18 — Recombination mechanisms
Chapter 14 treated recombination as a black box that emits recombinant chromatids at a rate proportional to distance. This chapter opens the box. What comes out is stranger than the black box suggested: the machinery that shuffles your is a pathway, and runs it by deliberately damaging the genome.
What you'll be able to do
- Trace a meiotic double-strand break from cleavage to a crossover or non-crossover, naming the intermediate at each step, and predict the flanking-marker and gene-conversion outcome of DSBR versus SDSA resolution
- Explain why the same pathway is a repair mechanism in a somatic cell and a shuffling mechanism in a germ cell, and what differs between the two
- Explain why , not , is the usual fate of a programmed break, and why its GC bias mimics weak selection closely enough to generate false positives in selection scans
- Derive why every must receive at least one crossover, and quantify what happens when one doesn't
- Explain the hotspot paradox and why 's rapid evolution resolves it
- Distinguish homologous, site-specific and transpositional recombination by their requirements and their products
- Explain how non-allelic between repeats generates recurrent deletions and duplications
The core idea
is, mechanically, a restore-from-replica operation. A DNA double-strand break destroys information on both strands at once, so there is nothing left locally to read across from (Ch 17). The only way to recover the sequence is to find another copy of that region elsewhere in the cell and copy from it.
A somatic cell does this defensively. Something broke; find the sister chromatid, which is an identical byte-for-byte replica; copy across; leave no trace. is an unwanted side effect and is actively suppressed.
runs the same pathway offensively. It makes the breaks on purpose, hundreds of them, then forbids the safe template — the sister — and forces repair from the homolog, which is a different version of the same region. The physical act of copying from the homolog, followed by cutting the resulting tangle the right way, exchanges the arms. That exchange is the crossover of Chapter 14. It is also, as we will see, mechanically required for the chromosomes to segregate at all.
Meiotic recombination is not a variation-generating mechanism that happens to borrow repair proteins. It is the repair pathway, deliberately triggered, with its template preference reversed and its outcome bias inverted. Everything else in this chapter follows from taking that literally.
1. One machine, two jobs
| Somatic HR (repair) | Meiotic HR (programmed) | |
|---|---|---|
| Trigger | Accidental break: radiation, collapse, nuclease | cuts on purpose, ~200–300 breaks per cell |
| Preferred template | Sister chromatid — identical, therefore error-free | Homolog — enforced by a barrier to sister repair (HORMAD proteins, ATR/CHK signalling on the axis) |
| Strand-exchange protein | RAD51 only | RAD51 and DMC1, the -specific that biases toward the homolog |
| Desired outcome | Non-crossover. Crossovers between homologs risk loss of heterozygosity | Crossover, at least one per chromosome pair, guaranteed |
| Timing | S/G2, when a sister exists | Prophase I, within the synaptonemal complex |
| When it fails | Chromosome rearrangements, cancer predisposition (BRCA1, BRCA2) | Non-disjunction, aneuploid gametes, infertility |
The protein list overlaps almost completely. RAD51, RPA, BRCA2, the MRE11–RAD50–NBS1 (MRN) complex, EXO1, the resolvases — all of them do both jobs. Meiosis adds a handful of components (SPO11, DMC1, MSH4–MSH5, and MLH3, which partners the general mismatch-repair protein MLH1) that redirect an existing pathway rather than build a new one. This is why BRCA2 is both a breast-cancer gene and an essential meiotic gene, and why tumours that have lost are killed by (Ch 56).
2. Breaking your own chromosomes on purpose
is a -like enzyme. Topoisomerases relieve torsional stress by cutting DNA, letting it swivel, and resealing — the cut is transient and the enzyme stays covalently attached to the DNA end while the break is open. SPO11 performs the first half of that reaction and never performs the second. It cuts, stays attached to the 5' end via a covalent tyrosine link, and leaves.
A cell entering therefore inflicts on itself ~200–300 double-strand breaks — the single most dangerous lesion in DNA biology — at a moment when it has 46 to keep track of. Estimated from RAD51/DMC1 focus counts in mouse and human spermatocytes, that is roughly one break every 10–15 Mb.
Why accept that risk? Because the break is not only the initiator of exchange, it is the search signal. Homologs have to find each other in a crowded nucleus before they can be pulled apart correctly, and the resected single-stranded tail from a SPO11 break is the probe that performs that search. Knock out SPO11 and homologs fail to pair and synapse, meiosis arrests, and the animal is sterile. The damage is the mechanism.
Note the outcome asymmetry this creates immediately: the chromatid that gets cut is the one that loses sequence, because it is repaired using the uncut homolog as template. Hold that thought; it detonates in §7.
3. From break to invasion
Three steps deserve unpacking.
Resection degrades the 5'-ended strand at each side of the break, leaving 3'-ended single-stranded tails hundreds to a couple of thousand nucleotides long. This is the committing step: once resected, the break can no longer be blunt-end joined (Ch 17). Resection is the switch that selects -directed repair over non-homologous end joining.
Filament formation. RPA binds the ssDNA first, removing secondary structure; mediators (BRCA2 with PALB2 in humans) then exchange RPA for RAD51/DMC1, which polymerise into a filament that stretches the DNA ~1.5-fold and presents the bases ready to test for complementarity.
Homology search is the step that should bother a programmer. There is no index. The filament finds its match by colliding with duplex DNA, transiently opening it, testing ~8 bases, and letting go if the test fails — a scan by random collision over a 3.1 Gb search space, run ~250 times in parallel, completed in hours. It terminates only because the search space is pre-partitioned by nuclear architecture: territories and tethering of breaks to the chromosome axis cut the effective space by orders of magnitude before the probe starts. When a biological search looks impossibly slow, the geometry usually did most of the work first.
A successful search produces the D-loop: the invading 3' end base-paired to the , with the donor's identical strand displaced as a single-stranded bubble. That invading end is now a primer, and a polymerase extends it using the homolog as template. From here the broken chromatid is copying sequence it did not previously have.
4. Two endings, and what decides between them
Everything above is shared. The branch comes now.
SDSA — synthesis-dependent strand annealing. The extended invading strand is stripped back out of the D-loop by a and anneals to the ssDNA on the other side of the break. Gaps fill in, ends ligate. The two duplexes were never covalently joined, so flanking markers are not exchanged. The only trace is that a patch of sequence at the break site was copied from the homolog. This is a non-crossover with .
DSBR — double-strand break repair via a double . The displaced strand of the D-loop captures the second broken end, synthesis and ligation proceed on both sides, and you end with two four-way junctions flanking the repaired region. This intermediate can go either way.
A Holliday junction is four strand-ends meeting at one branch point. The branch is mobile: it migrates along the duplexes without breaking anything, extending or shrinking the region of heteroduplex DNA where one strand comes from each parent. Getting rid of the junction requires cutting exactly two of the four strands — a two-bit decision — and which pair you cut sets the topology:
Single junction, two ways to cut:
cut the pair of strands that crossed → duplexes separate as they went in
flanking markers UNCHANGED
cut the other pair → duplexes separate swapped
flanking markers EXCHANGED
Double Holliday junction — the outcome is the XOR of the two junctions:
both junctions cut the same way → NON-CROSSOVER (+ conversion tract)
the two junctions cut differently → CROSSOVER (+ conversion tract)
There is also a third exit that avoids cutting altogether: dissolution. The BLM helicase with IIIα (the BTR complex) migrates the two junctions toward each other until they merge, then decatenates the result. Dissolution can only ever produce a non-crossover. Loss of BLM causes Bloom syndrome, whose diagnostic cellular is a tenfold excess of sister-chromatid exchanges — crossovers that should have been dissolved and instead got resolved.
Here is the part that textbooks routinely get wrong. The XOR argument makes it look as though half of all double Holliday junctions should become crossovers. In they do not. Resolution is directed, not random: a dedicated meiotic resolvase, MutLγ (MLH1–MLH3) acting with EXO1 and marked by the MutSγ complex (MSH4–MSH5), resolves designated junctions almost exclusively toward crossover. Non-crossovers overwhelmingly arise by SDSA and never form a double Holliday junction at all. The coin-flip picture describes a chemical possibility, not the in vivo statistic.
The arithmetic falls out: ~250 breaks per human meiocyte, ~50 crossovers. Roughly 80–90% of programmed double-strand breaks are repaired as non-crossovers. is the rare outcome, and the abundant outcome is gene conversion.
5. Gene conversion: the non-reciprocal half
A crossover is reciprocal — chromatid 1 gets the right arm of chromatid 3 and vice versa, nothing is created or lost. But the few hundred base pairs around the initiating break are not reciprocal. That region was destroyed on the cut chromatid and rebuilt by copying the homolog. One has been overwritten by the other.
Two mechanisms produce it: copy from the donor (the synthesis step in the D-loop) and of heteroduplex (where the two annealed strands disagree, the repair machinery excises one and rewrites it using the other as template — resolving the conflict by overwriting rather than merging).
The classic evidence predates all of this molecular detail. Fungi such as Neurospora and yeast keep all four meiotic products in a single ascus, so you can count them directly. A heterozygote should give 2:2 at every locus. Occasionally an ascus segregates 3:1 (or 6:2 after the post-meiotic mitosis), and sometimes 5:3 — a ratio that is only possible if one product carried an unrepaired mismatch through the division. Mendel's first law is violated in a small, specific, mechanistically explicable fraction of meioses.
In humans, non-crossover conversion tracts are short, roughly 50–1,000 bp, and any given base is converted at about 5.9 × 10⁻⁶ per generation — several hundred times the point of ~1.3 × 10⁻⁸ (verified facts). is not a curiosity; it is the most common thing that happens to a heterozygous site during , apart from nothing.
Three consequences that matter downstream:
| Consequence | Why |
|---|---|
| Transmission bias | An allele that is preferentially converted is transmitted at a rate ≠ 0.5. Selection is not the only way a frequency can change systematically |
| Homogenisation of gene families | Tandem convert one another, so in an array stay more similar to each other than to their in other species — concerted evolution (Ch 35). rRNA arrays are the canonical case |
| Contamination of phylogenetic and selection signals | A converted region has a different history from its neighbours. Trees built across a conversion tract are wrong (Ch 34) |
GC-biased gene conversion
The mismatch-repair step is not neutral about which allele it keeps. Faced with a G:T or an A:C mismatch in heteroduplex, human repair preferentially restores the G:C pair. Measured from , when a conversion event covers a site heterozygous for one GC and one AT allele, the GC allele is transmitted about 68% of the time instead of 50%.
Formally this is a transmission bias with the same mathematical form as weak directional selection favouring GC — it enters the allele-frequency equations in the same place and behaves the same way, while having nothing to do with survival or reproduction. Its effects are exactly what you would predict from that:
- GC content tracks recombination rate across the genome. High-recombination regions — subtelomeric sequence, short , male-biased hotspot regions — are GC-rich. Much of the "isochore" structure of mammalian genomes is a fossil of where recombination has been.
- It can fix deleterious alleles. A weakly harmful AT→GC change in a hotspot can be driven up in frequency, because the bias acts on transmission regardless of .
- It generates false positives in selection scans. Regions of rapid, AT→GC-skewed substitution look like adaptive evolution and are frequently not. A large share of "human accelerated regions" sit in high-recombination sequence and show precisely the GC skew that gBGC predicts. Any test of selection run on recombining sequence has to control for it (Ch 33).
6. Why every bivalent gets a crossover
Crossovers are not distributed at random along , and not distributed at random between chromosomes either. Two departures from Poisson, both load-bearing.
. One crossover suppresses others nearby — Chapter 14 measures this as a coefficient of coincidence below 1. Mechanistically, the current leading model is coarsening: a limited pool of pro-crossover protein (HEI10/RNF212 in mammals) diffuses along the synaptonemal complex, and larger foci grow at the expense of smaller ones, the way large droplets grow at the expense of small ones in Ostwald ripening. The dynamics spontaneously produce one winner per region, well spaced. Crossovers designated this way are class I (MutSγ/MutLγ-dependent) and are ~90% of the total in mammals; the remainder are class II, resolved by MUS81–EME1 and showing no interference.
Assurance and homeostasis. Every receives at least one crossover, essentially without exception, and the total is buffered against the number of breaks: reduce activity several-fold in yeast and the crossover count barely moves. This is a control loop with a setpoint on the output, not the input.
Why is that required? Because a crossover is not only genetic — it is mechanical. The , held together by cohesion distal to the exchange point, is the only physical link between homologs at metaphase I. It is what lets the spindle pull them in opposite directions and feel resistance. No crossover means no link, means two univalents that segregate at random.
Derive the cost. Human male has a genetic map of about 2,600 cM, which is 26 Morgans per transmitted gamete and therefore about 52 crossovers per meiocyte (each crossover involves 2 of 4 chromatids, so one crossover per cell contributes 50 cM). Spread over 23 bivalents that is 2.26 per bivalent. If placement were Poisson:
P(a given bivalent gets zero crossovers) = e^-2.26 = 0.104
expected achiasmate bivalents per meiosis = 23 × 0.104 = 2.4
P(all 23 bivalents get at least one) = (1 - 0.104)^23 = 0.080
P(euploid gamete, univalents random) = (1 - 0.104/2)^23 = 0.29
Only 8% of meioses would have every bivalent chiasmate. That is not the same as the gamete being wrong the other 92% of the time: two univalents still segregate to opposite poles by luck half the time, so the model's euploid-gamete rate is the last line — about 29%. Still catastrophic. The real rate of is far lower than that, which tells you assurance is enforced rather than lucky. And the argument understates the problem: crossovers scale with chromosome length, so the smallest autosomes sit closest to the one-crossover floor. They are the ones that mis-segregate most often — chromosomes 21, 22 and 16 dominate the trisomies seen at conception. Which of those you ever see in a newborn is a separate question, settled by gene dosage rather than by : only trisomies 21, 18 and 13 carry few enough genes to be tolerated to term, while 16 — the single most common autosomal trisomy at conception — is lost in the first trimester without exception (Ch 20). Maternal trisomy 21 is associated with reduced or absent recombination on chromosome 21, or with crossovers placed too near the to hold. The same explains why the X and Y, sharing only a few megabases of pseudoautosomal sequence, must nevertheless cross over there in every male meiosis (Ch 13).
7. Hotspots, PRDM9, and a paradox
Recombination is not uniform along the genome. From -disequilibrium mapping, more than 25,000 hotspots in the human genome, typically 1–2 kb wide, carry ~80% of all recombination in ~10–20% of the sequence. Rate varies by orders of magnitude over kilobase scales, which is why LD blocks have edges (Ch 29).
In humans and mice, hotspot positions are set by one protein. has a C2H2 zinc-finger array that binds a specific DNA motif — a degenerate 13-mer, CCNCCNTNNCCNC, is found at roughly 40% of human hotspots — and a PR/SET domain that trimethylates nearby histone H3 on lysine 4 and lysine 36. is directed to sites carrying those marks. Hotspots are therefore not a property of the DNA sequence alone; they are a property of a protein's binding preference, and different PRDM9 specify different hotspot maps. Two humans can have measurably different recombination landscapes.
Now the paradox, and it is a real one. Recall §2: the that is cut is the one that loses its sequence. A chromosome carrying an intact PRDM9 binding motif gets cut; it is repaired from the homolog, which may carry a disrupted motif; the intact motif is converted to the disrupted one. The active allele destroys itself, at a rate proportional to how well it works. Hotspots are self-limiting by construction — yet they exist everywhere, and they are strong.
The resolution: hotspots are not persistent; the system that makes them is. PRDM9's zinc-finger array is a tandem minisatellite that mutates rapidly by unequal exchange between repeats, and the DNA-contacting residues within it are among the fastest-evolving in the genome, under strong positive selection. Dozens of PRDM9 alleles segregate in humans, each with a different binding specificity. As soon as one allele has eroded its own target sites, a variant that targets a new set has an advantage. Hotspots turn over on a timescale of hundreds of thousands of years, which is why human and chimpanzee hotspot positions essentially do not overlap despite ~99% sequence identity between the genomes.
The control experiment is provided by species that have lost PRDM9 — and there are many, because the gene is ancestral to vertebrates ( run from jawless and cartilaginous fish through coelacanth, turtles, snakes and lizards to mammals) and has been lost repeatedly. Dogs, wolves and coyotes carry a ; birds and crocodilians lost it outright. In those lineages, recombination is targeted to and — an open-chromatin default requiring no sequence-specific targeting protein — and hotspot positions are stable over evolutionary time, because promoters cannot be eroded away without cost. PRDM9 is best understood as a system for placing recombination away from functional elements, at the price of permanent instability.
PRDM9 as a speciation gene. Two mouse subspecies, Mus musculus musculus and M. m. domesticus, produce sterile F1 males in one direction of the cross. The responsible locus, mapped as Hst1 and identified by positional cloning as Prdm9, remains the only hybrid-sterility gene identified in vertebrates. The mechanism follows directly from erosion: within each subspecies, PRDM9 target sites on its own chromosomes have been degraded by conversion, while the sites recognised on the other subspecies' chromosomes are intact. In the hybrid, PRDM9 therefore binds asymmetrically — mostly to one homolog — so the break frequently occurs at a position where the partner chromosome lacks the corresponding site. Repair and synapsis fail, arrests, the male is sterile. A gene-conversion bias operating over tens of thousands of generations becomes a reproductive barrier.
That barrier has a name and a general form, supplied in Ch 35A §3: this is a , and PRDM9 is the only one identified to the gene in any vertebrate. Both requirements of that model are met here and are worth checking against the mechanism above — each allele is derived, and the two sit in different lineages, so the protein–DNA pairing that fails was never assembled inside any population where selection could have seen it. Note also what the interacting partners are: a protein and a binding site, not two proteins. The hotspot paradox and the speciation gene are one mechanism read at two timescales.
8. Recombination that is not homologous
"Recombination" names several unrelated chemistries. Keeping them apart matters, because they have different requirements, different products, and different failure modes.
| Homologous | Site-specific | Transpositional | |
|---|---|---|---|
| Requires sequence homology? | Yes, extensive (hundreds of bp) | No — recognises short defined sites | No |
| Recognition | Base-pairing during homology search | Protein binds a specific short sequence | Protein binds the element's own ends |
| Enzymes | RAD51/DMC1/RecA + resolvases | Tyrosine or serine recombinases; RAG1/RAG2 | Transposase / integrase |
| Product | Crossover or between allelic copies | Integration, excision, between defined sites | Element copied or moved to a new location |
| Examples | Meiotic ; DSB repair | Phage λ integration; Cre–lox; V(D)J | , Alu, DNA transposons (Ch 19) |
Phage λ integration is the clean case. Integrase, a tyrosine recombinase, plus the host factor IHF, aligns attP on the phage with attB on the E. coli — sharing only a 15 bp core — and swaps strands to insert the phage genome, generating attL and attR. Adding the Xis protein runs the reaction backwards. No homology search, no resection, no repair synthesis; a defined reaction on defined substrates. The domesticated versions, Cre–lox and FLP–FRT, are the standard tools for conditional genetics (Ch 37).
V(D)J recombination is site-specific recombination running as a diversity generator. The antibody and T-cell-receptor loci contain arrays of V, D and J segments, each flanked by a recombination signal sequence: a conserved heptamer (CACAGTG), a spacer of exactly 12 or 23 bp, and a nonamer (ACAAAAACC). RAG1/RAG2 will only join a 12-spacer site to a 23-spacer site — the 12/23 rule — which is what enforces V-to-J rather than V-to-V joining. RAG cuts precisely at the heptamer border, generating hairpinned coding ends; those are opened by Artemis, chewed and extended with random nucleotides by terminal transferase, and joined by non-homologous end joining. Combinatorial choice of segments gives ~10⁶ combinations; the sloppy junction adds several more orders of magnitude. Note what this implies: a controlled, programmed double-strand break, repaired by the error-prone pathway, where the errors are the point. RAG1/RAG2 is itself descended from a domesticated transposon, which is exactly what its chemistry looks like.
9. NAHR: homology in the wrong place
identifies its template by sequence similarity. The human genome contains millions of repeated sequences with high identity — , ~1.1 million Alu elements, ~46% transposable-element-derived sequence (verified facts). The alignment step cannot always tell an allelic copy from a paralogous one.
Non-allelic homologous recombination is recombination between two repeats that are similar in sequence but not at the same locus. The requirement is a few hundred base pairs of near-perfect identity — a minimum efficient processing segment — which most low-copy repeats far exceed, typically running 10–400 kb at 95–99% identity.
The product depends on the geometry of the repeats:
DIRECT repeats on homologous chromosomes (misaligned pairing, then crossover):
chr A ──[REP]═════ X ═════[REP]── crossover between the mis-paired REPs
chr B ──[REP]═════ X ═════[REP]── |
v
product 1 ──[REP]── DELETION of X (one copy of the region)
product 2 ──[REP]═════ X ═════ X ═════[REP]── DUPLICATION (three copies)
INVERTED repeats on the same chromosome → INVERSION of the intervening segment
Repeats on non-homologous chromosomes → RECIPROCAL TRANSLOCATION
Two features make this clinically distinctive. First, the products are reciprocal: the deletion and the duplication are made in the same event, in the same , and both are seen as human syndromes. Second, the breakpoints are recurrent — they fall inside the repeats, so unrelated patients have essentially the same rearrangement, to within the length of the repeat.
The textbook example is 17p12. A 1.4 Mb segment containing PMP22 is flanked by 24 kb repeats (CMT1A-REP) sharing ~98.7% identity. between them deletes or the segment. Three copies of PMP22 cause Charcot–Marie–Tooth disease type 1A; one copy causes hereditary neuropathy with liability to pressure palsies. Same event, opposite products, different disease — a pure gene-dosage effect. The same architecture explains the recurrent deletions at 22q11.2, 7q11.23, 15q11–q13 and 16p11.2 (Ch 20).
For a programmer the analogy is exact: NAHR is a read mismapping to a paralogous repeat, except that the aligner is a nuclease and the consequence is written back to disk.
Common misconceptions
| What people think | What's actually true |
|---|---|
| Crossovers exist to generate variation | They are mechanically required for homolog — the is the physical link the spindle pulls against. Variation is a consequence, and organisms with achiasmate show what happens without one |
| Meiotic recombination starts when accidentally break | makes ~200–300 double-strand breaks deliberately. Without them, homologs do not pair, synapse or segregate |
| Recombination is a reciprocal exchange | The crossover is reciprocal; the few hundred bp around the break are not. That patch is copied from the homolog — , non-reciprocal, and the more common outcome |
| Half of double resolve to crossovers | That is the geometry, not the biology. Meiotic resolution is directed by MutLγ almost entirely toward crossover, and most non-crossovers arise by SDSA without ever forming a junction |
| Gene conversion is a rare oddity | ~80–90% of programmed meiotic breaks resolve as non-crossover conversions. Per base per generation, conversion is hundreds of times more likely than point |
| are encoded in the DNA | In humans they are specified by binding. Different PRDM9 give different hotspot maps, so the recombination landscape differs between individuals — and dogs, lacking PRDM9, use instead |
| is a form of selection | It is a meiotic transmission bias, independent of . It enters the equations where selection does, which is exactly why it produces false positives in selection scans |
| Recombination is purely a shuffling process, not a mutational one | Crossover sites carry elevated rates, and between repeats generates the recurrent deletions and duplications behind many genomic disorders |
Worked example: one break, tracked to both endings
A , heterozygous at two flanking markers and at the hotspot itself. H is a binding motif that is intact; h is the same site with the motif disrupted by a point , so PRDM9 does not bind it.
marker hotspot marker
homolog 1 ────────── A ─────────── H ─────────── B ────── chromatid 1
────────── A ─────────── H ─────────── B ────── chromatid 2
homolog 2 ────────── a ─────────── h ─────────── b ────── chromatid 3
────────── a ─────────── h ─────────── b ────── chromatid 4
Step 1. PRDM9 binds only the intact motif, so H3K4me3 is deposited on homolog 1 and cuts chromatid 1 at the hotspot. Chromatid 3 is not cut.
Step 2. MRN releases SPO11 with a short oligonucleotide; EXO1 resects to 3' tails. The H on chromatid 1 is now physically gone.
Step 3. RAD51/DMC1 filament invades homolog 2. Repair synthesis uses h as template.
Step 4a — SDSA (the common outcome, ~80–90% of breaks):
chromatid 1 ── A ─────── h ─────── B ── converted H→h, flanks NOT exchanged
chromatid 2 ── A ─────── H ─────── B ── untouched
chromatid 3 ── a ─────── h ─────── b ── donor, unchanged
chromatid 4 ── a ─────── h ─────── b ── untouched
Step 4b — DSBR with the double resolved to crossover:
chromatid 1 ── A ─────── h ─────── b ── converted AND flanks exchanged
chromatid 3 ── a ─────── h ─────── B ── reciprocal crossover product
chromatid 2 ── A ─────── H ─────── B ── parental
chromatid 4 ── a ─────── h ─────── b ── parental
Step 5 — count alleles at the hotspot. Before: 2 H : 2 h. After, in both outcomes: 1 H : 3 h. The crossover/non-crossover decision changes the flanking markers and nothing about the conversion.
Step 6 — the transmission consequence. In a where the hotspot fires, P(transmit H) = 1/4 rather than 1/2. If the hotspot fires in a fraction f of heterozygous meioses:
E[P(transmit H)] = f·(1/4) + (1-f)·(1/2) = 1/2 − f/4
which is a transmission deficit of f/4 against the allele that is doing the work. Even f = 0.02 gives a deficit of 0.005 — comparable to a of 5 × 10⁻³, one to two orders of magnitude above the threshold 1/(2Nₑ) for humans.
And it acts fast. A deficit d applied in every heterozygote gives Δp = −2pqd, so the log-odds of the active allele fall linearly at 2d = 0.01 per generation, taking it from p = 0.5 to p = 0.01 in ln(99)/0.01 ≈ 460 generations. The active hotspot allele is removed on a timescale of roughly a thousand generations at f = 0.02, and tens of thousands only if the site fires in a far smaller fraction of meioses.
Step 7 — resolve the paradox. Hotspots cannot persist; only the mechanism for creating new ones can. That mechanism is PRDM9's rapidly mutating zinc-finger array, and the prediction it makes — that hotspot locations should turn over faster than species diverge — is confirmed by the near-total non-overlap of human and chimpanzee hotspot maps.
Connections
- Back to: Ch 09 ( and the synaptonemal complex); Ch 14 (the crossover as a measured recombination fraction — this chapter supplies its mechanism); Ch 17 ( as one of the DSB repair options); Ch 16 (the this chapter compares conversion against)
- Forward to: Ch 19 (transposition, and the repeats uses); Ch 20 (NAHR-driven genomic disorders and non-disjunction); Ch 20A (RecA-mediated strand invasion as an assay: every transferred fragment must recombine in to be inherited, which is what makes cotransduction and interrupted mating into rulers — plus λ site-specific recombination, a second mechanism that needs no at all); Ch 29 (hotspots are why LD has block structure); Ch 33 (gBGC as a confound in selection tests); Ch 35 (concerted evolution of gene families by conversion); Ch 35A §3 (§7's mouse hybrid sterility named as a and derived as one — why both must be derived and in different lineages, and why such incompatibilities accumulate with the square of divergence rather than in proportion to it); Ch 38 (HDR-based editing is this pathway, hijacked); Ch 56 (HR deficiency, BRCA1/2, )