variant at a time
• Sample size of one cannot pass significance
• Misses rare, high-impact mutations entirely
• For rare variants, switch to burden tests

The rare-variant problem

• A variant is found in 1 person out of 50,000
• Sample size of one · no per-variant statistics
• But many such variants together may converge on one gene

The burden test idea

Don't test variants. Test genes.

• For each gene · count damaging variants in cases vs controls
• Aggregate across the gene
• Excess in cases → gene is a risk factor

How burden tests work

1. Sequence cases and controls (~10,000 + 10,000)
2. Filter to damaging variants — LoF, predicted-deleterious missense
3. Filter to rare — MAF < 0.1%
4. Per gene · count carriers in cases vs controls
5. Statistical test · genome-wide threshold p < 2.5e-6 (~20,000 genes)

Real example · autism risk genes

Satterstrom et al. 2020, Cell

Where burden tests shine

• De novo mutations — recurrent across unrelated cases
• Ultra-rare private variants — aggregated across a gene
• Recessive diseases — compound heterozygotes / homozygotes
• Gene intolerance — depleted in healthy controls (gnomAD pLI)

Two methods · two regimes

Volume knob vs light switch

GWAS findings

• 287 genome-wide significant loci
• Each variant: OR 1.03 – 1.15 (3 – 15% risk)
• Together: ~20% of liability variance
• 95% in non-coding regions
• Trubetskoy et al. 2022, Nature

Burden findings

• 10 high-confidence risk genes
• Each LoF mutation: OR 3 – 50 (3- to 50-fold risk)
• Frequency: ~0.01 – 0.1% of cases per gene
• Top genes: SETD1A · GRIN2A · GRIA3 · TRIO
• Singh et al. 2022, Nature

Biological convergence

Both common and rare variants converge on the same pathways.

• Synaptic function — glutamate signalling
• Chromatin regulation — gene expression control
• Neurodevelopment — early cortical patterning

Association is not causation

• GWAS or burden hit → candidate gene
• Need to verify that the gene actually causes phenotype
• What does it do? How does it act? Can we rescue it?
• The next half is reverse genetics

What to take away

• Forward genetics: phenotype → gene
• GWAS finds common, small-effect, mostly non-coding variants
• Burden tests find rare, large-effect, protein-coding variants
• The two methods complement, not compete
• Together → full architecture of disease (Schizophrenia)
• Association requires reverse genetics for mechanism

From candidate gene
to mechanism

Chapter 26 · Reverse Genetics — From Gene to Function