Contents
04 — DNA replication Part 1 — Molecular foundations

04 — DNA replication

Before this: Ch 01 · Ch 02 · Ch 03 · Time: ~40 min

What you'll be able to do

The core idea

Copying a genome looks like it should be the easy part: separate the two complementary strands, read each, write the complement. A memcpy with a checksum. Three things make it hard, and essentially all of the mechanism is a response to them.

The copying enzyme only works in one direction. extends a chain at its 3′ end and nowhere else. The strands are antiparallel, so at any moving fork one new strand can be extended toward the fork and the other cannot. The second has to be built backwards, in pieces, and stitched together. That is not an implementation detail — it is the reason half the machinery exists.

The copy must be almost perfect, and no single step is close. The observed error rate is around one substitution per 10⁹–10¹⁰ bases. No enzyme achieves that. Three individually mediocre mechanisms are applied in series, and the rates multiply.

It has to happen exactly once per cell cycle, on a substrate simultaneously being transcribed, packaged into , chemically damaged, and twisted by the act of unwinding itself. Replication is a copy operation running concurrently with everything else the genome is doing, with no locks.


1. Semiconservative replication, and how it was proved

Base pairing implies that each strand can template the other. It does not tell you what happens to the parental strands. Three models were live in the 1950s:

ModelAfter one round of copying, each daughter duplex is…
ConservativeEither both-old or both-new. The parental duplex survives intact
SemiconservativeOne old strand + one new strand
DispersiveBoth strands are patchworks of old and new segments

Meselson and Stahl (1958) separated these with one idea: make old and new DNA physically distinguishable by weight, then weigh it.

Grow E. coli for many generations in medium whose only nitrogen source carries the heavy isotope ¹⁵N, so every base in every genome is heavy. Then transfer to ordinary ¹⁴N medium and sample after each generation. Anything made after the switch is light; anything inherited is heavy.

The measurement is the beautiful part. Spin a caesium chloride solution hard enough and long enough and the CsCl itself redistributes into a smooth density gradient — dense at the bottom of the tube, light at the top. DNA migrates until it reaches the depth where the solution's density matches its own buoyant density, and stops. It is an equilibrium method, not a rate method: band position is a physical constant of the molecule, not a function of how long you spun. ¹⁵N- and ¹⁴N-DNA differ in density by under 1%, and that is enough — the gradient is shallow, the bands are sharp, and the two resolve cleanly.

     LIGHT ◄───────────── buoyant density ─────────────► HEAVY
            (all ¹⁴N)         (hybrid)          (all ¹⁵N)

  gen 0                                          ████████
  gen 1                        ████████
  gen 2      ████████          ████████
  gen 3     ████████████         ████
  gen 4    ██████████████         ██

           bar width ∝ fraction of total DNA in that band

Now read the discrimination.

Generation 1 kills the conservative model. Conservative replication predicts two bands — a surviving heavy parental duplex and a fully light daughter, in equal amounts. What appears is a single band at exactly intermediate density, with no heavy DNA left.

Generation 2 kills the dispersive model. Both surviving models explain generation 1: a hybrid duplex is half-heavy, and so is a uniformly dispersed patchwork. They diverge at generation 2. Dispersive predicts one band that keeps getting lighter — always one band, never two. Semiconservative predicts two discrete bands, half the DNA hybrid and half fully light, with the hybrid band never moving. Two bands appeared.

The instructive feature of the design is that generation 1 was not enough and the authors knew it. An experiment that separates your favoured hypothesis only from the most obviously wrong alternative has not tested anything. They also ran the confirmation: heat the generation-1 hybrid to separate the strands and re-run. Semiconservative predicts two single-strand bands, one heavy and one light; dispersive predicts every single strand to be intermediate. Two bands.

2. Origins, bubbles, and bidirectional forks

Replication does not start at the end of a and run to the other end. Do the arithmetic and you can see why not.

A human moves at roughly 1–2 kb per minute — call it 1.5. Chromosome 1 is 249 Mb (GRCh38). One origin at the centre, two forks travelling outward, each covering 124.5 Mb:

124,500,000 bp ÷ 1,500 bp/min = 83,000 min ≈ 58 days

Two months for one chromosome. S in a human cell is about 8 hours. The gap is more than two orders of magnitude — 83,000 minutes needed against 480 available, a factor of ~170 — and it is closed the only way it can be: massive parallelism.

Replication initiates at many origins simultaneously. Each origin opens a replication bubble with a fork at each end, travelling in opposite directions, until each fork meets a fork coming the other way from a neighbouring origin.

   origin                          origin
      │                               │
 ═════╪═══════════════════════════════╪═════   parental duplex

 ════<══>═══════════════════════<══>══════     t₁  bubbles open,
      ←  →                       ←  →              two forks each

 ═<════════>═══════════════<══════════>══      t₂  forks diverge

 <══════════════><══════════════════════>      t₃  forks meet and
                 ▲                                 terminate here

How many origins? Each fork runs for at most the length of S phase: 1.5 kb/min × 480 min = 720 kb. A bidirectional origin therefore covers ~1.44 Mb at best. A genome of 6.2 Gb needs at least 6.2 × 10⁹ / 1.44 × 10⁶ ≈ 4,300 origins even under the impossible assumption that every fork runs the entire S phase without stopping. In reality origins fire in waves throughout S phase, forks terminate early against their neighbours, and the measured figure is 30,000–50,000 origins fired per cell cycle.

Where they are is a genuinely different story in bacteria and eukaryotes:

Bacteria (E. coli)Budding yeastHuman
Origins per genomeone (oriC, ~245 bp)a few hundred (ARS, ~11 bp consensus)30,000–50,000 fired, far more licensed
Defined bysequencesequencechromatin context, not sequence
Genome / time4.6 Mb in ~40 min12 Mb3.1 Gb (haploid) in ~8 h

That middle row matters computationally: you cannot grep for a human replication origin. There is no consensus motif. Origin choice is set by -free regions, ORC binding, G-quadruplex-forming sequence and chromatin state, it varies between cell types, and it is stochastic between individual cells in the same population — only ~5–10% of licensed origins fire in any given cycle.

E. coli with a single origin replicates its 4.6 Mb chromosome in ~38 minutes (2.3 Mb per fork at ~1,000 bp/s). It can nevertheless divide every 20 minutes, by starting the next round of replication before the current one finishes — nested forks, several rounds in flight at once.

3. The constraint that shapes everything: 5′→3′ only

Here is the reaction. The 3′-OH at the end of the growing chain attacks the innermost phosphate of an incoming deoxynucleoside triphosphate. A new phosphodiester bond forms and pyrophosphate leaves; hydrolysis of that pyrophosphate makes the step effectively irreversible.

Two consequences fall straight out.

A polymerase cannot start a chain. It can only extend an existing 3′-OH. Something else must lay down the first few nucleotides — that is primase, an , which can start from nothing. The primer is RNA and not DNA for a good reason: RNA is chemically marked as foreign in a DNA molecule, so it can be found and removed later. The least accurate nucleotides in every fragment — the ones laid down without a template-matched predecessor to check against — are deliberately made of a material that is guaranteed to be excised.

Synthesis cannot run 3′→5′. The energy for each addition comes from the triphosphate on the incoming nucleotide. Reverse the direction and the triphosphate would have to sit on the growing chain's 5′ end — in which case removing a wrongly incorporated terminal nucleotide would leave a monophosphate and permanently dead-end the chain. and 5′→3′ synthesis are the same design decision. A hypothetical 3′→5′ polymerase could not proofread, and a polymerase that cannot proofread is 10⁵ times too inaccurate to build a genome.

Now combine that with antiparallel strands. At a fork moving rightward:

                                             fork moves ──►
  lagging-strand template
  5'──────────────────────────────────────────────────╮
                                                      │
     3'◄═══════5'  3'◄══════5'  3'◄═════5'            │
      fragment 1    fragment 2   fragment 3           │  helicase
      (oldest)                   (newest, at the fork)│  unwinds here
                                                      │
     5'════════════════════════════════════════►3'    │
      leading strand — one continuous chain           │
                                                      │
  3'──────────────────────────────────────────────────╯
  leading-strand template

The leading strand is extended toward the fork and is synthesised continuously. The must grow away from the fork, so it is made in short pieces — — each newly primed at the position closest to the fork and extended backwards into the previous fragment.

Both strands are made 5′→3′. There is no exception, anywhere, in any organism.

Fragment sizes: 1,000–2,000 nt in bacteria, 100–200 nt in eukaryotes. The eukaryotic figure is not arbitrary — Okazaki fragment ends map onto midpoints, because the fragment terminates when the polymerase runs into the next re-deposited histone octamer. Chromatin sets the fragment length.

Count them. Replicating a 6.2 Gb genome means synthesising 1.24 × 10¹⁰ new nucleotides, half of them on lagging strands. At ~150 nt per fragment that is roughly 40 million prime–extend–clean-up cycles per S , each needing a primer laid, a clamp loaded, a polymerase engaged, the primer removed and the nick sealed. Every one of them completed correctly.

4. The machinery at the fork

ComponentBacteriaEukaryotesJob — and what fails without it
DnaB (5′→3′ on lagging template)CMG: Cdc45–MCM2-7–GINS (3′→5′ on leading template)Breaks the base pairs ahead of the fork. Without it, nothing opens
Single-strand bindingSSBRPACoats exposed single strands, stops re-annealing and hairpin formation. RPA-coated ssDNA is also the signal that a fork is in trouble (§10)
PrimaseDnaGPol α–primaseLays the RNA primer. Needed once per leading strand, ~40 million times on
Replicative polymerasePol III holoenzymePol ε (leading), Pol δ (lagging)Extends. Bare polymerase falls off after ~10 nt
Sliding clampβ clamp (dimer)PCNA (trimer)A ring encircling DNA that tethers the polymerase. Raises processivity from ~10 nt to >50 kb
Clamp loaderγ/τ complexRFCATP-driven; cracks the clamp open and closes it around each new primer
Primer removalPol I 5′→3′ exonuclease (nick )Pol δ strand displacement + FEN1 flap endonuclease (Dna2 for long flaps)Excises RNA and replaces it with DNA
LigaseLigA (NAD⁺-dependent)Ligase 1 (ATP-dependent)Seals the remaining nick. Without it the lagging strand stays as fragments
Topo I; DNA gyrase (Topo II)Topo I, Topo IIRelieves torsional stress ahead of the fork; Topo II also disentangles the finished sister chromatids

Two of these deserve more than a table row.

The sliding clamp is the reason replication is fast. The clamp is a closed ring threaded onto the DNA; the polymerase grips the ring rather than the DNA, so it slides freely along the template while being topologically unable to leave — the difference between holding a rope and being handcuffed to it. On the lagging strand a clamp must be loaded and unloaded for every fragment, which makes the loader one of the busiest machines in the cell.

Torsion is a quantitative problem, not a hand-wave. DNA has ~10.5 bp per turn, so unwinding at 1,000 bp/s in E. coli requires the duplex ahead of the fork to rotate about 95 times per second — or to accumulate that many positive supercoils. A circular cannot rotate freely, and neither can a several-hundred-megabase chromosome anchored in a nucleus. Topoisomerases cut the backbone (one strand for Topo I, both for the ATP-driven Topo II), pass DNA through the break, and reseal. That transient covalent enzyme–DNA intermediate is a major drug target: fluoroquinolones trap bacterial gyrase, etoposide and irinotecan trap human Topo II and Topo I. All of them work by blocking the resealing step, converting a routine intermediate into a double-strand break when the next fork arrives.

Geometry forces one further trick. Both polymerases belong to a single complex moving in one physical direction, which is only possible if the lagging-strand template loops back on itself — the "trombone" model, the loop growing as a fragment is extended and collapsing when it is finished.

stateDiagram-v2 direction LR [*] --> Prime Prime : primase lays a ~10 nt RNA primer Prime --> Load : RFC opens PCNA around the primer Load --> Extend : Pol δ extends ~150 nt, always 5-to-3 Extend --> Displace : runs into the previous fragment Displace --> Cleave : strand displacement lifts a flap, FEN1 cuts it Cleave --> Seal : Ligase 1 seals the nick Seal --> Prime : clamp unloaded, repeat ~40M times per S phase

5. Polymerases: prokaryotic and eukaryotic

Bacteria manage with a small set. E. coli has five ; two do replication.

EnzymeRoleProofreads?
Pol IIIThe replicase. Both leading and , held together by the clamp loader as one holoenzyme — classically drawn as a dimer, but live-cell counting finds three polymerase cores per replisome (Reyes-Lamothe et al., Science 2010)Yes (ε subunit)
Pol INot a replicase. Its 5′→3′ exonuclease chews out RNA primers while its polymerase fills behind — "nick ". Also a repair enzymeYes

Eukaryotes split the job three ways, and the split is informative:

EnzymeRoleProofreads?
Pol α–primaseStarts every fragment: ~10 nt of RNA, then ~20 nt of DNA. Then hands offNo
Pol εExtends the leading strandYes
Pol δExtends the lagging strand; also does Okazaki maturation and fills gaps in repairYes

Pol α cannot proofread, which is exactly why it is allowed to make only the first ~30 nucleotides — and why most of those are later removed and resynthesised by an enzyme that can. The Pol ε / Pol δ division of labour was established by tracking where each enzyme's characteristic errors land in -deficient mutants; the assignment is well supported, though the details of hand-off remain active work.

6. Fidelity is a product of filters, not a property of an enzyme

This is the argument made qualitatively in Ch 01 §5, now with numbers.

Hydrogen-bonding energetics alone distinguish a correct base pair from an incorrect one by only about a factor of 10–100 — nowhere near enough. Three filters are applied in series:

FilterMechanismError rate afterFactor gained
Base selectionThe polymerase active site is a geometric template. A correct Watson–Crick pair fits and triggers a conformational change that permits catalysis; a mispair does not fit and catalysis is ~10³-fold slower~10⁻⁵~10⁵
A mispaired 3′ end frays, the DNA transfers to a separate 3′→5′ exonuclease site 3–4 nm away, the wrong nucleotide is excised, and the DNA returns~10⁻⁷~10²
Post-replicative. Scans for the helix distortion a mismatch causes, identifies which strand is new, excises a tract around the error and resynthesises~10⁻⁹–10⁻¹⁰10²–10³

Multiply: 10⁻⁵ × 10⁻² × 10⁻³ ≈ 10⁻¹⁰ per base per replication.

No step in replication is accurate enough to copy a genome. The observed fidelity is a product of three independent, individually unimpressive filters. This has a sharp consequence: lose any one of them and you lose its entire factor, not some fraction of it. Fidelity does not degrade gracefully.

Humans confirm that prediction brutally. Inherit a broken mismatch-repair gene (MLH1, MSH2, MSH6, PMS2) and you have Lynch syndrome — tumours with 10²–10³-fold elevated and unstable microsatellites. Acquire a in the exonuclease domain of POLE and you get "ultramutated" tumours carrying more than 100 mutations per megabase, over a hundred times the normal somatic burden (Ch 56).

Note what the third filter requires: mismatch repair must know which strand is new, or it will "correct" the template half the time and make the error permanent. Bacteria label the strand with Dam methylation — GATC sites are methylated on both strands, but for a few minutes after the fork passes only the parental strand is, and repair reads that hemimethylated state. Eukaryotes use the nicks transiently present in the nascent strand — the un-ligated Okazaki junctions themselves — plus the orientation of PCNA.

Sanity-check against the pinned rate: ~10⁻¹⁰ per base per division, multiplied through the few hundred divisions separating a male zygote from his sperm, gives a few × 10⁻⁸, against a measured rate of ~1.1–1.3 × 10⁻⁸ per bp per generation (verified-facts). Note the direction: the estimate overshoots the measurement by a factor of two or three. That is the expected failure mode, because 10⁻¹⁰ is an order-of-magnitude bound rather than a measured constant — the true per-division fidelity is somewhat better than that, or the effective division count somewhat lower. Agreement within a small factor is all this calculation is entitled to claim. Separately, and pushing the other way, the germline total is not purely replicative at all: a substantial share of germline mutation comes from unrepaired chemical damage rather than polymerase error (Ch 16).

7. The ends of linear chromosomes

Circular bacterial have no ends and no problem. Linear ones do.

The last on a chromosome is primed some distance from the terminus. When its RNA primer is removed, the resulting gap cannot be filled, because filling it needs a 3′-OH on the far side — which would belong to a fragment primed off the end of the molecule.

   lagging-strand template                       chromosome end
   5'───────────────────────────────────────────────────────┫ 3'
       ◄══════════  ◄══════════  ◄════════rrr               ┃
        fragment 1   fragment 2   last fragment  ▲          ┃
                                                 RNA primer ┃
                                                 (removed)  ┃
   result 5'─────────────────────────────────           ────┫
                                                └── gap: unfillable,
                                    no 3'-OH exists to its right

Every round of lagging-strand synthesis therefore loses the length of a terminal primer. The leading strand could in principle be copied to the last base — but it is not left blunt: nucleases deliberately resect the new strand's 5′ end to generate a single-stranded 3′ overhang. Both ends shorten, and part of the shortening is a choice rather than a limitation.

The solution is to make the ends expendable. Human chromosomes terminate in : thousands of tandem copies of the hexamer TTAGGG, totalling roughly 10–15 kb at birth, ending in a 3′ overhang of 50–300 nt. The overhang folds back and invades the duplex to form a t-loop, bound by the six-protein shelterin complex. Two jobs: supply sequence that is safe to lose, and hide the chromosome end from repair machinery that would otherwise read a natural terminus as a double-strand break and fuse it to another chromosome.

Telomeres shorten by ~50–200 bp per division, so somatic cells carry a division counter. After 40–60 population doublings — the Hayflick limit, measured in cultured human fibroblasts in 1961 — the ends can no longer be protected, a DNA-damage response fires, and the cell enters irreversible senescence.

resets the counter. It is a reverse transcriptase (protein subunit TERT) carrying its own RNA template (TERC), which contains the complement of the telomere repeat: it binds the 3′ overhang, copies a repeat from its internal template, translocates, and repeats — extending the chromosome without any external template. It is active in the germ line, in stem cells, and transiently in activated lymphocytes, and off in most somatic cells.

That off-switch is a tumour suppressor: a cell needing hundreds of divisions to form a tumour runs out of telomere first, unless it restores telomere maintenance. It nearly always does — 85–90% of cancers reactivate telomerase and 10–15% use ALT, a recombination-based mechanism. Two point in the TERT are among the commonest routes, at GRCh38 chr5:1,295,113 and chr5:1,295,135 (conventionally named C228T and C250T; the C>T is written on the TERT coding strand, which is the minus strand of chromosome 5). Each creates a new binding site for an ETS-family and raises TERT several-fold. They are among the most frequent non-coding known, present in roughly 70–80% of glioblastomas and melanomas.

The awkward part is that telomere attrition is also pro-cancer. A pre-malignant cell that reaches critically short telomeres before reactivating telomerase suffers end-to-end chromosome fusions, and a fused dicentric chromosome is torn apart at the next division — the breakage–fusion–bridge cycle, one of the classic engines of chaotic tumour (Ch 20).

8. Once and only once: licensing

Replicating any region twice in one cycle would produce copy-number gain and, at a fork, a broken . Getting exactly-once semantics out of tens of thousands of independent, stochastically firing origins is a hard concurrency problem, and the solution is elegant: make licensing and firing require mutually exclusive conditions, separated in time.

graph LR G1["<b>G1</b><br/>CDK activity LOW"] -->|"ORC + Cdc6 + Cdt1 load<br/>MCM2-7 double hexamer<br/><i>inactive</i>"| LIC["origin <b>licensed</b>"] LIC -->|"S phase: CDK + DDK<br/>recruit Cdc45 + GINS"| FIRE["CMG helicase assembles<br/>origin <b>fires</b>"] FIRE --> REP["bidirectional forks"] S["<b>S/G2/M</b><br/>CDK activity HIGH"] -->|"blocks re-loading:<br/>ORC/Cdc6/Cdt1 inhibited,<br/>geminin sequesters Cdt1"| NOLIC["re-licensing impossible"]

In G1, CDK activity is low, and that is the only window in which the MCM2-7 can be loaded — as an inactive double hexamer. Entering S raises CDK activity, and the same signal does two things: it activates loaded helicases and it blocks any further loading. A spent licence cannot be reissued until the cell passes through mitosis and CDK activity collapses again.

There is no counter and no lock here. It is a phase : permission to license and the trigger to fire can never be true simultaneously. Deregulate it — overexpress Cdt1, lose geminin — and cells re-replicate, producing gene amplification.

Licensing is deliberately generous. MCM2-7 is loaded at 10–20× more sites than ever fire; the excess are dormant origins, reserve capacity that lets a cell finish S phase when forks stall (§10).

9. Timing, chromatin, and where mutations land

Origins do not all fire at once. Firing is spread across S in a reproducible pattern of replication timing domains a few hundred kilobases to a megabase wide, mappable by sequencing newly synthesised DNA at intervals through S phase (Repli-seq). Timing tracks chromatin state almost exactly:

Early-replicatingLate-replicating
Open chromatin, accessibleCompact
Gene-rich, actively transcribedGene-poor, silent
A compartment (Ch 50)B compartment; often at the nuclear lamina
GC-richAT-rich
Lower Higher mutation rate

That last row is the one with consequences. Late-replicating regions accumulate substantially more , in the germ line and far more visibly in tumours. Several mechanisms contribute: nucleotide pools are depleted by late S phase, so polymerases misincorporate more; is coupled to replication and works better early; late forks expose more persistent single-stranded DNA.

The effect is large enough to break naive analysis. Chromatin features plus replication timing explain up to ~86% of the variance in density at megabase scale across cancer genomes (Polak et al., Nature 2015). A method that asks "is this gene mutated more than chance predicts?" against a genome-wide average background will nominate long, late-replicating, lowly expressed genes as drivers purely for sitting in high-background neighbourhoods. That is why modern driver detection models a local background rate (Lawrence et al., Nature 2013).

Replication also has to rebuild chromatin behind itself: parental histones are recycled onto both daughter strands, and new ones are deposited by chaperones recruited via PCNA. That partial, diluted inheritance of histone marks is one of the mechanisms — an imperfect one — by which chromatin state survives cell division (Ch 23).

10. Replication stress and fork collapse

A fork stalls whenever it meets something it cannot pass: a chemical lesion, a G-quadruplex, a repeat that forms a hairpin, a tightly bound protein, an R-loop where has left RNA paired with DNA, a trapped , or simply a shortage of dNTPs. Collectively this is replication stress, and it is routine rather than exceptional.

A stalled fork is not yet a disaster. The keeps unwinding after the polymerase stops, exposing long stretches of RPA-coated single-stranded DNA — which is the signal that activates the ATR–CHK1 checkpoint. ATR stabilises the fork, suppresses global origin firing while permitting local dormant origins to rescue the unreplicated region, and delays mitosis. Forks can also reverse, backing up and re-annealing the two nascent strands into a four-way "chicken foot" that protects the end while the obstacle is cleared.

A collapsed fork is a disaster: the replisome disassembles and the exposed end becomes a one-ended double-strand break, with no second end to ligate to. Repair requires recombination with the sister chromatid (Ch 18), and when that goes wrong the products are , deletions and copy-number changes.

Common fragile sites are large, late-replicating, gene-poor regions — often containing very long genes — where replication routinely fails to finish before mitosis. FRA3B, inside FHIT, is the archetype; they break reproducibly under mild replication stress and are recurrent sites of deletion in tumours.

-induced replication stress is an early event in cancer. Activating MYC or RAS forces cells into S and fires too many origins at once, exhausting nucleotide pools and RPA. The resulting fork collapse generates the damage that activates p53 — which is why loss of the damage response is so often the next step in tumour evolution (Ch 56). It also opens a therapeutic window: cells already running at the edge of replication catastrophe are unusually sensitive to ATR, CHK1 and WEE1 inhibitors.


Common misconceptions

What people thinkWhat's actually true
unwinds the double helixIt does not. unwinds; polymerase can only copy an already-separated template. Removing helicase stops replication before polymerase is even relevant
Replication starts at one end of a It starts at tens of thousands of internal origins, each opening a bubble with two forks running in opposite directions. Sequential copying of chromosome 1 would take about two months
The is synthesised 3′→5′No strand, anywhere, is ever synthesised 3′→5′. Each is made 5′→3′; the series of fragments moves away from the fork. The direction of the fragments and the direction of the strand's growth are different things
is what makes replication accurateProofreading contributes one factor of ~100 out of ~10¹⁰. Base selection contributes 10⁵ and another 10²–10³. Any one of the three, alone, is hopelessly inadequate
Meselson and Stahl proved semiconservative replication in one generationGeneration 1 only excluded the conservative model. Dispersive replication predicts an identical intermediate band. The second generation — two discrete bands rather than one drifting band — is what settled it
shorten because polymerase "can't copy the last bit"The lagging-strand terminal-primer gap is real, but the leading-strand end is also deliberately resected by nucleases to create the 3′ overhang. Both ends shorten, and part of the shortening is an active choice, not a limitation
Longer telomeres are healthier; is an anti-ageing enzymeIt is a division counter with a genuine trade-off. variants that shorten telomeres cause bone-marrow failure and pulmonary fibrosis; germline variants that lengthen them raise the risk of melanoma, glioma and other cancers. Reactivated telomerase is a near-universal feature of tumours
Every origin fires in every cell cycleOnly ~5–10% do, and which ones is stochastic and differs between individual cells. The 10–20× excess of licensed origins is deliberate reserve capacity for when forks stall
Replication and politely take turnsThey collide, especially head-on, and those collisions are a major source of R-loops, fork stalling and . Highly transcribed genes tend to be oriented so that collisions are co-directional

Worked example: reading a Meselson–Stahl gradient

Part (a) — predict all three models, quantitatively.

Setup: one duplex, both strands ¹⁵N-heavy, transferred to ¹⁴N medium at generation 0.

After n rounds there are 2ⁿ duplexes, hence 2ⁿ⁺¹ strands, of which exactly 2 are the original heavy strands (strands are conserved; they are never destroyed) and 2ⁿ⁺¹ − 2 are new and light.

Semiconservative. Each of the 2 original heavy strands sits in a duplex with a light partner → exactly 2 hybrid duplexes, always. Everything else is fully light.

fraction hybrid = 2 / 2ⁿ = 2^(1−n)
fraction light  = 1 − 2^(1−n)

Conservative. The one original duplex survives unchanged; everything else is fully light.

fraction heavy = 1 / 2ⁿ ,   fraction light = 1 − 1/2ⁿ ,   no hybrid, ever

Dispersive. Old material is distributed evenly among all duplexes, so every duplex is identical and its ¹⁵N fraction is 1/2ⁿ. One band always, drifting continuously toward light.

GenerationSemiconservativeConservativeDispersive
0100% heavy100% heavy100% heavy
1100% hybrid50% heavy + 50% light100% at ½-heavy
250% hybrid + 50% light25% heavy + 75% light100% at ¼-heavy
325% hybrid + 75% light12.5% heavy + 87.5% light100% at ⅛-heavy
412.5% hybrid + 87.5% light6.25% heavy + 93.75% light100% at 1⁄16-heavy

Now compare against what was observed — generation 1: a single band at intermediate density, no heavy band; generation 2: two bands, equal intensity, at intermediate and light positions.

Only semiconservative survives both.

Part (b) — the inverse problem.

You are handed an unlabelled gradient from the same experiment. Densitometry gives 12.5% of the DNA in the intermediate band and 87.5% in the light band. Which generation is it?

2^(1−n) = 0.125 = 2⁻³
1 − n = −3
n = 4

Four generations. Note what makes this inference possible: the number of bands identifies the mechanism, and the ratio of the bands identifies the generation. Two independent readouts from one photograph.

Part (c) — the confirmatory control.

Suppose a colleague reports the generation-2 gradient as one band at ¼ density and concludes dispersive replication. Ask them to denature the sample and re-run it. Semiconservative replication predicts that heating hybrid duplexes yields one fully heavy and one fully light single strand — two bands at the extremes. Dispersive replication predicts that every single strand is a patchwork, so all of them band together in the middle. This was Meselson and Stahl's own control, and it is the cleaner experiment: it interrogates the composition of individual strands rather than of duplexes.

Connections