Contents
17 — DNA repair Part 3 — Genome instability

17 — DNA repair

Before this: Ch 04 — DNA replication · Ch 16 — Mutation · Time: ~40 min

What you'll be able to do

The core idea

Chapter 16 gave you a of about 1.1–1.3 × 10⁻⁸ per base pair per generation (verified facts). It is tempting to read that as a statement about how often DNA gets damaged. It is not. Damage happens roughly ten million times more often.

A human cell sustains on the order of 10⁴–10⁵ chemical lesions per day: bases falling off the backbone, cytosines deaminating, oxygen radicals from the cell's own metabolism attacking guanine, UV photons welding adjacent thymines, and 10–50 double-strand breaks. None of that is pathological. It is Tuesday.

The mutation rate you measure is what survives a stack of error-correcting filters:

  ~10⁵ lesions/day ──►  [ direct reversal  ]
                        [ base excision    ]
                        [ nucleotide exc.  ]  ──►  ~10⁻² inherited
                        [ mismatch repair  ]        mutations/day
                        [ DSB repair       ]

So the useful question is not "why do happen?" — the chemistry makes that inevitable — but "why are there so few?" That reframing has a practical payoff: every property of the mutation rate you care about downstream (why it varies across the genome, why some tumours die to a drug that healthy cells shrug off) is a property of repair, not of damage.

Repair does not fix mutations. It fixes damage. A lesion is chemically abnormal DNA — a base that isn't one of the four, a break in the backbone, a position where the two strands disagree. All of that is detectable, because DNA is redundant: something else says what the answer should be. A mutation is a chemically normal, correctly paired base pair that merely differs from the ancestral one. There is nothing left to detect. Every pathway below is a race against replication to fix the damage while the evidence still exists.


1. What every pathway exploits

To correct an error you must detect it and recover the original. Biology has exactly two sources of redundancy.

The complementary strand. Damage to one strand leaves the other intact: cut out a stretch of the damaged strand and re-synthesise from the survivor. Single-strand damage is in principle always correctable.

The sister chromatid. A double-strand break destroys both copies at once, so the first redundancy is gone. The only remaining template is the identical sister made during replication — which exists only in S and G2 (Ch 09). Everything strange about break repair follows from that .

The pathways differ mainly in how much sequence they remove, which is set by how large a chemical disturbance they must clear:

PathwayLesion classPatch removedTemplate
Direct reversalone specific modified base0 ntnone — modification undone in place
small, non-distorting base damage1 nt, or 2–13 nt long-patchother strand
bulky, helix-distorting lesions24–32 nt in humans (12–13 in E. coli)other strand
replication errors — normal bases, wrong pairingup to ~1 kbother (parental) strand
double-strand breakvariablesister chromatid
Non-homologous end joiningdouble-strand break0 to tens of nt, unpredictablenone
graph TD L["DNA lesion"] --> Q1{"Both strands<br/>broken?"} Q1 -->|yes| DSB["DSB repair<br/>NHEJ / HR / MMEJ"] Q1 -->|no| Q2{"Bases chemically<br/>normal?"} Q2 -->|"no — bulky, distorts helix"| NER["Nucleotide excision repair"] Q2 -->|"no — small, non-distorting"| Q3{"Directly<br/>reversible?"} Q3 -->|yes| DR["Direct reversal<br/>MGMT, ALKBH"] Q3 -->|no| BER["Base excision repair"] Q2 -->|"yes — but mispaired"| MMR["Mismatch repair"]

2. Direct reversal: the cheapest repair, and the strangest enzyme in the cell

Sometimes the modification can simply be undone — no cutting, no template, no resynthesis.

Photolyase absorbs a blue-light photon and uses its energy to split the covalent bond joining a UV-induced dimer. One enzyme, one photon, lesion gone. It is present in bacteria, plants, fungi, insects, fish and marsupials, and absent in placental mammals. You do not have it. Every UV dimer in your skin must go through the far more expensive nucleotide excision pathway.

MGMT (O⁶-methylguanine-DNA methyltransferase) removes a methyl group from the O⁶ position of guanine — the alkylation lesion that matters most, because O⁶-methylguanine pairs with thymine and so causes G:C → A:T transitions. MGMT transfers the methyl onto one of its own cysteines, which irreversibly inactivates the protein. It is a suicide enzyme: strictly stoichiometric, one protein molecule consumed per lesion, not a catalyst at all.

That economics has a clinical consequence. Alkylating chemotherapy (temozolomide) kills glioblastoma by creating exactly this lesion. Tumours expressing MGMT mop it up and survive; tumours that have epigenetically silenced MGMT by methylation cannot, and respond. MGMT promoter methylation is a routine predictive biomarker in glioblastoma — an epigenetic mark on a repair gene predicting drug response (Hegi et al., NEJM 2005; Ch 23).

3. Base excision repair: the workhorse

BER handles the highest-volume damage. Its signature move is that recognition is outsourced to a family of specialists — eleven human DNA glycosylases, each recognising one kind of damaged base and nothing else. A dispatch table keyed on lesion type.

GlycosylaseRecognisesArising from
UNGuracil in DNAcytosine deamination (~100–500/cell/day)
OGG18-oxoguanineoxidative damage from normal metabolism
MUTYHadenine mispaired opposite 8-oxoGthe replication error 8-oxoG causes
TDGthymine opposite Gdeamination of 5-methylcytosine

Once a glycosylase finds its substrate the mechanism is uniform:

                     ↓ spontaneous deamination of C
5'- A C G T A U G C A T -3'      U:G — a U where a C should be
3'- T G C A T G C G T A -5'

  1. UNG flips U out of the helix, cleaves the base–sugar bond
5'- A C G T A _ G C A T -3'      abasic (AP) site: sugar stays, base gone
3'- T G C A T G C G T A -5'

  2. APE1 nicks the backbone 5' of the AP site
5'- A C G T A|_ G C A T -3'
3'- T G C A T G C G T A -5'

  3. Pol beta removes the sugar remnant, reads the intact strand, inserts C
5'- A C G T A C G C A T -3'
3'- T G C A T G C G T A -5'

  4. LIG3–XRCC1 seals the nick. Restored.

Two things to carry forward. The intermediates are more dangerous than the lesion — an abasic site, then a nicked backbone, are both worse than a uracil — so BER is a committed hand-off chain in which each enzyme recruits the next and the intermediate is never released.

And TDG's job exposes a design flaw in the epigenome: 5-methylcytosine deaminates to thymine, a perfectly normal DNA base. Uracil is obviously foreign and trivially excised; thymine is not, so this repair is intrinsically unreliable. That is why CpG dinucleotides are mutational hotspots roughly an order of magnitude above background, and why CpGs are depleted from the genome over evolutionary time (Ch 16).

4. Nucleotide excision repair: recognise the distortion, not the lesion

Bulky lesions — UV dimers, tobacco-smoke adducts, cisplatin crosslinks — are too varied for a dispatch table. NER instead recognises a structural property: a lesion that distorts the helix and disrupts base pairing. One recognition mechanism, unlimited substrate scope, which is why NER is the pathway for essentially every environmental carcinogen.

Recognition happens two ways, and the split matters:

Global-genome NER-coupled NER
SensorXPC–RAD23B patrolling the genome II stalling at a lesion
everywhere, slowlytranscribed strand of expressed genes, fast
Loss causesxeroderma pigmentosumCockayne syndrome

Both converge on TFIIH — the same complex that opens during transcription — whose XPB and XPD subunits unwind ~25 bp around the lesion. XPA and RPA verify the damage; then XPF–ERCC1 incises ~20 nt 5' of the lesion and XPG incises ~5 nt 3' of it, releasing a 24–32-nucleotide oligomer containing the damage. Pol δ/ε fill from the intact strand.

Xeroderma pigmentosum is the loss-of-function . Seven complementation groups (XP-A to XP-G) correspond to NER components, and they do not all sit in the same branch: XP-C and XP-E lose the global-genome sensor specifically, while XP-A, XP-B, XP-D, XP-F and XP-G are core incision factors shared by both branches and so knock out global-genome and transcription-coupled NER together. Patients carry a >10,000-fold increased risk of non-melanoma skin cancer and >2,000-fold for melanoma, with median age at first skin cancer around 9 years against ~60 in the general population. Incidence ranges from ~1 in 20,000 in Japan to ~1 in 250,000 in the USA.

The eighth group is instructive. XP-V patients have entirely normal NER; their defect is in POLH, a translesion polymerase (§8). Same clinical phenotype, different pathway — phenotype does not identify mechanism.

The contrast with Cockayne syndrome is the sharpest teaching point here. CS patients lose only transcription-coupled NER. They have severe neurodegeneration, developmental failure and dramatic premature ageing — and no cancer predisposition at all. The reason is what happens to the cell after the failure: lesions that block transcription trigger death and tissue attrition (ageing), whereas lesions that are tolerated and bypassed produce (cancer). Repair deficiency causes cancer only in cells that survive it.

5. Mismatch repair: the strand-discrimination problem

Replication is already accurate: polymerase base selection gives ~10⁻⁴–10⁻⁵ errors per base, and 3'→5' catches ~99% of those (Ch 04). MMR is the third filter, worth another 100–1000-fold. But it faces a problem the other pathways do not:

                      ↓ replication error
5'- A C G T A G G C A T -3'   newly synthesised: G inserted opposite T
3'- T G C A T T C G T A -5'   parental template
                      G:T — both bases chemically normal

There is no damage here. There is a disagreement. Excising the G restores the original; excising the T fixes a into both strands permanently. A strand-blind system would not be merely useless — it would be actively mutagenic in half of all events.

Bacteria use methylation. E. coli Dam methylase methylates adenine in GATC sites, but only after a delay, so immediately behind the fork the parental strand is methylated and the nascent strand is not. MutS finds the mismatch, MutL couples, and MutH nicks the unmethylated strand. The methyl mark is the strand label.

Eukaryotes use nicks. There is no equivalent mark. Instead the discontinuity of synthesis marks the new strand: the is full of gaps between , the leading strand has 5' ends from ribonucleotide processing. MutSα (MSH2–MSH6, single-base mismatches and small loops) or MutSβ (MSH2–MSH3, larger insertion–deletion loops) finds the mismatch; MutLα (MLH1–PMS2) carries a latent endonuclease activated by PCNA — and PCNA is a ring loaded at the fork with a defined orientation, so the sliding clamp carries the strand information. EXO1 excises from the nick past the mismatch, sometimes across a kilobase.

Lynch syndrome is heterozygosity for MLH1, MSH2, MSH6, PMS2, or EPCAM deletions that silence MSH2. Population prevalence is roughly 1 in 300 to 1 in 2,000 depending on the study; lifetime colorectal cancer risk runs ~40–80% for MLH1/MSH2 carriers and appreciably lower for MSH6 and PMS2, a gradient that drives gene-specific surveillance (Ch 54).

Microsatellite instability is the diagnostic readout. A microsatellite is a short tandem repeat — (A)₂₅, (CA)₁₅ — where polymerase slippage is common, producing insertion–deletion loops that MutSβ normally removes. Without MMR, slippage becomes permanent, so every microsatellite in the tumour to a new length. Compare tumour against normal at a panel (classically BAT-25, BAT-26, D2S123, D5S346, D17S250; instability at ≥2 of 5 is MSI-high), or read the same signal from sequencing across hundreds of loci (Ch 46).

MSI is the cleanest case in medicine of a broken pathway leaving a machine-readable fingerprint, and the fingerprint now determines treatment: MMR-deficient tumours accumulate enormous mutation burdens, therefore many neoantigens, therefore respond to checkpoint blockade. Pembrolizumab's 2017 approval for MSI-H/dMMR tumours regardless of tissue of origin was the first tissue-agnostic cancer indication ever granted — a drug prescribed on a defective repair pathway rather than an organ.

6. Double-strand breaks: two pathways, one commitment step

Non-homologous end joining
Templatenonesister chromatid
Cell cycleall of itS/G2 only
Speedminuteshours
Fidelityerror-prone: at the junctionessentially error-free
Core factorsKu70/80, DNA-PKcs, Artemis, XRCC4–LIG4, XLFMRN, CtIP, EXO1/DNA2–BLM, RPA, BRCA1, PALB2, BRCA2, RAD51

is not a sloppy fallback the cell resorts to; it is the default. Ku70/80 is among the most abundant nuclear proteins (~10⁵–10⁶ copies) and binds free ends within seconds; DNA-PKcs follows, ends are trimmed and filled as necessary, LIG4–XRCC4 ligates. Because ends are usually damaged rather than clean, "as necessary" means a few bases lost or gained — a , which NHEJ accepts, because a with a small indel is enormously better than a chromosome in two pieces.

The cell exploits this deliberately. V(D)J recombination generates antibody and T-cell-receptor diversity by making programmed breaks and repairing them by NHEJ, using junctional imprecision as a source of diversity. Hence the of NHEJ deficiency: LIG4, PRKDC, NHEJ1, DCLRE1C mutations cause radiosensitive severe combined immunodeficiency — no immune repertoire plus extreme radiation sensitivity, from one broken pathway. It is also why -induced breaks produce indels by default (Ch 38).

Resection is the decision

Both pathways see the same substrate. What separates them is one irreversible step: whether to chew back the 5' ends and expose 3' single-stranded tails.

   double-strand break
   ────────────┤ ├────────────
        ┌────────┴────────┐
   no resection      resection (5'→3')
   ────────┤ ├────────    ────────┐        ┌────────
   Ku binds the ends           3' ssDNA tails exposed
        │                            │
      NHEJ                    RPA coats ssDNA → BRCA2 loads RAD51
   fast, any phase            homology search → HR (S/G2 only)

Once the ends are single-stranded, Ku cannot bind and NHEJ is off the table. So the regulation of resection is the regulation of pathway choice, and it is set by two antagonists:

On top sits the cell-cycle gate: CDK activity, low in G1 and high from S onward, phosphorylates CtIP and EXO1, and is required for resection. This is elegant engineering — the signal that licenses homologous recombination is the same signal that guarantees a sister chromatid exists to recombine with. HR is not preferred in S/G2 as a policy; it is physically gated on the availability of its template.

Mitosis is the exception that shows the gate is not simply "CDK high". CDK1–cyclin B activity is maximal in M, not in S/G2, and mitotic ends can indeed be resected — but the block moves downstream, with mitotic CDK1 preventing RAD51 from loading onto the resulting ssDNA. Mitotic breaks are therefore marked and deferred to the following G1 rather than repaired on the spindle, where joining two ends belonging to different chromosomes would fuse them.

After resection, RPA coats the ssDNA and BRCA2 — recruited via PALB2, which is itself bridged to BRCA1 — hands the DNA to RAD51, whose filament searches the sister for and invades (Ch 18). A third pathway, microhomology-mediated end joining driven by polymerase θ (POLQ), anneals 2–20 bp of flanking microhomology and therefore always deletes the sequence between.

7. BRCA, synthetic lethality, and PARP inhibitors

The genetic setup. BRCA1 and BRCA2 are required for HR. Carriers are heterozygous — one working copy, which suffices. Tumours arise when the second is lost somatically, so the tumour is HR-deficient while every normal cell in the patient is HR-proficient. That asymmetry is the therapeutic window, and it exists before any drug is given.

is a classical genetic relationship: two genes are synthetically lethal if losing either alone is survivable but losing both is not. We want a gene X with

   BRCA2⁻/⁻          viable   (true by construction — the tumour exists)
   X inhibited       viable   (normal cells tolerate it)
   BRCA2⁻/⁻ + X⁻     dead

Give a drug against X and you kill only cells that have already lost BRCA. PARP1 turned out to be such an X.

The mechanism, in the order it was understood. The original model (Bryant et al. and Farmer et al., both Nature 434, 2005): PARP1 detects single-strand breaks; inhibit it and SSBs persist; a running into an SSB collapses into a one-ended double-strand break; one-ended breaks are repairable only by HR; HR-deficient cells die. It made the right prediction, and has since been substantially revised — which is the interesting part:

Olaparib was approved in 2014 for BRCA-mutated ovarian cancer — the first cancer drug licensed alongside an FDA-approved germline companion diagnostic (Myriad's BRACAnalysis CDx) — and the class now spans breast, pancreatic and prostate cancer. Not the first drug licensed on a germline marker outright, though: ivacaftor was approved in 2012 restricted to cystic-fibrosis patients carrying the CFTR G551D allele.

Resistance proves the model. Tumours escape by restoring HR:

MechanismWhat it does
Reversion A second mutation restores the BRCA reading frame
Loss of 53BP1 or shieldinResection proceeds without BRCA1 — rescues BRCA1-mutant cells only, since BRCA2 acts downstream at RAD51 loading
Restored fork protectionGaps and nascent DNA protected without HR
Drug efflux, PARP1 mutationPharmacological, not genetic

The 53BP1 row is §6's argument reappearing as a clinical observation. Finally, the eligible population is larger than the mutation: tumours with the same downstream from PALB2, RAD51C/D or ATM loss or epigenetic BRCA1 silencing are called "BRCAness" and identified by genomic scars or SBS3 (Ch 56).

8. Damage tolerance: when repair is not an option

Replication cannot wait. A replicative polymerase — tight active site, accepts only correct Watson–Crick geometry — stalls at an unrepaired lesion, and a stalled fork left long enough collapses into a double-strand break, which is worse than the lesion.

Translesion synthesis is the licensed workaround. RAD6–RAD18 monoubiquitinates PCNA at K164, swapping the replicative polymerase for a Y-family enzyme — Pol η (POLH), Pol ι, Pol κ, REV1, or B-family Pol ζ. These have open active sites that accommodate a distorted template and no : error rates of 10⁻¹ to 10⁻³, three to six orders of magnitude worse than the replicative enzymes. They insert a few bases across the lesion and hand back.

This is not repair — the lesion is still there. Tolerance buys fork progression at the price of , and the cell limits the damage by using the sloppy enzyme for only a few nucleotides. Pol η is the elegant case: it bypasses a UV thymine–thymine dimer accurately, because its active site happens to fit that lesion and insert AA. Accurate on its cognate substrate, mutagenic on everything else. Lose it (XP-V) and dimers get bypassed by the other Y-family polymerases, which insert the wrong base — full xeroderma pigmentosum from a tolerance defect. The error-free alternative, template switching, uses the newly made sister strand to get past the lesion, but is slower and needs HR machinery.

9. The DNA damage response: sensing, stalling, deciding

Repair enzymes act locally. The DNA damage response is the global signalling layer that notices damage, halts the cell cycle so repair has time, and removes the cell if the damage is too great.

graph TD DSB["Double-strand break"] --> MRN["MRN sensor<br/>MRE11–RAD50–NBS1"] --> ATM["<b>ATM</b>"] SS["Resected ends,<br/>stalled forks:<br/>RPA-coated ssDNA"] --> ATRIP["ATRIP sensor"] --> ATR["<b>ATR</b>"] ATM --> H2AX["γH2AX spreads over Mb<br/>MDC1 recruits more MRN"] H2AX -.->|positive feedback| ATM ATM --> CHK2["CHK2"] ATR --> CHK1["CHK1"] CHK1 --> CDC25["CDC25 phosphatases<br/>inactivated"] CHK2 --> CDC25 CHK2 --> P53["<b>p53</b> stabilised"] CDC25 --> ARREST["CDK off →<br/>cell-cycle arrest"] P53 --> P21["p21 → G1/S arrest"] P53 --> SEN["senescence"] P53 --> APO["apoptosis<br/>PUMA, NOXA, BAX"]

ATM responds to double-strand breaks via MRN. ATR responds to RPA-coated single-stranded DNA — which appears at resected breaks and at stalled forks, making ATR the general replication-stress kinase.

The amplification step is what makes one break visible. ATM phosphorylates the histone variant H2AX to make γH2AX, which spreads over up to megabases of flanking chromatin; MDC1 binds γH2AX and recruits more MRN and more ATM — positive feedback that converts a single lesion into a microscopic nuclear focus and a genome-wide signal.

Downstream, CHK1 and CHK2 inactivate the CDC25 , CDK activity falls, and the cycle arrests. In parallel ATM and CHK2 phosphorylate p53 and its degrader MDM2, breaking the loop that normally keeps p53 scarce. p53 is a , and which programme it activates decides the cell's fate:

OutcomeTriggerEffector
Arrest and repairmodest, repairable damagep21 (CDKN1A) inhibits CDKs
Senescencepersistent but survivable damagepermanent arrest; cell lives on, secreting inflammatory signals
Apoptosissevere or irreparable damagePUMA, NOXA, BAX → mitochondrial death pathway

Both are tumour suppression by sacrifice, which is why TP53 is the most frequently mutated gene in human cancer, disrupted in roughly half of all tumours. A cell that loses p53 does not repair worse — it stops caring, proceeding through the cycle with damaged DNA, so damage becomes . loss is Li-Fraumeni syndrome. And senescence, the non-lethal option, is a leading suspect in ageing — the usual explanation for why so many repair-deficiency syndromes look progeroid.

10. Deficiency syndromes, and why mutation rates are heterogeneous

SyndromeGene(s)Pathway lostHallmark
Xeroderma pigmentosumXPA–XPG; POLH (XP-V)NER; TLS for XP-V>10,000× skin cancer, first tumour ~age 9
Cockayne syndromeERCC6, ERCC8TC-NERProgeria, neurodegeneration, no cancer excess
TrichothiodystrophyERCC2, ERCC3, GTF2H5NER via TFIIH stabilityBrittle sulfur-deficient hair, developmental defects
Lynch syndromeMLH1, MSH2, MSH6, PMS2, EPCAMMMRColorectal + endometrial cancer; MSI-high tumours
CMMRDbiallelic MMR genesMMRChildhood brain and blood tumours; café-au-lait macules
MUTYH-associated polyposisMUTYHBER glycosylaseAdenomatous polyposis; excess G:C → T:A
Hereditary breast/ovarian cancerBRCA1, BRCA2, PALB2HRPARP-inhibitor sensitive tumours
Fanconi anaemia20+ FANC genes (BRCA2 = FANCD1)Interstrand crosslink repair / HRMarrow failure, AML, breakage on mitomycin C
telangiectasiaATMDSB signallingAtaxia, immunodeficiency, radiosensitivity, lymphoma
Nijmegen breakage syndromeNBNMRN sensorMicrocephaly, immunodeficiency, lymphoma
Bloom syndromeBLMRecQ in HR~10× elevated sister chromatid exchange
Werner syndromeWRNRecQ helicaseAdult-onset progeria
Radiosensitive SCIDLIG4, PRKDC, NHEJ1, DCLRE1CImmunodeficiency (V(D)J fails) + radiosensitivity

Mutation-rate heterogeneity is repair heterogeneity

If were driven by damage, mutation density would be roughly uniform — damage chemistry does not care where in the genome it lands. It is not uniform. density in cancer genomes varies several-fold at megabase scale, tracking features that are about repair access, not chemistry:

The consequence is unavoidable for anyone doing cancer genomics. A test asking "is this gene mutated more than expected by chance?" needs a per-region background rate; a genome-wide average produces confident nonsense — famously, very large, late-replicating, lowly expressed genes such as TTN and the olfactory receptor family surface as top "drivers". MutSigCV-style methods exist to model replication timing, expression and chromatin as covariates (Ch 56). The same caution applies to any genome-wide scan for elevated substitution rate, including tests of selection (Ch 33).


Common misconceptions

What people thinkWhat's actually true
fixes It fixes damage — abnormal or mispaired DNA, detectable because something else disagrees. A mutation is a normal, correctly paired base pair; nothing is left to detect
The tells you how often DNA is damagedIt tells you the residual. Damage runs ~10⁴–10⁵ lesions per cell per day; roughly one in 10⁷ becomes an inherited mutation
Repair is accurate — that's the point of it is mutagenic by design, MMEJ always deletes, translesion synthesis is licensed error. The cell repeatedly trades fidelity for survival
A repair-deficient cell just mutates faster, uniformlyEach pathway has a lesion specificity, so each defect leaves a characteristic signature — MMR loss gives MSI, MUTYH loss gives G:C→T:A, HR loss gives SBS3 and structural scars
BRCA1 and BRCA2 do the same jobBoth are needed for HR at different steps: BRCA1 licenses resection against 53BP1, BRCA2 loads RAD51 afterwards. Only BRCA1 loss is rescued by deleting 53BP1 — which is how we know
work by switching off a repair pathwayKilling tracks PARP trapping — the drug locking PARP1 onto DNA as a physical obstacle — far better than catalytic inhibition. Otherwise the class would be equipotent, and it isn't
p53 repairs DNAp53 is a at the end of the signalling chain. It arrests the cycle, or triggers senescence or apoptosis. It never touches a lesion
Every organism can undo UV damage with lightPhotolyase does exactly that, and placental mammals lost it. Every UV dimer in your skin goes through NER or gets bypassed
Mutation rate is a genome-wide constant you can use as a nullIt varies several-fold at megabase scale, driven by repair access. Ignoring that is how olfactory receptor genes end up on cancer-driver lists

Worked example: how efficient does repair have to be?

Derive the filter efficiency rather than accepting "very".

Step 1 — count lesions along the lineage that matters. Only the single chain of cells running from zygote to gamete transmits anything. Follow that chain in a 30-year-old father.

Step 2 — count what got through. From verified facts, 1.1–1.3 × 10⁻⁸ per bp per generation over a 6.2 × 10⁹ bp genome:

   1.1e-8 × 6.2e9 = 68
   1.3e-8 × 6.2e9 = 81      →  ~70–80 de novo mutations per offspring

About 80% are paternal in origin, so the paternal chain contributed roughly 55–65. Call it 60.

Step 3 — the residual per lesion.

   60 mutations / 5.5e8 lesions  ≈  1.1e-7

About one lesion in ten million ends up as an inherited base change.

Step 4 — cross-check from replication fidelity, independently. Three filters in series, each measured separately:

FilterError rate after it
Polymerase base selection10⁻⁴ – 10⁻⁵
+ 3'→5' (~100×)10⁻⁶ – 10⁻⁷
+ (~100–1000×)10⁻⁸ – 10⁻¹⁰ per bp per division

A 30-year-old male has undergone roughly 380–400 divisions (~30 to puberty, then ~23 per year), so:

   400 divisions × 1e-8  per bp per division  =  4e-6 per bp per generation
   400 divisions × 1e-10 per bp per division  =  4e-8 per bp per generation

against an observed 1.1–1.3 × 10⁻⁸. Read that honestly: the prediction spans 4 × 10⁻⁶ to 4 × 10⁻⁸, and only its accurate end — the ~10⁻¹⁰ per-division figure Ch 04 takes as its point estimate — lands within a factor of three of the measurement. The sloppy end overshoots ~300-fold. So the cross-check does not confirm the per-division fidelity figures; it constrains them: real per-division fidelity has to sit near the good end of the quoted range for the observed germline rate to come out, which is the same conclusion Ch 04 reaches from the other direction.

Step 5 — how wrong can this be? The damage estimate is the weak link, uncertain by an order of magnitude either way. Even so the residual lands in 10⁻⁶ to 10⁻⁸: repair removes six to eight orders of magnitude of error, against proofreading's two. Two caveats push the same direction — not every unrepaired lesion is mutagenic (a lesion in a quiescent cell can be repaired at leisure and cannot become a until a fork copies it), and germline stem cells cycle only about every 16 days. Both widen the gap between lesions that escape repair and lesions that become transmitted mutations — and the ratio above counts every escaped-but-silent lesion as a repair success. So 1.1 × 10⁻⁷ is a lower bound on repair's failure rate, and the six-to-eight-orders figure is a ceiling on repair's performance. The number flatters repair; it does not understate it.

The conclusion survives all of it: the is not a property of DNA chemistry. It is a property of .

Connections