00 — The whole story
Before this: nothing · Time: ~35 min
This chapter tells the entire subject at low resolution: how information is stored in a cell, how it becomes a working organism, how it passes to the next generation, how it changes, and how we read it. Every later chapter is a zoom into one region of this map.
Read it once now. Read it again after Part 4 — it will look completely different.
What you'll be able to do
- Given one strand of a duplex, write the other, and say what that single redundancy buys for copying, repair and templated reading
- Trace information from DNA to a working protein, and explain why the cell reads a disposable copy rather than the protected original
- Explain why every cell in your body has the same genome but does different things, and name the cell types that genuinely break the rule
- Compute how many new a child should carry from the per-base rate and the genome size, explain why trio studies report fewer, and distinguish what mutation and what recombination each contribute to variation
- Distinguish from , explain why "" says nothing about how common or how strong an is, and say why a trait running in a family is not yet evidence that it is genetic
- Explain what "evolution" means mechanically, in terms of allele frequencies, and classify a described change as the work of one of the four forces
- Explain why both "98% of the genome is junk" and "it's all functional" are wrong, from the protein-coding fraction and the non-coding gene counts
The core idea
Life faces an engineering problem: build a machine complex enough to survive, from instructions small enough to copy reliably, and copy them with enough fidelity to preserve what works but enough infidelity to keep exploring.
Every solution has converged on the same architecture. A long linear molecule stores information as a sequence of four chemical letters. That sequence is copied, in full, every time a cell divides. Portions of it are transcribed on demand into a working copy, which is translated into a protein — a molecular machine that does something. Which portions get read depends on the cell's state, which depends on which proteins are already present, which depends on which portions have been read. The system is a strange loop: the instructions build the machinery that reads the instructions.
That loop is the whole subject. Genetics is the study of how information moves through it across generations; genomics is what happened when we became able to read the entire instruction set at once.
A warning about the code analogy
You are going to be tempted — because you program — to think of the genome as source code. The analogy is useful and I will use it. But its failures are exactly where the biology lives, so let me spoil them now:
| The analogy suggests | Reality |
|---|---|
| Code is separate from data | No such separation. The same sequence is instruction, binding site, structural element, and raw material for the next |
| A program has an entry point and runs | No entry point, no clock, no scheduler. Millions of processes proceed concurrently and stochastically, at rates set by chemistry |
| A blueprint specifies the result | It's closer to a recipe than a blueprint. Nothing in the genome is a picture of a finger. Structure emerges from local rules interacting |
| Bugs cause crashes | Most changes do nothing at all. A few are catastrophic. Predicting which is a central unsolved problem |
| You can version and roll back | There is no rollback. Every lineage carries the accumulated consequences of every ancestor's accidents |
Hold the analogy loosely and it will help you. Hold it tightly and it will mislead you at every important juncture.
1. The storage medium
DNA is a long polymer. Its backbone is monotonous — the same sugar–phosphate unit repeating — and hanging off that backbone is a sequence of four bases: adenine, cytosine, guanine, thymine, written A, C, G, T.
The information is entirely in the order of those bases. Two bits per position.
The molecule is double-stranded, and this is the crucial structural fact: A always pairs with T, and G always pairs with C. The two strands are therefore not independent — each is a complete recipe for reconstructing the other. Written out:
5'- A T G C G T A C -3' "top" strand
| | | | | | | |
3'- T A C G C A T G -5' "bottom" strand — complementary, antiparallel
Everything else follows from that redundancy. Copying is possible because each strand templates the other. Repair is possible because damage to one strand can be corrected by reading across. Reading is possible because a machine can open the helix locally and use one strand as a template.
The 5' and 3' labels mark the chemical asymmetry of the backbone: the two ends of a DNA strand are chemically different, so a strand has direction, and the two strands run antiparallel. This is not pedantry — it dictates why replication is messy (Ch 04), why sequence is written 5'→3' by convention, and why a coordinate on the reverse strand needs care.
A human cell contains about 3.1 billion base pairs per haploid set, and you have two sets — one from each parent. At two bits a base that is roughly 750 MB of raw information, which is to say: the entire specification of a human being is smaller than a video file, and most of it is not specification at all.
2. Reading it: the central dogma
Information flows in a canonical direction:
copies a stretch of DNA into RNA — chemically similar, single-stranded, short-lived. This is a working copy, not the master. It can be made in thousands of identical units, used, and destroyed, without touching the original. Cheap, disposable instances of an expensive, protected source.
reads that RNA in non-overlapping groups of three bases — — and builds a chain of amino acids. There are 64 possible codons and 20 amino acids, so the code is redundant: several codons often mean the same amino acid. Three codons mean stop. One means start (and also methionine).
That chain of amino acids folds into a specific three-dimensional shape, and the shape is the function. Proteins are the actual machinery of the cell: they catalyse reactions, form structures, transport cargo, transmit signals, and — closing the loop — bind DNA to control which genes get transcribed next.
A gene is, loosely, a stretch of DNA that gets transcribed into a functional product. Loosely, because the definition frays under examination — genes overlap, nest inside one another, produce multiple different products, and thousands of them never make protein at all. Chapter 08 takes the definition apart properly.
"Dogma" is a historical misnomer, and the arrow is not one-way in the way the name implies. Retroviruses copy RNA back into DNA. RNA regulates and catalyses. But as a first-order description of where information goes, it holds.
3. The part everyone underestimates: regulation
Here is the fact that reorganises everything.
Every cell in your body contains the same genome. A neuron and a liver cell and a skin fibroblast all carry the same ~6.2 billion base pairs — two copies of the 3.1-billion-base haploid set. They are radically different in shape, behaviour and lifespan — and the difference is not in the information they hold. It is in which parts they read.
(One deliberate exception, worth knowing early: mature lymphocytes. B and T cells physically cut out and rejoin segments of their antibody and T-cell-receptor loci as they develop, so each one really does carry a genome that differs from the — the cell lineage that ends up making eggs and sperm, and the only lineage whose changes are passed on — at those loci (Ch 18). Somatic changes, meaning changes in any other lineage, die with the body that carries them; setting those aside, lymphocytes are the main programmed rearrangement of the sequence itself. Other cell types depart from the rule in cruder ways — red blood cells eject their nucleus as they mature and end up carrying no genome at all, and liver, heart-muscle and platelet-forming cells routinely carry extra whole copies of it.)
This is where the "blueprint" framing collapses and something more interesting appears. The genome is not a description of an organism. It is closer to a vast conditional program in which the running state determines which branches execute — and where the state is itself the accumulated product of which branches have executed so far.
About 20,000 human genes encode proteins (19,442 in GENCODE Release 50; reference/verified-facts.md). A given cell type expresses perhaps half of them, at wildly differing levels, and the specific combination is the cell type. Control is exerted at every stage: whether the DNA is physically accessible, whether initiates, how the RNA is processed, how long it survives, whether it is translated, and how long the protein lasts.
Two consequences worth carrying forward:
- Most of what distinguishes species is regulatory, not structural. Human and chimpanzee proteins are nearly identical. The interesting differences are in when and where genes are switched on.
- Most disease-associated genetic variation is regulatory. The variants that turn up in studies of common disease overwhelmingly sit outside protein-coding sequence (Ch 51). They change how much of a protein gets made, not what it looks like.
4. Packaging: chromosomes
Two metres of DNA has to fit in a nucleus about six micrometres across, remain accessible for reading, and be divided accurately when the cell splits. The first step of the solution is definite: DNA wraps around protein spools — — and that level is seen directly inside living cells. Above it, the familiar textbook ladder of ever-thicker coils is not established; imaging of intact nuclei shows an irregular, locally variable polymer, and the large-scale order comes instead from motor proteins that reel DNA into loops and from like associating with like. Only as the cell divides is the whole thing compacted into the visible under a microscope — and that too is a nested array of motor-made loops rather than a coiled coil (Ch 03 takes the textbook ladder apart).
Humans have 46 chromosomes: 23 pairs. One member of each pair came from each parent. Twenty-two pairs are autosomes; the twenty-third pair is the , X and Y.
Packaging is not just storage. Compacted DNA is unreadable, so the state of the packaging is itself a layer of regulation — a way of marking whole regions as available or off-limits, and of propagating that decision to daughter cells. That's the substance of (Ch 23), a word that has attracted more nonsense than any other in the field.
5. Copying and transmitting
Cells divide by mitosis: the genome is replicated and one complete copy goes to each daughter. Both are genetically identical to the parent cell. This is growth, repair, and maintenance.
Making gametes — eggs and sperm — requires something different, because two full genomes combining would double the count each generation. halves it. A cell with 46 chromosomes produces gametes with 23, so that fertilisation restores 46.
Meiosis does two things that matter enormously:
It shuffles whole chromosomes. Which member of each pair goes into a given gamete is independent for each of the 23 pairs. That alone yields 2²³ ≈ 8.4 million combinations.
It shuffles within chromosomes. Paired chromosomes physically exchange segments — recombination, or . The chromosome you pass on is a mosaic of the two you received.
parental chromosomes after recombination
▓▓▓▓▓▓▓▓▓▓▓▓ (from mother) ▓▓▓▓░░░░░░▓▓▓▓
░░░░░░░░░░░░ (from father) ░░░░▓▓▓▓▓▓░░░░
Recombination is why genes near each other on a chromosome tend to be inherited together and distant ones don't — the basis of genetic mapping (Ch 14), and the reason association studies work at all a century later (Ch 29).
6. Where variation comes from
Because come in pairs, you carry two copies of most genes. The variants are . Two identical copies makes you homozygous; two different ones, heterozygous.
Two distinct processes produce new alleles and new combinations of them, doing different jobs.
creates new variants. Replication is astonishingly accurate but not perfect, and DNA is chemically damaged constantly. The residual error rate is about 1.1–1.3 × 10⁻⁸ per base pair per generation. Multiply by a 6.2-billion-base genome and the naive product is roughly 68–81 new mutations per person that neither parent carried; trio studies actually report ~60–70, because short reads cannot call variants across the 10–15% of the genome that is too repetitive to read reliably (Ch 16 does the derivation properly). Most land in sequence where they do nothing. Mutation is the only source of genuinely new alleles, and it is undirected — mutations are not more likely to occur because they would be useful.
Recombination reshuffles existing variants. It creates no new alleles but constantly generates new combinations, letting a beneficial variant be separated from a harmful one it happened to sit beside.
Together: mutation supplies the raw material, recombination explores the combinations.
7. Genotype and phenotype
Your is the sequence you carry. Your is what you observably are. The relationship between them is the central problem of the field, and it is looser than intuition suggests.
When one 's effect masks the other's, it's called , and the masked one . This vocabulary causes more confusion than any other in genetics:
Dominant does not mean common, strong, better, or more likely to be inherited. It means only that the heterozygote resembles one homozygote rather than falling between them. A dominant allele can be vanishingly rare and severely harmful. Huntington's disease is dominant and rare; the allele for blood group O is recessive and the most common in most populations.
And for most traits the whole framework barely applies. Height, blood pressure, and disease risk are influenced by thousands of variants, each shifting the outcome slightly, together with environment and chance. These are quantitative or complex traits (Part 6) and they are the normal case. Single-gene traits are the exception — they are taught first because they are tractable, not because they are typical.
8. Populations, and what evolution actually is
Step back from individuals to a population, and describe it by frequencies: what fraction of all copies of a gene, across everyone, is each variant?
That reframing makes evolution precise and unmysterious:
Evolution is change in allele frequencies in a population over time.
Exactly four processes change them (Ch 27):
| Force | What it does |
|---|---|
| Introduces new alleles. Slow, but the only original source | |
| Selection | Variants affecting survival or reproduction change frequency systematically |
| Genetic | Random sampling between generations. Frequencies wander. in small populations |
| Migration | Movement between populations homogenises them |
Two things worth internalising early. Selection is not the only force, and often not the strongest — a great deal of molecular evolution is drift acting on variants that don't matter (Ch 33). And selection has no foresight: it cannot preserve a currently-useless variant for later usefulness. It has no objective function; it is not optimising anything. It is a bias in a sampling process.
9. Reading genomes: what genomics is
Genetics spent its first century inferring the contents of the genome indirectly — from patterns of inheritance, from crosses, from disease running through families. Genomics is what happened when we could just look.
Sequencing determines the order of bases in a DNA molecule. It has gone from heroic-effort-per-gene to routine-per-genome in about three decades, and the cost has fallen faster than semiconductor manufacturing did over the same period. The consequence is a shift in the binding : generating data is no longer the hard part; interpreting it is.
Modern platforms make different trade-offs — short reads that are cheap and accurate, long reads that resolve repetitive regions, and newer chemistries still being benchmarked (Ch 40). Every one produces fragments that must be reassembled or aligned computationally, which is why genomics is as much a computational discipline as a biological one, and why Part 9 will feel like algorithms.
What this bought us:
- The — a standard coordinate system for human sequence, only recently completed end to end, and now being generalised to a that represents many people rather than a composite of a few (Ch 45)
- Catalogues of variation — databases recording, across hundreds of thousands of people, how common every observed variant is. This is what makes clinical interpretation possible: a variant seen in thousands of healthy adults is not causing a severe childhood disease
- Functional readouts — sequencing not just DNA but which genes are being expressed, in which cells, with which regions accessible (Part 10)
- Association studies — scanning millions of variants across hundreds of thousands of people to find those statistically linked to a trait (Ch 51)
10. The map
Where each part of this curriculum sits in the story:
Common misconceptions
| What people think | What's actually true |
|---|---|
| Genes are a blueprint for the organism | Nothing in the genome corresponds to a picture of the result. It's a set of local rules whose interaction produces structure |
| There's "a gene for" intelligence, height, addiction | Almost all interesting traits involve thousands of variants of tiny individual effect, plus environment. "A gene for X" is nearly always a category error |
| are more common or stronger | Dominance describes only the heterozygote's appearance. It says nothing about frequency, severity, or |
| Most of the genome is useless junk | Only ~2% encodes protein, but much of the rest is regulatory, structural, or transcribed into non-coding RNA. Non-coding RNA genes outnumber coding ones better than 2:1, and about 3:1 once degraded are counted too. How much is functional is genuinely contested — but "junk" was always overconfident |
| are usually harmful | Most are neutral — they land where sequence doesn't matter much. Harmful ones are a minority; beneficial ones rarer still but not negligible |
| Evolution improves organisms toward a goal | It's change in allele frequencies. Selection is a sampling bias with no foresight and no objective. changes frequencies with no reference to fitness at all |
| Identical twins prove genes determine traits | They share a genome and are demonstrably not identical in most traits. Twin studies quantify the gap between and — they don't close it |
| Your DNA is fixed and identical in every cell | Cells accumulate throughout life. Your cells are a mosaic. This is central to cancer (Ch 56) |
Worked example: following one gene all the way through
Take LCT, the gene encoding lactase — the enzyme that digests lactose. Watch the whole subject appear in one example.
Molecular. LCT sits on 2. Transcribed into RNA, processed, translated into a protein that folds into an enzyme lining the small intestine and cleaving lactose into glucose and galactose. (Parts 1.)
Regulatory. In most mammals LCT is switched off after weaning — the enzyme is useless once milk stops. The off-switch is not in the gene at all: it's in a regulatory element about 14,000 bases away, inside a neighbouring gene. (Part 4, and a good illustration of why "the gene" is a slippery unit.)
Variation. A single-base change in that distant element keeps LCT switched on into adulthood. Carriers digest milk as adults — "lactase persistence". (Part 3.)
Transmission. The persistence behaves as a : one copy suffices, because one working switch is enough. (Part 2.)
Population. It is common in northern Europe and in some East African pastoralist groups, rare elsewhere. Different populations independently evolved different variants doing the same job — convergent evolution, visible in sequence. (Part 5.)
Selection. These variants carry among the strongest signals of recent selection in the human genome — lactase persistence is the most strongly selected single-gene trait known from the last 10,000 years — and the signature is directly readable: the allele sits on a long stretch of shared , because it rose in frequency faster than recombination could break the haplotype up. Why it rose is less settled than the textbook version implies. Ancient DNA finds the European persistence allele already present by ~4700 BC but rare for nearly three millennia, reaching appreciable frequency only around 2000 BC — long after dairying was widespread. Milk is what makes the allele matter, but it does not by itself explain the timing; famine and increased pathogen exposure have been proposed as the conditions that made lactase non-persistence costly (Evershed et al., Nature 608:336, 2022). Treat it as a very strong selection signal whose driver is still argued over. (Parts 5 and 7.)
Genomics. All of this was found by sequencing populations, mapping the association, noticing the regulatory element was nowhere near the gene, and confirming it functionally. (Parts 9–11.)
One gene, and every part of this curriculum is already implicated. That is the normal situation.