For years I told people my brother was adopted. He was taller than the rest of us and his eyes were a different colour. It was a joke, but it rested on an instinct everyone shares: children should look like their parents.
Geneticists can put a number on that instinct. Compare identical twins, who share all their DNA, with fraternal twins, who share about half, and you can estimate how much of the variation in a trait tracks with genes. The number is called heritability. For adult height, a study of twin cohorts in eight countries put it between 87 and 93 percent in men and between 68 and 84 percent in women.1 Round it to 80 percent and you won't be far off.
Notice what that number does not tell you. It says nothing about which genes are involved, how many there are, or where in the genome they sit. Twin studies measure the shadow, not the object casting it.
The object goes missing
By the mid-2000s, geneticists thought they could finally see the object. The human genome had been sequenced, and cheap genotyping chips could read hundreds of thousands of common DNA variants in a single person. The plan was simple: collect thousands of people, measure their height, and look for variants that turn up more often in tall people than short ones. This is a genome-wide association study, or GWAS.
In May 2008, three groups published the first large GWAS of height in the same issue of Nature Genetics. Between them they found about 40 regions of the genome linked to height. Added together, those regions explained roughly 5 percent of the differences in height between people.2
Five percent, against a twin estimate of 80. Height was not an outlier, either. The same pattern showed up for diabetes, heart disease and schizophrenia. Later that year a news feature in Nature gave the problem its name: the case of the missing heritability.3 A group of leading geneticists followed in 2009 with a list of places the rest might be hiding: rare variants too uncommon for the chips to see, structural changes the chips couldn't read at all, and interactions between genes.4
There was a more unsettling possibility. In 2012, Or Zuk and colleagues at the Broad Institute showed that if genes interact with each other, the twin estimates themselves can be inflated.5 Some of the heritability might never have existed. They called it phantom heritability.
Spread too thin to see
The first real clue came from Jian Yang, Peter Visscher and colleagues in 2010. Instead of hunting for individual variants that cleared a strict statistical bar, they asked a different question: if you take all 294,831 variants on the chip at once, how much of the variation in height do they capture together? In 3,925 unrelated people, the answer was 45 percent.6
That changed the shape of the mystery. Most of the heritability was not missing. It was spread across so many variants, each with such a small effect, that no single one stood out from the noise. Each one might nudge height by a millimetre or two. To see a millimetre, you need a very large crowd.
The crowd took fourteen years to assemble. In 2022, a consortium led by Loïc Yengo and Joel Hirschhorn published a GWAS of 5.4 million people. They found 12,111 independent variants associated with height, clustered in regions covering about a fifth of the genome. In people of European ancestry those variants explained about 40 percent of the variation in height, which is close to everything common variants can explain.7 For common variants and height, the map is essentially finished.
The same year, Pierrick Wainschtein and colleagues sequenced the whole genomes of 25,465 people rather than reading a chip. Whole-genome sequencing sees rare variants that chips miss. Their estimate of height heritability rose to 68 percent, much closer to what the twins had promised all along.8 Much of the remaining gap sat in rare variants, especially ones that change proteins.
What is still missing
So the story of height has a mostly happy ending. Two parts of it do not.
The first is ancestry. The same 12,111 variants that explain 40 percent of height variation in Europeans explain only about 10 to 20 percent in people of other ancestries.7 The biology of height is not different on other continents. The studies are. Most of the people who have been genotyped are of European descent, and the variants that tag causal DNA in one population don't always tag it in another. A genetic prediction that works well in Vancouver can fail in Lagos.
The second is the parts of the genome we couldn't read. Until 2022, about 8 percent of the human genome was simply absent from the reference sequence everyone compared against. The Telomere-to-Telomere consortium filled it in, adding nearly 200 million bases of sequence that included 1,956 gene predictions.9 A year later, the Human Pangenome Reference Consortium released 47 complete, phased genomes from genetically diverse people, adding another 119 million bases that the old reference lacked. Using it doubled the number of structural variants detected per haplotype.10
This is where my own work sits. Structural variants, the large inversions, duplications and rearrangements of DNA, are hard to see with short-read sequencing. They often sit in repetitive stretches where short reads can't be placed, and standard methods blend a person's two copies of each chromosome into one consensus sequence. Long-read sequencing and single-cell methods like Strand-seq can read the two copies separately and follow rearrangements across them. We are only starting to count how much heritability lives there.
Disease is also further behind than height. Twin studies estimate the heritability of schizophrenia at about 81 percent.11 The largest GWAS, comparing 76,755 people with schizophrenia to 243,649 without, found 287 regions of the genome linked to it.12 That is real progress, and it points clearly at genes active in neurons and synapses. But predicting schizophrenia risk from DNA remains far cruder than predicting height.
My brother is still taller than me. I can't tell you exactly which letters of his genome are responsible. But I can now tell you where most of them are, roughly how many there are, and why it took so long to find them. Fifteen years ago, nobody could.