Contents
19 — Transposable elements Part 3 — Genome instability

19 — Transposable elements

Before this: Ch 16 · Ch 18 · Ch 03 · Time: ~35 min

What you'll be able to do

The core idea

Roughly 46% of the human genome is derived from (verified-facts, ). That is not a fact about human biology. It is a fact about what happens when self-replicating sequences get access to a .

A transposable element encodes the means of making more copies of itself within the genome it already occupies. It does not need to help you. It needs only to raise its own copy number faster than the host removes it. Selection acts at two levels at once — on the element, favouring replication; on the organism, favouring suppression — and your genome is the standing equilibrium of that conflict.

A transposable element is not a feature of your genome. It is a lineage of self-replicating sequence resident in your ancestors' germline for tens of millions of years, whose interests and yours coincide only by accident. Almost every fact below — the copy numbers, the elaborate silencing machinery, the handful of domesticated genes — follows from taking that sentence literally.


1. Two ways to move

The primary split is the intermediate: RNA or DNA. Copy-and-paste versus cut-and-paste — and that is not an analogy, it is the standard vocabulary.

graph TD TE["<b>Transposable element</b>"] TE --> C1["<b>Class I — retrotransposon</b><br/>RNA intermediate<br/><i>copy-and-paste</i>: original stays"] TE --> C2["<b>Class II — DNA transposon</b><br/>DNA intermediate<br/><i>cut-and-paste</i>: original excised"] C1 --> LTR["<b>LTR / endogenous retrovirus</b><br/>reverse transcription in a<br/>cytoplasmic particle, then integration"] C1 --> NL["<b>non-LTR</b><br/>reverse transcription happens<br/><i>at</i> the insertion site"] NL --> LINE["<b>LINE</b> — autonomous<br/>encodes its own machinery"] NL --> SINE["<b>SINE</b> — non-autonomous<br/>encodes nothing; borrows"] C2 --> DNAT["hAT, Tc1/<i>mariner</i>, piggyBac…<br/>transposase + inverted repeats"]

The asymmetry matters. Class II transposition is copy-number-neutral by default — the element leaves one site and arrives at another. Class I transposition is strictly additive: the source is transcribed, never excised, so every success adds one copy. Given tens of millions of years and no ceiling other than host defence, an unbounded copying process is exactly what you would expect to dominate. It does.

An element is autonomous if it encodes the proteins that move it. Non-autonomous elements are parasites of parasites: they supply only a substrate the autonomous machinery will accept. As with any shared library, all that matters is matching the calling convention.

2. The human catalogue

ClassFamilyCopiesGenome shareAutonomous?
I, non-LTR (L1)~500,000~17%yes — the only one in humans
I, non-LTRAlu (SINE)~1.1 million~11%no
I, non-LTRSVA~2,700<0.2%no
I, non-LTRolder LINEs (L2), older SINEs (MIR)—~6%no (all dead)
I, LTRERVs and solo LTRs—~8%no (all dead in humans)
IIDNA transposons (hAT, Tc1/mariner…)—~3%no (none active ~37 Myr)

Those sum to ~45% on GRCh38; , which resolved the and acrocentric short arms, gives ~46%. Treat it as a lower bound — a copy that landed 200 million years ago has been decaying by point ever since and eventually stops being recognisable. The measured fraction is a function of detection sensitivity and has risen every time methods improved.

Only L1 is autonomous. Every new insertion in a living human — Alu, SVA, processed , or L1 itself — is catalysed by L1 protein. One family holds the engine for all of them, which is why silencing L1 is the load-bearing job in §6.

ERVs are fossils of infection. A retrovirus infecting a somatic cell is an evolutionary dead end; one that integrates in a germ cell becomes a heritable locus. Most human ERVs are now solo LTRs — the internal gag/pol/env was deleted by recombination between the element's own two LTRs, leaving one behind. That leftover LTR is still a , which matters in §7.

Human DNA transposons are extinct, last active ~37 million years ago. That is a fact about humans, not about DNA transposons — they are busy in maize, Drosophila and bacteria, and resurrected versions (Sleeping Beauty, piggyBac) are standard lab tools (Ch 37).

3. How L1 copies itself

A full-length L1 is ~6 kb: a 5' UTR carrying an internal II , then two , then a 3' poly(A) signal.

Both bind their own transcript preferentially — cis preference — which is why L1 copies itself far more readily than anything else, and why Alu has to be opportunistic.

The mechanism is target-primed reverse (TPRT). Follow it in detail, because every diagnostic signature falls out of it.

STEP 1  ORF2p endonuclease nicks one strand at a degenerate 5'-TTTT/AA-3' consensus

        5'- G C A T T T T ¦ A A G C T -3'      <-- nick, TOP strand as drawn
        3'- C G T A A A A   T T C G A -5'
                        ^  free 3'-OH, at the end of a run of T

STEP 2  the L1 mRNA's poly(A) tail base-pairs with that exposed T-rich strand

        5'- G C A T T T T-OH
                    | | | |
                3'-(A A A A ... L1 mRNA ... 5')

STEP 3  reverse transcriptase extends the DNA 3'-OH along the RNA template.
        The RNA is read 3'->5', so the cDNA is built from the element's
        3' END backwards toward its 5' end.

STEP 4  staggered second nick on the BOTTOM strand, 7–20 bp away;
        second-strand synthesis; ligation

        5'- G C A T T T T  A A G C T N N [(A)n══ new L1 ══]  A A G C T N N -3'
                           ^^^^^^^^^^^^^                     ^^^^^^^^^^^^^
                           target site duplication — the bases lying between
                           the two nicks, now on both flanks: identical, direct

        Both nicks on one strand would duplicate nothing. It is the stagger
        across the two strands that leaves single-stranded target sequence at
        each end, and filling that in is what writes the repeat twice — which
        is why the nick spacing IS the TSD length. The insert reads 3'-end
        first here: its poly(A) sits on the left, where the T-tract primed it.
SignatureWhy the mechanism produces it
Flanking direct repeats (TSDs), 7–20 bpThe two nicks are staggered; filling in the intervening bases
Poly(A) tail at the insert's 3' endIt was the primer
5' truncation, almost alwaysSynthesis starts at the 3' end, so any interruption leaves the 5' end unmade. The direction of failure is fixed by the chemistry
3' transductionsL1's own poly(A) signal is weak, so Pol II reads through into flanking DNA and carries it along — a lineage tag identifying which source element fired

The same machinery acting in trans on an ordinary mRNA produces a processed : an intronless genomic copy with a poly(A) tail and TSDs. GENCODE 50 annotates 14,702 pseudogenes; many arose exactly this way (Ch 35).

Alu is ~300 bp, derived from 7SL RNA (the RNA of the signal recognition particle), transcribed by RNA polymerase III from an internal promoter, and encodes nothing. Its whole strategy is to be an abundant poly(A)-tailed transcript sitting near translating — where ORF2p is newly made — and capture it before cis preference does. SVA is a hominid-specific composite (hexamer repeat + Alu-like region + VNTR + a HERV-K fragment), likewise L1-dependent.

Alu outnumbers L1 two to one with no engine of its own: short, cheap, abundantly transcribed, and individually less deleterious to insert. A payload optimised for the library it hijacks.

4. Why the graveyard is so large

Half a million L1 copies; roughly 80–100 retrotransposition-competent in a typical human genome, with about six "hot" elements accounting for most measured activity. Two mechanical causes.

5' truncation. TPRT builds the copy 3'-end-first, so an incomplete reaction deletes the 5' UTR — which contains the . A truncated L1 cannot be transcribed, and what cannot be transcribed cannot copy itself again. The failure mode destroys precisely the part needed for the next round. Only a few thousand of the ~500,000 copies are full length.

Mutational decay. From the moment it lands, a copy is ordinary neutral sequence accumulating substitutions at ~1.1–1.3 × 10⁻⁸ per bp per generation (Ch 16). Two intact ORFs are required; one nonsense or frameshift in either kills the element, and nothing selects against that. A copy that landed 10 million years ago has had ~400,000 generations to break.

A TE family therefore behaves like a branching process in which most lineages are sterile at birth and the rest sterilise stochastically. The 46% is overwhelmingly a fossil record, not an active threat — but the active fraction is not zero, and that is where the disease is.

5. McClintock, and why the field would not have it

Barbara McClintock worked out transposition in maize in the 1940s from cytogenetics and breeding alone, with no molecular biology at all. She identified Ds (Dissociation), which broke the at a specific site, and Ac (Activator), required for Ds to do anything — both of which changed position between generations. Ac encodes a transposase; Ds elements are internally deleted Ac derivatives retaining the ends but not the enzyme. Autonomous and non-autonomous, deduced from kernel colour. She published it in 1950.

The reception was poor, and the usual explanation — that she was dismissed for being a woman — is too simple to be useful. She was elected to the National Academy of Sciences in 1944 (the third woman ever) and became the first woman president of the Genetics Society of America in 1945, both before the transposition work, and was securely funded throughout. The resistance was substantive:

Vindication came from bacterial insertion sequences in the 1960s–70s and then from cloning, which made the elements directly visible. The Nobel Prize followed in 1983, unshared. The lesson is not that scientists are prejudiced; it is that an observation invalidating a field's core methodological assumption is extraordinarily expensive to accept, and will be resisted until it can be shown in a system the field already trusts.

6. Three ways TEs cause disease

Around 100–125 insertions causing single-gene disease are documented — Alu most numerous, then L1, then SVA. Certainly an undercount, because standard exome pipelines are close to blind to them (see the worked example).

(a) Insertional inactivation. In 1988 Kazazian and colleagues screened 240 unrelated haemophilia A patients and found, in two, a de novo L1 insertion into exon 14 of F8 — 3.8 kb in one patient, 2.3 kb in the other, both 5'-truncated and both ending in a poly(A) tract — absent from both parents and traceable to a full-length source L1 elsewhere. Since then: Alu into NF1 (neurofibromatosis), Alu into BTK ( agammaglobulinaemia), SVA into FKTN (Fukuyama muscular dystrophy, a Japanese founder ). Intronic insertions can be as damaging as exonic ones, because L1 sequence carries splice sites and premature poly(A) signals.

(b) Non-allelic . Quieter, quantitatively larger, and requiring no element to be active. A million Alu copies at ~11% divergence are a million dispersed substrates for homologous recombination (Ch 18).

    correct alignment                  misalignment between dispersed Alus
    ═Alu1═[ exons 3-8 ]═Alu2═          ═Alu1═[ exons 3-8 ]═Alu2═
    ═Alu1═[ exons 3-8 ]═Alu2═               ╲__________________╱
                                        ═Alu1═════════════════Alu2═
    crossover -> no change              crossover -> exons 3-8 deleted

Alu–Alu recombination causes deletions in LDLR (familial hypercholesterolaemia), MSH2 (Lynch syndrome), BRCA1 and VWF. The repeat is not a mutagen; it is a latent ambiguity in the alignment that recombination occasionally resolves the wrong way — the biological version of a diff over a file full of near-identical blocks.

(c) Exonisation. An intronic antisense Alu already supplies half an exon: read in the direction of its poly(A) appears as a poly(T) run, which the accepts as a polypyrimidine tract, so a usable cryptic 3' acceptor is already there. What is missing is the other end, and one point can supply it — in OAT, a G→C transversion created a new 5' donor site inside the element, which activated that waiting acceptor and spliced a 142-nt Alu-derived "exon" that never existed into the mRNA, causing gyrate atrophy. Roughly 5% of alternatively spliced human exons are Alu-derived, which makes exonisation one mechanism seen from two directions: a disease mechanism when it wrecks a gene, a source of new coding sequence when it does not.

7. Host defence as an arms race

insertions are permanent and heritable; somatic ones affect one lineage. The heavy machinery is accordingly germline-focused, and layered.

SystemMechanismProgrammer's view
piRNA24–31 nt PIWI-associated small RNAs transcribed from piRNA clusters — regions dense in broken TE fragments — guide cleavage of TE transcripts and self-amplify by "ping-pong"A signature database built by capturing the malware itself
KRAB zinc fingers~350 KZFPs, the largest human TF family, mostly clustered on 19; each recognises a TE family and recruits TRIM28/KAP1 → SETDB1 → H3K9me3 → Per-family signature matching, hard-coded and under selection
CpG methylation of the L1 5' UTR shuts its offPersistent config, copied through cell division
APOBEC3Seven cytidine deaminases clustered on chromosome 22 restrict L1 and AluCorrupting the payload in transit

KZFPs are the clearest Red Queen dynamic: a KZFP recognises a TE sequence, descendants that escape recognition replicate more, the KZFP locus and its DNA-contacting residues evolve rapidly to re-cover the escapees. KZFP clusters expand and turn over at a rate tracking the arrival of new TE families.

The failure mode of any signature-based defence is a family with no signature. In Drosophila, crossing a female from a strain lacking P elements to a male carrying them yields offspring whose maternally supplied piRNA pool matches nothing in P. P transposes freely, causing sterility and chromosome breakage. Hybrid dysgenesis is a defence system missing a database entry.

Two windows open the door in mammals: germ cell development and the early embryo, both of which globally erase and rewrite DNA methylation (Ch 23). L1 expression rises in exactly those windows, which is where new germline insertions come from. The other place silencing collapses is tumours (Ch 56).

8. Domestication

Occasionally the host stops fighting a sequence and starts using it — exaptation.

Syncytin. The placenta requires trophoblast cells to fuse into a continuous multinucleate layer. The fusion is performed by syncytin-1, encoded by ERVW-1, which is the env gene of a HERV-W — the protein a retrovirus uses to fuse with a target cell. A second, syncytin-2 (ERVFRD-1), comes from a different ERV; mice use syncytin-A and -B captured from yet others. The mammalian placenta has been re-engineered from viral fusion machinery independently in at least six lineages.

RAG1/RAG2. V(D)J recombination — the cut-and-paste that generates antibody diversity — is run by a domesticated transposase from a RAG-like (Transib-related) DNA transposon, and the recombination signal sequences it recognises are the transposon's terminal inverted repeats. The clincher is ProtoRAG in the lancelet: an intact ancestral element with convergently oriented RAG1-like and RAG2-like genes between TIRs, still capable of excision and transposition. Your adaptive immune system is a cut-and-paste transposon pointed at a single locus.

Regulatory rewiring is the largest effect by volume. A TE brings its own and -factor binding sites, so one family can install the same regulatory input at thousands of loci at once — a genome-wide find-and-replace no point could achieve. The best-characterised case: the MER41 ERV family carries STAT1 sites, and nearly a thousand MER41 copies in the human genome are STAT1-bound. One of them is the only STAT1 site within 50 kb of AIM2; delete it with and the cell can no longer induce AIM2 in response to interferon-γ. MER41-like elements with STAT1 sites colonised bats, carnivores and artiodactyls independently.

9. Parasite, or raw material?

Both, and the tension is the substance.

The parasite reading is correct mechanistically: TEs replicate at the host's expense, cause disease, and provoke an arms race that has consumed enormous evolutionary effort. The raw material reading is correct evolutionarily: TEs are the largest single source of new sequence, they install regulatory modules in parallel across thousands of loci, they supply the repeats that make duplication by possible, and they have been domesticated into the placenta and the immune system.

These are claims about different timescales. Over one generation a TE is a mutagen; over ten million generations, a mutagen with a distinctive, non-random, modular signature is a source of structured variation. Selection has no foresight (Ch 00) — nothing preserved a TE for future usefulness. An enormous number of insertions occurred, almost all neutral or bad, and the vanishing fraction that happened to be useful were retained afterwards by ordinary selection.

Which is why "" fails in both directions. Calling the 46% junk asserts that inert today means inert forever and ignores the domesticated fraction. Calling it all functional confuses being transcribed or bound with doing something the organism needs. The honest statement is uncomfortable and precise: most TE-derived sequence does nothing, a minority does something, the boundary moves over evolutionary time, and the only way to know which category a given copy is in is to test it.


Common misconceptions

What people thinkWhat's actually true
Transposons are constantly jumping around inside youIn almost every somatic cell nothing moves. Activity is confined to windows where silencing relaxes: germ cell development, the early embryo, some neurons, tumours. The 46% accumulated over hundreds of millions of years, not over your lifetime
cut themselves out and moveOnly Class II elements excise. A retrotransposon is transcribed and a new copy inserted elsewhere; the original never leaves. Hence a single Class I element can never relocate: transposition only ever adds copies. Copy number is a different matter — it falls again by deletion and by removing insertions, so the genome does not simply ratchet upward
Alu must be autonomous, given a million copiesCopy number measures success, not self-sufficiency. Alu encodes no protein at all. It is short, cheap, Pol III-transcribed at high abundance, and hijacks L1's ORF2p — the most successful parasite of a parasite in the genome
46% is the true TE contentIt is a lower bound set by detection. Copies older than a few hundred million years have decayed past recognition, and the estimate rises whenever alignment sensitivity or assembly completeness improves
ERVs are infections that "got stuck" in usSpecifically infections — a somatic one dies with the host. And most human ERVs are now solo LTRs: the internal genes recombined out between the element's own LTRs, leaving a lone
McClintock was ignored because she was a womanShe was in the National Academy by 1944 and led the Genetics Society by 1945. The binding was that transposition invalidated the fixed-locus model underpinning all genetic mapping, shown in a system the field was abandoning. Prejudice was real but not the main obstacle
Domestication shows TEs are "there for a reason"Exaptation is retrospective. Selection cannot retain a useless sequence against future use. Syncytin was captured independently in at least six mammalian lineages precisely because it was contingent — whatever env gene was lying around got used

Worked example

A 4-year-old boy has severe haemophilia A. Clinical of F8 reports no pathogenic variant. Where did the go?

F8 spans ~186 kb at Xq28 (GRCh38, chrX ≈ 154.8–155.0 Mb, minus strand). The is unambiguous and (Ch 13), so the negative result is a statement about the assay, not the patient.

Step 1 — enumerate what an exome cannot see. Exome capture reports substitutions and small inside captured intervals. It systematically misses deep intronic variants, the recurrent -22 (itself generated by between repeats — the commonest severe haemophilia A ), and large insertions. A read carrying a 3.8 kb insertion does not align across it: it soft-clips, and a caller expecting substitutions discards it.

Step 2 — re-examine the raw alignment, not the variant calls. In short-read WGS, look at exon 14 for two things:

reference  ... A C T G G A A T | T T T T A A G ... exon 14
reads      ─────────────────────┐
           ─────────────────────┤  <- many reads soft-clipped at the SAME base
           ─────────────────────┘     (a sharp boundary, not scattered mismatches)

           plus: read pairs whose mate maps to L1 consensus, or fails to map

That combination is the canonical mobile-element-insertion call (Ch 46).

Step 3 — assemble the clipped tails and check the mechanism's fingerprints. Suppose local assembly recovers, in order: L1 3'-end sequence, a poly(A) run, then flanking genome resuming.

CheckExpected under TPRTMeaning
Sequence just 5' of the insertloose match to 5'-TTTT/AA-3'ORF2p endonuclease site
Identical short repeat on both flanks7–20 bp direct TSDthe staggered second nick
3' end of the insertpoly(A) tractit was the primer
Length vs 6 kb full length~3.8 kb, missing the 5' UTRTPRT stopped before finishing

All four present: a de novo L1 insertion, not a or a viral integration.

Step 4 — assign the source and confirm inheritance. Non-L1 sequence downstream of the poly(A) would be a 3' transduction; aligning it locates the source element that fired. Sequencing both parents shows the insertion in neither — a de novo event (check the mother for before quoting a recurrence risk). Truncated L1 sequence inside exon 14 introduces stop and splice signals, no functional factor VIII is made, and the result is severe haemophilia A.

This is not hypothetical. It is what Kazazian and colleagues found in 1988, in two of 240 patients — the observation that established retrotransposition as an ongoing source of human mutation, thirty-eight years after McClintock's paper and five years after her Nobel.

Connections