A skin cell and a nerve cell in your body carry the same three billion letters of DNA, and they could hardly be less alike. One sheds and renews every few weeks; the other can last a lifetime and sends signals over a metre. Neither has lost a page of the manual. They differ in which pages are open.
Pour a cup of milk for a bacterium and something similar happens. A bacterium that has been living on glucose will switch on a handful of genes for digesting milk sugar, but only when the sugar is there and the better food is gone. Nobody told it. The switch is a molecule that sticks to a stretch of DNA.
A molecular biologist sees all of this as one story: information stored in DNA is copied, read, and switched on and off, by machines made of the very molecules they read. Everything in later biology, from inheritance to development to cancer, is a variation on that story.
In 1958 Francis Crick stated an idea about sequence information: it can pass from nucleic acids to protein, but once it is in a protein it cannot get back out. In a 1970 paper in Nature he restated it carefully, and the version he wrote is stricter and more interesting than the textbook cartoon "DNA makes RNA makes protein". He listed which transfers are general (DNA to DNA, DNA to RNA, RNA to protein), which are special (RNA to RNA, RNA to DNA, and DNA straight to protein), and which are unknown (from a protein to anything).
Some "special" cases turned out to be everywhere. Retroviruses such as HIV carry RNA and copy it into DNA using reverse transcriptase, found independently by Howard Temin and David Baltimore in 1970 (Nobel Prize in Physiology or Medicine, 1975). Many RNA viruses copy RNA straight into RNA. And RNA is not only a messenger: Thomas Cech and Sidney Altman showed that some RNAs are catalysts, enzymes made of RNA (Nobel Prize in Chemistry, 1989).
Prions are the case people think breaks the dogma. A misfolded protein makes other copies of the same protein misfold. Stanley Prusiner won the 1997 Nobel Prize for the idea. But no sequence is copied from protein to anything; the amino-acid order is still written in the host's gene. The dogma is about sequence, and prions pass on a shape.
Categories follow Crick (1970). "Seen" means observed in nature or, for the special DNA-to-protein case, in a test tube.
DNA is two strands, each the template for the other. Matthew Meselson and Franklin Stahl showed in 1958 that copying is semi-conservative: each new double helix keeps one old strand and builds one new one. The enzyme that builds is DNA polymerase, which only adds letters to an existing end, in one direction, using the old strand as the pattern: A pairs with T, and C with G.
The remarkable part is how rarely it slips. A polymerase choosing letters gets roughly one in 105 wrong. Many polymerases also carry a proofreading arm that backs up and removes the last letter if it is wrong, which cuts errors by a factor of about a hundred. After the copy, a separate mismatch repair system scans the new strand for the bumps where a wrong letter sits, and fixes most of what is left, by roughly another hundred-fold. Those are textbook orders of magnitude, not exact constants, but the structure is the point: three filters in a row, each cheap.
Illustrative orders of magnitude (10−5 raw, ÷100, ÷100). Genome sizes are approximate: phage lambda 48.5 kb, E. coli K-12 4.6 Mb, baker's yeast 12 Mb, human about 3.1 Gb per haploid set, so 6.2 Gb in a body cell.
With all three filters on, copying a human body cell's roughly 6.2 billion letters leaves about six mistakes. With none, about sixty thousand. The mistakes that remain are the raw material of mutation (see Genetics, 35/46, and Evolution, 35/47). Perfect copying would be a trap too: a species that never changed could not adapt.
To use a gene, the cell does not read DNA directly. RNA polymerase copies one strand of the gene into a messenger RNA, the same letters but with U instead of T. In bacteria the message goes straight to a ribosome. In humans and other eukaryotes it is first edited: non-coding stretches (introns) are cut out and the coding stretches (exons) joined, and cutting in different ways, alternative splicing, lets one gene make several proteins. Humans have roughly 20,000 protein-coding genes, a number that surprised people in 2001 and 2004 when the genome was sequenced, because the real richness is in splicing and regulation.
The ribosome reads the message three letters at a time. Each triplet, a codon, is matched by a transfer RNA carrying one amino acid, and the ribosome links them into a chain that folds into a protein. With four letters there are 43 = 64 codons for 20 amino acids: 61 codons code for an amino acid, and three (UAA, UAG, UGA) say stop. AUG both codes for methionine and is the usual start signal. The code is redundant but not ambiguous: leucine, serine and arginine each have six codons, while methionine and tryptophan have one each. Marshall Nirenberg and Heinrich Matthaei cracked the first word in 1961: an RNA made only of U's produced a chain of phenylalanine. The code is almost universal across life; a few exceptions exist, such as mitochondria reading UGA as tryptophan.
The worked example is the most famous single-letter change in medicine. Haemoglobin's beta chain begins with the codons GTG, CAC, CTG, ACT, CCT, GAG (after the opening ATG). The sixth is GAG, which gives glutamic acid. In sickle cell disease it is GTG, which gives valine. One letter, one amino acid, a protein that sticks to its neighbours when oxygen is low. Try it, and try breaking it other ways.
Sequence: the start of the human HBB gene, bases 51 to 92 of RefSeq NM_000518.5, plus a stop codon added for the example. The mRNA matches the coding strand with U for T; the template strand is its reverse complement. Standard genetic code, NCBI table 1.
Notice what the third preset does. Deleting one letter does not change one amino acid; it shifts every codon after it. That is a frameshift, and it usually ends in an early stop. The reading frame is why insertions and deletions of one or two letters are so much more damaging than one swapped letter.
François Jacob and Jacques Monod worked out the first complete switch in the early 1960s, in E. coli (Nobel Prize, 1965, shared with André Lwoff). The lac operon is a cluster of genes for taking up and digesting lactose, under two controls. A repressor protein sits on a stretch of DNA called the operator and blocks RNA polymerase. When lactose is around, a lactose by-product binds the repressor and it lets go. That is the first control, a "don't make the enzymes when there is nothing to digest" switch.
The second is an activator, CAP, that helps the polymerase but only when glucose is scarce, which the cell senses through a small molecule called cyclic AMP. So the logic is an AND gate with a twist: full expression needs lactose present and glucose absent. The cell does not waste effort on milk sugar while its preferred food is in the dish.
Expression numbers are illustrative relative levels (induced = 100), not measurements. Mutant strains as in the classic experiments: a broken repressor gene, or an operator the repressor cannot recognise.
Humans use the same ideas with far more parts. Instead of one operator, a gene may be controlled by many transcription factors that bind enhancers, DNA switches that can sit thousands, even a million letters from the gene. A limb enhancer for the gene Sonic hedgehog lies about a million letters away, and a single-letter change in it can cause extra fingers. A cell type is a particular pattern of factors turned on: the genome is a library, and the factors choose the books.
"Epigenetic" originally meant everything that makes a cell become what it is. Today it usually means chemical marks on DNA and on the histone proteins that DNA winds around, mainly DNA methylation and histone modifications, that change how readable a gene is without changing its letters. These marks can be copied when a cell divides, which is how a liver cell's daughters stay liver cells.
What is solid: marks hold cell identity; in females one X chromosome in each cell is silenced by marking (Mary Lyon, 1961); some genes are switched on from only the mother's or only the father's copy (imprinting); cancers carry abnormal methylation; and methylation patterns drift with age so reliably that "epigenetic clocks" (Steve Horvath, 2013) estimate age from a blood or tissue sample. Identical twins are more different epigenetically when they are older (Fraga and colleagues, 2005).
What is oversold: the claim that your grandparents' hunger or trauma is written into your genes. Epigenetic inheritance across generations is documented in plants and in some animals, including worms, but in mammals most marks are wiped in the early embryo and germ cells. Studies of people born around the Dutch Hunger Winter (1944 to 1945) found differences in methylation decades later, a real finding, but it shows the womb environment leaves marks, not that marks are passed to grandchildren. Whether and how much this happens in people is open. If a headline says "trauma inherited in DNA", read the study.
PCR, the polymerase chain reaction, was conceived by Kary Mullis in 1983 (Nobel Prize in Chemistry, 1993). It is replication, repeated: heat splits the double strand; short primers stick to the two ends of the stretch you want; polymerase builds new copies; repeat. Each cycle roughly doubles the number of copies of that stretch, so n cycles give about 2n times more. Thirty cycles is about a billion-fold. A heat-tolerant polymerase from a hot-spring bacterium, Thermus aquaticus, is what made the cycling practical. PCR is behind most infectious disease tests, forensic profiles and ancient DNA work.
copies = start × (1 + efficiency)cycles. Real reactions run out of ingredients and plateau; this page caps at an illustrative 1012.
CRISPR began as a puzzle in bacterial genomes: strange repeated stretches, noticed in 1987 by Yoshizumi Ishino's group in Osaka and characterised from the 1990s by Francisco Mojica in Alicante. In 2007 Rodolphe Barrangou, Philippe Horvath and colleagues showed it is an immune system: bacteria keep snippets of virus DNA as a "most wanted" list, and a protein (Cas9) guided by a matching RNA cuts any DNA that matches. In 2012 Jennifer Doudna, Emmanuelle Charpentier and colleagues showed it could be programmed, by changing the guide, to cut a chosen site; the pair shared the 2020 Nobel Prize in Chemistry. In 2013 teams including those of Feng Zhang and George Church made it work in human cells.
As an idea it is a search-and-cut tool: a guide of about 20 letters finds the one place in the genome it matches. Why does 20 work? The number of possible 20-letter strings is 420, about 1.1 trillion, much larger than the genome, so a random match is rare. Slide the guide length and see.
Expected chance matches = 2 × 3.1×109 ÷ 4L (both strands), assuming random sequence. Real genomes are not random and the tool tolerates some mismatches, which is why off-target cuts are a real engineering problem.
The first CRISPR-based medicine, exa-cel (Casgevy), received conditional approval in the UK in November 2023 and US approval in December 2023 for sickle cell disease and a related blood disorder. It edits a patient's own blood stem cells outside the body. A very different case: in 2018 the Chinese researcher He Jiankui announced that he had edited human embryos that became babies. It was condemned worldwide as premature and unethical, and he was later sentenced to prison in China. Editing eggs, sperm or embryos, which would pass changes on, is a policy debate, not a routine technique. None of this is a guide to doing it, and none of it is medical advice.
Information in DNA is copied with layered accuracy, read three letters at a time into proteins, and switched on and off by regulators that behave like logic gates. One genome, many cell types; a few strange cases such as prions and reverse transcription fit once the dogma is read as being about sequence. PCR and CRISPR are what you get when you turn replication and bacterial immunity into tools.
Try this: take any short word and read it in the wrong frame, moving the spaces by one letter. That is a frameshift, and it is why a one-letter deletion is worse than a swap.
Sources. Crick, "Central dogma of molecular biology", Nature 227 (1970); Meselson and Stahl, PNAS (1958); the Nobel Prizes of 1965, 1968, 1975, 1989, 1993, 1997 and 2020 (nobelprize.org); Jacob and Monod, J Mol Biol (1961); Jinek and others, Science (2012); Barrangou and others, Science (2007); NCBI RefSeq NM_000518.5 and the standard genetic code (NCBI table 1); Mullis, Nobel lecture (1993); Fraga and others, PNAS (2005); Horvath, Genome Biology (2013); Heijmans and others, PNAS (2008); UK MHRA (16 November 2023) and US FDA (December 2023) on Casgevy. Error rates are textbook orders of magnitude (Alberts, Molecular Biology of the Cell). Describes, and is not medical advice.