DNA Replication, Recombination, Mutation and Repair

Each dividing cell copies billions of base pairs with fewer than one error per billion, and repairs thousands of DNA injuries every day. When copying, recombination or repair goes wrong, the result is a mutation, the raw material of evolution and the root of many inherited diseases and cancers. This chapter shows how the molecular machinery works and, just as important, how geneticists reasoned their way to knowing it.

Section 1Semiconservative replication and its enzymes

Three models and one decisive experiment

Watson and Crick noted in 1953 that base pairing suggests a copying mechanism. Semiconservative replication predicts that each daughter molecule keeps one old strand and gains one new strand. Two alternatives were also on the table. In conservative replication the parental duplex would stay intact and a wholly new duplex would be built beside it. In dispersive replication each strand of every daughter molecule would be a patchwork of old and new segments.

In 1958 Matthew Meselson and Franklin Stahl distinguished these models with a density label. They grew Escherichia coli for many generations in medium whose only nitrogen source was the heavy isotope ¹⁵N, so every DNA molecule was "heavy." They then moved the cells to medium containing ordinary ¹⁴N and took samples after each round of division. DNA from each sample was spun in a cesium chloride (CsCl) density gradient, where molecules settle into a band at the position matching their own buoyant density. Heavy DNA bands lower in the tube, light DNA higher, and a hybrid of one heavy and one light strand exactly in between.

After one generation in ¹⁴N, all the DNA formed a single band at the hybrid position. That result alone ruled out conservative replication, which would have given one heavy and one light band. After two generations they saw two bands of equal intensity, one hybrid and one light. A dispersive mechanism would have produced a single band that drifted steadily lighter, never a discrete light band, so the second generation ruled out dispersive replication. Only the semiconservative model predicts both observations.

Origins, forks and the direction of synthesis

Replication begins at specific sites called origins of replication. The circular E. coli chromosome, about 4.6 million base pairs (4.6 Mb), has a single origin named oriC, about 245 bp long. It contains repeated 9 bp binding sites for the initiator protein DnaA and a cluster of AT-rich 13 bp repeats. DnaA bound to ATP wraps the origin DNA and melts the AT-rich region, which opens easily because A:T pairs have only two hydrogen bonds. The helicase loader DnaC then places two hexameric DnaB helicases on the single strands, one facing each direction. Replication is therefore bidirectional: two replication forks move away from the origin around the circle until they meet roughly opposite, in a terminus region flanked by Ter sites bound by the Tus protein, which act as one-way traps for forks. Because the replicating circle looks like the Greek letter theta, this is called theta replication; some plasmids and phages instead use rolling circle replication.

Every DNA polymerase obeys two rules that shape the fork. First, it adds nucleotides only to a free 3'-OH, so synthesis always runs 5' to 3' and the template is read 3' to 5'. Second, no DNA polymerase can start a chain from nothing; it needs a primer. In cells the primer is a short stretch of RNA, about 10 nucleotides in bacteria, laid down by primase (DnaG in E. coli). Each dNTP is joined by its innermost phosphate, releasing pyrophosphate.

Because the two strands of the helix are antiparallel, these rules force an asymmetry. On one template the fork opens in the same direction that the polymerase travels, so a single primer is enough and the new strand, the leading strand, grows continuously toward the fork. On the other template the polymerase must move away from the fork. That strand, the lagging strand, is made discontinuously as a series of short pieces, each started by its own primer. These pieces are Okazaki fragments, about 1,000 to 2,000 nucleotides long in bacteria and only 100 to 200 nucleotides in eukaryotes.

The cast of enzymes at the fork

Unwinding the helix creates two problems. The separated strands want to snap back together, and the duplex ahead of the fork becomes overwound. Single-strand binding proteins (SSB) coat exposed single strands, keeping them extended and protected. DNA gyrase, a type II topoisomerase, works ahead of the fork, cutting both strands, passing a segment of duplex through the break and resealing it to introduce negative supercoils that relieve the strain. When replication of the circle is finished, the two daughter circles are interlocked like chain links, and topoisomerase IV separates (decatenates) them.

The main replicative enzyme in E. coli is DNA polymerase III holoenzyme, a multi-subunit machine. Its alpha subunit performs polymerization; its epsilon subunit is a 3' to 5' exonuclease that removes a mismatched nucleotide just added, a process called proofreading; and its beta clamp, a ring that encircles the DNA, holds the polymerase on the template so it can add thousands of nucleotides without falling off. This property, called processivity, lets pol III synthesize at about 1,000 nucleotides per second. A clamp loader complex opens the beta ring and closes it around each primer. One holoenzyme replicates both strands, the lagging template looping out so both cores move together (the trombone model).

DNA polymerase I finishes the lagging strand. It has a 5' to 3' exonuclease activity that removes the RNA primer of the fragment ahead while its polymerase activity fills the gap with DNA, a combined process called nick translation. Finally, DNA ligase seals the remaining nick by forming the last phosphodiester bond. Bacterial ligase uses NAD⁺; eukaryotic ligases use ATP.

Fidelity: three layers of accuracy

Replication is astonishingly accurate, and the accuracy comes in layers. Base selection by the polymerase active site, which fits only a correctly shaped Watson and Crick pair, gives roughly one error per 10⁴ to 10⁵ nucleotides. Proofreading by the 3' to 5' exonuclease improves this about 100-fold, to roughly 10^-7. Mismatch repair (Section 5) catches most of the remaining errors after the fork has passed, bringing the final rate to about one error per 10⁹ to 10¹⁰ nucleotides replicated.

Section 2Replication in eukaryotes and telomeres

Many origins, licensed once per cycle

Eukaryotic chromosomes are linear, far larger than bacterial chromosomes, packaged in chromatin, and replicated by forks that move more slowly, roughly 50 nucleotides per second compared with about 1,000 in E. coli. A single origin per chromosome would never finish in time, so each chromosome uses many origins. The stretch of DNA replicated from one origin is a replicon. Yeast origins are short defined sequences called autonomously replicating sequences (ARS); human origins lack a simple consensus, and tens of thousands fire in each S phase.

A genome with thousands of origins faces a bookkeeping problem: every segment must be copied exactly once per cell cycle. Over-replication creates extra copies of genes and broken chromosomes. Cells solve this by separating two steps in time. During G1, when cyclin-dependent kinase (CDK) activity is low, the origin recognition complex (ORC) binds origins and, with the helpers Cdc6 and Cdt1, loads inactive double rings of the MCM2 to 7 helicase. This loaded pre-replicative complex marks the origin as licensed. When S phase begins, rising CDK and the kinase DDK (Dbf4-dependent kinase) activate licensed origins by recruiting Cdc45 and the GINS complex, which together with MCM form the active CMG helicase. The same high CDK activity that fires origins also prevents new licensing (Cdc6 is degraded or exported and Cdt1 is inhibited by the protein geminin), so no origin can be loaded again until the cell passes through mitosis and returns to G1. Active euchromatin generally replicates early in S phase and heterochromatin late.

Eukaryotic polymerases and lagging strand maturation

Eukaryotes divide the work among several polymerases. Pol α-primase starts every chain: its primase subunit makes about 10 nucleotides of RNA and pol α then adds about 20 nucleotides of DNA. Pol α lacks proofreading and has low processivity, so it is quickly replaced. Pol ε synthesizes most of the leading strand and pol δ most of the lagging strand; both proofread. The ring-shaped clamp PCNA (proliferating cell nuclear antigen), loaded by RFC, plays the role of the bacterial beta clamp. Because eukaryotic polymerases lack the 5' to 3' exonuclease of bacterial pol I, primers are removed differently: pol δ runs into the previous fragment and displaces its 5' end as a single-stranded flap, which the flap endonuclease FEN1 cuts off (with help from RNase H, which degrades RNA in RNA:DNA hybrids). DNA ligase I then seals the nick. Mitochondrial DNA is copied by pol γ.

Chromatin must also be copied. Nucleosomes ahead of the fork are disassembled and parental histones are passed to both daughter duplexes, while new histones are deposited by the chaperone CAF-1, which is recruited to PCNA. Parental histones keep their chemical marks, which helps pass on gene activity patterns (Chapter 9).

The end replication problem

Linear chromosomes have a weakness that circular ones do not. At the very end of a chromosome, the lagging strand template has no place for a final primer that can be replaced with DNA: when the last RNA primer is removed, nothing upstream can supply a 3'-OH to fill the gap. In addition, nucleases process the ends to leave a single-stranded overhang. The result is that each daughter molecule would be a little shorter than its parent, and after enough divisions genes near the ends would be lost. This is the end replication problem.

Chromosome ends are protected by telomeres, long arrays of a short repeat. In humans and all other vertebrates the repeat is 5'-TTAGGG-3' on the strand running 5' to 3' toward the chromosome end, repeated to a total length of roughly 10 to 15 kb at birth. The G-rich strand ends in a 3' single-stranded overhang of about 50 to 300 nucleotides. That overhang tucks back and invades the double-stranded repeats to form a protective lasso called the t-loop, held in place by a six-protein complex called shelterin (including TRF1, TRF2 and POT1). The t-loop hides the end so the cell does not mistake it for a broken chromosome and try to repair it by joining chromosomes together.

Telomerase

The enzyme that restores telomere length is telomerase, discovered by Elizabeth Blackburn and Carol Greider in the ciliate Tetrahymena. Telomerase is a ribonucleoprotein with two essential parts: a protein subunit, TERT (telomerase reverse transcriptase), and an RNA, TERC (also called TR), which carries a short template complementary to the telomere repeat. In humans the template region reads 3'-CAAUCCCAAUC-5'. Telomerase binds the 3' overhang so that the end of the DNA pairs with the 3' part of the template, then TERT copies the rest of the template, adding one GGTTAG unit (completing a TTAGGG repeat). It then slides forward and repeats. Telomerase is thus a reverse transcriptase carrying its own template. Once the G-rich strand has been lengthened, pol α-primase fills in the complementary C-rich strand by ordinary lagging strand synthesis.

Telomeres, aging and cancer

Most human somatic cells express little or no TERT, so their telomeres shorten by roughly 50 to 200 bp with each division. When telomeres become critically short, the exposed ends trigger a DNA damage response and the cells enter permanent growth arrest, called replicative senescence. This is the molecular basis of the Hayflick limit, the observation that normal human fibroblasts divide only about 40 to 60 times in culture. Germ cells and many stem cells do express telomerase. About 85 to 90 percent of human cancers reactivate telomerase, often through mutations in the TERT promoter, and most of the rest maintain telomeres by recombination between telomeres, a pathway called alternative lengthening of telomeres (ALT). Inherited telomerase defects cause dyskeratosis congenita, with bone marrow failure.

Section 3Homologous recombination

What recombination is for

Homologous recombination is the exchange of genetic information between two DNA molecules that share a long stretch of identical or nearly identical sequence. You already know one of its products: the crossovers between homologous chromosomes in meiosis that reshuffle alleles and that geneticists use to build linkage maps (Chapter 4). Its most ancient job, though, is repair. A double-strand break severs a chromosome completely, and the only way to restore the lost or damaged sequence without error is to copy it from an intact homologous template, usually the sister chromatid after replication or the homolog in meiosis. It also restarts collapsed forks and, in bacteria, integrates DNA brought in by transformation, transduction and conjugation (Chapter 5).

Homologous recombination should be distinguished from two other kinds of DNA rearrangement. Site-specific recombination joins two short, specific sequences recognized by a dedicated enzyme (phage lambda integration, or Cre recombinase acting on loxP sites). Transposition moves a mobile element into a new site that needs no homology at all (Section 5). Only homologous recombination depends on base pairing between long matching sequences.

The Holliday model and the Holliday junction

In 1964 Robin Holliday proposed that single-strand nicks at the same position in two homologous duplexes let the nicked strands swap partners, each pairing with the complementary strand of the other duplex. The result is a four-armed, cross-shaped structure now called a Holliday junction. Wherever a strand from one parent pairs with a strand from the other, the duplex is a heteroduplex; if the two parents differ at a site in that region, the heteroduplex contains a mismatch. The junction can slide along the DNA, a process called branch migration, lengthening the heteroduplex. Finally, the junction is cut by a resolvase. A Holliday junction can be cut in either of two planes. Cutting the two strands that were exchanged restores the original partners at flanking markers, a noncrossover that leaves only a patch of heteroduplex. Cutting the two strands that were not exchanged joins the left part of one chromosome to the right part of the other, a crossover.

The double-strand break repair model

Yeast experiments later showed that meiotic recombination begins with a double-strand break and that the broken chromosome receives information. The resulting double-strand break repair (DSBR) model is now the textbook framework. Its steps are:

1. Break. In meiosis, the topoisomerase-like protein Spo11 cuts both strands deliberately at hundreds of sites; in mitotic cells breaks come from radiation, chemicals or collapsed forks.
2. Resection. Nucleases (the MRN complex with CtIP, then Exo1 and Dna2 in eukaryotes) degrade the 5' ends on both sides of the break, leaving long 3' single-stranded tails. RPA coats the tails at first.
3. Strand invasion. The recombinase Rad51 (and in meiosis also its relative Dmc1) replaces RPA, forming a helical nucleoprotein filament on a 3' tail. Loading Rad51 requires the mediator protein BRCA2. The filament finds a matching sequence and invades it, displacing one strand as a D-loop.
4. Synthesis. A DNA polymerase extends the invading 3' end, using the homolog as template. This copies any sequence lost at the break.
5. Second end capture. The enlarged D-loop pairs with the other 3' tail, which is also extended. Ligation of the ends produces two Holliday junctions, a double Holliday junction.
6. Resolution. The two junctions are cut by resolvases or, in somatic cells, dissolved by a helicase and topoisomerase complex. If the two junctions are cut in the same orientation the outcome is a noncrossover; if they are cut in opposite orientations the outcome is a crossover.

Many breaks never form a double Holliday junction. In synthesis-dependent strand annealing (SDSA), the invading strand is extended, released from the D-loop and annealed back to the other tail on its own chromosome. SDSA always gives a noncrossover, which is why most double-strand breaks in mitotic cells are repaired without exchanging chromosome arms. This matters because a mitotic crossover can cause loss of heterozygosity, unmasking a recessive tumor suppressor mutation (Chapter 13).

The bacterial machinery

Much of what we know about recombination enzymes was worked out in E. coli. The RecBCD complex binds a double-stranded end and travels along it, unwinding and degrading DNA, until it meets a short sequence called Chi (5'-GCTGGTGG-3'). At Chi, RecBCD stops degrading the 3' strand and loads RecA onto it. RecA is the bacterial counterpart of Rad51: it polymerizes on single-stranded DNA, searches for homology and catalyzes strand invasion. The RuvA and RuvB proteins bind Holliday junctions and drive branch migration, and RuvC resolves them by symmetric nicks.

Gene conversion

Mendel's first law predicts that a heterozygote A/a produces meiotic products in a 2:2 ratio of A to a. In fungi such as Neurospora and Sordaria, where all products of a single meiosis stay together in an ascus, and where a postmeiotic mitosis turns the four products into eight spores (an octad), the normal ratio is 4:4. Occasionally, however, an ascus shows 6:2 or 2:6, meaning one allele was converted into the other, or 5:3 and 3:5, meaning one spore pair contains both alleles. This nonreciprocal transfer of information is gene conversion.

Recombination explains both kinds of aberrant ascus. When a heteroduplex forms across a site where the parents differ, the two strands of that region carry different alleles and form a mismatch. If mismatch repair corrects the mismatch, it may restore the original allele (giving a normal 4:4) or convert it to match the invading strand (giving 6:2 or 2:6). If the mismatch escapes repair, the two strands separate at the next DNA replication, the postmeiotic mitosis, and one spore pair becomes a non-identical pair, giving 5:3 or 3:5, an event called postmeiotic segregation. Gap repair in the DSBR model also converts: sequence lost by resection at the break is replaced by copying from the homolog.

Section 4Mutation types, causes and rates

Kinds of mutation

A mutation is any heritable change in DNA sequence. Mutations in germ-line cells can be passed to offspring; somatic mutations arise in body cells and are passed only to that cell's descendants, as in cancer. Here we treat changes of one or a few nucleotides; larger changes were covered in Chapter 6.

The simplest gene mutation is a base substitution. A transition replaces a purine with the other purine or a pyrimidine with the other pyrimidine (A:T to G:C or G:C to A:T). A transversion replaces a purine with a pyrimidine or the reverse (for example G:C to T:A). Although there are twice as many possible transversions, transitions are more common, because deamination and tautomeric mispairing produce transitions. Insertions and deletions (indels) add or remove nucleotides. If the number added or removed is not a multiple of three in a coding region, the reading frame shifts and every downstream codon is misread; this is a frameshift mutation, and it nearly always destroys protein function, usually by reaching a premature stop codon soon after the shift.

Mutations are also classified by their effect on the protein:

• A synonymous (silent) mutation changes a codon to another codon for the same amino acid, most often at the third position.
• A missense mutation changes one amino acid to another. A conservative missense change swaps similar amino acids (leucine to isoleucine) and may be harmless; a nonconservative change swaps dissimilar ones and is more likely to disrupt structure.
• A nonsense mutation changes an amino acid codon into a stop codon, truncating the protein. Messenger RNAs with premature stop codons are often destroyed by nonsense-mediated decay (Chapter 8).
• A splice-site mutation alters the signals for intron removal, often causing exon skipping or intron retention.
• A regulatory mutation in a promoter or enhancer changes how much protein is made rather than what kind.

A forward mutation changes the wild type to a mutant; a reverse mutation (reversion) restores the wild-type phenotype, either by restoring the original sequence or by a second change. A suppressor mutation is a second mutation at a different site that hides the effect of the first; it may be intragenic (a compensating change in the same gene, such as a +1 frameshift near a −1 frameshift) or intergenic (for example, a mutant tRNA that reads a stop codon as an amino acid). Loss-of-function mutations are usually recessive; gain-of-function mutations, which give a protein a new or excessive activity, are usually dominant.

Expanding repeats

In trinucleotide repeat expansions, a run of a three-base repeat becomes unstable past a threshold length and tends to grow in each generation. In Huntington disease, a CAG repeat encoding polyglutamine in the huntingtin gene causes disease at 40 or more copies (36 to 39 copies give incomplete penetrance). In fragile X syndrome, a CGG repeat in the 5' untranslated region of FMR1 is normal below about 45 copies, a premutation from about 55 to 200, and a full mutation above 200, at which point the gene is methylated and silenced. Myotonic dystrophy type 1 involves a CTG repeat in a 3' untranslated region. Because longer repeats are more unstable and cause earlier, more severe disease, these disorders show anticipation: symptoms appear earlier or more severely in successive generations.

Spontaneous mutation

Many mutations arise without any outside agent:

• Replication errors. Despite proofreading, polymerases occasionally insert the wrong base. Rare tautomeric shifts (the amino to imino form of adenine or cytosine, the keto to enol form of guanine or thymine) and wobble pairing allow mispairs such as G:T and A:C, which become transitions after the next round of replication.
• Strand slippage. In runs of a repeated base or short repeat, the new or template strand can loop out, producing small insertions or deletions. These create mutational hot spots in repetitive sequences.
• Depurination. The bond joining a purine base to deoxyribose breaks spontaneously, roughly 10,000 times per human cell per day, leaving an apurinic (AP) site with no base to guide replication.
• Deamination. Cytosine loses its amino group to become uracil, which pairs with adenine; if unrepaired, G:C becomes A:T. Cells recognize uracil as foreign in DNA and remove it efficiently. But deamination of 5-methylcytosine yields thymine, a normal DNA base, so the resulting G:T mismatch is repaired less reliably. That is why methylated CpG sites are the strongest mutational hot spots in mammalian genomes, mutating 10 to 50 times faster than other sites.
• Oxidative damage. Reactive oxygen species from normal metabolism convert guanine to 8-oxoguanine, which pairs with adenine and so causes G:C to T:A transversions.

Induced mutation and mutagens

Mutagens raise the mutation rate, and knowing how each acts predicts the kind of change it causes.

The Ames test, developed by Bruce Ames in the 1970s, screens chemicals for mutagenic potential. It uses strains of Salmonella carrying mutations in a histidine synthesis gene (his⁻), so they cannot grow without added histidine. Tester strains carry substitution or frameshift alleles and lack excision repair. Because many compounds become mutagenic only after liver metabolism, the chemical is mixed with a liver extract (S9 fraction). The bacteria are plated on medium without histidine, and only cells in which a new mutation has reverted the his⁻ allele can form colonies. More revertant colonies than on a control plate indicates a mutagen, and most chemical carcinogens score as mutagens.

Mutations are random: the fluctuation test

Do mutations arise at random, or are they induced by the selective condition that reveals them? In 1943 Salvador Luria and Max Delbrück addressed this with bacteria and phage T1. They grew many small, separate cultures of E. coli from a few cells each, then plated each entire culture on phage-coated plates and counted resistant colonies. If resistance were induced by contact with phage, every culture would give a similar small number, with a variance about equal to the mean (a Poisson distribution). Instead, the counts fluctuated wildly: most cultures had zero or a few resistant cells, while a few had hundreds. This "jackpot" pattern is expected if mutations occur randomly during growth: an early mutation leaves many resistant descendants, a late one only a few. Joshua and Esther Lederberg confirmed this with replica plating, showing resistant colonies existed before exposure to phage.

The fluctuation test also gives a way to measure the mutation rate, the probability of a mutation per cell per division (or per gene per generation). The simplest version, the P₀ method, uses the fraction of cultures with no mutants. If mutations occur at random, the number of mutational events per culture follows a Poisson distribution with mean m, so the fraction of cultures with no mutational events is P₀ = e^−m, and m = −ln P₀. Dividing m by the number of cell divisions in a culture (approximately the final number of cells) gives the rate.

How often mutations happen

Mutation rates are usually quoted at two scales. Per gene, forward mutation rates typically fall between 10^-5 and 10^-7 per generation. Per nucleotide, the human germ-line rate is about 1.2 × 10^-8 per base pair per generation. With a diploid genome of about 6.2 × 10⁹ bp, that predicts 1.2 × 10^-8 × 6.2 × 10⁹ = about 74 new point mutations in each child, in line with the 60 to 75 measured by sequencing parents and children. Most arise in the father, whose sperm precursors keep dividing throughout life. RNA viruses, whose polymerases lack proofreading, mutate on the order of 10^-4 per nucleotide per replication, which is why influenza and HIV evolve so rapidly.

Section 5DNA repair systems and transposable elements

Direct reversal

The simplest repair undoes the damage chemically without removing anything. Photolyase uses the energy of visible light to split cyclobutane pyrimidine dimers back into separate bases, a process called photoreactivation; it works in bacteria, fungi, plants and many animals but is absent from placental mammals. O⁶-methylguanine DNA methyltransferase (MGMT) transfers the methyl group from O⁶-methylguanine onto one of its own cysteines. The enzyme is inactivated by the reaction, so it is called a suicide enzyme.

Excision repair

Base excision repair (BER) handles small lesions that do not distort the helix: uracil, 8-oxoguanine, alkylated and deaminated bases. A DNA glycosylase specific for one kind of damage flips the base out and cuts the bond to its sugar, leaving an AP site. AP endonuclease nicks the backbone, DNA polymerase (pol β in mammals) inserts the correct nucleotide, and ligase seals the strand. Using thymine rather than uracil in DNA lets cells treat any uracil as deaminated cytosine.

Nucleotide excision repair (NER) handles bulky lesions that distort the helix, chiefly UV dimers and chemical adducts. In E. coli, UvrA and UvrB find the distortion, UvrC cuts the damaged strand on both sides, and UvrD helicase removes an oligonucleotide of 12 to 13 nucleotides; pol I and ligase fill the gap. In humans the same logic uses about 30 proteins, including the XP proteins and the helicases of TFIIH, and removes 24 to 32 nucleotides. Transcription-coupled NER repairs lesions that stall RNA polymerase faster than global genome NER.

Mismatch repair

Mismatch repair corrects replication errors that escaped proofreading. The key problem is deciding which strand is wrong. In E. coli the answer comes from methylation: the Dam methylase adds methyl groups to the A of every GATC sequence, but only after a delay, so just behind the fork the new strand is briefly unmethylated. MutS recognizes the mismatch, MutL links it to MutH, and MutH nicks the unmethylated (new) strand at a nearby GATC. An exonuclease removes the new strand past the mismatch and pol III resynthesizes it. Eukaryotes use MutS and MutL homologs (MSH2, MLH1 and others) and recognize the new strand by its nicks.

Double-strand breaks and damage tolerance

Double-strand breaks are repaired by two main routes. Homologous recombination (Section 3) is accurate but needs a sister chromatid, so it operates in S and G2. Nonhomologous end joining (NHEJ) works throughout the cycle: the Ku70/Ku80 heterodimer binds the broken ends, recruits DNA-PKcs, and the ends are trimmed and joined by DNA ligase IV. NHEJ often adds or loses a few nucleotides, so it is error prone. Lesions that block replication can also be bypassed rather than repaired. In E. coli, single-stranded DNA at stalled forks activates RecA, which stimulates self-cleavage of the LexA repressor and induces more than 40 genes of the SOS response, including the error-prone translesion polymerases pol IV and pol V. Humans have several translesion polymerases; pol η correctly inserts two A's opposite a thymine dimer.

Transposable elements

Transposable elements are DNA sequences that move from one place in the genome to another. Barbara McClintock discovered them in maize in the 1940s, when the Ds (Dissociation) element moved only in the presence of Ac (Activator), which supplies the enzyme; Ds is a defective copy. Two main classes exist.

• Class II, DNA transposons, move as DNA. Each carries short terminal inverted repeats recognized by the transposase it encodes. Bacterial insertion sequences (IS elements), 0.7 to 2.5 kb, carry only a transposase gene. Composite transposons such as Tn10 are two IS elements flanking other genes, often antibiotic resistance genes. Most DNA transposons move by cut-and-paste (conservative) transposition; others, such as Tn3, use replicative transposition, leaving a copy behind.
• Class I, retrotransposons, move through an RNA intermediate that is reverse transcribed, so they copy and paste. LTR retrotransposons (yeast Ty1) resemble retroviruses. LINEs such as human L1 (about 6 kb, roughly 17 percent of the human genome) encode their own reverse transcriptase. SINEs such as Alu (about 300 bp, over one million copies, about 10 percent of the genome) encode nothing and borrow L1 enzymes.

Almost every insertion creates a target site duplication: the enzyme makes staggered cuts in the target, the element is joined to the overhangs, and repair synthesis copies the single-stranded regions, leaving a short direct repeat (typically 2 to 12 bp) on each side. Nearly half of the human genome derives from transposable elements, although only a few dozen L1 copies remain active.

Transposons mutate genes by inserting into coding sequence or regulatory regions, and they drive larger rearrangements when recombination occurs between copies. Mendel's wrinkled pea allele is caused by a transposon-like insertion in a starch branching enzyme gene, and new L1 insertions into the factor VIII gene have caused hemophilia A.

Words worth knowing