D3 — Repeat-expansion disorders
Before this: Ch 07 The genetic code · Ch 11 Beyond Mendel · Ch 15 Pedigrees · Ch 16 Mutation · Ch 54 Rare variants and Mendelian disease · Time: ~105 min (2 sessions)
Statistics needed: S2 Distributions · S4 Hypothesis testing · S5 Variance and regression
Track note: part of the specialisation (D1 · D2 · D4 · D5); read after D1 and D2.
What you'll be able to do
- Classify any repeat-expansion disorder by its mechanism class — coding polyglutamine, silencing, transcriptional blockade, RNA gain-of-function, , regulatory mis-setting — from nothing but the motif and its genomic location, and say which classifications that rule of thumb gets wrong.
- Explain why a repeat tract in a neuron that will never divide again can still expand, and why the bias is towards expansion rather than contraction.
- Reconstruct the argument of the Huntington disease onset-modifier — a genome-wide association study run on a fully Mendelian disease — and say what it means that every replicated modifier locus is a DNA-repair gene.
- Distinguish an , a and a reduced- , with the canonical example of each, and say what each one means for the carrier versus for the carrier's children.
- State what is quantitatively, which disorders show a parent-of-origin asymmetry in which direction, and why the ascertainment-bias objection was methodologically correct even though it turned out to be wrong.
- Explain why this entire disease class was invisible to the short-read diagnostic pipeline of Ch 54 — not missed, but invisible by construction — and what replaced guesswork.
- Read a with molecular repeat sizes on it and infer instability, transmission bias and risk, including the risks the sizing data cannot give you.
The core idea
Ch 16 §9 gave you the seed of this chapter in one section: slippage at a tandem repeat, a threshold beyond which the tract becomes unstable, a positive feedback in which the is a function of the current length. Four diseases fitted in one table. This chapter is what that section grows into when you give it room, because the four-row table is the trailer for one of the strangest and most instructive corners of human genetics: several dozen diseases, most of them neurological, all caused by the same kind of — a run of a short motif that got longer — and yet running on molecular mechanisms so different that calling them one class is almost a pun.
Here is the tension to hold from the start. The mutation is the same everywhere; the disease is decided by the address. A CAG tract inside a reading frame becomes a run of glutamines in a protein, and the disease is a poisoned protein. The same CAG tract in an untranslated region becomes a run of CUGs or CAGs in an RNA that never meets a in the normal way, and the disease is a poisoned RNA — or no RNA at all, or too much of one. One chemistry of instability, and at least six distinct molecular crimes across the class — three of them from the CAG motif alone — depending on where in the gene the tract happens to sit. Position, not motif, is destiny — and the disorder this track is heading towards, , is the awkward case that proves it: a CAG expansion that causes a dominantly inherited without the repeat being translated as part of the reading frame, because it sits in the 5′ regulatory region of the gene, upstream of the start site (Ch 05 §6 for what a transcription start site and a pre-initiation complex actually are).
The second theme is that a repeat allele is not a fixed thing you inherit and keep. It is a process. It changes between generations — that is , and you have met it in Ch 11 §11. What Ch 11 could not tell you is that it also changes within you, tissue by tissue, cell by cell, across your lifetime — and that the best current evidence says this ongoing somatic expansion is not a curiosity but the disease clock itself.
A threshold is a convention, not a constant. Every "pathogenic threshold" in this chapter is a clinical-laboratory convention — a line drawn through messy data, revised whenever someone reports a shorter allele in a symptomatic person. Three of the thresholds below moved within the last decade and are still moving. Whenever this chapter states one, it states whose it is. A course that reports a threshold as though it were the mass of the electron is teaching a convention as a measurement.
1. The taxonomy: organising the zoo by mechanism
Do not memorise the tables that follow. Read them the way you would read a well-organised codebase: the organising principle is the content. The principle here is mechanism class, and there are six of them — I, II, II′, III, IV and V, where II′ is a variant on its neighbour and V is the awkward case this track is heading towards.
1.1 Class I — coding repeats: the polyglutamine (and polyalanine, and polyglycine) disorders
The repeat is inside an exon, in frame, and is translated. CAG reads as glutamine, so an expanded CAG tract becomes an expanded polyglutamine (polyQ) tract in the protein, which misfolds and aggregates — a gain of toxic function (Ch 08 §10 gives the loss/gain/poison framework; Ch 08 §8 the folding background).
All values are repeat units, not base pairs — a 40-repeat CAG is 120 bp of tract. Thresholds are from GeneReviews (fetched 2026-08-25) via reference/verified-facts.md, except SCA17 (secondary read of NBK1438), SCA4 (Nat Genet 2024 and STRipy) and OPMD (STRipy only) — the last three are corroborated-secondary and are stated as approximate; MOI = mode of inheritance.
| Disorder | Gene | Motif | Location | Normal | Intermediate / reduced | Pathogenic | MOI |
|---|---|---|---|---|---|---|---|
| Huntington disease (HD) | HTT | CAG | coding, exon 1 | ≤26 | 27–35 intermediate; 36–39 reduced penetrance | ≥40 | AD |
| SBMA (Kennedy disease) | AR | CAG | coding | ≤34 | 35 uncertain; 36–37 reduced penetrance | ≥38 | XL |
| DRPLA | ATN1 | CAG | coding | 6–35 | 35–47 | 48–93 | AD |
| SCA1 | ATXN1 | CAG | coding | 6–35 uninterrupted, or 36–44 with CAT interruptions | 36–38 uninterrupted mutable | 39–44 uninterrupted (46–70 reported) | AD |
| SCA2 | ATXN2 | CAG | coding | ≤31 | 32 unclassifiable; 33–34 at risk | ≥35 | AD |
| SCA3 (Machado–Joseph) | ATXN3 | CAG | coding | 12–44 | 45–59 | ~60–87; smallest full-penetrance allele not well defined | AD |
| SCA6 | CACNA1A | CAG | coding, 3′ end | ≤18 | 19 uncertain | 20–33 | AD |
| SCA7 | ATXN7 | CAG | coding | 7–27 | 28–33 mutable; 34–36 reduced penetrance | 37–460 | AD |
| SCA17 | TBP | CAG/CAA | coding | 25–40 | 41–48 reduced penetrance | ≥49 | AD |
| SCA4 (2024) | ZFHX3 | GGC → polyglycine | coding | 14–31 | — | ≥42 | AD |
| OPMD | PABPN1 | GCG → polyalanine | coding | 6 | — | ≥7 | AD/AR |
Three things to notice before moving on. First, the polyQ and HD are clinically different diseases attacking different neurons — HD the striatum, the SCAs the — from what is superficially the same in different genes; is D1's subject. Second, SCA4 joined this table in 2024: a GGC expansion translated as polyglycine, a whole new chemistry, solving a locus mapped since the 1990s — the table is not closed. Third, SCA6 is the standing exception to nearly every generalisation in this chapter: its pathogenic range (20–33) overlaps the normal range of most other polyQ genes, and is effectively absent — GeneReviews states plainly that expansions are not commonly observed in parent-to-child transmission. Whatever makes CAG unstable, it is not the CAG alone; the flanking context of each locus matters, and SCA6 is the locus that proves it.
1.2 Classes II–V — non-coding repeats: the same mutation, four other crimes
When the repeat sits outside the reading frame, it cannot make a polyQ protein — but it is not thereby harmless. It simply commits different crimes.
| Class | The crime | Type specimen | How it works, in one line |
|---|---|---|---|
| II | via silencing | Fragile X syndrome (FMR1) | The expanded CGG (>200) becomes hypermethylated; the gene is transcriptionally silenced |
| II′ | Loss of function via transcriptional blockade | Friedreich ataxia (FXN) | The expanded GAA forms triplex/R-loop structures and repressive chromatin that block elongation |
| III | RNA gain-of-function | Myotonic dystrophy type 1 (DMPK) | The expanded CUG RNA accumulates in nuclear foci and sequesters factors |
| IV | C9orf72 ALS/FTD | The repeat RNA is translated without a start , in multiple frames, into aggregating dipeptide-repeat proteins | |
| V | Regulatory mis-setting | () | The repeat sits in the 5′ regulatory region and changes how much of the gene is expressed |
And the loci, with thresholds:
| Disorder | Gene | Motif | Location | Normal | Intermediate / | Pathogenic | MOI |
|---|---|---|---|---|---|---|---|
| DM1 | DMPK | CTG | 3′ UTR | 5–34 | 35–49 premutation | >50; congenital usually >1,000 | AD |
| DM2 | CNBP | CCTG | 1 | ≤30 uninterrupted | ~30–54 mutable; ~55–74 uncertain | ~75–11,000 (mean ~5,000) | AD |
| Fragile X (FXS) | FMR1 | CGG | 5′ UTR | ~5–44 | 45–54 grey zone; 55–200 premutation | >200 + methylation | XL |
| FXTAS | FMR1 | CGG | 5′ UTR | — | the 55–200 premutation is the disease allele | the full mutation does not cause FXTAS | XL |
| Friedreich ataxia | FXN | GAA | intron 1 | 5–33 | 34–65 mutable normal | 66–~1,300 | AR |
| C9orf72 ALS/FTD | C9orf72 | GGGGCC | intron 1 / region | 2–24 | 25–60 uncertain | 61–>4,000, age-dependent reduced penetrance | AD |
| SCA8 | ATXN8OS/ATXN8 | CTG·CAG | 3′ UTR of one gene, sense ORF of the overlapping gene | 15–50 | 51–70 unclear | 54–250 typical; 71–>1,300 in affected and unaffected | AD, markedly reduced penetrance |
| SCA10 | ATXN10 | ATTCT | intron 9 | 10–32 | 280–850 reduced penetrance | 800–4,500 | AD |
| SCA12 | PPP2R2B | CAG | 5′ region / probable promoter, ~133 nt upstream of the reported TSS (5q32) | 4–32 | 43–50 argued pathogenic | ≥51 classical convention; ≥43 argued — contested | AD |
| SCA27B | FGF14 | GAA | deep intron 1 | 8–249 | 250–300 reduced penetrance | ≥300 | AD |
| SCA31 | BEAN1 | (TGGAA)n insertion | intron | 0 TGGAA | — | insertion ≈ 2.5–3.8 kb (~500–760 pentamers) | AD |
| SCA37 | DAB1 | (ATTTC)n inserted inside (ATTTT)n | intron | 0 ATTTC | — | (ATTTC)₃₁–₇₅ | AD |
| CANVAS | RFC1 | AAGGG replacing reference AAAAG | intron 2 | (AAAAG)n benign at any length | — | biallelic (AAGGG)₄₀₀–₂₀₀₀₊ | AR |
| FAME/BAFME | SAMD12 and 5 other genes | (TTTCA)n inserted inside (TTTTA)n | intron | 0 TTTCA | — | locus-specific | AD |
The last four rows are a different kind of mutation again: in SCA31, SCA37, CANVAS and the familial myoclonic epilepsies, the pathogenic event is not that an existing repeat got longer but that a new motif appeared inside or in place of a benign one — (ATTTC)n inserted inside an (ATTTT)n tract, AAGGG replacing AAAAG. The reference motif is harmless at essentially any length; the variant motif is the disease. Motif identity, not motif length, is the mutation. Keep that in mind when you meet the repeat callers in lab 11 — a tool that reports only a length at these loci has missed the variable that matters.
1.3 Location is destiny: hold the motif fixed, vary the address
Now collapse the two tables along a single axis — hold the motif fixed and vary the address:
| Locus | Motif | Where the repeat sits | What the expansion does |
|---|---|---|---|
| HTT | CAG | coding exon 1 | Translated → polyQ protein , aggregation |
| ATXN8OS/ATXN8 | CTG·CAG | 3′ UTR of one gene and sense ORF of the overlapping gene | CUG RNA gain of function and polyQ and RAN products — one repeat, three products |
| DMPK | CTG | 3′ UTR | Untranslated CUG RNA → nuclear foci → splicing factor sequestration |
| PPP2R2B (SCA12) | CAG | 5′ region, ~133 nt upstream of the TSS | No polyQ from the PPP2R2B reading frame. A cis-regulatory element that alters PPP2R2B expression — but see §6.6, which is less tidy than this row |
| FXN | GAA | intron 1 | Triplex/R-loop → blocked → loss of frataxin |
| FGF14 (SCA27B) | GAA | deep intron 1 | Reduced FGF14 RNA and protein — same motif and direction as FRDA, different gene, |
| FMR1 | CGG | 5′ UTR | Two diseases from one repeat, decided by size: 55–200 → toxic RNA + RAN product (FXTAS); >200 → methylation and silencing (FXS) |
Note that CAG and CTG are the same repeat read from opposite strands — HTT's CAG and DMPK's CTG are one chemical entity in two genomic contexts. And read the FMR1 row twice, because it makes the location principle's companion point from the other side: identical motif, identical position, different size — and the mechanism flips from RNA toxicity to transcriptional silencing as the tract grows past ~200. Size does not just tune severity; past a point it changes which disease you have.
The SCA12 hook, in one sentence: HTT and PPP2R2B both carry an expanded CAG, but HTT's is inside the reading frame and PPP2R2B's is ~133 nucleotides upstream of the transcription start site — so one disease is a poisoned protein and the other is a mis-set dial on a . That dial, and what it controls, is D2's and D4's territory. What matters here is that SCA12 does not fit cleanly into any of classes I–IV, and the field's uncertainty about its threshold (§4) is partly a symptom of that mechanistic homelessness.
1.4 Whose threshold? Where the sources disagree
Compare two curated sources for the same loci — GeneReviews chapters against the STRipy STR database, both read on the same day (2026-08-25) — and the boundaries move:
| Locus | STRipy says | GeneReviews says | What to conclude |
|---|---|---|---|
| ATXN7 (SCA7) | pathogenic ≥36 | reduced penetrance 34–36, full ≥37 | 34–36 is a genuine grey zone |
| ATXN3 (SCA3) | pathogenic ≥56 | "smallest full-penetrance allele is not well defined" | The floor is unknown — not 56, and not 60 |
| C9orf72 | pathogenic ≥24 | normal 2–24, uncertain 25–60, pathogenic ≥61 | A 37-repeat-wide disagreement at the decisive boundary |
| AR (SBMA) | pathogenic ≥40 | ≥38 | Two repeats apart, exactly where a diagnosis is made |
| PPP2R2B (SCA12) | pathogenic ≥51 | (chapter NBK1202, last updated 2011, retired 2018; while current it gave ≥51 as diagnostic — and said in the same breath that the threshold itself was "not clear"); Srivastava et al., Brain 2017 argue ≥43 | The authority withdrew; the field has not converged. This is D4's opening problem |
This is not sloppiness to be cleaned up; it is what the underlying biology looks like. Penetrance falls off continuously with repeat length (Ch 11 §8), instability rises continuously, and any line drawn through two continuous functions is a decision, not a discovery. It is also why the ACMG framework of Ch 55, built for variants that are present or absent, fits this disease class so awkwardly — a theme D5 returns to.
2. Instability mechanics: beyond slippage
Ch 16 §9 gave you slippage and the positive feedback; this section is what it left out. Take the slippage mechanism as read — misalignment of repetitive strands during synthesis — and ask the two questions Ch 16 could not answer: why do some motifs expand and not others, and how does a repeat expand in a cell that never replicates its DNA again?
Why motifs differ: secondary structure
The unifying model (Pearson, Edamura & Cleary, Nat Rev Genet 2005) is that repeat tracts are unstable because the transiently single-stranded DNA generated during replication, repair, and recombination can misalign and fold. A folded intermediate that survives to the next round of synthesis is copied as a length change. What differs between motifs is what they fold into:
| Motif | Structure formed | Consequence |
|---|---|---|
| CAG·CTG | Hairpins (imperfect — one mismatch per repeat unit) | Slipped-strand intermediates; substrate for MutSβ (§3); expansion-biased |
| CGG·CCG | Hairpins and G-quadruplexes; heavily methylatable CpG content | Replication stalling; once expanded, and silencing |
| GAA·TTC | Intramolecular triplex (H-DNA), "sticky DNA", R-loops | Blocks transcription elongation, not just replication |
| GGGGCC | DNA and RNA G-quadruplexes, RNA·DNA hybrids | Length-dependent accumulation of abortive transcripts; nucleolar stress (Haeusler et al., Nature 2014) |
The load-bearing contrast. CAG folds into a hairpin, which is a replication-and-repair problem, and the diseases that follow are mostly gain-of-function — a poisoned protein or RNA. GAA folds into a triplex and R-loop, which is a transcription problem, and the diseases that follow are loss-of-function — the gene cannot be read. Same "trinucleotide repeat disorder" heading; two different molecular crimes, and you can predict which from the physical chemistry of the fold. For GAA the correlation is strikingly tight: the repeat length that inhibits transcription in vivo and in vitro (>59 units) is the length at which the tract adopts the sticky-DNA conformation.
The structure column also explains the thresholds. A hairpin or triplex needs a minimum tract length to be thermodynamically stable; below it, slippage still happens (that is the ordinary microsatellite polymorphism of Ch 16) but the fold that protects a slipped intermediate from correction cannot form, so changes stay at ±1 unit. The threshold in the disease tables is, to a first approximation, the length at which the escape structure becomes stable — and the positive feedback begins.
Germline versus somatic instability: two clocks
The distinction of Ch 16 §1 — versus somatic — now pays off twice over, because repeat instability runs on both clocks and they answer different questions:
- Germline instability is measured between generations. It is what changes the your child inherits relative to yours, it is motif- and parent-of-origin-specific (§5), and it is the substrate of .
- is measured within one person. The repeat length in your blood, your liver and your neurons diverges over your lifetime — you are a mosaic of repeat lengths, and the mosaic is not random. In HD, the CAG tract undergoes progressive length increases over time preferentially in the brain regions that degenerate (Swami et al., Hum Mol Genet 2009) — and, in that study of human HD brain, longer somatic expansions were associated with earlier onset.
That last clause should stop you. It suggests somatic expansion is not a side effect of disease but a cause — a suggestion §3 will make precise.
The postmitotic puzzle: expansion without replication
Here is the puzzle stated plainly: a striatal or cerebellar neuron (D1) is postmitotic. It exited the cell cycle before you were born and will never pass a over its genome again. Slippage-during-replication cannot be how its repeats change. Yet they do — and they change in the most expansion-biased way of any tissue. The resolution is that the instability engine in neurons is repair, not replication. Two mechanisms are on the table, and they are not mutually exclusive:
- Oxidative-damage-initiated . Kovtun et al. (Nature 2007) showed that OGG1 — the glycosylase that excises the oxidised base 8-oxoguanine — initiates age-dependent CAG expansion in somatic cells. The proposed "toxic oxidation" cycle: oxidative damage lands in the repeat; OGG1 excises the lesion; the repair gap is filled by strand-displacement synthesis that captures a hairpin; the tract is now longer, and a longer tract is a bigger target for the next oxidative hit. An escalating cycle, running on nothing but ordinary metabolism and repair — which is why it runs perfectly well in a neuron, and why it accelerates with age.
- gone wrong — the engine of §3.
Both mechanisms are biased towards addition, for a structural reason worth internalising: the intermediate that gets fixed into the genome is an extra loop of DNA stabilised as a hairpin, not a missing one. Repair that resolves the intermediate in favour of the loop adds repeats. The neuron's repeat tract is a ratchet, and repair is the pawl.
3. The mismatch-repair engine
This section is one of modern human genetics' best stories, and it is worth telling in the order it happened, because the order is the argument.
The strange GWAS
Huntington disease is as Mendelian as a disease gets: one gene, one type, , fully penetrant at ≥40 repeats. There is nothing for a genome-wide association study to find about whether you get it. But CAG length only partially predicts when — it accounts for roughly 50–70% of the variance in onset age, the exact figure depending on cohort and model (GeneReviews says "up to 70%"; the GeM-HD 2019 cohort estimate is ~57–60%), and an estimated 10–20% of the residual variability is itself heritable. Somewhere in the genome are common variants that shift the onset of a Mendelian disease. So the GeM-HD Consortium ran a — the machinery of Ch 51, applied not to case–control status but to residual onset age among people who all carry the mutation.
The 2015 result (Cell 2015): a locus on 15 carrying two independent signals in opposite directions — one hastening onset by 6.1 years, one delaying it by 1.4 — a locus on chromosome 8 hastening onset by 1.6 years, and association at MLH1, with pathway analysis pointing at DNA handling and repair.
The 2019 result (Cell 2019, 9,064 HD subjects) turned a hint into a list, and the list is the point. The replicated modifier loci carry, gene after gene: FAN1 (the top locus), MSH3, PMS1, MLH1, PMS2, LIG1. Look at that list with Ch 17 §5 open. It is not a list of neuronal genes, synaptic genes or protein-quality-control genes. It is the mismatch-repair pathway, plus a nuclease that regulates it. The genome, asked an open-ended question about what controls HD onset, answered: the machinery that decides whether repeat tracts grow. (Several loci carry multiple independent acting in opposite directions — some hastening, some delaying — exactly what you would expect if repair activity is a dial that population variation turns both ways. The per-haplotype effect sizes in years are published in the 2019 paper's Table 1, but this course has not verified them line-by-line and will not quote numbers it has not checked; the 2015 figures above are the ones verified here.)
The mechanism: MutSβ as an accomplice
Why would — the pathway that protects you from microsatellite instability in Ch 17 — drive ? Because the CAG hairpin is a trap for it.
MutSβ is the MSH2–MSH3 heterodimer, the loop-recognising arm of MMR — the one that normally handles small insertion/deletion loops. Its failure mode on disease-length repeats, as currently understood:
- MutSβ binds the CAG/CTG hairpin but cannot process it properly: on (CAG)₁₃ and (CTG)₁₃ hairpins its ATP hydrolysis is reduced, so the reaction stalls at recognition instead of proceeding to excision. The hairpin is protected rather than removed.
- MutSβ physically recruits β to the hairpin and stimulates its retention: Polβ extends from the hairpin, locking the extra repeats in (Zhao et al., Cell Res 2016).
- Stoichiometry sets the dial: MSH3 competes with MSH6 for MSH2, so when MSH3 is abundant the balance tips from MutSα towards MutSβ — and towards expansion. That is precisely the axis the MSH3 modifier haplotypes from the GWAS sit on.
FAN1 is the counterweight — a nuclease that opposes expansion, reported both to control MMR complex assembly at the repeat via MLH1 retention and to process and pause on slipped-DNA structures. The top HD-modifier locus in the genome, FAN1, is the brake on the expansion engine. The circle closes.
Read the logic of this argument, not just its conclusion. No one hypothesised MutSβ into the GWAS. The GWAS was unbiased — millions of variants, no favourites (Ch 51) — and it converged on a pathway that mechanistic work (in mice and cells, largely done earlier and separately) had implicated. Two independent lines of evidence meeting at the same pathway is how you know you are near the machine, and it is a far stronger epistemic position than either line alone.
The two-step model of onset
Put §2 and §3 together and a model assembles itself:
Step 1: the inherited expands somatically in vulnerable neurons — slowly, over decades, at a rate set by the person's DNA-repair . Step 2: once a given cell's tract crosses a critical threshold, a toxic mechanism (§6) fires in that cell and it dies. Onset is then set by the rate of somatic expansion; the inherited length matters chiefly because it sets how far each cell has to travel.
The strongest current evidence is single-cell: Handsaker et al. (Cell 2025) measured repeat lengths cell-by-cell in postmortem HD brain and found that striatal projection neurons — precisely the cells HD kills — carry frequent somatic expansions of the disease allele, while most other cell types show limited somatic variation. The transcriptional collapse begins at around 150 CAG: cells below it look essentially normal, cells beyond it lose their identity. On the authors' reading, neurons spend the great majority of life expanding quietly below threshold, and only a short final above it. Two honest caveats, because this is a fast-moving front: the ~150 figure is corroborated by multiple independent reports of the paper but the claim that crossing it is "necessary and sufficient" is the authors' claim about one cell type, not yet a field-wide consensus about the whole disease; and whether somatic expansion is the entire story of onset — rather than the dominant term — is exactly what is being tested now.
If the model is even approximately right, the therapeutic implication is startling: you would not need to touch the toxic mechanism at all — slow the expansion engine, and you delay onset of every downstream event, whatever the toxic mechanism turns out to be. MSH3 is the obvious dial to reach for: the GWAS has already shown that naturally occurring variation in it shifts onset. That idea, and what it might mean for specifically, is planted here and harvested in D5.
4. Thresholds, premutations, intermediate alleles and penetrance
Three different concepts hide under the word "borderline", and conflating them causes real counselling errors. Learn them as a triple with a canonical example each:
- Intermediate (mutable-normal) — does not cause the classical disease in the carrier, but is meiotically unstable, so the carrier's children are at risk of inheriting an expansion. Canonical: HD 27–35 CAG; also DM1 35–49, SCA1 36–38 uninterrupted, FMR1 45–54. The risk belongs to the next generation, not the person in front of you.
- with its own — not the disease allele for the classical disorder, but pathogenic in its own right, causing a different disease. Canonical: FMR1 55–200, which causes FXTAS (a late-onset tremor/ syndrome) and FXPOI — by a toxic-RNA mechanism the full cannot use, because the full mutation is methylated and silent (§6). "Premutation" is a historical name and a misleading one: for FXTAS it is simply the mutation.
- Reduced- allele — genuinely disease-causing, but not everyone who carries it manifests within a lifetime. Canonical: HD 36–39 CAG.
The HD 36–39 range, in numbers
This is the canonical reduced-penetrance range in all of human genetics, and its numbers repay staring at (all from GeneReviews NBK1305 via reference/verified-facts.md):
| Quantity | Value |
|---|---|
| Reduced-penetrance range | 36–39 CAG |
| Lifetime penetrance of a 36–38 allele | approximately 0.2%–2% |
| Frequency of ≥36 CAG alleles, European-ancestry populations | as many as 1:400 — the majority being 36–39 |
| Clinical prevalence of HD, European ancestry | 9.71 : 100,000 (up to 17 : 100,000 with multi-source ascertainment) |
| Clinical prevalence, East Asian and African populations | 0.1–2 : 100,000 |
Now do the division. Roughly 1 in 400 people of European ancestry carries an HTT allele in the disease-causing range, against a disease prevalence of roughly 1 in 10,000 — a ~25-fold gap, and nearly all of it is 36–39-repeat alleles that never declare themselves. The same shape recurs in SBMA: clinical prevalence ~1:300,000 males, yet 1 in 6,887 males carries a pathogenic AR expansion. Hold this against Ch 54 §11: penetrance estimated in clinically ascertained families is always an overestimate of penetrance in the population, and repeat disorders exhibit the gap in its purest form. If population screening ever reports "HD mutation carrier" from biobank-scale data, the majority of the people so labelled — mostly 36–39 carriers — will die of something else, never knowing.
Two more entries for your collection of penetrance shapes: C9orf72 expansions show age-dependent reduced penetrance even at frankly pathogenic sizes; and SCA8 shows reduced penetrance at all sizes — alleles from 71 to over 1,300 repeats occur in affected and unaffected people alike, which is as strong a warning as exists against reading a repeat size as a verdict.
Interruptions: the hidden variable under every threshold
Every threshold above is really a threshold on uninterrupted repeat length. A repeat tract broken by occasional variant units is chemically a different object — the interruption breaks the register of the hairpin or triplex — and both stability and pathogenicity follow the pure tract, not the total:
| Locus | Interrupting unit | Effect |
|---|---|---|
| ATXN1 | CAT | Stabilises: a 36–44 allele with CAT interruptions is normal; the same length without is mutable or pathogenic |
| HTT | CAA (the CAACAG) | Loss of the CAACAG → onset earlier than the glutamine count predicts; a CAACAG duplication — more glutamines — → onset later (GeM-HD 2019) |
| ATXN2 | CAA | Codes glutamine too, so does not reduce pathogenicity — but may enhance meiotic stability |
| FMR1 | AGG, normally every 9–10 CGG | Presence reduces the risk of maternal expansion to full mutation for alleles below ~100 repeats |
| ATXN10 | ATCCT | Interrupted expansions associate with more seizures — and show with paradoxical repeat contraction |
| RFC1 | motif substitution, not interruption | (AAAAG)n benign at any length; (AAGGG)n pathogenic — the "interruption" is the whole mutation |
The HTT CAA row is the most beautiful natural experiment in the field, and §6 will lean on it. For now, note the practical consequence: two alleles that an electrophoresis-based assay calls "40 repeats" can differ in interruption structure and therefore in stability and effect — one reason sequence-level genotyping, not just sizing, is where lab 11 is headed.
5. Anticipation, quantitatively
Ch 11 §11 defined and Ch 16 §9 made the ascertainment-bias caution — the objection was methodologically correct, it was raised by the right instinct, and only molecular measurement of repeat length settled it. Neither is repeated here; go back if the caution is not vivid. What this section adds is the quantitative structure.
Repeat length versus onset: regressions, and their honest scatter
The relationship that rescued anticipation from the ascertainment-bias objection is the within-locus inverse correlation between repeat length and onset age. Its canonical quantitative form is the Langbehn model for HD (Langbehn et al., Clin Genet 2004): a parametric survival model fitted to 2,913 individuals from 40 centres, in which the probability of diagnosis by a given age is logistic conditional on CAG count, yielding onset-probability curves for CAG 36–56. The shape is what to carry: steeply falling and convex — each added repeat costs more years at the short end than the long end — and flattening above ~50 repeats. (The published coefficients are not reproduced here; if you need the curve, read them out of the paper rather than reconstructing them, because a misremembered survival coefficient in a counselling context is not a rounding error.)
But the correlation's strength varies enormously by locus, and the variation is informative:
| Locus | Length ↔ onset relationship |
|---|---|
| HD | CAG length explains ~50–70% of onset variance, cohort-dependent (§3) |
| SCA1 | 36–70% of onset variance — wide, and honestly reported as wide |
| SBMA | ~60% of clinical variability |
| Pearson r = −0.65, P < 10⁻⁴, n = 124 unrelated patients (Srivastava et al., Brain 2017) — so r² ≈ 0.42: less than half the variance | |
| FRDA | The shorter (GAA1) predicts onset: <700 repeats → mean onset 18 y; >700 → mean onset 9.7 y; <500 → onset after 25; <300 → onset after 40 |
| SCA31 | insertion length inversely correlated with onset |
| DM2 | no significant correlation at all between CCTG size and onset or severity — the standing counterexample |
Note FRDA's twist: it is , both alleles are expanded, and it is the shorter one that sets the — because the shorter allele sets the residual of FXN, and the disease is loss-of-function (§6). The regression you fit follows from the mechanism class. And keep DM2 pinned where you can see it: a repeat disorder with a pathogenic range spanning more than two orders of magnitude (75 to 11,000) and no length–severity correlation, with anticipation not confirmed. Any story you tell about repeat length must survive DM2 (statistical footing for these correlations: S4 for what the P-value does and does not license, S5 for reading r²).
Parent-of-origin asymmetries, and the gametogenesis logic
Anticipation is not symmetric in the transmitting parent, and the direction flips by locus:
| Disorder | Bias | The numbers |
|---|---|---|
| HD | Paternal | Anticipation far more common in paternal transmission; large expansions (>7 CAG) almost exclusively paternal |
| DRPLA | Paternal ≫ maternal | Offspring onset ~26–29 y earlier than affected fathers; ~14–15 y earlier than affected mothers |
| SCA2 | Paternal | Large expansions almost exclusively through the paternal |
| SCA7 | Paternal, dramatically | Infantile-onset cases at 200–400 repeats from affected fathers |
| SCA1 | Paternal for expansion | Contractions more typical of maternal transmission |
| DM1 | Both, differently | Congenital DM1 is most often maternally inherited — yet paternally transmitted expansions are larger (median +425 repeats, range 70–2,000, vs maternal median +200, range 57–1,400) |
| SCA8 | Maternal — the reversal | Expands maternally; tends to contract paternally, often into the reduced- range |
| FMR1 | Maternal only | Expansion to full occurs only through the mother |
| SCA6 | none | Anticipation not observed |
| DM2 | none confirmed | — |
| C9orf72 | disputed | Reported and contested; confounded by observation bias and by the difficulty of sizing the repeat in blood |
The first-order logic comes straight from Ch 09: spermatogenesis is continuous — spermatogonia keep dividing from puberty onwards, stacking up replication opportunities per transmitted genome — while oogenesis completes its replications before the mother's birth and the oocyte then waits, arrested, for decades. More replication, more slippage opportunity: hence the paternal expansion bias at most CAG loci, and it is the same asymmetry that makes new point mutations predominantly paternal (Ch 16 §7).
But do not let the tidy logic own you, because the table refuses to obey it uniformly. Congenital DM1 comes through the mother despite paternal jumps being larger — the largest paternal alleles appear to be selected against somewhere in spermatogenesis or transmission, while the maternal germline (and possibly early embryo) permits the giant alleles. FMR1 expands to full mutation only maternally. SCA8 inverts the rule entirely. And §2 told you that replication count cannot be the whole story anyway, since the most dramatic expansion environment in the body is a cell that never replicates. Parent-of-origin bias is real, quantitative and clinically decisive — a father with an FMR1 and a mother with the same allele give their children categorically different risks — but it is locus-specific empirical knowledge, not a law you can derive once and apply everywhere.
Suggested stopping point (Session 1 of 2). §§1–5 cover repeat instability, somatic expansion, anticipation, and inheritance patterns. §6 onward works through the toxicity mechanisms by class — a natural break for a second sitting.
6. Toxicity mechanisms in depth: one crime per class, one type specimen each
Each mechanism class earns its place in the taxonomy through one disorder where the evidence is cleanest — the type specimen. Learn the specimen and the observation that established it, and you own the class.
6.1 Class I: the poisoned protein — HD
The expanded polyQ tract makes the protein aggregate: DiFiglia et al. (Science 1997) found huntingtin aggregated in neuronal intranuclear inclusions and dystrophic neurites in HD brain, and inclusions have been the visible signature of polyQ disease since. But the field's understanding has a sting in it. The GeM-HD 2019 CAA-interruption result (§4) showed that individuals with more glutamines but a shorter uninterrupted CAG tract have later onset than their glutamine count predicts — so uninterrupted CAG length at the DNA level, not polyglutamine length at the protein level, determines onset timing. The cleanest reading, via §3: the DNA tract length drives the somatic-expansion clock, and the protein's toxicity fires downstream once the clock runs out. Aggregation is real and replicated; what it explains, on its own, is smaller than it looks.
6.2 Class II: silencing — fragile X
Above ~200 CGG repeats, the FMR1 tract and its surrounding become hypermethylated and the gene is transcriptionally silenced: no mRNA, no FMRP protein, and the syndrome is a straight (Ch 16 §9 introduced this; the chromatin machinery is Ch 22's). The proof of the logic is the exception: the 55–200 is not silenced — it is transcribed at elevated levels, 2–8-fold — and causes a completely different disease (FXTAS) via its RNA (§6.4) and a RAN product (§6.5). One locus, one motif; below ~200 a toxic-RNA disease, above it a silencing disease. The full cannot cause FXTAS because it is silent. No single observation in this chapter separates "the repeat" from "what the repeat does" more sharply.
6.3 Class II′: transcriptional blockade — Friedreich ataxia
FRDA is loss-of-function by a different route: the expanded GAA tract does not silence FXN through methylation of a promoter so much as physically and epigenetically obstruct its elongation — repressive chromatin forms around the expanded repeat, and the triplex/sticky-DNA and R-loop structures of §2 interfere with through the tract. Both mechanisms are stated side by side by GeneReviews; they are not competing so much as compounding. The consequences you can now predict rather than memorise: inheritance (a partial loss needs both copies hit before the appears — and note the contrast with SCA27B, where a GAA expansion reducing FGF14 expression is , because different genes tolerate halving differently); and the shorter- regression of §5, because residual frataxin output is set by the less-blocked allele.
6.4 Class III: the poisoned RNA — myotonic dystrophy
DM1's expanded CUG RNA never leaves the nucleus normally; it accumulates in foci and recruits the muscleblind (MBNL) family of regulators — Miller et al. (EMBO J 2000) showed MBNL proteins co-localising with the expanded-repeat transcripts. Sequestered MBNL cannot do its job, which is directing developmentally regulated (Ch 06 §9); dozens of transcripts revert to fetal splice isoforms in adult tissue, and the multi-system phenotype — myotonia, cardiac conduction defects, cataracts — reads as a catalogue of mis-spliced targets. The clincher for the mechanism is DM2: a different motif (CCTG) in a different gene (CNBP) producing foci that sequester the same MBNL proteins — Miller et al. saw MBNL in DM2 foci too — and yielding an overlapping multi-system phenotype — myotonia, cataracts, cardiac conduction disease, insulin resistance — though not an identical one: DM2 is milder, proximally rather than distally weak, and has no congenital form. The shared agent is the sequestered factor, not the locus.
DM1 also supplied the observation that repeats refuse to stay in one gene's story: the CTG tract is transcribed from both strands — there is an antisense transcript across the repeat, with constraining it and the local (Cho et al., Mol Cell 2005). Bidirectional transcription at repeat loci turns out to be common (Ch 24 §7) — C9orf72's antisense transcript is also translated (§6.5), and bidirectional transcription has been reported at the locus, a thread D4 picks up. Stop thinking of a repeat as sitting inside one gene; think of it as a genomic address that several transcripts may cross.
6.5 Class IV: RAN translation — read honestly
In 2011, Zu, Ranum and colleagues reported something that contradicted the textbook definition of : expansion constructs expressed homopolymeric proteins — polyglutamine, polyalanine and polyserine, from the same CAG construct in different frames — with no ATG start anywhere (Zu et al., PNAS 2011). They named it repeat-associated non-ATG (RAN) translation. The expanded repeat's RNA structure itself appears to license initiation, in multiple frames, on sense and antisense transcripts alike.
Sit with why that was a shock, because the shock is the content. Eukaryotic initiation, as Ch 07 sets out in its section on the and the translation cycle, is a scanning process: eIF4E binds the 5′ m⁷G cap, the small subunit loads at the extreme 5′ end of the message, and it travels along until it meets an AUG in good Kozak context — a at −3 and a G at +4 doing most of the work. Every part of that model makes a start site a property of one position in one sequence: a particular codon, plus a handful of flanking bases, met by a complex moving in one direction from one entry point. Ch 07 also gives you the model's two sanctioned ways to bend, and both are worth holding next to . Leaky scanning past a weak-context AUG lets some complexes run on to a downstream AUG, yielding two N-terminal isoforms from one mRNA; upstream ORFs in the 5′ UTR capture scanning subunits and throttle the main ORF below them — which is also, incidentally, why a repeat tract sitting in a 5′ UTR was never expected to be translationally inert. But both of those bend only the rule about which AUG gets used. Neither touches the rule that initiation happens at an AUG at all, and that is the rule the 2011 constructs broke — in several frames at once, with no ATG anywhere in the construct to appeal to.
So what is the repeat doing to the scanning model? The honest answer comes first: no mechanism for RAN initiation is established, and this chapter will not manufacture one. What can be said is where the proposal comes from, and why the phrasing above — that the repeat's RNA structure appears to license initiation — is deliberately weak and should stay weak. §2's structural chemistry was about DNA, but the same motifs fold as RNA: CAG and CUG into hairpins, CGG and GGGGCC into G-quadruplexes. A scanning complex meeting a stable fold inside a 5′ UTR is not in the situation the clean-message model describes, and the multi-frame output is the part that no single, correctly-chosen start codon explains at all. What the fold actually does to the complex — stall it, hold it over the tract, or hand the message to a route that never scanned in the first place — is not settled. Grade the two claims separately, on the three-tier ladder this track uses (D4): that RAN products exist is Established; that a repeat's fold subverts cap-dependent scanning in order to make them is Conjectured — argued from adjacent facts, not measured. And whether any of it happens at the SCA12 locus, in which frames, and whether it happens anywhere outside a dish, is not settled in §6.6 below either — that is D4's business, in its Hypothesis C section on RNA-level toxicity and RAN translation, where the SCA12 evidence is weighed frame by frame.
The famous case is C9orf72. The GGGGCC expansion — an intronic repeat, which classical logic says should produce no protein — yields dipeptide-repeat proteins (DPRs) from multiple frames, found as insoluble aggregates in patient CNS tissue by two independent groups in the same year (Mori et al., Science 2013: poly-GA most abundant, plus poly-GP and poly-GR; Ash et al., Neuron 2013). The antisense transcript is RAN-translated too, adding further DPR species. And FXTAS gets a RAN chapter of its own: the FMR1 premutation's 5′ UTR yields FMRpolyG, toxic in cell and fly models.
Now the honest ledger, because RAN translation attracts more confidence than the evidence has yet earned:
- Established: RAN translation happens — in constructs, in cells, and the predicted products are physically present in patient tissue as aggregates, for C9orf72 and FXTAS among others. The citations above are load-bearing and replicated.
- Assumed, or still open: how much of the pathology the RAN products cause. For C9orf72, GeneReviews lists three mechanisms side by side — (the expansion reduces C9orf72 expression), RNA toxicity via sequestration (the G-quadruplex-rich transcripts and abortive products of §2), and DPR toxicity — and declines to adjudicate. That non-adjudication is the current state of the field, not a gap in the summary. Model systems can make each mechanism sufficient at sufficient dose; which dominates in patients, at patient doses, over patient decades, is unresolved.
A disease is allowed to run several mechanisms at once, and most of the non-coding disorders probably do. SCA8 is the compact demonstration: one locus expressing a CUG RNA from ATXN8OS (class III), a polyQ protein from the overlapping sense ORF of ATXN8 (class I), and RAN products (class IV) simultaneously. The taxonomy of §1 classifies mechanisms, and diseases are free to subscribe to more than one.
6.6 Class V: the mis-set dial — where SCA12 will sit
The remaining class has one occupant, and it is ours — though "clean" is exactly the word it will not earn. In SCA12 the CAG tract sits ~133 nt upstream of the reported transcription start site of — in cis-regulatory territory (Ch 22 §2) — and the working model is that the expansion acts as a cis element that alters the gene's expression (with the evidence pointing towards up-regulation of at least some isoforms, plus reported bidirectional transcription at the locus). The picture is a promoter with its dial mis-set, feeding a different amount of a — B55β, the product of PPP2R2B — into neurons whose signalling depends on that dial (D2).
But keep the ledger honest, because "class V" is a slot in a taxonomy, not a certificate of exclusivity. What is not established for SCA12 is a polyQ tract from the PPP2R2B reading frame and methylation-driven silencing. What is now reported is more than the class-V label suggests: nuclear CAG RNA foci in patient iPSC-derived neural stem cells, sequestering a set of proteins that binds the expanded repeat and not the normal one; RAN translation at the locus in multiple frames, in cells and iPSC-derived neurons — polyglutamine and polyserine from the sense strand, polyalanine from the antisense PPP2R2B-AS1 ORF, with the frames not yet reconciled between the two labs reporting them (Kumar et al., iScience 2024; Zhou et al., Mov Disord 2023). Neither has been demonstrated in SCA12 human brain, and the same iPSC work finds most PPP2R2B isoforms down-regulated in mature neurons, which cuts against the up-regulation model. So the working model is cis-regulatory mis-setting, and SCA12 may simultaneously be running class III and class IV mechanisms — which is exactly the multi-mechanism point §6.5 has just made for SCA8 and C9orf72, and is why this class is the one to hold most loosely. Which mechanism, if any, is the one that matters in a patient's is precisely the business of D4.
Here SCA12 completes the taxonomy in the weakest sense that is still true: the CAG motif alone has now been implicated in classes I, III (as CTG on the other strand), IV and V, which is the "location is destiny" principle with all its cards face-up — and a reminder that one address can be charged with more than one crime.
7. Why this class evaded the pipeline
Ch 54 walked you through the modern rare-disease pipeline — exome or genome, filter, match — and its §9 table of "where the undiagnosed cases go" contains one row that this chapter has been building towards:
| | Invisible by construction — the read is shorter than the repeat | Targeted repeat genotypers, long reads |
Sit with "invisible by construction". A 150-bp short read that starts inside a 900-repeat GAA tract contains nothing but GAA. It carries no unique sequence, aligns nowhere in particular or to the reference-length , and produces no variant call — not a wrong call, no call (Ch 46 §10 gives the alignment logic; note also Ch 46 §13: the benchmark truth sets used to validate pipelines exclude these regions from their confident intervals, so the pipeline's advertised accuracy was never measured where these diseases live). An expansion is not a needle in a haystack; it is a needle the detector was built not to register. A family could have , genome sequencing and a negative report, while carrying a of a class known since 1991.
The consequences ran in both directions. Diagnostically, repeat disorders stayed on the candidate-gene treadmill — clinicians ordered locus-by-locus tests guided by , which works only when the phenotype names the locus. And gene discovery stalled in the same blind spot: SCA27B — a deep-intronic GAA expansion in FGF14, reported in 2023 to be among the commonest causes of late-onset , found in 61% of French-Canadian index patients in the discovery cohort — and the RFC1 expansions behind CANVAS, now implicated in 14–22% of adult-onset ataxia, both sat undiscovered through three decades of molecular ataxia genetics and fifteen years of short-read sequencing of exactly the right patients. The answer was in the DNA the whole time; the reads were too short to spell it.
What changed: long reads that span the tract entirely (Ch 40 §3; Ch 40 §10 is the platform-by-question logic this is a textbook case of) — sizing it directly, catching the interruption structure of §4, and even reading somatic molecule-by-molecule — and, for the short-read archives, targeted repeat genotypers that recover expansion signal from read pairs and off-target repeat content rather than from alignment. Population-scale repeat calling is now turning the field from families to biobanks — repeat instability has recently been measured across roughly 900,000 biobank participants, including loci where inherited variants modulate somatic expansion, the §3 story at population scale. How all of these actually behave on real data — including the wet-lab methods (, Southern blotting) that remain the clinical arbiters, and the ways each method fails — is lab 11, which follows this chapter.
Common misconceptions
| What people think | What's actually true |
|---|---|
| CAG expansion means a polyglutamine disease | Only if the tract is translated. The same motif causes RNA-mediated disease from a UTR (as CTG in DMPK, the same duplex read from the other strand) and expression-level disease from a region (, ~133 nt upstream of the TSS). Position, not motif, sets the mechanism (§1.3) |
| A person's repeat number is a single, fixed value | You are a mosaic. The tract expands somatically, tissue by tissue, throughout life — fastest in the vulnerable neurons themselves, where HD striatal projection neurons reach ~150 CAG from an inherited ~40. The number on a blood report is one tissue's snapshot, not the number (§2–3) |
| means acquired severity is inherited | Nothing acquired is inherited. The DNA itself changes at gametogenesis — the child inherits a longer , which is a new (larger) , not the parent's experience. Ch 11 made this point; the molecular sizing data settle it |
| Pathogenic thresholds are measured biological constants | They are laboratory conventions drawn through continuous data, they differ between authorities (C9orf72: "pathogenic" starts at 24 or at 61 depending on whom you read), and they move — SCA12's proposed floor shifted from 51 to 43 in 2017 and the field has not converged (§1.4, §4) |
| A is a milder version of the disease allele | FMR1 55–200 causes a different disease (FXTAS), by a toxic-RNA and RAN mechanism the full mutation cannot use — the full mutation is silenced. "Pre-" describes the allele's future children, not its own effect (§4, §6.2) |
| Longer repeat always means earlier and worse disease | Within most loci, on average, yes — but DM2 shows no length–severity correlation across a 75–11,000 range, SCA8 alleles of all sizes occur in unaffected people, and SCA6 shows no anticipation. The regression is locus-specific, never universal (§5) |
| Repeats always expand through the father | HD, DRPLA, SCA2, SCA7: paternal, yes. But FMR1 expands to full mutation only maternally, congenital DM1 is maternally transmitted, and SCA8 contracts paternally. The asymmetry is real but its direction is locus-specific empirical fact (§5) |
| Repeat diseases are , gain-of-function conditions | FRDA and CANVAS are loss-of-function; fragile X is a loss-of-function silencing disorder. The class spans every inheritance pattern and both signs of effect (§1) |
| Polyglutamine length determines HD onset | Uninterrupted CAG length at the DNA level does: alleles with a CAACAG duplication carry more glutamines yet later onset. The DNA tract drives the somatic-expansion clock; the protein is downstream (§3, §6.1) |
| A negative exome and genome rules out a Mendelian diagnosis | are invisible to standard short-read calling by construction — the read is shorter than the repeat, and benchmark regions exclude these loci. A negative report is silent, not exculpatory, about this entire class (§7) |
Worked example: reading a pedigree with sizing data on it
A family, written out in the house convention of Ch 15 (text, not drawn). An HTT CAG-repeat test has been run on blood-derived DNA for the individuals shown; sizes are repeat units, larger allele first.
Generation I
I-1 male onset of chorea age 63; died 71 41 / 17
I-2 female unaffected, died 84 22 / 19
Generation II (children of I-1 x I-2)
II-1 male onset age 45; now 58 44 / 22
II-2 female unaffected at 66 38 / 22
Generation III
III-1 male son of II-1; unaffected at 27 47 / 19
III-2 female daughter of II-2; age 40, declines
testing; asks for risk figures not tested
Step 1 — classify every allele before interpreting anything. Against the HD allele classes (normal ≤26; intermediate 27–35; reduced 36–39; full penetrance ≥40): the 17, 19 and 22 alleles are normal. I-1's 41 and II-1's 44 and III-1's 47 are full-penetrance HD alleles. II-2's 38 is a reduced-penetrance allele — disease-causing in class, with a lifetime penetrance of roughly 0.2–2%.
Step 2 — trace the transmissions and read the instability. The disease allele's path is I-1 → II-1 → III-1 and I-1 → II-2:
I-1 (41) --> II-1 (44) +3 paternal transmission, expansion
I-1 (41) --> II-2 (38) -3 paternal transmission, contraction
II-1 (44) --> III-1 (47) +3 paternal transmission, expansion
Both expansions are paternal, consistent with the HD paternal bias of §5 — but I-1 → II-2 is a paternal contraction, and it happened in the same that produced a +3 expansion. Instability is a distribution, not a rule (S2): the paternal bias shifts the mean of transmitted change upwards; it does not make every draw positive. One family is a handful of draws from that distribution, and this family happens to display both tails.
Step 3 — the onset pattern, and what you may not conclude from it. Onset 63 (41 repeats) → onset 45 (44 repeats) looks like textbook , and the direction matches the inverse length–onset correlation of §5. But two people are not a regression, and Ch 16 §9's ascertainment caution applies with full force to any single family that came to attention because a younger member presented early. What the molecular data add — and this is exactly what rescued anticipation historically — is the mechanistic observation: the repeat measurably grew, +3 and +3, along the transmissions. You may say: the substrate of anticipation is present and moving in this family. You may not say: this family demonstrates an onset trend with meaningful precision.
Step 4 — III-1, the 27-year-old with 47 repeats. His allele is fully penetrant; the Langbehn-type survival models of §5 (logistic in CAG, fitted on 2,913 individuals for CAG 36–56) would give his probability of onset by each age. We do not reproduce those coefficients here (§5 said why), but the qualitative shape is enough for the counselling reality: at 47 repeats he is well into the range where onset is expected in mid-adulthood, and far enough along the length axis that further repeats would change that expectation comparatively little (§5: the curve flattens above ~50) — and — per §3 — a somatic-expansion clock already running in his striatal neurons, its rate part-set by his DNA-repair . He is also the family member for whom the D5 class of therapies-in-development (suppress the expansion engine) is most relevant; see D5.
Step 5 — III-2's question, in a Bayes table. III-2's mother II-2 carries the 38-repeat allele, so III-2's prior for having inherited it is 1/2 (Ch 15 §5). She is unaffected at 40 — but for a 38-repeat allele that is almost no information, since the allele's lifetime penetrance is only ~0.2–2%. Take the top of that range, and (as an explicit teaching stipulation, not a published figure) assume essentially none of that small lifetime risk is spent by age 40:
| Carries 38 allele | Does not | |
|---|---|---|
| Prior | 0.5 | 0.5 |
| Conditional: unaffected at 40 | ≈ 1 (penetrance ≤ 0.02 lifetime) | 1 |
| Joint | ≈ 0.5 | 0.5 |
| Posterior | ≈ 0.5 | ≈ 0.5 |
Being unaffected moves nothing, because the allele almost never manifests anyway. Her risk of developing HD is therefore at most 0.5 × ~0.02 ≈ 1%, and likely nearer 0.5 × 0.002 ≈ 0.1%. Contrast Ch 15's fully penetrant examples, where surviving unaffected steadily pays down the posterior: with reduced penetrance, absence of disease is weak evidence, and the prior just sits there.
Step 6 — the risk the sizing data cannot give you, which is the real question. III-2's own risk is small. But if she carries the 38, each of her children draws from the transmission distribution of Step 2 — and a +2 draw makes a 40, fully penetrant. This family has already shown ±3 changes in two generations. The honest counselling statement: your own risk of HD is on the order of 1% or less; your risk of transmitting an allele that could expand into the fully penetrant range is real but not precisely quantifiable from this — maternal expansion of intermediate-range HTT alleles is less likely than paternal, but this is a 38, already disease-class, and locus-specific published transmission data (not this family's three observed transmissions) would be the basis for any number offered. Note also the caveat that runs under the whole example: every genotype here is a blood measurement, and blood is not where HTT expansion runs fastest — the expansion bias sits in the brain regions that degenerate (§2–3), and the germline samples a distribution of its own. The pedigree's numbers are lower bounds on the biology, which is a limit of the method — and of every repeat report you will ever read.
Connections
Back to:
- D1 — The neuron, the cerebellum and selective vulnerability — the that decides which postmitotic neurons bear the somatic-expansion burden of §2–3.
- D2 — Kinases, phosphatases and PP2A — the holoenzyme and the B55β dial that 's class-V mechanism mis-sets.
- Ch 05 — Transcription — §6's start site and pre-initiation complex are the machinery SCA12's repeat sits ~133 nt upstream of (§1.3, §6.6).
- Ch 07 — The genetic code and translation — its section on the and the cycle gives the cap-dependent scanning model, Kozak context, leaky scanning and upstream ORFs that §6.5's contradicts.
- Ch 09 — Mitosis and meiosis — the continuous-spermatogenesis / arrested-oogenesis asymmetry behind parent-of-origin expansion bias (§5).
- Ch 11 — Beyond Mendel — defined, age-dependent , and the misconception table this chapter's §4–5 quantify.
- Ch 15 — Pedigrees — the Bayes table the worked example runs on reduced-penetrance .
- Ch 16 — Mutation — §9 is the seed of this whole chapter: slippage, thresholds, the four-disease table, and the ascertainment-bias history that is deliberately not repeated here.
- Ch 17 — DNA repair — §5's and microsatellite instability are the same MutSβ machinery that §3 catches driving expansions.
- Ch 51 — GWAS — the association machinery the GeM-HD consortium pointed at onset age instead of case status.
- Ch 54 — Rare variants and Mendelian disease — §9's "invisible by construction" row, which §7 unpacked into a field-sized blind spot.
Forward to:
- D4 — SCA12 I: from repeat to phenotype — the awkward case of §1.3 and §6.6 given a full chapter: the contested threshold, the expression model and its gaps.
- D5 — SCA12 II: population, clinic, therapy — the therapeutic implication planted in §3: target the expansion engine, not the toxic product.
- Lab 11 — Genotyping repeat expansions — sizing repeats for real: , Southern blots, short-read repeat genotypers and long reads, with their failure modes on display.
- Ch 55 — Clinical variant interpretation — read its SNV-shaped evidence framework against §1.4's moving thresholds and ask which criteria even apply.