The idea was PCR, the polymerase chain reaction, and before I explain it I should clear up a confusion I carried for embarrassingly long, because plenty of people share it.
Reading DNA and copying DNA are not the same thing. Reading means working out the order of the letters. Copying means making more of it. Different jobs. The second one is the quiet workhorse that makes almost everything else possible, the reading included.
The problem it solves is that real-world DNA turns up in useless quantities. A single hair. A drop of blood. Far too little to study. Mullis worked out how to borrow the cell's own copying trick. Heat the DNA so the strands separate, add the raw materials and an enzyme that rebuilds each half, and now you have two. Again, four. Again, eight. Thirty rounds turn one molecule into a billion before lunch.
Mullis won the Nobel Prize in Chemistry in 1993. He also spent much of his later life arguing that HIV does not cause AIDS. The man who built the machine for turning a trace into evidence was, on the subject that mattered most in his lifetime, unmoved by evidence. Genius and judgment do not always travel in the same car.
You have relied on PCR without knowing it. The COVID test that laboratories ran was PCR, copying a trace of virus until there was enough to detect. Before you can read a faint signal, you copy it up to something a machine can read. No copying, no reading.
In the labs I work with, copying rarely shows up as plain PCR. It arrives in variants, each with a job. qPCR to measure how much is there. RT-PCR to turn RNA into DNA before anything can read it. ddPCR to partition a sample into thousands of tiny reactions and count what is present when there is almost nothing to count. Different tools, one purpose. Take a faint trace and amplify it until it is impossible to miss.
The part that nearly killed it
The method needs heat, a lot of it, to prise the strands apart each cycle. But the enzyme doing the rebuilding is a protein, and heat wrecks proteins the way it wrecks an egg.
So in the early days a person stood at the bench and added fresh enzyme by hand, every single cycle. Thirty cycles, thirty top-ups, an entire day surrendered to the world's most tedious cooking.
The fix came from a hot spring in Yellowstone. In the 1960s the microbiologist Thomas Brock noticed bacteria thriving in near-boiling water where, by every reasonable expectation, nothing should live. One of them, Thermus aquaticus, carried a polymerase built for that heat. It did not cook. It got on with the job.
Borrow that enzyme and PCR becomes a sealed tube that cycles its own temperature while you go home. Taq is why the technique became a button you press instead of a day you lose. It had been isolated and described in 1976, seven years before Mullis pulled over on Highway 128 with a problem it happened to solve.
The story usually ends there, with the tidy lesson that the missing piece was sitting somewhere nobody thought worth looking. It is a good lesson and it is not the one I keep running into.
What amplification cannot do
Amplification is a solved problem. You can take a molecule that was almost not there and make a billion of it, cheaply, in a sealed tube, overnight. One molecule is a rumour. A billion is evidence.
Evidence of what, though, is a separate question, and it is not answered by chemistry.
A variant of uncertain significance, a VUS, is a spelling the reference dictionary has not catalogued yet. You can amplify it perfectly. You can sequence it to any depth you like. You will still be looking at a change whose meaning nobody has established, and you will be looking at it in front of a clinician who has to decide something.
Most of the world's reference data comes from a thin slice of humanity. India has been badly underrepresented in it. The first phase of the GenomeIndia project sequenced 9,772 individuals across 83 population groups and reported around 180 million variants, 130 million of them on the autosomes and 50 million on the sex chromosomes. A large share of what it found is rare, population-specific, or absent from the databases the rest of the world has been reading against.
This is the part I run into every week. You find a variant in an Indian patient and the global databases shrug, because they were built from people who do not carry it. As more Indian genomes enter the reference, some of those question marks resolve into real warnings. Others resolve the other way. A variant flagged as dangerous only because a Western database had never seen it turns out to be ordinary here, and the alarm was false the whole time.
The chemistry for that second half does not need inventing. Mullis and a Yellowstone bacterium settled it forty years ago and it now costs very little. What is missing is not a method. It is entries in a database, contributed one cohort at a time, by laboratories that have to be funded and staffed and kept running for years before the reference is any good.
That is a slower business than a night drive on Highway 128.
Sources
Mullis KB. Nobel Prize in Chemistry 1993. His account of the Highway 128 drive appears in his Nobel lecture and in Dancing Naked in the Mind Field (1998).
Brock TD, Freeze H. Thermus aquaticus gen. n. and sp. n., a nonsporulating extreme thermophile. Journal of Bacteriology 98(1), 289-297 (1969).
Chien A, Edgar DB, Trela JM. Deoxyribonucleic acid polymerase from the extreme thermophile Thermus aquaticus. Journal of Bacteriology 127(3), 1550-1557 (1976). The isolation of Taq polymerase, seven years before Mullis had the idea that would need it.
Saiki RK, Gelfand DH, Stoffel S, Scharf SJ, Higuchi R, Horn GT, Mullis KB, Erlich HA. Primer-directed enzymatic amplification of DNA with a thermostable DNA polymerase. Science 239(4839), 487-491 (1988).
Mapping genetic diversity with the GenomeIndia project. Nature Genetics 57, 767-773 (April 2025). 9,772 individuals across 83 population groups. Approximately 180 million variants, comprising 130 million on the autosomes and 50 million on the sex chromosomes.
A note on one figure that is not used here. Press coverage of GenomeIndia widely reports 44 million variants not previously recorded in global databases. That figure does not appear in the Nature Genetics marker paper cited above. It appears to originate in the project's detailed manuscript, "An Atlas of Indian Genetic Diversity", posted on medRxiv. If the number is wanted in this piece, cite the preprint and say that it is a preprint.