Problem set 04 — Pedigrees and risk
Covers Ch 15.
Attempt before revealing. Each of these is a model-selection exercise on a tiny, non-randomly ascertained sample: the answer is never the label, it is the number and the assumptions under it.
are written out rather than drawn so there is nothing to misread. Roughly in order of difficulty; ★ marks the two worth returning to.
1. Assigning a mode, and killing the alternatives
A family is ascertained through a young man with a progressive neurological condition.
I I-1 male AFFECTED × I-2 female, unaffected, married in
II children of I-1 × I-2: II-1 M AFF | II-2 F | II-3 F AFF | II-4 M
married in: II-5 F (partner of II-1), II-6 M (partner of II-3)
III children of II-1 × II-5: III-1 M AFF | III-2 M | III-3 F AFF
children of II-3 × II-6: III-4 M AFF | III-5 F
(a) Assign the most likely mode of inheritance. (b) For each of the other four modes, state the specific observation that excludes it, and say whether the exclusion is logical or merely probabilistic. (c) Autosomal is not logically excluded. Quantify how badly it is disfavoured, assuming a population of 1/50. (d) II-4 is an unaffected man of 40. is 0.8 by age 40. What is the probability he carries the ? [trap]
Solution
(a) Autosomal . Affected people in all three generations (vertical), both sexes affected in roughly equal numbers (4 male, 2 female), and every affected person has an affected parent. Among the children of an affected parent, 5 of 9 are affected — consistent with 1/2.
(b) Name the observation, not the vibe:
- recessive — killed by male-to-male transmission, twice: I-1 → II-1 and II-1 → III-1. A father gives his son a Y, never an X. Also by the affected females (II-3, III-3), who under XLR would each need an affected father and a carrier mother.
- X-linked dominant — killed by the same two transmissions, and independently by I-1's daughters: an affected father transmits to all daughters and no sons. I-1 has one affected daughter (II-3), one unaffected (II-2), and an affected son (II-1). Both halves fail.
- Mitochondrial — killed by transmission through affected males. An affected father transmits to nobody; I-1 and II-1 transmitted to 4 of 7 children between them.
- Autosomal recessive — not logically excluded, see (c).
The male-to-male observations are probabilistic, not logical: you saw an affected father with an affected son, not the transmission of his allele. Misattributed parentage, or a common X-linked allele entering through the mothers, gives the same picture — which is why two independent instances matter and one would not be enough.
(c) For AR to produce this , every unaffected spouse who married in — I-2, II-5, II-6 — would have to be a carrier ("pseudo-dominance"). With carrier frequency 1/50 = 0.02:
P(all three carriers) = 0.02³ = 8 × 10⁻⁶
Under AD those three spouses are unremarkable, probability ≈ 1, so
LR = 1 / (8 × 10⁻⁶) = 125,000 : 1 in favour of AD
before any data is used. That is the honest form of "it's dominant" — a likelihood ratio, not a proof.
(d) [trap] The trap is answering 0. That reflex is correct only at penetrance 1, and it gets carried over to the reduced-penetrance case where it is wrong.
Prior: I-1 is heterozygous, I-2 homozygous normal, so each child is a carrier with probability 1/2.
| Carrier | Not a carrier | |
|---|---|---|
| Prior | 1/2 | 1/2 |
| Conditional: unaffected at 40 | 1 − 0.8 = 0.2 | 1 |
| Joint | 0.5 × 0.2 = 0.10 | 0.5 × 1 = 0.50 |
| Posterior | 0.10 / 0.60 = 1/6 ≈ 16.7% | 0.50 / 0.60 = 5/6 |
16.7%, not 0 — and his children are each at 1/6 × 1/2 × 0.8 ≈ 6.7% of being affected by 40. Incomplete penetrance is why dominant conditions appear to skip generations (Ch 11).
2. When two modes both fit
A different family, ascertained through III-1.
I I-1 female AFFECTED × I-2 male, unaffected, married in
II children of I-1 × I-2: II-1 F AFF | II-2 M | II-3 M AFF | II-4 F
married in: II-5 M (partner of II-1)
III children of II-1 × II-5: III-1 M AFF | III-2 F | III-3 F AFF
(a) Which two modes remain live, and why can you not separate them from this ? (b) A colleague says "no male-to-male transmission — so it's ." What is wrong with that inference here? [trap] (c) Name the single additional observation that would discriminate, and say why it is the sharpest one available. (d) II-3 later has three daughters and two sons. All three daughters are affected; neither son is. Compute the likelihood ratio. Then combine it with prior odds of 20:1 favouring autosomal (such conditions being much the commoner class) and give the posterior.
Solution
(a) Autosomal dominant and X-linked dominant. Every transmission here is from a mother — I-1 and II-1 — and through mothers the two modes are identical: a heterozygous mother passes to 1/2 of her children, sexes exchangeable. Observed: I-1 gave 2 of 4, II-1 gave 2 of 3.
Autosomal needs two carrier spouses marrying in, disfavoured as in problem 1. Mitochondrial predicts all of an affected woman's children are at risk, not half; here 2 of 4 and 1 of 3 are unaffected — strong evidence, though keeps it evidence rather than exclusion. X-linked recessive is dead: it cannot give affected females without affected fathers.
(b) [trap] The pedigree gave male-to-male transmission zero opportunities to appear — the only affected males, II-3 and III-1, have had no children. Count the chances a signature had before crediting its absence; here the count is zero, so LR = 1 and the observation is worthless.
(c) Find an affected male with children of both sexes and record every child's status. Through mothers the hypotheses are indistinguishable; through an affected father the sex-by-affection table is degenerate under XLD — one diagonal is structurally zero where AD predicts an even split. It is the only place the two models disagree.
(d) Five children, all daughters affected, no sons affected.
- P(outcome | XLD) = 1 — the only outcome XLD permits.
- P(outcome | AD) = (1/2)⁵ = 1/32 — five independent coin flips landing the right way.
LR = 1 ÷ (1/32) = 32 : 1 in favour of XLD
With prior odds 1:20 (XLD : AD):
Posterior odds = 32 × (1/20) = 1.6 : 1 → P(XLD) = 1.6 / 2.6 = 0.615 ≈ 62%
Two lessons. k children of an affected father give LR = 2ᵏ — one bit per child — so five buy a factor of 32, which against a sceptical prior leaves you at a wobbly 62%, not certainty. And the asymmetry: had one son been affected, P(data | XLD) = 0 and XLD would be dead outright. Falsifying XLD takes one observation; supporting it takes many.
3. Recessive recurrence and the 2/3 correction
A man's sister has phenylketonuria (autosomal ). He is unaffected. His partner is unrelated, from a population with a PKU of 1/50.
(a) What is his prior probability of being a carrier, and why is it not 1/2? (b) What is the probability their first child is affected? (c) The first child is unaffected. What is the probability the second child is affected? [trap] (d) A relative says: "you've had one healthy child, so the 1-in-4 has been used up — the next is more likely to be affected." Diagnose the error, and note which direction the real effect runs.
Solution
(a) 2/3. His parents must both be carriers (they produced an affected daughter), so his four equally likely are AA, Aa, aA, aa. "Unaffected" deletes aa and renormalises over the remaining three, two of which are heterozygous: P(carrier | unaffected) = 2/3. The 1/2 answer forgets to condition on the fact that put him in the problem — the most commonly dropped step in recessive risk work.
(b) Three independent requirements: he is a carrier, she is a carrier, and both transmit.
P = (2/3) × (1/50) × (1/4) = 2/600 = 1/300 ≈ 0.333%
(c) [trap] The unaffected child is evidence, and it points the opposite way from intuition. Set up the hypothesis as "both parents are carriers", since that is the only state in which children are at risk.
Prior: P(both carriers) = (2/3)(1/50) = 1/75; P(not both) = 74/75.
| Both carriers | Not both | |
|---|---|---|
| Prior | 1/75 | 74/75 |
| Conditional: one unaffected child | 3/4 | 1 |
| Joint | (1/75)(3/4) = 3/300 | 74/75 = 296/300 |
| Posterior | (3/300)/(299/300) = 3/299 | 296/299 |
P(second child affected) = (3/299) × (1/4) = 3/1196 ≈ 0.251% ≈ 1 in 399
Down from 1 in 300 — a ~25% reduction in risk (24.7% exactly), purely from one free observation.
(d) The gambler's fallacy applied to : each conception is an independent draw, nothing is used up. But "no change" is wrong too — the real effect runs opposite to the fallacy, because an unaffected child is mild evidence the parents are not both carriers, so risk falls, 1/300 → 1/399.
The sharpest version: had both parents already tested positive, the unaffected child would change nothing — risk stays exactly 1/4 for every subsequent child, forever. The update in (c) happens only because carrier status was uncertain. Data updates the hypothesis, never the coin.
4. The full Bayesian table
A woman's maternal grandfather had haemophilia B ( , F9). She has four sons, all healthy and all past the age at which the condition declares itself. She has no molecular testing.
(a) Justify her prior carrier probability from the alone. (b) Build the full prior / conditional / joint / posterior table. (c) What is the probability that her next son is affected? Her next child, sex unknown? (d) She also has two unaffected daughters. Redo the table including them. [trap] (e) Derive the general posterior for n unaffected sons, and find how many sons it would take to bring her carrier probability below 1%.
Solution
(a) An affected male transmits his single X — the one carrying the variant — to every daughter and no son. Her mother is therefore an obligate carrier, no test required, and transmitted one of her two X at random: P(carrier) = 1/2.
(b) Each son of a carrier gets the mutant X with probability 1/2, so is unaffected with probability 1/2; four sons are independent draws.
P(4 unaffected sons | carrier) = (1/2)⁴ = 1/16 P(4 unaffected sons | not a carrier) = 1
| Carrier | Not a carrier | |
|---|---|---|
| Prior | 1/2 | 1/2 |
| Conditional: 4 unaffected sons | 1/16 | 1 |
| Joint | (1/2)(1/16) = 1/32 | (1/2)(1) = 1/2 = 16/32 |
| Posterior | (1/32) ÷ (17/32) = 1/17 ≈ 5.88% | 16/17 ≈ 94.12% |
Four sons, no laboratory, no cost — and the risk has fallen from 50% to 5.9%.
(c) Convert carrier risk into the question actually asked:
P(next son affected) = (1/17) × (1/2) = 1/34 ≈ 2.94% P(next child affected, sex unknown) = (1/17) × (1/4) = 1/68 ≈ 1.47%
The 1/4 is 1/2 for "the child is male" times 1/2 for "he gets the mutant X".
(d) [trap] The daughters contribute nothing. Redo the conditional row honestly:
P(daughter unaffected | mother is a carrier) = 1 P(daughter unaffected | mother is not a carrier) = 1
Her partner is unaffected, so every daughter gets a normal X from him and is unaffected whether or not she received the mutant maternal X. LR = 1/1 = 1, both joint cells are multiplied by 1, and the posterior stays at 1/17.
The habit to build: evidence is only evidence when the two hypotheses assign it different probabilities. (Each daughter is nonetheless a carrier with probability (1/17) × (1/2) = 1/34 ≈ 2.94% — she just does not inform her mother's status.)
(e) With n unaffected sons:
posterior = [(1/2)(1/2ⁿ)] ÷ [(1/2)(1/2ⁿ) + (1/2)(1)] = (1/2ⁿ) ÷ (1/2ⁿ + 1) = 1 / (1 + 2ⁿ)
Check: n = 4 gives 1/17 ✓.
Require 1/(1 + 2ⁿ) < 0.01 ⇒ 1 + 2ⁿ > 100 ⇒ 2ⁿ > 99 ⇒ n = 7.
- n = 6: 1/65 = 1.54%
- n = 7: 1/129 = 0.775% ✓
Seven unaffected sons. Note how slowly this converges — one bit per son, and the posterior never reaches 0. Compare problem 5, where a single test at LR 20:1 outperforms four sons.
5. A negative test with imperfect sensitivity
A woman's brother has spinal muscular atrophy (autosomal , SMN1). She has SMN1 dosage testing, which detects a carrier with probability 0.95; her result is negative. Her partner has no family history; the population is 1/50. He has the same test, also negative.
(a) Show exactly where the sensitivity of 0.95 enters the table, and why the non-carrier column gets a 1. (b) Compute her posterior carrier probability. (c) Compute his. (d) Probability their child is affected. Compare with the no-testing figure. (e) The same test, same sensitivity, moved her prior by a factor of 7.3 and his by a factor of 19.6. Explain the asymmetry. [trap]
Solution
(a) Sensitivity is P(test positive | carrier) = 0.95. The conditional row needs P(observed result | hypothesis), and the observed result is negative:
P(negative | carrier) = 1 − 0.95 = 0.05 — the false-negative rate. Sensitivity enters as its complement. P(negative | not a carrier) = 1 — this is the specificity, taken as 1.0. Such assays essentially never call a true non-carrier positive; were specificity 0.98 you would put 0.98 here.
(b) Her prior is 2/3 — unaffected sibling of an affected person, as in problem 3(a).
| Carrier | Not a carrier | |
|---|---|---|
| Prior | 2/3 | 1/3 |
| Conditional: negative test | 0.05 | 1 |
| Joint | (2/3)(0.05) = 1/30 | 1/3 = 10/30 |
| Posterior | (1/30) ÷ (11/30) = 1/11 ≈ 9.09% | 10/11 ≈ 90.9% |
(c) His prior is the population carrier frequency, 1/50 = 0.02.
| Carrier | Not a carrier | |
|---|---|---|
| Prior | 0.02 | 0.98 |
| Conditional: negative test | 0.05 | 1 |
| Joint | 0.02 × 0.05 = 0.001 | 0.98 × 1 = 0.98 |
| Posterior | 0.001 / 0.981 = 1/981 ≈ 0.102% | 0.980/0.981 ≈ 99.9% |
(d) P(affected child) = (1/11) × (1/981) × (1/4)
Denominator: 11 × 981 = 10,791; × 4 = 43,164
= 1/43,164 ≈ 2.32 × 10⁻⁵, about 1 in 43,000.
Without any testing it would have been (2/3)(1/50)(1/4) = 1/300. The two negative tests cut the risk by 43,164 / 300 = 143.9-fold — but not to zero.
(e) [trap] The tempting reading is that the test "worked better" on him. It did not — the likelihood ratio is identical. Odds form makes this transparent:
LR(negative) = 0.05 / 1 = 1/20 for everybody.
- Her prior odds 2 : 1 → posterior odds 2/20 = 1 : 10 → P = 1/11 ✓
- His prior odds 1 : 49 → posterior odds 1 : 980 → P = 1/981 ✓
Odds are multiplied by exactly 1/20 in both cases. What differs is the probability reduction, because converting odds back to probability renormalises:
- Hers: (2/3) ÷ (1/11) = 22/3 = 7.33-fold
- His: (1/50) ÷ (1/981) = 981/50 = 19.62-fold
With a small prior the denominator barely moves (0.981 ≈ 1), so the probability falls by nearly the full LR of 20; with a large prior the denominator falls too and the reduction is much less. The likelihood ratio belongs to the test; the fold-change in probability belongs to the test and the prior. "This test cuts your risk 20-fold" is therefore wrong for exactly the high-risk patients who most want to hear it.
The residual is not an arithmetic artefact. 1 in 43,000 rather than 0 reflects the 5% of carriers the assay misses — for SMN1, the "2+0" configuration, two copies on one and none on the other, reads out as normal dosage in a true carrier — plus the few per cent of cases from a de novo deletion in a genuinely non-carrier parent.
6. Consanguinity by path counting
A couple are first cousins once removed. Precisely: the husband's mother is a first cousin of the wife, so the husband is one generation further down the same loop. The common ancestors are a single couple, A and B, both non-inbred.
(a) Draw the loops and compute F for their child by path counting. State the classic error that gives exactly twice the right answer. [trap] (b) A condition has a of 1/60 in their community. Compute the child's risk, and the fold increase over an unrelated couple. (c) Repeat for a much rarer , q = 1/1000, and explain the pattern. (d) The couple ask what their overall risk of a child with a serious genetic condition is. Why is your answer to (b) not that number? [trap]
Solution
(a) The genealogy: A × B are the ancestor couple; P1 and P2 their children (full sibs); C1 is P1's child, C2 is P2's child, so C1 and C2 are first cousins; D is C2's child by an unrelated partner; C1 × D → X.
Closed loops linking X's two parents through a common ancestor:
Path through A: C1 – P1 – A – P2 – C2 – D = 6 individuals Path through B: C1 – P1 – B – P2 – C2 – D = 6 individuals
F = Σ (1/2)ⁿ (1 + Fₐ) with n = 6 and Fₐ = 0:
F = (1/2)⁶ + (1/2)⁶ = 1/64 + 1/64 = 2/64 = 1/32 = 0.03125
Matching the standard table ✓ — half the first-cousin 1/16, each loop being one individual longer.
[trap] The classic error is counting only meioses. A 6-individual loop contains 5 meioses, giving (1/2)⁵ per path and 2/32 = 1/16 — exactly twice the right answer. The missing 1/2 is not a : it is the probability that the common ancestor sends the same one of its two alleles down both branches. That is its self-kinship ½(1 + Fₐ) — the source of the (1 + Fₐ).
(b) Carrier frequency 1/60 ≈ 2q, so q = 1/120 = 0.008333.
P(aa) = F·q + (1 − F)q²
- F·q = 0.03125 × 0.008333 = 2.6042 × 10⁻⁴
- q² = (0.008333)² = 6.9444 × 10⁻⁵
- (1 − F)q² = 0.96875 × 6.9444 × 10⁻⁵ = 6.7274 × 10⁻⁵
- Total = 2.6042 × 10⁻⁴ + 0.6727 × 10⁻⁴ = 3.2769 × 10⁻⁴ ≈ 1 in 3,052
Unrelated couple: q² = 6.9444 × 10⁻⁵ = 1 in 14,400
Fold increase = 3.2769 × 10⁻⁴ ÷ 6.9444 × 10⁻⁵ = 4.72×
Check it algebraically rather than trusting the division: the ratio is F/q + (1 − F) = 3.75 + 0.96875 = 4.71875 ✓
(c) With q = 1/1000: F·q = 0.03125 × 0.001 = 3.125 × 10⁻⁵ and (1 − F)q² = 0.96875 × 10⁻⁶ = 9.6875 × 10⁻⁷, total 3.2219 × 10⁻⁵ ≈ 1 in 31,038, against an outbred 1 in 1,000,000. Fold increase = 31.25 + 0.96875 = 32.2×.
The IBD term is linear in q, the outbred term quadratic, so their ratio ≈ F/q and grows without bound as the allele gets rarer. barely matters for common recessives and enormously for rare ones — hence the enrichment of consanguineous families among parents of children with rare recessive disease, and hence homozygosity mapping.
(d) [trap] Part (b) is a per-locus number, for one allele you already know the family carries. Their question is genome-wide, dominated by aggregate recessive load — the several dozen rare pathogenic alleles everyone carries heterozygously, none of them identified.
Empirically: against a background of roughly 2–3% for a significant congenital anomaly or genetic disorder in any pregnancy, first cousins (F = 1/16) carry an additional 1.7–2.8% absolute. First cousins once removed sit at half that F, so of order 1% extra on a 2–3% base. Real, and not catastrophic; saying only one of those out loud is a counselling failure.
7. ★ Two affected children, unaffected parents
A healthy, unrelated couple have two children with osteogenesis imperfecta. Sequencing finds the same heterozygous COL1A1 variant in both children. Neither parent carries it in blood at standard depth. A junior colleague writes "autosomal , recurrence risk 25%".
(a) Give two reasons the molecular data refutes autosomal recessive, and name the one loose end you must close first. (b) Derive a per-gamete rate for a new pathogenic COL1A1 variant from the , watching the units. COL1A1 has ~4,400 bp of coding sequence; take roughly one in five new coding substitutions there to be pathogenic. [trap] (c) Condition on the first affected child (that is why the family is here) and compute the posterior odds that a parent is a germline mosaic, taking a prior of 0.08 and a mosaic gamete fraction of 0.05. (d) Sperm testing shows the variant at a variant fraction of 4% in the father's semen. What is the recurrence risk? [trap] (e) Compare the three numbers the family could have been given, and give the affected child's own offspring risk.
Solution
(a) Two reasons:
- Both affected children are heterozygous for a single variant. Autosomal recessive needs two hits — homozygous, or compound heterozygous. One pathogenic allele plus disease is a -acting allele by definition.
- Neither parent carries it. Under AR both would be obligate carriers of a transmitted allele.
The loose end: exclude a second, undetected COL1A1 or COL1A2 allele — a deep intronic or short-read would miss. Until that is done, (1) is an inference, not an observation.
(b) [trap] The unit trap. Reach for the germline rate, not the replication-fidelity figure. Replication fidelity is ~10⁻¹⁰ per base per replication; the germline SNV rate is ~1.2 × 10⁻⁸ per base per generation. They differ ~100-fold and are not interchangeable. Here we need per generation, because a gamete is the end of a generation, not of one cell division.
Coding substitutions in COL1A1 per gamete per generation:
4,400 bp × 1.2 × 10⁻⁸ per bp per generation = 5.28 × 10⁻⁵
Pathogenic fraction ≈ 1/5:
5.28 × 10⁻⁵ ÷ 5 ≈ 1.06 × 10⁻⁵ ≈ 1 × 10⁻⁵ per gamete
(c) Two hypotheses, both conditioned on child 1 being affected:
- Hₘ: one parent is a germline mosaic, gamete fraction m = 0.05. P(child 2 also affected | Hₘ) = 0.05
- Hᵢ: no ; each event is independent de novo. Mind the units again: (b) gave a rate per gamete, and a conception samples two gametes — either parent's could carry the new variant — so the per-conception rate is twice it. P(child 2 also affected | Hᵢ) = 2 × 1.06 × 10⁻⁵ ≈ 2.1 × 10⁻⁵
LR = 0.05 ÷ 2.1 × 10⁻⁵ ≈ 2,400
Prior odds = 0.08 : 0.92 = 0.08696
Posterior odds = 0.08696 × 2,400 ≈ 210 : 1, i.e. P(mosaic) ≈ 99.5%
That understates it: both children carry the identical nucleotide substitution, and two independent events hitting the same base is rarer still than two independent pathogenic events anywhere in the gene.
Two affected children of unaffected parents is not a recessive signature. It is equally the signature of for a dominant allele, and the molecular data — one heterozygous variant, absent from parental blood — settles which.
(d) [trap] 4% in sperm means 4% of gametes carry the variant: recurrence risk 4% per pregnancy.
The trap is halving it. In a somatic sample a heterozygous variant present in a fraction c of cells reads out at VAF ≈ c/2 — one mutant allele among two per carrying cell. Sperm are haploid: one out of one, so VAF equals the cell fraction directly. Applying the diploid correction to sperm halves the risk you quote the family.
(e) Three numbers, same family, same variant:
| The call | Recurrence risk quoted | Error |
|---|---|---|
| "Autosomal recessive" | 25% | overstates by 25/4 = 6.25× |
| "De novo — so it won't happen again" | ~2 × 10⁻⁵ per conception | understates by 0.04 ÷ (2 × 10⁻⁵) = 2,000× |
| Germline mosaicism, measured | 4% | — |
The affected child's own offspring risk, should they reproduce, is 1/2: they are a constitutional heterozygote, not a mosaic, so every gamete derives from a carrying cell and is the ordinary dominant 1/2. Their parents' risk is 4%. Two figures about one variant in one family differing 12.5-fold, and which applies depends on who is sitting in front of you.
A negative parental blood test converts a high risk into a low one. Never into zero.
8. ★★ Imprinting: one deletion, two syndromes
The 15q11.2–q13 region contains genes expressed only from the paternal (the SNRPN/SNORD116 cluster) and UBE3A, expressed only from the maternal chromosome in neurons. Two syndromes arise from it: Prader–Willi (PWS) and Angelman (AS).
(a) A child has a ~5 Mb deletion of 15q11.2–q13. Which syndrome? Justify from expression, not from memory. (b) A second child has a normal microarray and normal UBE3A sequence, but abnormal methylation, and SNP genotyping shows both copies of chromosome 15 came from the mother. Which syndrome, and by what mechanism? (c) Maternal UPD15 usually follows rescue. Compute the probability that rescuing a trisomy produces UPD, and use it to explain why maternal UPD15 accounts for ~25% of PWS while paternal UPD15 accounts for only ~5% of AS. (d) Using deletion 70%, UPD 25%, imprinting-centre defect 2% for PWS, and deletion 70%, UPD 5%, IC defect 3%, UBE3A point 11% for AS: in 100 consecutive patients of each, how many would a methylation test detect? [trap] (e) An unaffected woman carries a deletion of the imprinting centre element required to establish the maternal imprint, on the chromosome 15 she inherited from her father. Explain why she is unaffected, and give the risk and syndrome for her children.
Solution
(a) It depends entirely on which parent's chromosome carries the deletion, and only a parent-of-origin test can say.
- Paternal deletion → Prader–Willi. The maternal SNORD116 copies are already silenced by imprinting, so deleting the paternal copy takes expression to zero.
- Maternal deletion → Angelman. The paternal UBE3A is already silenced in neurons, so deleting the maternal copy removes the only expressed copy.
At an imprinted locus you are functionally : one deletion is a total loss, and which genes go to zero depends on the parent of origin. The five-mode sieve has no slot for that.
(b) Prader–Willi, by maternal 15. Both chromosome 15s carry the maternal imprint, so the paternally expressed genes are silenced on both — functionally identical to a paternal deletion, with no deletion anywhere. Hence array-normal, sequence-normal cases, and hence methylation as the correct first-line test.
(c) A trisomic conceptus from maternal meiotic has two maternal and one paternal chromosome 15. Rescue drops one of the three at random:
- Lose the paternal one: P = 1/3 → surviving pair both maternal → maternal UPD15
- Lose either maternal one: P = 2/3 → normal biparental disomy, no consequence
So 1/3 of rescued trisomy-15 conceptuses have maternal UPD.
The 5:1 asymmetry (25% of PWS vs 5% of AS) follows because the input is asymmetric: maternal meiotic nondisjunction is far commoner than paternal — oocytes sit arrested for decades, sperm do not — so trisomy 15 is overwhelmingly maternal and rescue overwhelmingly yields maternal UPD. Paternal UPD15 needs a rarer route (monosomy rescue duplicating the lone paternal chromosome). Maternal UPD15 therefore carries a maternal-age effect — the only PWS mechanism that does.
(d) [trap] Methylation testing reads the parent-of-origin methylation pattern. It is abnormal for deletion, UPD and IC defect — all three change which parental pattern is present — and normal for a UBE3A point mutation, which leaves imprinting untouched.
PWS, 100 patients: 70 + 25 + 2 = 97 detected. All three mechanisms disturb methylation, so this is effectively a complete first-line test (>99% in practice; the 3-patient shortfall is midpoint rounding of the published ranges, not an undetectable class).
AS, 100 patients: 70 + 5 + 3 = 78 detected. The 11 UBE3A point mutations methylate normally, and ~11 more have no identified mechanism at all.
Sensitivity: ~97–99% for PWS, ~78% for AS. The same first-line test, on two syndromes from the same 5 Mb of DNA, misses roughly one AS patient in five and essentially no PWS patient. A normal methylation result rules PWS out; it does not rule AS out. Next step: UBE3A sequencing.
(e) The deleted element is only needed on a chromosome that must acquire the maternal imprint. Hers came from her father, so it had to acquire the paternal imprint — which it did, the paternal machinery not using this element. Her other chromosome 15 is normal and maternal. One working paternal contribution, one working maternal: no .
Her children are another matter. In her every chromosome 15 she transmits must be given the maternal imprint, and on the deleted one she cannot establish it. A child inheriting it gets a maternal chromosome carrying a paternal-type imprint plus a genuinely paternal one from the father — two paternal patterns, no maternal UBE3A expression in neurons.
Risk: 1/2 of her children, affected with Angelman syndrome.
Note the shape: unaffected transmitting parent, abrupt 50% risk, and a mode that reverses with the transmitter's sex. The mirror is exact — a man carrying the PWS-side IC deletion on his maternally inherited chromosome 15 is likewise unaffected, and 1/2 of his children have Prader–Willi.
So mechanism, not diagnosis, sets recurrence risk. Same clinical label: de novo deletion or de novo UPD, under 1%; inherited IC deletion or maternally inherited UBE3A variant, 50%. A 50-fold range. Classify the mechanism before quoting a number.
Where you went wrong
| Error pattern | What to re-read |
|---|---|
| Read "no male-to-male transmission" as evidence without counting the opportunities | Problem 2(b) |
| Used 1/2 instead of 2/3 for the unaffected sibling of an affected person | Problems 3(a), 5(b) |
| Said an unaffected child of an affected parent cannot be a carrier | Problem 1(d) — true only at 1 |
| Counted unaffected daughters as evidence in an table | Problem 4(d) — LR exactly 1 |
| Put sensitivity rather than (1 − sensitivity) in the conditional cell | Problem 5(a) |
| Quoted a test's fold risk-reduction as a property of the test | Problem 5(e) |
| Computed F as (1/2)^(meioses) and got exactly twice the answer | Problem 6(a) |
| Gave the per-locus risk as the couple's overall risk | Problem 6(d) |
| Read two affected children of unaffected parents as recessive | Problem 7(c) |
| Used the ~10⁻¹⁰ replication fidelity where a per-generation rate was needed | Problem 7(b) — ~100× apart |
| Halved a sperm as though the sample were | Problem 7(d) |
| Assigned a syndrome from a deletion without asking which parent it came from | Problem 8(a) |
| Treated a normal methylation result as excluding Angelman syndrome | Problem 8(d) — ~78% sensitive |
| Quoted a recurrence risk from the diagnosis rather than the mechanism | Problem 8(e) — <1% versus 50% |