Replication: how DNA copies itself

When a cell divides in two, each daughter needs a complete copy of the genome. Not a summary, not most of it, the whole thing, every base in the right order. A human cell copies roughly three billion bases before it divides, in a few hours, with an error rate so low it is hard to believe at first. This lesson is how.

The surprising part is how little cleverness the copying needs, because the hard part was already solved by the structure of DNA itself. You met the idea in S3.2: the two strands of the double helix are complementary, so each strand is a backup of the other. That one fact is what makes faithful copying not just possible but almost easy. Copying DNA is mostly a matter of getting out of the way and letting complementarity do the work.

The trick is hidden in the pairing rule

Recall the pairing rule. Adenine (A) pairs only with thymine (T), and cytosine (C) pairs only with guanine (G). The pairing is exclusive: a base can accept its one partner and nothing else. And the two strands run in opposite directions, one 5 prime to 3 prime, the other 3 prime to 5 prime, which is part of why they fit.

Now look at the consequence. If the pairing is exclusive, then one strand completely determines the other. Show me the sequence of one strand and I can write the other with no extra information, just by applying the rule base by base. The genome is not two independent messages. It is one message stored twice, each copy the photographic negative of the other.

That is the whole basis of copying. To duplicate DNA you do not read a strand and transcribe a matching one from some outside manual. The strand is the manual. You separate the two negatives and let each one develop its missing partner.

Look at the double helix below. Pick either strand and read a few bases along it. Notice that the opposite strand told you nothing new: once you chose one strand, the other was already fixed by the pairing rule. That redundancy is the feature we are about to exploit.

Step one: unzip

The two strands are held together by many small bonds between the paired bases, called hydrogen bonds. Each one is weak. A single base pair would come apart from a nudge. But a strand has thousands or millions of them stacked in a row, and collectively they hold the helix shut tight. This is a deliberate balance: individually weak so the strands can be pulled apart on demand, collectively strong so they never fall apart on their own.

To start copying, an enzyme called a helicase runs along the helix and breaks those cross-strand bonds, unzipping it like a zipper opening into a Y shape. It does not cut either strand. It only separates them. Behind the moving helicase you now have two bare single strands, each with its bases exposed to the surrounding fluid. A bare strand being used as a guide to build a new one is called a template.

Step two: let the rule fill each template

The cell is full of loose nucleotides drifting around, the single A, C, G, and T building blocks. When one bumps into an exposed base on a template, it either fits or it does not. An exposed T can hold onto an incoming A and nothing else. An exposed C can hold a G. The pairing rule is doing the selecting, physically, by shape.

An enzyme called DNA polymerase walks along the template and locks each correctly matched incoming nucleotide onto the end of the new strand it is building. It does not decide the sequence. The template decides, one base at a time, and the polymerase makes it official. Feed it a template reading A, C, C, G and the new strand can only come out T, G, G, C. There is no choice in it.

Here is the payoff, and it is worth sitting with for a second. The new strand is the complement of the template. But the template's original partner, the strand it used to be zipped to, was also its complement. The complement of a complement is the original. So the new strand you just built is identical to the partner that floated away. Each template rebuilds the exact partner it lost. That is why the copy is faithful and not merely close.

The analogy, and where it breaks

If you have set up disk redundancy, this will feel familiar. A RAID-1 mirror keeps the same data on two disks, so when one dies you rebuild it from the survivor. DNA is that idea baked into the molecule: two strands, either one enough to regenerate the pair. Copying is a scheduled rebuild from a mirror the cell always keeps.

Here is where the analogy misleads you, and it is worth stating plainly. A RAID mirror stores identical bytes on both disks, and a rebuild copies those bytes across unchanged. DNA does not store identical strands. It stores complementary ones, mirror images, not duplicates. So the rebuild does not copy the surviving strand, it inverts it through the pairing rule. And a RAID rebuild is triggered by a failure, something broke. DNA replication is not a recovery from damage at all. It is a routine, deliberate act the cell performs on healthy DNA, on purpose, every time it divides. Use the mirror picture for the redundancy. Drop it the moment you start thinking rebuilds only happen after a crash.

Polymerase only runs one way

There is one hard constraint on DNA polymerase, and it shapes how replication actually looks. Polymerase can only add a new nucleotide to the 3 prime end of the strand it is building. It cannot add to the 5 prime end. So a new strand can only grow in the 5 prime to 3 prime direction, always, no exceptions.

That would be tidy if both templates pointed the same way. They do not. Remember the two strands are antiparallel, they run in opposite directions. So at the unzipping fork, one template is oriented for the polymerase to follow the opening smoothly, building its new strand in one continuous run. The other template points the wrong way. Its new strand still has to be built 5 prime to 3 prime, so the polymerase ends up making it in short backward pieces, each started as more template is exposed, and the pieces are stitched together afterward.

Two helices, each half old

When the fork has run the length of the molecule, you have two complete double helices where there was one. Now ask a precise question about each of them. What is a new helix made of? One strand of it is an original strand from the parent DNA, the template that was there all along. The other strand is freshly built. Every daughter helix is exactly half old and half new.

This pattern has a name: semi-conservative replication. Semi, meaning half. Each copy conserves one strand of the original. It is not that one daughter cell gets the old DNA and the other gets brand-new DNA. Both daughters get a hybrid, one old strand paired with one new one. That is a direct result of the mechanism. You never threw the templates away, you built new partners onto them, so the old strands are still right there in the products.

How we actually know replication is semi-conservative

In 1958 Matthew Meselson and Franklin Stahl grew bacteria on a heavy form of nitrogen so their DNA was measurably denser than normal, then switched them to light nitrogen and let them copy their DNA once. If copying were semi-conservative, every DNA molecule after one round should be exactly half heavy and half light, a single in-between density, and that is precisely what they saw. After a second round they saw a mix of half-heavy and fully light molecules, again exactly what the half-old-half-new model predicts and what the competing models did not. It is one of the cleanest experiments in biology, which is why semi-conservative is a settled fact rather than a nice story.

Why the copy is so nearly perfect

Base pairing by itself is good but not great. Left to shape-fitting alone, the wrong nucleotide slips in roughly once in every ten thousand bases. For a three-billion-base genome that would be hundreds of thousands of errors per copy, which no organism could survive. So the cell layers extra checks on top.

First, DNA polymerase proofreads as it goes. Just after it adds a base, it can sense whether the new pair sits correctly. If the base is wrong, the polymerase backs up, clips out the mistake, and tries again. Proofreading alone pushes the error rate down to roughly one in ten million.

Second, after replication a separate crew called mismatch repair scans the fresh DNA for the tell-tale bulge where two bases failed to pair cleanly, cuts out the bad stretch, and refills it from the correct strand. Add that layer and the final error rate lands around one in a billion bases, sometimes better.

Notice the shape of this. It is defense in depth. A cheap fast filter (base pairing) catches most errors, a second check (proofreading) catches most of what leaks through, and a third pass (mismatch repair) catches most of the rest. Each layer is imperfect, and stacked together they are astonishingly good. Any engineer who has built a validation pipeline will recognize the design.

When a slip survives

A copying error that escapes both proofreading and mismatch repair becomes permanent. The next time that DNA is copied, the machinery reads the changed base as if it had always belonged there and dutifully pairs a partner to it. The mistake is now baked into the sequence, and every descendant cell inherits it. A permanent change in the DNA sequence is called a mutation, and it is the subject of S7.

Hold two things about mutations at once, because the next module develops both. Most mutations do nothing, or nearly nothing. A few are harmful. And a rare few change something in a way that matters, which, spread across a population and across time, is the raw material evolution works on. Faithful copying with a tiny, non-zero error rate is not a flaw the cell failed to fix. It is the exact balance life runs on: stable enough to stay itself, changeable enough to adapt.

Key terms

complementary base pairing

The exclusive rule that A pairs only with T and C only with G, which lets one strand fully determine the other.

template

A separated single strand used as the guide for building a new complementary strand.

helicase

The enzyme that unzips the double helix by breaking the hydrogen bonds between the two strands, without cutting either one.

DNA polymerase

The enzyme that builds a new strand by adding matched nucleotides onto a template, working only in the 5 prime to 3 prime direction.

semi-conservative replication

The result that each new double helix is one old parental strand paired with one newly built strand, so no copy is all old or all new.

proofreading

DNA polymerase's ability to sense a just-added wrong base, remove it, and retry, lowering the error rate as it copies.

mismatch repair

A separate post-replication system that finds mispaired bases, cuts out the bad stretch, and refills it correctly.

mutation

A permanent change in the DNA sequence, such as a copying error that escaped every check and is inherited by all descendant cells.